diff --git "a/checkpoint-1593/trainer_state.json" "b/checkpoint-1593/trainer_state.json"
--- "a/checkpoint-1593/trainer_state.json"
+++ "b/checkpoint-1593/trainer_state.json"
@@ -10,11441 +10,11441 @@
   "log_history": [
     {
       "epoch": 0.005649717514124294,
-      "grad_norm": 2.53125,
+      "grad_norm": 2.1875,
       "learning_rate": 2e-05,
-      "loss": 1.7472,
+      "loss": 1.7628,
       "step": 1
     },
     {
       "epoch": 0.005649717514124294,
-      "eval_loss": 2.0531885623931885,
-      "eval_runtime": 6.0231,
-      "eval_samples_per_second": 24.738,
-      "eval_steps_per_second": 3.155,
+      "eval_loss": 1.9544461965560913,
+      "eval_runtime": 5.7039,
+      "eval_samples_per_second": 26.298,
+      "eval_steps_per_second": 3.331,
       "step": 1
     },
     {
       "epoch": 0.011299435028248588,
-      "grad_norm": 1.8203125,
+      "grad_norm": 2.140625,
       "learning_rate": 4e-05,
-      "loss": 1.8233,
+      "loss": 1.8127,
       "step": 2
     },
     {
       "epoch": 0.01694915254237288,
-      "grad_norm": 2.671875,
+      "grad_norm": 1.7578125,
       "learning_rate": 6e-05,
-      "loss": 1.9262,
+      "loss": 1.6358,
       "step": 3
     },
     {
       "epoch": 0.022598870056497175,
-      "grad_norm": 1.765625,
+      "grad_norm": 1.609375,
       "learning_rate": 8e-05,
-      "loss": 1.7232,
+      "loss": 1.592,
       "step": 4
     },
     {
       "epoch": 0.02824858757062147,
-      "grad_norm": 1.078125,
+      "grad_norm": 0.89453125,
       "learning_rate": 0.0001,
-      "loss": 1.6063,
+      "loss": 1.4602,
       "step": 5
     },
     {
       "epoch": 0.03389830508474576,
-      "grad_norm": 0.546875,
+      "grad_norm": 0.48828125,
       "learning_rate": 0.00012,
-      "loss": 1.3681,
+      "loss": 1.2401,
       "step": 6
     },
     {
       "epoch": 0.03954802259887006,
-      "grad_norm": 0.5390625,
+      "grad_norm": 0.451171875,
       "learning_rate": 0.00014,
-      "loss": 1.4529,
+      "loss": 1.3002,
       "step": 7
     },
     {
       "epoch": 0.04519774011299435,
-      "grad_norm": 0.62890625,
+      "grad_norm": 0.58984375,
       "learning_rate": 0.00016,
-      "loss": 1.372,
+      "loss": 1.3004,
       "step": 8
     },
     {
       "epoch": 0.05084745762711865,
-      "grad_norm": 0.609375,
+      "grad_norm": 0.60546875,
       "learning_rate": 0.00018,
-      "loss": 1.3344,
+      "loss": 1.2782,
       "step": 9
     },
     {
       "epoch": 0.05649717514124294,
-      "grad_norm": 0.6640625,
+      "grad_norm": 0.48046875,
       "learning_rate": 0.0002,
-      "loss": 1.4499,
+      "loss": 1.3755,
       "step": 10
     },
     {
       "epoch": 0.062146892655367235,
-      "grad_norm": 0.427734375,
+      "grad_norm": 0.419921875,
       "learning_rate": 0.0001999998406895374,
-      "loss": 1.4297,
+      "loss": 1.1181,
       "step": 11
     },
     {
       "epoch": 0.06779661016949153,
-      "grad_norm": 0.41796875,
+      "grad_norm": 0.42578125,
       "learning_rate": 0.00019999936275865712,
-      "loss": 1.1761,
+      "loss": 1.4226,
       "step": 12
     },
     {
       "epoch": 0.07344632768361582,
-      "grad_norm": 0.412109375,
+      "grad_norm": 0.390625,
       "learning_rate": 0.00019999856620888195,
-      "loss": 1.2343,
+      "loss": 1.1685,
       "step": 13
     },
     {
       "epoch": 0.07909604519774012,
-      "grad_norm": 0.447265625,
+      "grad_norm": 0.357421875,
       "learning_rate": 0.00019999745104274993,
-      "loss": 1.2532,
+      "loss": 1.2037,
       "step": 14
     },
     {
       "epoch": 0.0847457627118644,
-      "grad_norm": 0.38671875,
+      "grad_norm": 0.328125,
       "learning_rate": 0.00019999601726381413,
-      "loss": 1.1204,
+      "loss": 1.1093,
       "step": 15
     },
     {
       "epoch": 0.0903954802259887,
-      "grad_norm": 0.357421875,
+      "grad_norm": 0.408203125,
       "learning_rate": 0.00019999426487664297,
-      "loss": 1.1571,
+      "loss": 1.3148,
       "step": 16
     },
     {
       "epoch": 0.096045197740113,
-      "grad_norm": 0.388671875,
+      "grad_norm": 0.33984375,
       "learning_rate": 0.0001999921938868198,
-      "loss": 1.0381,
+      "loss": 1.1924,
       "step": 17
     },
     {
       "epoch": 0.1016949152542373,
-      "grad_norm": 0.41796875,
+      "grad_norm": 0.373046875,
       "learning_rate": 0.00019998980430094334,
-      "loss": 1.1143,
+      "loss": 1.2372,
       "step": 18
     },
     {
       "epoch": 0.10734463276836158,
-      "grad_norm": 0.44140625,
+      "grad_norm": 0.447265625,
       "learning_rate": 0.00019998709612662724,
-      "loss": 1.341,
+      "loss": 1.3177,
       "step": 19
     },
     {
       "epoch": 0.11299435028248588,
-      "grad_norm": 0.431640625,
+      "grad_norm": 0.26953125,
       "learning_rate": 0.00019998406937250034,
-      "loss": 1.2264,
+      "loss": 0.9893,
       "step": 20
     },
     {
       "epoch": 0.11864406779661017,
-      "grad_norm": 0.40625,
+      "grad_norm": 0.4140625,
       "learning_rate": 0.0001999807240482065,
-      "loss": 1.4006,
+      "loss": 1.2532,
       "step": 21
     },
     {
       "epoch": 0.12429378531073447,
-      "grad_norm": 0.42578125,
+      "grad_norm": 0.35546875,
       "learning_rate": 0.0001999770601644046,
-      "loss": 1.23,
+      "loss": 1.2103,
       "step": 22
     },
     {
       "epoch": 0.12994350282485875,
-      "grad_norm": 0.328125,
+      "grad_norm": 0.400390625,
       "learning_rate": 0.00019997307773276862,
-      "loss": 1.1496,
+      "loss": 1.115,
       "step": 23
     },
     {
       "epoch": 0.13559322033898305,
-      "grad_norm": 0.357421875,
+      "grad_norm": 0.41015625,
       "learning_rate": 0.00019996877676598733,
-      "loss": 1.0949,
+      "loss": 1.2606,
       "step": 24
     },
     {
       "epoch": 0.14124293785310735,
-      "grad_norm": 0.384765625,
+      "grad_norm": 0.423828125,
       "learning_rate": 0.00019996415727776455,
-      "loss": 1.0633,
+      "loss": 1.25,
       "step": 25
     },
     {
       "epoch": 0.14689265536723164,
-      "grad_norm": 0.42578125,
+      "grad_norm": 0.3984375,
       "learning_rate": 0.00019995921928281894,
-      "loss": 1.2307,
+      "loss": 1.1588,
       "step": 26
     },
     {
       "epoch": 0.15254237288135594,
-      "grad_norm": 0.4453125,
+      "grad_norm": 0.455078125,
       "learning_rate": 0.00019995396279688397,
-      "loss": 1.2097,
+      "loss": 1.1455,
       "step": 27
     },
     {
       "epoch": 0.15819209039548024,
-      "grad_norm": 0.3671875,
+      "grad_norm": 0.345703125,
       "learning_rate": 0.00019994838783670793,
-      "loss": 1.0494,
+      "loss": 1.0924,
       "step": 28
     },
     {
       "epoch": 0.1638418079096045,
-      "grad_norm": 0.46875,
+      "grad_norm": 0.48046875,
       "learning_rate": 0.00019994249442005377,
-      "loss": 1.3792,
+      "loss": 1.1255,
       "step": 29
     },
     {
       "epoch": 0.1694915254237288,
-      "grad_norm": 0.453125,
+      "grad_norm": 0.421875,
       "learning_rate": 0.0001999362825656992,
-      "loss": 1.1919,
+      "loss": 1.2381,
       "step": 30
     },
     {
       "epoch": 0.1751412429378531,
-      "grad_norm": 0.419921875,
+      "grad_norm": 0.3828125,
       "learning_rate": 0.00019992975229343642,
-      "loss": 1.2022,
+      "loss": 1.2144,
       "step": 31
     },
     {
       "epoch": 0.1807909604519774,
-      "grad_norm": 0.373046875,
+      "grad_norm": 0.4375,
       "learning_rate": 0.0001999229036240723,
-      "loss": 1.1325,
+      "loss": 1.3607,
       "step": 32
     },
     {
       "epoch": 0.1864406779661017,
-      "grad_norm": 0.375,
+      "grad_norm": 0.40234375,
       "learning_rate": 0.00019991573657942812,
-      "loss": 1.0697,
+      "loss": 1.1471,
       "step": 33
     },
     {
       "epoch": 0.192090395480226,
-      "grad_norm": 0.384765625,
+      "grad_norm": 0.376953125,
       "learning_rate": 0.00019990825118233957,
-      "loss": 1.1981,
+      "loss": 1.1821,
       "step": 34
     },
     {
       "epoch": 0.1977401129943503,
-      "grad_norm": 0.396484375,
+      "grad_norm": 0.365234375,
       "learning_rate": 0.00019990044745665672,
-      "loss": 1.1623,
+      "loss": 1.0378,
       "step": 35
     },
     {
       "epoch": 0.2033898305084746,
-      "grad_norm": 0.38671875,
+      "grad_norm": 0.42578125,
       "learning_rate": 0.00019989232542724384,
-      "loss": 1.0098,
+      "loss": 1.1688,
       "step": 36
     },
     {
       "epoch": 0.20903954802259886,
-      "grad_norm": 0.388671875,
+      "grad_norm": 0.33984375,
       "learning_rate": 0.00019988388511997944,
-      "loss": 1.1908,
+      "loss": 1.0936,
       "step": 37
     },
     {
       "epoch": 0.21468926553672316,
-      "grad_norm": 0.384765625,
+      "grad_norm": 0.439453125,
       "learning_rate": 0.0001998751265617561,
-      "loss": 1.1398,
+      "loss": 1.1055,
       "step": 38
     },
     {
       "epoch": 0.22033898305084745,
-      "grad_norm": 0.4296875,
+      "grad_norm": 0.4140625,
       "learning_rate": 0.00019986604978048042,
-      "loss": 1.2397,
+      "loss": 1.2482,
       "step": 39
     },
     {
       "epoch": 0.22598870056497175,
-      "grad_norm": 0.32421875,
+      "grad_norm": 0.435546875,
       "learning_rate": 0.0001998566548050729,
-      "loss": 0.9737,
+      "loss": 1.1231,
       "step": 40
     },
     {
       "epoch": 0.23163841807909605,
-      "grad_norm": 0.439453125,
+      "grad_norm": 0.41015625,
       "learning_rate": 0.00019984694166546793,
-      "loss": 1.2387,
+      "loss": 1.1507,
       "step": 41
     },
     {
       "epoch": 0.23728813559322035,
-      "grad_norm": 0.43359375,
+      "grad_norm": 0.412109375,
       "learning_rate": 0.00019983691039261357,
-      "loss": 1.2435,
+      "loss": 1.1808,
       "step": 42
     },
     {
       "epoch": 0.24293785310734464,
-      "grad_norm": 0.462890625,
+      "grad_norm": 0.380859375,
       "learning_rate": 0.00019982656101847162,
-      "loss": 1.2568,
+      "loss": 1.005,
       "step": 43
     },
     {
       "epoch": 0.24858757062146894,
-      "grad_norm": 0.4609375,
+      "grad_norm": 0.35546875,
       "learning_rate": 0.00019981589357601727,
-      "loss": 1.2027,
+      "loss": 0.9458,
       "step": 44
     },
     {
       "epoch": 0.2542372881355932,
-      "grad_norm": 0.447265625,
+      "grad_norm": 0.392578125,
       "learning_rate": 0.00019980490809923926,
-      "loss": 1.1662,
+      "loss": 1.1905,
       "step": 45
     },
     {
       "epoch": 0.2542372881355932,
-      "eval_loss": 1.271873116493225,
-      "eval_runtime": 5.9261,
-      "eval_samples_per_second": 25.143,
-      "eval_steps_per_second": 3.206,
+      "eval_loss": 1.2649747133255005,
+      "eval_runtime": 5.7291,
+      "eval_samples_per_second": 26.182,
+      "eval_steps_per_second": 3.316,
       "step": 45
     },
     {
       "epoch": 0.2598870056497175,
-      "grad_norm": 0.341796875,
+      "grad_norm": 0.3984375,
       "learning_rate": 0.00019979360462313962,
-      "loss": 0.995,
+      "loss": 1.1615,
       "step": 46
     },
     {
       "epoch": 0.2655367231638418,
-      "grad_norm": 0.4453125,
+      "grad_norm": 0.546875,
       "learning_rate": 0.00019978198318373362,
-      "loss": 1.3499,
+      "loss": 1.3074,
       "step": 47
     },
     {
       "epoch": 0.2711864406779661,
-      "grad_norm": 0.458984375,
+      "grad_norm": 0.42578125,
       "learning_rate": 0.00019977004381804952,
-      "loss": 1.1659,
+      "loss": 1.1342,
       "step": 48
     },
     {
       "epoch": 0.2768361581920904,
-      "grad_norm": 0.376953125,
+      "grad_norm": 0.423828125,
       "learning_rate": 0.00019975778656412872,
-      "loss": 1.0402,
+      "loss": 1.0531,
       "step": 49
     },
     {
       "epoch": 0.2824858757062147,
-      "grad_norm": 0.390625,
+      "grad_norm": 0.4921875,
       "learning_rate": 0.00019974521146102537,
-      "loss": 1.115,
+      "loss": 1.0751,
       "step": 50
     },
     {
       "epoch": 0.288135593220339,
-      "grad_norm": 0.37109375,
+      "grad_norm": 0.42578125,
       "learning_rate": 0.00019973231854880633,
-      "loss": 1.0301,
+      "loss": 1.0091,
       "step": 51
     },
     {
       "epoch": 0.2937853107344633,
-      "grad_norm": 0.427734375,
+      "grad_norm": 0.318359375,
       "learning_rate": 0.0001997191078685512,
-      "loss": 1.0565,
+      "loss": 0.776,
       "step": 52
     },
     {
       "epoch": 0.2994350282485876,
-      "grad_norm": 0.41015625,
+      "grad_norm": 0.455078125,
       "learning_rate": 0.0001997055794623519,
-      "loss": 0.9576,
+      "loss": 1.101,
       "step": 53
     },
     {
       "epoch": 0.3050847457627119,
-      "grad_norm": 0.412109375,
+      "grad_norm": 0.43359375,
       "learning_rate": 0.0001996917333733128,
-      "loss": 1.0076,
+      "loss": 1.0977,
       "step": 54
     },
     {
       "epoch": 0.3107344632768362,
-      "grad_norm": 0.416015625,
+      "grad_norm": 0.578125,
       "learning_rate": 0.00019967756964555045,
-      "loss": 1.127,
+      "loss": 1.2489,
       "step": 55
     },
     {
       "epoch": 0.3163841807909605,
-      "grad_norm": 0.3515625,
+      "grad_norm": 0.578125,
       "learning_rate": 0.0001996630883241934,
-      "loss": 1.0786,
+      "loss": 1.246,
       "step": 56
     },
     {
       "epoch": 0.3220338983050847,
-      "grad_norm": 0.431640625,
+      "grad_norm": 0.392578125,
       "learning_rate": 0.0001996482894553822,
-      "loss": 1.2096,
+      "loss": 1.0312,
       "step": 57
     },
     {
       "epoch": 0.327683615819209,
-      "grad_norm": 0.37890625,
+      "grad_norm": 0.41015625,
       "learning_rate": 0.00019963317308626914,
-      "loss": 1.0681,
+      "loss": 0.9253,
       "step": 58
     },
     {
       "epoch": 0.3333333333333333,
-      "grad_norm": 0.3671875,
+      "grad_norm": 0.486328125,
       "learning_rate": 0.00019961773926501818,
-      "loss": 0.9736,
+      "loss": 1.092,
       "step": 59
     },
     {
       "epoch": 0.3389830508474576,
-      "grad_norm": 0.39453125,
+      "grad_norm": 0.494140625,
       "learning_rate": 0.0001996019880408046,
-      "loss": 1.0853,
+      "loss": 1.1731,
       "step": 60
     },
     {
       "epoch": 0.3446327683615819,
-      "grad_norm": 0.3828125,
+      "grad_norm": 0.490234375,
       "learning_rate": 0.00019958591946381518,
-      "loss": 0.9952,
+      "loss": 1.0973,
       "step": 61
     },
     {
       "epoch": 0.3502824858757062,
-      "grad_norm": 0.423828125,
+      "grad_norm": 0.42578125,
       "learning_rate": 0.00019956953358524774,
-      "loss": 0.9737,
+      "loss": 1.0246,
       "step": 62
     },
     {
       "epoch": 0.3559322033898305,
-      "grad_norm": 0.4296875,
+      "grad_norm": 0.455078125,
       "learning_rate": 0.00019955283045731116,
-      "loss": 1.1173,
+      "loss": 1.0232,
       "step": 63
     },
     {
       "epoch": 0.3615819209039548,
-      "grad_norm": 0.474609375,
+      "grad_norm": 0.470703125,
       "learning_rate": 0.00019953581013322502,
-      "loss": 1.1305,
+      "loss": 1.06,
       "step": 64
     },
     {
       "epoch": 0.3672316384180791,
-      "grad_norm": 0.5,
+      "grad_norm": 0.484375,
       "learning_rate": 0.0001995184726672197,
-      "loss": 1.1714,
+      "loss": 1.1433,
       "step": 65
     },
     {
       "epoch": 0.3728813559322034,
-      "grad_norm": 0.419921875,
+      "grad_norm": 0.466796875,
       "learning_rate": 0.00019950081811453597,
-      "loss": 1.089,
+      "loss": 1.1356,
       "step": 66
     },
     {
       "epoch": 0.3785310734463277,
-      "grad_norm": 0.43359375,
+      "grad_norm": 0.4765625,
       "learning_rate": 0.00019948284653142494,
-      "loss": 1.0704,
+      "loss": 1.1103,
       "step": 67
     },
     {
       "epoch": 0.384180790960452,
-      "grad_norm": 0.408203125,
+      "grad_norm": 0.453125,
       "learning_rate": 0.0001994645579751478,
-      "loss": 1.0772,
+      "loss": 1.0773,
       "step": 68
     },
     {
       "epoch": 0.3898305084745763,
-      "grad_norm": 0.416015625,
+      "grad_norm": 0.51171875,
       "learning_rate": 0.00019944595250397578,
-      "loss": 1.2268,
+      "loss": 0.9772,
       "step": 69
     },
     {
       "epoch": 0.3954802259887006,
-      "grad_norm": 0.376953125,
+      "grad_norm": 0.38671875,
       "learning_rate": 0.00019942703017718975,
-      "loss": 1.0934,
+      "loss": 1.0252,
       "step": 70
     },
     {
       "epoch": 0.4011299435028249,
-      "grad_norm": 0.46875,
+      "grad_norm": 0.44921875,
       "learning_rate": 0.00019940779105508022,
-      "loss": 0.8819,
+      "loss": 1.0256,
       "step": 71
     },
     {
       "epoch": 0.4067796610169492,
-      "grad_norm": 0.41015625,
+      "grad_norm": 0.400390625,
       "learning_rate": 0.00019938823519894707,
-      "loss": 0.9737,
+      "loss": 0.9186,
       "step": 72
     },
     {
       "epoch": 0.4124293785310734,
-      "grad_norm": 0.54296875,
+      "grad_norm": 0.55859375,
       "learning_rate": 0.00019936836267109932,
-      "loss": 1.1224,
+      "loss": 1.2676,
       "step": 73
     },
     {
       "epoch": 0.4180790960451977,
-      "grad_norm": 0.4140625,
+      "grad_norm": 0.5,
       "learning_rate": 0.00019934817353485501,
-      "loss": 1.0209,
+      "loss": 1.053,
       "step": 74
     },
     {
       "epoch": 0.423728813559322,
-      "grad_norm": 0.42578125,
+      "grad_norm": 0.462890625,
       "learning_rate": 0.000199327667854541,
-      "loss": 1.0283,
+      "loss": 1.1316,
       "step": 75
     },
     {
       "epoch": 0.4293785310734463,
-      "grad_norm": 0.470703125,
+      "grad_norm": 0.435546875,
       "learning_rate": 0.00019930684569549264,
-      "loss": 1.0984,
+      "loss": 1.0186,
       "step": 76
     },
     {
       "epoch": 0.4350282485875706,
-      "grad_norm": 0.46875,
+      "grad_norm": 0.408203125,
       "learning_rate": 0.00019928570712405367,
-      "loss": 1.0828,
+      "loss": 0.9959,
       "step": 77
     },
     {
       "epoch": 0.4406779661016949,
-      "grad_norm": 0.39453125,
+      "grad_norm": 0.443359375,
       "learning_rate": 0.00019926425220757605,
-      "loss": 0.8265,
+      "loss": 1.0721,
       "step": 78
     },
     {
       "epoch": 0.4463276836158192,
-      "grad_norm": 0.486328125,
+      "grad_norm": 0.419921875,
       "learning_rate": 0.00019924248101441957,
-      "loss": 1.2117,
+      "loss": 0.9215,
       "step": 79
     },
     {
       "epoch": 0.4519774011299435,
-      "grad_norm": 0.3671875,
+      "grad_norm": 0.51171875,
       "learning_rate": 0.00019922039361395185,
-      "loss": 0.7824,
+      "loss": 1.1617,
       "step": 80
     },
     {
       "epoch": 0.4576271186440678,
-      "grad_norm": 0.462890625,
+      "grad_norm": 0.490234375,
       "learning_rate": 0.00019919799007654798,
-      "loss": 1.1149,
+      "loss": 0.9922,
       "step": 81
     },
     {
       "epoch": 0.4632768361581921,
-      "grad_norm": 0.484375,
+      "grad_norm": 0.40625,
       "learning_rate": 0.00019917527047359028,
-      "loss": 1.0127,
+      "loss": 0.9527,
       "step": 82
     },
     {
       "epoch": 0.4689265536723164,
-      "grad_norm": 0.45703125,
+      "grad_norm": 0.52734375,
       "learning_rate": 0.0001991522348774682,
-      "loss": 1.0921,
+      "loss": 1.0917,
       "step": 83
     },
     {
       "epoch": 0.4745762711864407,
-      "grad_norm": 0.470703125,
+      "grad_norm": 0.62109375,
       "learning_rate": 0.00019912888336157792,
-      "loss": 0.9566,
+      "loss": 1.1829,
       "step": 84
     },
     {
       "epoch": 0.480225988700565,
-      "grad_norm": 0.4453125,
+      "grad_norm": 0.48828125,
       "learning_rate": 0.00019910521600032227,
-      "loss": 1.0421,
+      "loss": 1.0053,
       "step": 85
     },
     {
       "epoch": 0.4858757062146893,
-      "grad_norm": 0.43359375,
+      "grad_norm": 0.3828125,
       "learning_rate": 0.00019908123286911046,
-      "loss": 0.9244,
+      "loss": 0.9311,
       "step": 86
     },
     {
       "epoch": 0.4915254237288136,
-      "grad_norm": 0.47265625,
+      "grad_norm": 0.41796875,
       "learning_rate": 0.00019905693404435773,
-      "loss": 1.0229,
+      "loss": 0.9428,
       "step": 87
     },
     {
       "epoch": 0.4971751412429379,
-      "grad_norm": 0.5234375,
+      "grad_norm": 0.42578125,
       "learning_rate": 0.00019903231960348523,
-      "loss": 0.9554,
+      "loss": 0.9485,
       "step": 88
     },
     {
       "epoch": 0.5028248587570622,
-      "grad_norm": 0.51171875,
+      "grad_norm": 0.41796875,
       "learning_rate": 0.0001990073896249197,
-      "loss": 1.0098,
+      "loss": 0.981,
       "step": 89
     },
     {
       "epoch": 0.5084745762711864,
-      "grad_norm": 0.39453125,
+      "grad_norm": 0.458984375,
       "learning_rate": 0.0001989821441880933,
-      "loss": 0.8512,
+      "loss": 1.0583,
       "step": 90
     },
     {
       "epoch": 0.5084745762711864,
-      "eval_loss": 1.1145851612091064,
-      "eval_runtime": 5.7225,
-      "eval_samples_per_second": 26.038,
-      "eval_steps_per_second": 3.32,
+      "eval_loss": 1.128895878791809,
+      "eval_runtime": 5.7026,
+      "eval_samples_per_second": 26.304,
+      "eval_steps_per_second": 3.332,
       "step": 90
     },
     {
       "epoch": 0.5141242937853108,
-      "grad_norm": 0.55078125,
+      "grad_norm": 0.404296875,
       "learning_rate": 0.00019895658337344322,
-      "loss": 0.9078,
+      "loss": 0.9323,
       "step": 91
     },
     {
       "epoch": 0.519774011299435,
-      "grad_norm": 0.50390625,
+      "grad_norm": 0.42578125,
       "learning_rate": 0.00019893070726241164,
-      "loss": 1.1472,
+      "loss": 0.9144,
       "step": 92
     },
     {
       "epoch": 0.5254237288135594,
-      "grad_norm": 0.41015625,
+      "grad_norm": 0.50390625,
       "learning_rate": 0.00019890451593744518,
-      "loss": 0.8285,
+      "loss": 1.0559,
       "step": 93
     },
     {
       "epoch": 0.5310734463276836,
-      "grad_norm": 0.470703125,
+      "grad_norm": 0.4375,
       "learning_rate": 0.00019887800948199496,
-      "loss": 0.9879,
+      "loss": 0.9494,
       "step": 94
     },
     {
       "epoch": 0.536723163841808,
-      "grad_norm": 0.49609375,
+      "grad_norm": 0.4375,
       "learning_rate": 0.00019885118798051605,
-      "loss": 0.9555,
+      "loss": 0.8916,
       "step": 95
     },
     {
       "epoch": 0.5423728813559322,
-      "grad_norm": 0.58203125,
+      "grad_norm": 0.51953125,
       "learning_rate": 0.0001988240515184674,
-      "loss": 1.0681,
+      "loss": 1.005,
       "step": 96
     },
     {
       "epoch": 0.5480225988700564,
-      "grad_norm": 0.44921875,
+      "grad_norm": 0.4921875,
       "learning_rate": 0.0001987966001823114,
-      "loss": 0.9245,
+      "loss": 0.797,
       "step": 97
     },
     {
       "epoch": 0.5536723163841808,
-      "grad_norm": 0.4453125,
+      "grad_norm": 0.5078125,
       "learning_rate": 0.00019876883405951377,
-      "loss": 0.9083,
+      "loss": 0.8724,
       "step": 98
     },
     {
       "epoch": 0.559322033898305,
-      "grad_norm": 0.55859375,
+      "grad_norm": 0.455078125,
       "learning_rate": 0.00019874075323854325,
-      "loss": 1.0449,
+      "loss": 0.9111,
       "step": 99
     },
     {
       "epoch": 0.5649717514124294,
-      "grad_norm": 0.51171875,
+      "grad_norm": 0.52734375,
       "learning_rate": 0.00019871235780887113,
-      "loss": 1.0546,
+      "loss": 1.088,
       "step": 100
     },
     {
       "epoch": 0.5706214689265536,
-      "grad_norm": 0.51953125,
+      "grad_norm": 0.55859375,
       "learning_rate": 0.00019868364786097123,
-      "loss": 0.8278,
+      "loss": 1.0186,
       "step": 101
     },
     {
       "epoch": 0.576271186440678,
-      "grad_norm": 0.50390625,
+      "grad_norm": 0.486328125,
       "learning_rate": 0.00019865462348631944,
-      "loss": 1.0551,
+      "loss": 0.9854,
       "step": 102
     },
     {
       "epoch": 0.5819209039548022,
-      "grad_norm": 0.51171875,
+      "grad_norm": 0.515625,
       "learning_rate": 0.0001986252847773935,
-      "loss": 1.009,
+      "loss": 0.9578,
       "step": 103
     },
     {
       "epoch": 0.5875706214689266,
-      "grad_norm": 0.45703125,
+      "grad_norm": 0.5078125,
       "learning_rate": 0.00019859563182767268,
-      "loss": 0.9314,
+      "loss": 0.9375,
       "step": 104
     },
     {
       "epoch": 0.5932203389830508,
-      "grad_norm": 0.53515625,
+      "grad_norm": 0.54296875,
       "learning_rate": 0.00019856566473163746,
-      "loss": 0.9617,
+      "loss": 1.0428,
       "step": 105
     },
     {
       "epoch": 0.5988700564971752,
-      "grad_norm": 0.46484375,
+      "grad_norm": 0.412109375,
       "learning_rate": 0.00019853538358476932,
-      "loss": 0.8029,
+      "loss": 0.8436,
       "step": 106
     },
     {
       "epoch": 0.6045197740112994,
-      "grad_norm": 0.55078125,
+      "grad_norm": 0.515625,
       "learning_rate": 0.00019850478848355027,
-      "loss": 0.9028,
+      "loss": 0.9373,
       "step": 107
     },
     {
       "epoch": 0.6101694915254238,
-      "grad_norm": 0.546875,
+      "grad_norm": 0.515625,
       "learning_rate": 0.00019847387952546278,
-      "loss": 0.9104,
+      "loss": 0.9945,
       "step": 108
     },
     {
       "epoch": 0.615819209039548,
-      "grad_norm": 0.63671875,
+      "grad_norm": 0.462890625,
       "learning_rate": 0.00019844265680898918,
-      "loss": 1.0671,
+      "loss": 0.8323,
       "step": 109
     },
     {
       "epoch": 0.6214689265536724,
-      "grad_norm": 0.455078125,
+      "grad_norm": 0.466796875,
       "learning_rate": 0.0001984111204336116,
-      "loss": 0.9423,
+      "loss": 0.9229,
       "step": 110
     },
     {
       "epoch": 0.6271186440677966,
-      "grad_norm": 0.51171875,
+      "grad_norm": 0.515625,
       "learning_rate": 0.0001983792704998116,
-      "loss": 0.919,
+      "loss": 0.9112,
       "step": 111
     },
     {
       "epoch": 0.632768361581921,
-      "grad_norm": 0.53125,
+      "grad_norm": 0.51171875,
       "learning_rate": 0.00019834710710906964,
-      "loss": 0.8353,
+      "loss": 1.0333,
       "step": 112
     },
     {
       "epoch": 0.6384180790960452,
-      "grad_norm": 0.5078125,
+      "grad_norm": 0.6171875,
       "learning_rate": 0.00019831463036386508,
-      "loss": 0.8803,
+      "loss": 0.9753,
       "step": 113
     },
     {
       "epoch": 0.6440677966101694,
-      "grad_norm": 0.6171875,
+      "grad_norm": 0.5859375,
       "learning_rate": 0.00019828184036767556,
-      "loss": 0.9161,
+      "loss": 0.8798,
       "step": 114
     },
     {
       "epoch": 0.6497175141242938,
-      "grad_norm": 0.61328125,
+      "grad_norm": 0.6640625,
       "learning_rate": 0.00019824873722497694,
-      "loss": 1.1127,
+      "loss": 0.9572,
       "step": 115
     },
     {
       "epoch": 0.655367231638418,
-      "grad_norm": 0.470703125,
+      "grad_norm": 0.58203125,
       "learning_rate": 0.00019821532104124273,
-      "loss": 0.8086,
+      "loss": 0.9029,
       "step": 116
     },
     {
       "epoch": 0.6610169491525424,
-      "grad_norm": 0.4453125,
+      "grad_norm": 0.6484375,
       "learning_rate": 0.00019818159192294387,
-      "loss": 0.9012,
+      "loss": 0.98,
       "step": 117
     },
     {
       "epoch": 0.6666666666666666,
-      "grad_norm": 0.48828125,
+      "grad_norm": 0.609375,
       "learning_rate": 0.0001981475499775484,
-      "loss": 0.7488,
+      "loss": 0.9489,
       "step": 118
     },
     {
       "epoch": 0.672316384180791,
-      "grad_norm": 0.51171875,
+      "grad_norm": 0.5234375,
       "learning_rate": 0.00019811319531352109,
-      "loss": 0.999,
+      "loss": 0.8836,
       "step": 119
     },
     {
       "epoch": 0.6779661016949152,
-      "grad_norm": 0.5703125,
+      "grad_norm": 0.546875,
       "learning_rate": 0.00019807852804032305,
-      "loss": 0.895,
+      "loss": 0.8237,
       "step": 120
     },
     {
       "epoch": 0.6836158192090396,
-      "grad_norm": 0.55078125,
+      "grad_norm": 0.5546875,
       "learning_rate": 0.00019804354826841153,
-      "loss": 0.7695,
+      "loss": 0.8586,
       "step": 121
     },
     {
       "epoch": 0.6892655367231638,
-      "grad_norm": 0.5390625,
+      "grad_norm": 0.5,
       "learning_rate": 0.00019800825610923934,
-      "loss": 0.8331,
+      "loss": 0.9908,
       "step": 122
     },
     {
       "epoch": 0.6949152542372882,
-      "grad_norm": 0.6328125,
+      "grad_norm": 0.470703125,
       "learning_rate": 0.00019797265167525473,
-      "loss": 0.7878,
+      "loss": 0.7604,
       "step": 123
     },
     {
       "epoch": 0.7005649717514124,
-      "grad_norm": 0.76171875,
+      "grad_norm": 0.640625,
       "learning_rate": 0.00019793673507990088,
-      "loss": 0.9084,
+      "loss": 0.9742,
       "step": 124
     },
     {
       "epoch": 0.7062146892655368,
-      "grad_norm": 0.5546875,
+      "grad_norm": 0.51953125,
       "learning_rate": 0.00019790050643761552,
-      "loss": 0.8761,
+      "loss": 0.7734,
       "step": 125
     },
     {
       "epoch": 0.711864406779661,
-      "grad_norm": 0.53515625,
+      "grad_norm": 0.474609375,
       "learning_rate": 0.00019786396586383075,
-      "loss": 0.8607,
+      "loss": 0.7948,
       "step": 126
     },
     {
       "epoch": 0.7175141242937854,
-      "grad_norm": 0.6328125,
+      "grad_norm": 0.5078125,
       "learning_rate": 0.00019782711347497248,
-      "loss": 0.8591,
+      "loss": 0.7784,
       "step": 127
     },
     {
       "epoch": 0.7231638418079096,
-      "grad_norm": 0.5390625,
+      "grad_norm": 0.5703125,
       "learning_rate": 0.00019778994938846007,
-      "loss": 0.8919,
+      "loss": 0.922,
       "step": 128
     },
     {
       "epoch": 0.7288135593220338,
-      "grad_norm": 0.71484375,
+      "grad_norm": 0.63671875,
       "learning_rate": 0.00019775247372270612,
-      "loss": 0.9822,
+      "loss": 0.8655,
       "step": 129
     },
     {
       "epoch": 0.7344632768361582,
       "grad_norm": 0.578125,
       "learning_rate": 0.00019771468659711595,
-      "loss": 0.8809,
+      "loss": 0.8844,
       "step": 130
     },
     {
       "epoch": 0.7401129943502824,
-      "grad_norm": 0.62890625,
+      "grad_norm": 0.56640625,
       "learning_rate": 0.00019767658813208726,
-      "loss": 0.8194,
+      "loss": 0.914,
       "step": 131
     },
     {
       "epoch": 0.7457627118644068,
-      "grad_norm": 0.59765625,
+      "grad_norm": 0.54296875,
       "learning_rate": 0.0001976381784490097,
-      "loss": 0.9644,
+      "loss": 0.9704,
       "step": 132
     },
     {
       "epoch": 0.751412429378531,
-      "grad_norm": 0.52734375,
+      "grad_norm": 0.49609375,
       "learning_rate": 0.00019759945767026452,
-      "loss": 0.9229,
+      "loss": 0.8511,
       "step": 133
     },
     {
       "epoch": 0.7570621468926554,
-      "grad_norm": 0.61328125,
+      "grad_norm": 0.578125,
       "learning_rate": 0.00019756042591922434,
-      "loss": 0.786,
+      "loss": 0.8498,
       "step": 134
     },
     {
       "epoch": 0.7627118644067796,
-      "grad_norm": 0.6875,
+      "grad_norm": 0.62890625,
       "learning_rate": 0.0001975210833202524,
-      "loss": 1.141,
+      "loss": 0.9094,
       "step": 135
     },
     {
       "epoch": 0.7627118644067796,
-      "eval_loss": 0.9756787419319153,
-      "eval_runtime": 5.8135,
-      "eval_samples_per_second": 25.63,
-      "eval_steps_per_second": 3.268,
+      "eval_loss": 0.9717230200767517,
+      "eval_runtime": 5.9691,
+      "eval_samples_per_second": 25.129,
+      "eval_steps_per_second": 3.183,
       "step": 135
     },
     {
       "epoch": 0.768361581920904,
-      "grad_norm": 0.455078125,
+      "grad_norm": 0.59765625,
       "learning_rate": 0.00019748142999870245,
-      "loss": 0.6275,
+      "loss": 0.9191,
       "step": 136
     },
     {
       "epoch": 0.7740112994350282,
-      "grad_norm": 0.63671875,
+      "grad_norm": 0.55859375,
       "learning_rate": 0.00019744146608091832,
-      "loss": 0.8421,
+      "loss": 0.811,
       "step": 137
     },
     {
       "epoch": 0.7796610169491526,
-      "grad_norm": 0.51953125,
+      "grad_norm": 0.57421875,
       "learning_rate": 0.00019740119169423337,
-      "loss": 0.7829,
+      "loss": 0.8093,
       "step": 138
     },
     {
       "epoch": 0.7853107344632768,
-      "grad_norm": 0.6328125,
+      "grad_norm": 0.73046875,
       "learning_rate": 0.00019736060696697022,
-      "loss": 0.7404,
+      "loss": 0.9262,
       "step": 139
     },
     {
       "epoch": 0.7909604519774012,
-      "grad_norm": 0.6875,
+      "grad_norm": 0.62109375,
       "learning_rate": 0.00019731971202844036,
-      "loss": 0.9372,
+      "loss": 0.7015,
       "step": 140
     },
     {
       "epoch": 0.7966101694915254,
-      "grad_norm": 0.75390625,
+      "grad_norm": 0.7890625,
       "learning_rate": 0.00019727850700894357,
-      "loss": 0.7863,
+      "loss": 0.8984,
       "step": 141
     },
     {
       "epoch": 0.8022598870056498,
-      "grad_norm": 0.6328125,
+      "grad_norm": 0.640625,
       "learning_rate": 0.00019723699203976766,
-      "loss": 0.846,
+      "loss": 0.8673,
       "step": 142
     },
     {
       "epoch": 0.807909604519774,
-      "grad_norm": 0.6640625,
+      "grad_norm": 0.69140625,
       "learning_rate": 0.00019719516725318805,
-      "loss": 1.0144,
+      "loss": 0.8079,
       "step": 143
     },
     {
       "epoch": 0.8135593220338984,
-      "grad_norm": 0.62109375,
+      "grad_norm": 0.66015625,
       "learning_rate": 0.00019715303278246724,
-      "loss": 0.7371,
+      "loss": 0.8473,
       "step": 144
     },
     {
       "epoch": 0.8192090395480226,
-      "grad_norm": 0.703125,
+      "grad_norm": 0.76953125,
       "learning_rate": 0.00019711058876185447,
-      "loss": 0.8283,
+      "loss": 0.9935,
       "step": 145
     },
     {
       "epoch": 0.8248587570621468,
-      "grad_norm": 0.6875,
+      "grad_norm": 0.64453125,
       "learning_rate": 0.00019706783532658526,
-      "loss": 0.872,
+      "loss": 0.8659,
       "step": 146
     },
     {
       "epoch": 0.8305084745762712,
-      "grad_norm": 0.7890625,
+      "grad_norm": 0.466796875,
       "learning_rate": 0.00019702477261288102,
-      "loss": 0.7923,
+      "loss": 0.7009,
       "step": 147
     },
     {
       "epoch": 0.8361581920903954,
-      "grad_norm": 0.6640625,
+      "grad_norm": 0.53125,
       "learning_rate": 0.00019698140075794854,
-      "loss": 0.8278,
+      "loss": 0.7009,
       "step": 148
     },
     {
       "epoch": 0.8418079096045198,
-      "grad_norm": 0.76953125,
+      "grad_norm": 0.77734375,
       "learning_rate": 0.0001969377198999797,
-      "loss": 0.9674,
+      "loss": 0.9978,
       "step": 149
     },
     {
       "epoch": 0.847457627118644,
-      "grad_norm": 0.578125,
+      "grad_norm": 0.69921875,
       "learning_rate": 0.00019689373017815073,
-      "loss": 0.7383,
+      "loss": 0.9561,
       "step": 150
     },
     {
       "epoch": 0.8531073446327684,
-      "grad_norm": 0.7109375,
+      "grad_norm": 0.66015625,
       "learning_rate": 0.0001968494317326222,
-      "loss": 0.9629,
+      "loss": 0.7592,
       "step": 151
     },
     {
       "epoch": 0.8587570621468926,
-      "grad_norm": 0.5859375,
+      "grad_norm": 0.56640625,
       "learning_rate": 0.00019680482470453816,
-      "loss": 0.8607,
+      "loss": 0.7073,
       "step": 152
     },
     {
       "epoch": 0.864406779661017,
-      "grad_norm": 0.69140625,
+      "grad_norm": 0.6953125,
       "learning_rate": 0.00019675990923602598,
-      "loss": 0.7762,
+      "loss": 0.9811,
       "step": 153
     },
     {
       "epoch": 0.8700564971751412,
-      "grad_norm": 0.58203125,
+      "grad_norm": 0.66796875,
       "learning_rate": 0.00019671468547019573,
-      "loss": 0.8593,
+      "loss": 0.7197,
       "step": 154
     },
     {
       "epoch": 0.8757062146892656,
-      "grad_norm": 0.73828125,
+      "grad_norm": 0.625,
       "learning_rate": 0.00019666915355113975,
-      "loss": 0.8494,
+      "loss": 0.8884,
       "step": 155
     },
     {
       "epoch": 0.8813559322033898,
-      "grad_norm": 0.671875,
+      "grad_norm": 0.68359375,
       "learning_rate": 0.00019662331362393234,
-      "loss": 0.7774,
+      "loss": 0.6698,
       "step": 156
     },
     {
       "epoch": 0.8870056497175142,
-      "grad_norm": 0.6640625,
+      "grad_norm": 0.92578125,
       "learning_rate": 0.00019657716583462901,
-      "loss": 0.7656,
+      "loss": 0.9329,
       "step": 157
     },
     {
       "epoch": 0.8926553672316384,
-      "grad_norm": 0.73046875,
+      "grad_norm": 0.63671875,
       "learning_rate": 0.00019653071033026633,
-      "loss": 0.9002,
+      "loss": 0.8287,
       "step": 158
     },
     {
       "epoch": 0.8983050847457628,
-      "grad_norm": 0.57421875,
+      "grad_norm": 0.56640625,
       "learning_rate": 0.00019648394725886126,
-      "loss": 0.7957,
+      "loss": 0.7756,
       "step": 159
     },
     {
       "epoch": 0.903954802259887,
-      "grad_norm": 0.71875,
+      "grad_norm": 0.73046875,
       "learning_rate": 0.00019643687676941068,
-      "loss": 0.9368,
+      "loss": 1.0596,
       "step": 160
     },
     {
       "epoch": 0.9096045197740112,
-      "grad_norm": 0.6171875,
+      "grad_norm": 0.66015625,
       "learning_rate": 0.0001963894990118911,
-      "loss": 0.8406,
+      "loss": 0.7007,
       "step": 161
     },
     {
       "epoch": 0.9152542372881356,
-      "grad_norm": 0.73046875,
+      "grad_norm": 0.6953125,
       "learning_rate": 0.0001963418141372579,
-      "loss": 0.6971,
+      "loss": 0.8837,
       "step": 162
     },
     {
       "epoch": 0.9209039548022598,
-      "grad_norm": 0.5234375,
+      "grad_norm": 0.67578125,
       "learning_rate": 0.00019629382229744511,
-      "loss": 0.8306,
+      "loss": 0.7926,
       "step": 163
     },
     {
       "epoch": 0.9265536723163842,
-      "grad_norm": 0.609375,
+      "grad_norm": 0.66796875,
       "learning_rate": 0.00019624552364536473,
-      "loss": 0.7188,
+      "loss": 0.863,
       "step": 164
     },
     {
       "epoch": 0.9322033898305084,
-      "grad_norm": 0.75,
+      "grad_norm": 0.72265625,
       "learning_rate": 0.00019619691833490643,
-      "loss": 0.8002,
+      "loss": 0.7391,
       "step": 165
     },
     {
       "epoch": 0.9378531073446328,
-      "grad_norm": 0.72265625,
+      "grad_norm": 0.6484375,
       "learning_rate": 0.00019614800652093686,
-      "loss": 0.7433,
+      "loss": 0.6982,
       "step": 166
     },
     {
       "epoch": 0.943502824858757,
-      "grad_norm": 0.78125,
+      "grad_norm": 0.7890625,
       "learning_rate": 0.0001960987883592993,
-      "loss": 0.8501,
+      "loss": 0.7521,
       "step": 167
     },
     {
       "epoch": 0.9491525423728814,
-      "grad_norm": 0.67578125,
+      "grad_norm": 0.76171875,
       "learning_rate": 0.00019604926400681308,
-      "loss": 0.6454,
+      "loss": 0.7907,
       "step": 168
     },
     {
       "epoch": 0.9548022598870056,
-      "grad_norm": 0.7890625,
+      "grad_norm": 0.82421875,
       "learning_rate": 0.00019599943362127325,
-      "loss": 0.7914,
+      "loss": 0.7764,
       "step": 169
     },
     {
       "epoch": 0.96045197740113,
-      "grad_norm": 0.7421875,
+      "grad_norm": 0.890625,
       "learning_rate": 0.00019594929736144976,
-      "loss": 0.7926,
+      "loss": 0.6525,
       "step": 170
     },
     {
       "epoch": 0.9661016949152542,
-      "grad_norm": 0.7265625,
+      "grad_norm": 0.6875,
       "learning_rate": 0.00019589885538708724,
-      "loss": 0.7469,
+      "loss": 0.767,
       "step": 171
     },
     {
       "epoch": 0.9717514124293786,
-      "grad_norm": 0.8046875,
+      "grad_norm": 0.6640625,
       "learning_rate": 0.00019584810785890438,
-      "loss": 0.7581,
+      "loss": 0.8074,
       "step": 172
     },
     {
       "epoch": 0.9774011299435028,
-      "grad_norm": 0.69140625,
+      "grad_norm": 0.70703125,
       "learning_rate": 0.0001957970549385934,
-      "loss": 0.7232,
+      "loss": 0.7375,
       "step": 173
     },
     {
       "epoch": 0.9830508474576272,
-      "grad_norm": 0.76953125,
+      "grad_norm": 0.6328125,
       "learning_rate": 0.00019574569678881964,
-      "loss": 0.7175,
+      "loss": 0.7003,
       "step": 174
     },
     {
       "epoch": 0.9887005649717514,
-      "grad_norm": 0.62890625,
+      "grad_norm": 0.703125,
       "learning_rate": 0.0001956940335732209,
-      "loss": 0.6937,
+      "loss": 0.7334,
       "step": 175
     },
     {
       "epoch": 0.9943502824858758,
-      "grad_norm": 0.91015625,
+      "grad_norm": 0.72265625,
       "learning_rate": 0.00019564206545640697,
-      "loss": 0.8942,
+      "loss": 0.7634,
       "step": 176
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.62109375,
+      "grad_norm": 0.87109375,
       "learning_rate": 0.00019558979260395916,
-      "loss": 0.6134,
+      "loss": 0.6792,
       "step": 177
     },
     {
       "epoch": 1.0056497175141244,
-      "grad_norm": 0.71484375,
+      "grad_norm": 0.66015625,
       "learning_rate": 0.00019553721518242968,
-      "loss": 0.5471,
+      "loss": 0.5653,
       "step": 178
     },
     {
       "epoch": 1.0112994350282485,
-      "grad_norm": 0.828125,
+      "grad_norm": 0.6875,
       "learning_rate": 0.00019548433335934124,
-      "loss": 0.6652,
+      "loss": 0.4815,
       "step": 179
     },
     {
       "epoch": 1.0169491525423728,
-      "grad_norm": 0.796875,
+      "grad_norm": 0.75390625,
       "learning_rate": 0.0001954311473031864,
-      "loss": 0.5009,
+      "loss": 0.6033,
       "step": 180
     },
     {
       "epoch": 1.0169491525423728,
-      "eval_loss": 0.7862020134925842,
-      "eval_runtime": 5.8105,
-      "eval_samples_per_second": 25.643,
-      "eval_steps_per_second": 3.27,
+      "eval_loss": 0.7865353226661682,
+      "eval_runtime": 5.744,
+      "eval_samples_per_second": 26.114,
+      "eval_steps_per_second": 3.308,
       "step": 180
     },
     {
       "epoch": 1.0225988700564972,
-      "grad_norm": 0.77734375,
+      "grad_norm": 0.78125,
       "learning_rate": 0.00019537765718342703,
-      "loss": 0.5277,
+      "loss": 0.4774,
       "step": 181
     },
     {
       "epoch": 1.0282485875706215,
-      "grad_norm": 0.8203125,
+      "grad_norm": 0.7734375,
       "learning_rate": 0.00019532386317049383,
-      "loss": 0.6077,
+      "loss": 0.4991,
       "step": 182
     },
     {
       "epoch": 1.0338983050847457,
-      "grad_norm": 0.73828125,
+      "grad_norm": 0.80078125,
       "learning_rate": 0.00019526976543578588,
-      "loss": 0.6444,
+      "loss": 0.6054,
       "step": 183
     },
     {
       "epoch": 1.03954802259887,
-      "grad_norm": 0.765625,
+      "grad_norm": 0.94140625,
       "learning_rate": 0.00019521536415166978,
-      "loss": 0.5582,
+      "loss": 0.5636,
       "step": 184
     },
     {
       "epoch": 1.0451977401129944,
-      "grad_norm": 0.79296875,
+      "grad_norm": 0.8359375,
       "learning_rate": 0.00019516065949147943,
-      "loss": 0.6193,
+      "loss": 0.6032,
       "step": 185
     },
     {
       "epoch": 1.0508474576271187,
-      "grad_norm": 0.8671875,
+      "grad_norm": 0.78125,
       "learning_rate": 0.00019510565162951537,
-      "loss": 0.4961,
+      "loss": 0.6017,
       "step": 186
     },
     {
       "epoch": 1.0564971751412429,
-      "grad_norm": 0.8046875,
+      "grad_norm": 0.609375,
       "learning_rate": 0.0001950503407410441,
-      "loss": 0.5411,
+      "loss": 0.643,
       "step": 187
     },
     {
       "epoch": 1.0621468926553672,
-      "grad_norm": 0.78515625,
+      "grad_norm": 0.61328125,
       "learning_rate": 0.0001949947270022977,
-      "loss": 0.5288,
+      "loss": 0.4719,
       "step": 188
     },
     {
       "epoch": 1.0677966101694916,
-      "grad_norm": 0.8203125,
+      "grad_norm": 0.5703125,
       "learning_rate": 0.0001949388105904732,
-      "loss": 0.6722,
+      "loss": 0.537,
       "step": 189
     },
     {
       "epoch": 1.073446327683616,
-      "grad_norm": 0.79296875,
+      "grad_norm": 0.75390625,
       "learning_rate": 0.00019488259168373197,
-      "loss": 0.6451,
+      "loss": 0.5237,
       "step": 190
     },
     {
       "epoch": 1.07909604519774,
-      "grad_norm": 0.765625,
+      "grad_norm": 0.80859375,
       "learning_rate": 0.00019482607046119923,
-      "loss": 0.4888,
+      "loss": 0.5253,
       "step": 191
     },
     {
       "epoch": 1.0847457627118644,
-      "grad_norm": 0.73046875,
+      "grad_norm": 0.8203125,
       "learning_rate": 0.00019476924710296338,
-      "loss": 0.4958,
+      "loss": 0.5397,
       "step": 192
     },
     {
       "epoch": 1.0903954802259888,
-      "grad_norm": 0.77734375,
+      "grad_norm": 0.96484375,
       "learning_rate": 0.0001947121217900756,
-      "loss": 0.4037,
+      "loss": 0.6271,
       "step": 193
     },
     {
       "epoch": 1.0960451977401129,
-      "grad_norm": 0.8515625,
+      "grad_norm": 0.72265625,
       "learning_rate": 0.000194654694704549,
-      "loss": 0.5689,
+      "loss": 0.5593,
       "step": 194
     },
     {
       "epoch": 1.1016949152542372,
-      "grad_norm": 0.69921875,
+      "grad_norm": 0.72265625,
       "learning_rate": 0.00019459696602935837,
-      "loss": 0.4434,
+      "loss": 0.5093,
       "step": 195
     },
     {
       "epoch": 1.1073446327683616,
-      "grad_norm": 0.86328125,
+      "grad_norm": 0.6953125,
       "learning_rate": 0.00019453893594843932,
-      "loss": 0.4725,
+      "loss": 0.4297,
       "step": 196
     },
     {
       "epoch": 1.112994350282486,
-      "grad_norm": 0.94921875,
+      "grad_norm": 0.765625,
       "learning_rate": 0.00019448060464668783,
-      "loss": 0.5757,
+      "loss": 0.5947,
       "step": 197
     },
     {
       "epoch": 1.11864406779661,
-      "grad_norm": 0.8515625,
+      "grad_norm": 0.82421875,
       "learning_rate": 0.0001944219723099596,
-      "loss": 0.6088,
+      "loss": 0.4453,
       "step": 198
     },
     {
       "epoch": 1.1242937853107344,
-      "grad_norm": 0.59375,
+      "grad_norm": 0.7734375,
       "learning_rate": 0.00019436303912506958,
-      "loss": 0.4702,
+      "loss": 0.5101,
       "step": 199
     },
     {
       "epoch": 1.1299435028248588,
-      "grad_norm": 0.765625,
+      "grad_norm": 0.73046875,
       "learning_rate": 0.00019430380527979123,
-      "loss": 0.6775,
+      "loss": 0.5062,
       "step": 200
     },
     {
       "epoch": 1.1355932203389831,
-      "grad_norm": 0.74609375,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.0001942442709628559,
-      "loss": 0.5321,
+      "loss": 0.4489,
       "step": 201
     },
     {
       "epoch": 1.1412429378531073,
-      "grad_norm": 0.8828125,
+      "grad_norm": 0.89453125,
       "learning_rate": 0.00019418443636395248,
-      "loss": 0.5679,
+      "loss": 0.5033,
       "step": 202
     },
     {
       "epoch": 1.1468926553672316,
-      "grad_norm": 0.73828125,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.00019412430167372647,
-      "loss": 0.5675,
+      "loss": 0.4087,
       "step": 203
     },
     {
       "epoch": 1.152542372881356,
-      "grad_norm": 0.78515625,
+      "grad_norm": 0.89453125,
       "learning_rate": 0.00019406386708377955,
-      "loss": 0.4511,
+      "loss": 0.5266,
       "step": 204
     },
     {
       "epoch": 1.1581920903954803,
-      "grad_norm": 0.8515625,
+      "grad_norm": 0.82421875,
       "learning_rate": 0.00019400313278666902,
-      "loss": 0.3674,
+      "loss": 0.4917,
       "step": 205
     },
     {
       "epoch": 1.1638418079096045,
-      "grad_norm": 0.796875,
+      "grad_norm": 0.76953125,
       "learning_rate": 0.00019394209897590704,
-      "loss": 0.4944,
+      "loss": 0.4094,
       "step": 206
     },
     {
       "epoch": 1.1694915254237288,
-      "grad_norm": 0.90625,
+      "grad_norm": 0.73046875,
       "learning_rate": 0.00019388076584596014,
-      "loss": 0.4534,
+      "loss": 0.5372,
       "step": 207
     },
     {
       "epoch": 1.1751412429378532,
-      "grad_norm": 0.90625,
+      "grad_norm": 0.77734375,
       "learning_rate": 0.00019381913359224842,
-      "loss": 0.5257,
+      "loss": 0.5364,
       "step": 208
     },
     {
       "epoch": 1.1807909604519775,
-      "grad_norm": 0.7890625,
+      "grad_norm": 0.8359375,
       "learning_rate": 0.0001937572024111452,
-      "loss": 0.5251,
+      "loss": 0.5772,
       "step": 209
     },
     {
       "epoch": 1.1864406779661016,
-      "grad_norm": 0.72265625,
+      "grad_norm": 0.7734375,
       "learning_rate": 0.0001936949724999762,
-      "loss": 0.3538,
+      "loss": 0.4831,
       "step": 210
     },
     {
       "epoch": 1.192090395480226,
-      "grad_norm": 0.7734375,
+      "grad_norm": 0.65234375,
       "learning_rate": 0.0001936324440570189,
-      "loss": 0.4683,
+      "loss": 0.4307,
       "step": 211
     },
     {
       "epoch": 1.1977401129943503,
-      "grad_norm": 0.82421875,
+      "grad_norm": 0.8828125,
       "learning_rate": 0.00019356961728150197,
-      "loss": 0.397,
+      "loss": 0.4707,
       "step": 212
     },
     {
       "epoch": 1.2033898305084745,
-      "grad_norm": 0.78515625,
+      "grad_norm": 0.94921875,
       "learning_rate": 0.00019350649237360472,
-      "loss": 0.5801,
+      "loss": 0.4271,
       "step": 213
     },
     {
       "epoch": 1.2090395480225988,
-      "grad_norm": 0.796875,
+      "grad_norm": 0.88671875,
       "learning_rate": 0.00019344306953445633,
-      "loss": 0.4672,
+      "loss": 0.3369,
       "step": 214
     },
     {
       "epoch": 1.2146892655367232,
-      "grad_norm": 0.84375,
+      "grad_norm": 0.796875,
       "learning_rate": 0.00019337934896613516,
-      "loss": 0.4278,
+      "loss": 0.433,
       "step": 215
     },
     {
       "epoch": 1.2203389830508475,
-      "grad_norm": 0.67578125,
+      "grad_norm": 0.8828125,
       "learning_rate": 0.00019331533087166836,
-      "loss": 0.5201,
+      "loss": 0.4465,
       "step": 216
     },
     {
       "epoch": 1.2259887005649717,
-      "grad_norm": 0.95703125,
+      "grad_norm": 0.9609375,
       "learning_rate": 0.00019325101545503092,
-      "loss": 0.489,
+      "loss": 0.4866,
       "step": 217
     },
     {
       "epoch": 1.231638418079096,
-      "grad_norm": 0.765625,
+      "grad_norm": 0.94921875,
       "learning_rate": 0.00019318640292114524,
-      "loss": 0.5382,
+      "loss": 0.6098,
       "step": 218
     },
     {
       "epoch": 1.2372881355932204,
-      "grad_norm": 0.84375,
+      "grad_norm": 0.70703125,
       "learning_rate": 0.00019312149347588037,
-      "loss": 0.4704,
+      "loss": 0.4366,
       "step": 219
     },
     {
       "epoch": 1.2429378531073447,
-      "grad_norm": 0.87109375,
+      "grad_norm": 0.75,
       "learning_rate": 0.00019305628732605137,
-      "loss": 0.4391,
+      "loss": 0.5957,
       "step": 220
     },
     {
       "epoch": 1.2485875706214689,
-      "grad_norm": 0.81640625,
+      "grad_norm": 0.796875,
       "learning_rate": 0.0001929907846794187,
-      "loss": 0.5068,
+      "loss": 0.5193,
       "step": 221
     },
     {
       "epoch": 1.2542372881355932,
-      "grad_norm": 0.7734375,
+      "grad_norm": 0.828125,
       "learning_rate": 0.00019292498574468748,
-      "loss": 0.4771,
+      "loss": 0.54,
       "step": 222
     },
     {
       "epoch": 1.2598870056497176,
-      "grad_norm": 0.8359375,
+      "grad_norm": 0.765625,
       "learning_rate": 0.00019285889073150688,
-      "loss": 0.4658,
+      "loss": 0.5294,
       "step": 223
     },
     {
       "epoch": 1.2655367231638417,
-      "grad_norm": 0.87109375,
+      "grad_norm": 0.8828125,
       "learning_rate": 0.00019279249985046948,
-      "loss": 0.4557,
+      "loss": 0.3968,
       "step": 224
     },
     {
       "epoch": 1.271186440677966,
-      "grad_norm": 0.94921875,
+      "grad_norm": 0.921875,
       "learning_rate": 0.00019272581331311047,
-      "loss": 0.4804,
+      "loss": 0.6043,
       "step": 225
     },
     {
       "epoch": 1.271186440677966,
-      "eval_loss": 0.6073141098022461,
-      "eval_runtime": 5.8072,
-      "eval_samples_per_second": 25.658,
-      "eval_steps_per_second": 3.272,
+      "eval_loss": 0.6347160935401917,
+      "eval_runtime": 5.764,
+      "eval_samples_per_second": 26.024,
+      "eval_steps_per_second": 3.296,
       "step": 225
     },
     {
       "epoch": 1.2768361581920904,
-      "grad_norm": 0.8984375,
+      "grad_norm": 0.90234375,
       "learning_rate": 0.00019265883133190713,
-      "loss": 0.453,
+      "loss": 0.4236,
       "step": 226
     },
     {
       "epoch": 1.2824858757062148,
-      "grad_norm": 1.171875,
+      "grad_norm": 1.0703125,
       "learning_rate": 0.0001925915541202781,
-      "loss": 0.5031,
+      "loss": 0.4837,
       "step": 227
     },
     {
       "epoch": 1.288135593220339,
-      "grad_norm": 0.72265625,
+      "grad_norm": 0.875,
       "learning_rate": 0.00019252398189258262,
-      "loss": 0.4964,
+      "loss": 0.5258,
       "step": 228
     },
     {
       "epoch": 1.2937853107344632,
-      "grad_norm": 0.7421875,
+      "grad_norm": 0.67578125,
       "learning_rate": 0.00019245611486411995,
-      "loss": 0.4322,
+      "loss": 0.4926,
       "step": 229
     },
     {
       "epoch": 1.2994350282485876,
-      "grad_norm": 0.80078125,
+      "grad_norm": 0.76171875,
       "learning_rate": 0.0001923879532511287,
-      "loss": 0.4621,
+      "loss": 0.557,
       "step": 230
     },
     {
       "epoch": 1.305084745762712,
-      "grad_norm": 0.91015625,
+      "grad_norm": 0.87109375,
       "learning_rate": 0.00019231949727078592,
-      "loss": 0.6447,
+      "loss": 0.5387,
       "step": 231
     },
     {
       "epoch": 1.310734463276836,
-      "grad_norm": 0.7578125,
+      "grad_norm": 0.70703125,
       "learning_rate": 0.00019225074714120677,
-      "loss": 0.3503,
+      "loss": 0.5381,
       "step": 232
     },
     {
       "epoch": 1.3163841807909604,
-      "grad_norm": 0.6953125,
+      "grad_norm": 0.72265625,
       "learning_rate": 0.00019218170308144355,
-      "loss": 0.3809,
+      "loss": 0.4008,
       "step": 233
     },
     {
       "epoch": 1.3220338983050848,
-      "grad_norm": 0.84765625,
+      "grad_norm": 0.76953125,
       "learning_rate": 0.000192112365311485,
-      "loss": 0.4375,
+      "loss": 0.3518,
       "step": 234
     },
     {
       "epoch": 1.327683615819209,
-      "grad_norm": 0.9296875,
+      "grad_norm": 0.765625,
       "learning_rate": 0.00019204273405225587,
-      "loss": 0.4247,
+      "loss": 0.4325,
       "step": 235
     },
     {
       "epoch": 1.3333333333333333,
-      "grad_norm": 0.7890625,
+      "grad_norm": 0.84375,
       "learning_rate": 0.00019197280952561586,
-      "loss": 0.4423,
+      "loss": 0.4445,
       "step": 236
     },
     {
       "epoch": 1.3389830508474576,
-      "grad_norm": 0.99609375,
+      "grad_norm": 0.9375,
       "learning_rate": 0.0001919025919543592,
-      "loss": 0.5912,
+      "loss": 0.4403,
       "step": 237
     },
     {
       "epoch": 1.344632768361582,
-      "grad_norm": 0.796875,
+      "grad_norm": 0.78515625,
       "learning_rate": 0.0001918320815622137,
-      "loss": 0.3741,
+      "loss": 0.4219,
       "step": 238
     },
     {
       "epoch": 1.3502824858757063,
-      "grad_norm": 0.8359375,
+      "grad_norm": 0.9296875,
       "learning_rate": 0.00019176127857384025,
-      "loss": 0.383,
+      "loss": 0.4117,
       "step": 239
     },
     {
       "epoch": 1.3559322033898304,
-      "grad_norm": 0.87890625,
+      "grad_norm": 0.88671875,
       "learning_rate": 0.00019169018321483198,
-      "loss": 0.3871,
+      "loss": 0.5112,
       "step": 240
     },
     {
       "epoch": 1.3615819209039548,
-      "grad_norm": 0.76171875,
+      "grad_norm": 0.83203125,
       "learning_rate": 0.00019161879571171362,
-      "loss": 0.5199,
+      "loss": 0.4838,
       "step": 241
     },
     {
       "epoch": 1.3672316384180792,
-      "grad_norm": 0.59375,
+      "grad_norm": 0.734375,
       "learning_rate": 0.00019154711629194062,
-      "loss": 0.3418,
+      "loss": 0.469,
       "step": 242
     },
     {
       "epoch": 1.3728813559322033,
-      "grad_norm": 0.8671875,
+      "grad_norm": 0.7578125,
       "learning_rate": 0.00019147514518389865,
-      "loss": 0.4909,
+      "loss": 0.4528,
       "step": 243
     },
     {
       "epoch": 1.3785310734463276,
-      "grad_norm": 0.859375,
+      "grad_norm": 0.96875,
       "learning_rate": 0.00019140288261690276,
-      "loss": 0.3602,
+      "loss": 0.3523,
       "step": 244
     },
     {
       "epoch": 1.384180790960452,
-      "grad_norm": 0.8984375,
+      "grad_norm": 0.73046875,
       "learning_rate": 0.00019133032882119653,
-      "loss": 0.4764,
+      "loss": 0.5253,
       "step": 245
     },
     {
       "epoch": 1.3898305084745763,
-      "grad_norm": 0.9296875,
+      "grad_norm": 0.734375,
       "learning_rate": 0.0001912574840279516,
-      "loss": 0.4677,
+      "loss": 0.4438,
       "step": 246
     },
     {
       "epoch": 1.3954802259887007,
-      "grad_norm": 0.765625,
+      "grad_norm": 0.83984375,
       "learning_rate": 0.00019118434846926668,
-      "loss": 0.3683,
+      "loss": 0.3639,
       "step": 247
     },
     {
       "epoch": 1.4011299435028248,
-      "grad_norm": 0.76171875,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.00019111092237816704,
-      "loss": 0.5077,
+      "loss": 0.4627,
       "step": 248
     },
     {
       "epoch": 1.4067796610169492,
-      "grad_norm": 0.859375,
+      "grad_norm": 0.9375,
       "learning_rate": 0.00019103720598860347,
-      "loss": 0.4067,
+      "loss": 0.5523,
       "step": 249
     },
     {
       "epoch": 1.4124293785310735,
-      "grad_norm": 0.97265625,
+      "grad_norm": 0.7578125,
       "learning_rate": 0.00019096319953545185,
-      "loss": 0.6655,
+      "loss": 0.3542,
       "step": 250
     },
     {
       "epoch": 1.4180790960451977,
-      "grad_norm": 0.86328125,
+      "grad_norm": 1.078125,
       "learning_rate": 0.00019088890325451226,
-      "loss": 0.2742,
+      "loss": 0.4747,
       "step": 251
     },
     {
       "epoch": 1.423728813559322,
-      "grad_norm": 1.03125,
+      "grad_norm": 0.90625,
       "learning_rate": 0.00019081431738250814,
-      "loss": 0.5036,
+      "loss": 0.3684,
       "step": 252
     },
     {
       "epoch": 1.4293785310734464,
-      "grad_norm": 0.84375,
+      "grad_norm": 1.015625,
       "learning_rate": 0.00019073944215708576,
-      "loss": 0.2973,
+      "loss": 0.4241,
       "step": 253
     },
     {
       "epoch": 1.4350282485875705,
-      "grad_norm": 0.8359375,
+      "grad_norm": 0.828125,
       "learning_rate": 0.00019066427781681315,
-      "loss": 0.4488,
+      "loss": 0.4017,
       "step": 254
     },
     {
       "epoch": 1.4406779661016949,
-      "grad_norm": 0.83203125,
+      "grad_norm": 1.125,
       "learning_rate": 0.00019058882460117974,
-      "loss": 0.3439,
+      "loss": 0.5719,
       "step": 255
     },
     {
       "epoch": 1.4463276836158192,
-      "grad_norm": 0.88671875,
+      "grad_norm": 0.91796875,
       "learning_rate": 0.0001905130827505952,
-      "loss": 0.4107,
+      "loss": 0.4189,
       "step": 256
     },
     {
       "epoch": 1.4519774011299436,
-      "grad_norm": 0.93359375,
+      "grad_norm": 0.76953125,
       "learning_rate": 0.0001904370525063889,
-      "loss": 0.388,
+      "loss": 0.3854,
       "step": 257
     },
     {
       "epoch": 1.457627118644068,
-      "grad_norm": 0.76171875,
+      "grad_norm": 0.64453125,
       "learning_rate": 0.00019036073411080916,
-      "loss": 0.3808,
+      "loss": 0.3311,
       "step": 258
     },
     {
       "epoch": 1.463276836158192,
-      "grad_norm": 0.8046875,
+      "grad_norm": 0.82421875,
       "learning_rate": 0.00019028412780702233,
-      "loss": 0.3394,
+      "loss": 0.3012,
       "step": 259
     },
     {
       "epoch": 1.4689265536723164,
-      "grad_norm": 0.7109375,
+      "grad_norm": 0.75390625,
       "learning_rate": 0.00019020723383911215,
-      "loss": 0.3153,
+      "loss": 0.3146,
       "step": 260
     },
     {
       "epoch": 1.4745762711864407,
-      "grad_norm": 0.9375,
+      "grad_norm": 0.70703125,
       "learning_rate": 0.00019013005245207886,
-      "loss": 0.2748,
+      "loss": 0.4174,
       "step": 261
     },
     {
       "epoch": 1.4802259887005649,
-      "grad_norm": 0.85546875,
+      "grad_norm": 0.79296875,
       "learning_rate": 0.00019005258389183851,
-      "loss": 0.4371,
+      "loss": 0.411,
       "step": 262
     },
     {
       "epoch": 1.4858757062146892,
-      "grad_norm": 0.921875,
+      "grad_norm": 0.84765625,
       "learning_rate": 0.00018997482840522217,
-      "loss": 0.3296,
+      "loss": 0.2602,
       "step": 263
     },
     {
       "epoch": 1.4915254237288136,
-      "grad_norm": 1.0546875,
+      "grad_norm": 0.98828125,
       "learning_rate": 0.00018989678623997503,
-      "loss": 0.5094,
+      "loss": 0.4686,
       "step": 264
     },
     {
       "epoch": 1.497175141242938,
-      "grad_norm": 0.96484375,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.00018981845764475582,
-      "loss": 0.3585,
+      "loss": 0.3445,
       "step": 265
     },
     {
       "epoch": 1.5028248587570623,
-      "grad_norm": 1.3359375,
+      "grad_norm": 0.796875,
       "learning_rate": 0.00018973984286913584,
-      "loss": 0.4485,
+      "loss": 0.3938,
       "step": 266
     },
     {
       "epoch": 1.5084745762711864,
-      "grad_norm": 0.92578125,
+      "grad_norm": 0.7265625,
       "learning_rate": 0.0001896609421635982,
-      "loss": 0.4715,
+      "loss": 0.4191,
       "step": 267
     },
     {
       "epoch": 1.5141242937853108,
-      "grad_norm": 0.765625,
+      "grad_norm": 0.8203125,
       "learning_rate": 0.00018958175577953702,
-      "loss": 0.2965,
+      "loss": 0.3094,
       "step": 268
     },
     {
       "epoch": 1.5197740112994351,
-      "grad_norm": 0.75,
+      "grad_norm": 0.58203125,
       "learning_rate": 0.00018950228396925672,
-      "loss": 0.3528,
+      "loss": 0.2907,
       "step": 269
     },
     {
       "epoch": 1.5254237288135593,
-      "grad_norm": 0.7890625,
+      "grad_norm": 1.03125,
       "learning_rate": 0.00018942252698597113,
-      "loss": 0.3472,
+      "loss": 0.3525,
       "step": 270
     },
     {
       "epoch": 1.5254237288135593,
-      "eval_loss": 0.4267495274543762,
-      "eval_runtime": 5.9521,
-      "eval_samples_per_second": 25.033,
-      "eval_steps_per_second": 3.192,
+      "eval_loss": 0.4456014037132263,
+      "eval_runtime": 5.9612,
+      "eval_samples_per_second": 25.163,
+      "eval_steps_per_second": 3.187,
       "step": 270
     },
     {
       "epoch": 1.5310734463276836,
-      "grad_norm": 0.71875,
+      "grad_norm": 0.9140625,
       "learning_rate": 0.00018934248508380267,
-      "loss": 0.3361,
+      "loss": 0.3351,
       "step": 271
     },
     {
       "epoch": 1.536723163841808,
-      "grad_norm": 0.88671875,
+      "grad_norm": 0.921875,
       "learning_rate": 0.00018926215851778163,
-      "loss": 0.3199,
+      "loss": 0.3904,
       "step": 272
     },
     {
       "epoch": 1.542372881355932,
-      "grad_norm": 0.77734375,
+      "grad_norm": 0.9765625,
       "learning_rate": 0.00018918154754384516,
-      "loss": 0.2592,
+      "loss": 0.3872,
       "step": 273
     },
     {
       "epoch": 1.5480225988700564,
-      "grad_norm": 0.69921875,
+      "grad_norm": 1.0,
       "learning_rate": 0.0001891006524188368,
-      "loss": 0.2166,
+      "loss": 0.4266,
       "step": 274
     },
     {
       "epoch": 1.5536723163841808,
-      "grad_norm": 0.78515625,
+      "grad_norm": 0.640625,
       "learning_rate": 0.00018901947340050528,
-      "loss": 0.363,
+      "loss": 0.2674,
       "step": 275
     },
     {
       "epoch": 1.559322033898305,
-      "grad_norm": 1.3125,
+      "grad_norm": 0.87890625,
       "learning_rate": 0.00018893801074750397,
-      "loss": 0.3793,
+      "loss": 0.3672,
       "step": 276
     },
     {
       "epoch": 1.5649717514124295,
-      "grad_norm": 1.09375,
+      "grad_norm": 0.703125,
       "learning_rate": 0.0001888562647193899,
-      "loss": 0.4374,
+      "loss": 0.3551,
       "step": 277
     },
     {
       "epoch": 1.5706214689265536,
-      "grad_norm": 1.109375,
+      "grad_norm": 0.66015625,
       "learning_rate": 0.00018877423557662305,
-      "loss": 0.3223,
+      "loss": 0.3174,
       "step": 278
     },
     {
       "epoch": 1.576271186440678,
-      "grad_norm": 0.9453125,
+      "grad_norm": 0.81640625,
       "learning_rate": 0.00018869192358056543,
-      "loss": 0.4218,
+      "loss": 0.33,
       "step": 279
     },
     {
       "epoch": 1.5819209039548023,
-      "grad_norm": 0.97265625,
+      "grad_norm": 0.82421875,
       "learning_rate": 0.00018860932899348028,
-      "loss": 0.2986,
+      "loss": 0.3079,
       "step": 280
     },
     {
       "epoch": 1.5875706214689265,
-      "grad_norm": 0.7421875,
+      "grad_norm": 0.90625,
       "learning_rate": 0.00018852645207853122,
-      "loss": 0.3206,
+      "loss": 0.3886,
       "step": 281
     },
     {
       "epoch": 1.5932203389830508,
-      "grad_norm": 0.73046875,
+      "grad_norm": 0.87890625,
       "learning_rate": 0.00018844329309978145,
-      "loss": 0.3514,
+      "loss": 0.3028,
       "step": 282
     },
     {
       "epoch": 1.5988700564971752,
-      "grad_norm": 0.75,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.0001883598523221929,
-      "loss": 0.3074,
+      "loss": 0.3191,
       "step": 283
     },
     {
       "epoch": 1.6045197740112993,
-      "grad_norm": 0.99609375,
+      "grad_norm": 0.734375,
       "learning_rate": 0.00018827613001162532,
-      "loss": 0.35,
+      "loss": 0.2464,
       "step": 284
     },
     {
       "epoch": 1.6101694915254239,
-      "grad_norm": 0.9453125,
+      "grad_norm": 1.171875,
       "learning_rate": 0.0001881921264348355,
-      "loss": 0.5249,
+      "loss": 0.4712,
       "step": 285
     },
     {
       "epoch": 1.615819209039548,
-      "grad_norm": 0.78125,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.00018810784185947646,
-      "loss": 0.2587,
+      "loss": 0.2817,
       "step": 286
     },
     {
       "epoch": 1.6214689265536724,
-      "grad_norm": 0.859375,
+      "grad_norm": 0.8515625,
       "learning_rate": 0.0001880232765540965,
-      "loss": 0.2479,
+      "loss": 0.2373,
       "step": 287
     },
     {
       "epoch": 1.6271186440677967,
-      "grad_norm": 0.9375,
+      "grad_norm": 0.76171875,
       "learning_rate": 0.0001879384307881383,
-      "loss": 0.3061,
+      "loss": 0.3821,
       "step": 288
     },
     {
       "epoch": 1.6327683615819208,
-      "grad_norm": 0.9765625,
+      "grad_norm": 0.8203125,
       "learning_rate": 0.0001878533048319383,
-      "loss": 0.3466,
+      "loss": 0.2772,
       "step": 289
     },
     {
       "epoch": 1.6384180790960452,
-      "grad_norm": 0.84765625,
+      "grad_norm": 0.71875,
       "learning_rate": 0.00018776789895672558,
-      "loss": 0.2282,
+      "loss": 0.3451,
       "step": 290
     },
     {
       "epoch": 1.6440677966101696,
-      "grad_norm": 0.91796875,
+      "grad_norm": 0.7890625,
       "learning_rate": 0.0001876822134346211,
-      "loss": 0.4825,
+      "loss": 0.339,
       "step": 291
     },
     {
       "epoch": 1.6497175141242937,
-      "grad_norm": 0.8359375,
+      "grad_norm": 0.75390625,
       "learning_rate": 0.00018759624853863693,
-      "loss": 0.3023,
+      "loss": 0.2863,
       "step": 292
     },
     {
       "epoch": 1.655367231638418,
-      "grad_norm": 0.83203125,
+      "grad_norm": 0.8125,
       "learning_rate": 0.0001875100045426752,
-      "loss": 0.2739,
+      "loss": 0.2714,
       "step": 293
     },
     {
       "epoch": 1.6610169491525424,
-      "grad_norm": 0.71875,
+      "grad_norm": 0.8203125,
       "learning_rate": 0.00018742348172152726,
-      "loss": 0.2698,
+      "loss": 0.2924,
       "step": 294
     },
     {
       "epoch": 1.6666666666666665,
-      "grad_norm": 0.76953125,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.00018733668035087302,
-      "loss": 0.2553,
+      "loss": 0.3953,
       "step": 295
     },
     {
       "epoch": 1.672316384180791,
-      "grad_norm": 0.78125,
+      "grad_norm": 0.953125,
       "learning_rate": 0.00018724960070727972,
-      "loss": 0.3388,
+      "loss": 0.381,
       "step": 296
     },
     {
       "epoch": 1.6779661016949152,
-      "grad_norm": 0.83984375,
+      "grad_norm": 0.828125,
       "learning_rate": 0.0001871622430682014,
-      "loss": 0.2911,
+      "loss": 0.2733,
       "step": 297
     },
     {
       "epoch": 1.6836158192090396,
-      "grad_norm": 0.8203125,
+      "grad_norm": 0.7421875,
       "learning_rate": 0.00018707460771197774,
-      "loss": 0.3212,
+      "loss": 0.2464,
       "step": 298
     },
     {
       "epoch": 1.689265536723164,
-      "grad_norm": 0.72265625,
+      "grad_norm": 0.69921875,
       "learning_rate": 0.0001869866949178333,
-      "loss": 0.2602,
+      "loss": 0.2804,
       "step": 299
     },
     {
       "epoch": 1.694915254237288,
-      "grad_norm": 0.8828125,
+      "grad_norm": 0.58984375,
       "learning_rate": 0.00018689850496587674,
-      "loss": 0.2849,
+      "loss": 0.2216,
       "step": 300
     },
     {
       "epoch": 1.7005649717514124,
-      "grad_norm": 0.765625,
+      "grad_norm": 0.890625,
       "learning_rate": 0.00018681003813709955,
-      "loss": 0.2604,
+      "loss": 0.3101,
       "step": 301
     },
     {
       "epoch": 1.7062146892655368,
-      "grad_norm": 1.078125,
+      "grad_norm": 0.91796875,
       "learning_rate": 0.00018672129471337568,
-      "loss": 0.3604,
+      "loss": 0.2657,
       "step": 302
     },
     {
       "epoch": 1.711864406779661,
-      "grad_norm": 0.859375,
+      "grad_norm": 0.8828125,
       "learning_rate": 0.0001866322749774602,
-      "loss": 0.181,
+      "loss": 0.222,
       "step": 303
     },
     {
       "epoch": 1.7175141242937855,
-      "grad_norm": 0.8828125,
+      "grad_norm": 0.78515625,
       "learning_rate": 0.00018654297921298863,
-      "loss": 0.4089,
+      "loss": 0.2698,
       "step": 304
     },
     {
       "epoch": 1.7231638418079096,
-      "grad_norm": 0.875,
+      "grad_norm": 0.9375,
       "learning_rate": 0.00018645340770447595,
-      "loss": 0.247,
+      "loss": 0.2853,
       "step": 305
     },
     {
       "epoch": 1.7288135593220337,
-      "grad_norm": 1.015625,
+      "grad_norm": 0.69140625,
       "learning_rate": 0.0001863635607373157,
-      "loss": 0.3652,
+      "loss": 0.3367,
       "step": 306
     },
     {
       "epoch": 1.7344632768361583,
-      "grad_norm": 1.0078125,
+      "grad_norm": 0.62890625,
       "learning_rate": 0.0001862734385977792,
-      "loss": 0.273,
+      "loss": 0.321,
       "step": 307
     },
     {
       "epoch": 1.7401129943502824,
-      "grad_norm": 0.8359375,
+      "grad_norm": 0.6796875,
       "learning_rate": 0.00018618304157301435,
-      "loss": 0.3634,
+      "loss": 0.243,
       "step": 308
     },
     {
       "epoch": 1.7457627118644068,
-      "grad_norm": 0.90625,
+      "grad_norm": 0.6875,
       "learning_rate": 0.00018609236995104507,
-      "loss": 0.2714,
+      "loss": 0.3415,
       "step": 309
     },
     {
       "epoch": 1.7514124293785311,
-      "grad_norm": 0.828125,
+      "grad_norm": 0.859375,
       "learning_rate": 0.00018600142402077006,
-      "loss": 0.325,
+      "loss": 0.269,
       "step": 310
     },
     {
       "epoch": 1.7570621468926553,
-      "grad_norm": 0.625,
+      "grad_norm": 0.95703125,
       "learning_rate": 0.00018591020407196212,
-      "loss": 0.248,
+      "loss": 0.2751,
       "step": 311
     },
     {
       "epoch": 1.7627118644067796,
-      "grad_norm": 0.81640625,
+      "grad_norm": 0.96484375,
       "learning_rate": 0.00018581871039526705,
-      "loss": 0.2274,
+      "loss": 0.3347,
       "step": 312
     },
     {
       "epoch": 1.768361581920904,
-      "grad_norm": 0.8671875,
+      "grad_norm": 0.79296875,
       "learning_rate": 0.00018572694328220293,
-      "loss": 0.3641,
+      "loss": 0.2166,
       "step": 313
     },
     {
       "epoch": 1.774011299435028,
-      "grad_norm": 0.76953125,
+      "grad_norm": 0.80078125,
       "learning_rate": 0.0001856349030251589,
-      "loss": 0.2969,
+      "loss": 0.2862,
       "step": 314
     },
     {
       "epoch": 1.7796610169491527,
-      "grad_norm": 0.78125,
+      "grad_norm": 0.61328125,
       "learning_rate": 0.00018554258991739452,
-      "loss": 0.2733,
+      "loss": 0.1879,
       "step": 315
     },
     {
       "epoch": 1.7796610169491527,
-      "eval_loss": 0.28075358271598816,
-      "eval_runtime": 5.698,
-      "eval_samples_per_second": 26.149,
-      "eval_steps_per_second": 3.334,
+      "eval_loss": 0.291787713766098,
+      "eval_runtime": 5.8866,
+      "eval_samples_per_second": 25.482,
+      "eval_steps_per_second": 3.228,
       "step": 315
     },
     {
       "epoch": 1.7853107344632768,
-      "grad_norm": 0.828125,
+      "grad_norm": 0.81640625,
       "learning_rate": 0.00018545000425303866,
-      "loss": 0.2434,
+      "loss": 0.234,
       "step": 316
     },
     {
       "epoch": 1.7909604519774012,
-      "grad_norm": 0.7734375,
+      "grad_norm": 0.96875,
       "learning_rate": 0.00018535714632708866,
-      "loss": 0.3304,
+      "loss": 0.2441,
       "step": 317
     },
     {
       "epoch": 1.7966101694915255,
-      "grad_norm": 0.80078125,
+      "grad_norm": 0.859375,
       "learning_rate": 0.00018526401643540922,
-      "loss": 0.3409,
+      "loss": 0.2339,
       "step": 318
     },
     {
       "epoch": 1.8022598870056497,
-      "grad_norm": 0.91796875,
+      "grad_norm": 0.76171875,
       "learning_rate": 0.00018517061487473174,
-      "loss": 0.2544,
+      "loss": 0.2795,
       "step": 319
     },
     {
       "epoch": 1.807909604519774,
-      "grad_norm": 0.828125,
+      "grad_norm": 0.86328125,
       "learning_rate": 0.0001850769419426531,
-      "loss": 0.2051,
+      "loss": 0.2469,
       "step": 320
     },
     {
       "epoch": 1.8135593220338984,
-      "grad_norm": 0.734375,
+      "grad_norm": 0.9921875,
       "learning_rate": 0.00018498299793763486,
-      "loss": 0.3484,
+      "loss": 0.2697,
       "step": 321
     },
     {
       "epoch": 1.8192090395480225,
-      "grad_norm": 0.796875,
+      "grad_norm": 0.91796875,
       "learning_rate": 0.00018488878315900227,
-      "loss": 0.3038,
+      "loss": 0.2888,
       "step": 322
     },
     {
       "epoch": 1.8248587570621468,
-      "grad_norm": 0.64453125,
+      "grad_norm": 0.80859375,
       "learning_rate": 0.00018479429790694337,
-      "loss": 0.2392,
+      "loss": 0.2804,
       "step": 323
     },
     {
       "epoch": 1.8305084745762712,
-      "grad_norm": 0.5859375,
+      "grad_norm": 0.7734375,
       "learning_rate": 0.0001846995424825079,
-      "loss": 0.1665,
+      "loss": 0.236,
       "step": 324
     },
     {
       "epoch": 1.8361581920903953,
-      "grad_norm": 1.09375,
+      "grad_norm": 1.265625,
       "learning_rate": 0.0001846045171876065,
-      "loss": 0.3119,
+      "loss": 0.2808,
       "step": 325
     },
     {
       "epoch": 1.84180790960452,
-      "grad_norm": 0.94140625,
+      "grad_norm": 1.203125,
       "learning_rate": 0.00018450922232500966,
-      "loss": 0.3059,
+      "loss": 0.2501,
       "step": 326
     },
     {
       "epoch": 1.847457627118644,
-      "grad_norm": 1.0078125,
+      "grad_norm": 0.8203125,
       "learning_rate": 0.00018441365819834668,
-      "loss": 0.2562,
+      "loss": 0.2069,
       "step": 327
     },
     {
       "epoch": 1.8531073446327684,
-      "grad_norm": 0.97265625,
+      "grad_norm": 0.91796875,
       "learning_rate": 0.00018431782511210495,
-      "loss": 0.2682,
+      "loss": 0.2639,
       "step": 328
     },
     {
       "epoch": 1.8587570621468927,
-      "grad_norm": 0.92578125,
+      "grad_norm": 0.7421875,
       "learning_rate": 0.00018422172337162867,
-      "loss": 0.2648,
+      "loss": 0.2253,
       "step": 329
     },
     {
       "epoch": 1.8644067796610169,
-      "grad_norm": 0.8984375,
+      "grad_norm": 0.63671875,
       "learning_rate": 0.00018412535328311814,
-      "loss": 0.1985,
+      "loss": 0.2423,
       "step": 330
     },
     {
       "epoch": 1.8700564971751412,
-      "grad_norm": 0.85546875,
+      "grad_norm": 0.875,
       "learning_rate": 0.00018402871515362857,
-      "loss": 0.2098,
+      "loss": 0.2312,
       "step": 331
     },
     {
       "epoch": 1.8757062146892656,
-      "grad_norm": 0.78125,
+      "grad_norm": 0.796875,
       "learning_rate": 0.00018393180929106935,
-      "loss": 0.179,
+      "loss": 0.2348,
       "step": 332
     },
     {
       "epoch": 1.8813559322033897,
-      "grad_norm": 0.89453125,
+      "grad_norm": 0.9140625,
       "learning_rate": 0.0001838346360042028,
-      "loss": 0.2455,
+      "loss": 0.2329,
       "step": 333
     },
     {
       "epoch": 1.8870056497175143,
-      "grad_norm": 0.66796875,
+      "grad_norm": 0.80859375,
       "learning_rate": 0.00018373719560264327,
-      "loss": 0.2149,
+      "loss": 0.1895,
       "step": 334
     },
     {
       "epoch": 1.8926553672316384,
-      "grad_norm": 0.80078125,
+      "grad_norm": 0.97265625,
       "learning_rate": 0.00018363948839685636,
-      "loss": 0.2022,
+      "loss": 0.1865,
       "step": 335
     },
     {
       "epoch": 1.8983050847457628,
-      "grad_norm": 0.8984375,
+      "grad_norm": 0.8125,
       "learning_rate": 0.00018354151469815767,
-      "loss": 0.2229,
+      "loss": 0.356,
       "step": 336
     },
     {
       "epoch": 1.9039548022598871,
-      "grad_norm": 0.68359375,
+      "grad_norm": 0.85546875,
       "learning_rate": 0.00018344327481871185,
-      "loss": 0.2049,
+      "loss": 0.2414,
       "step": 337
     },
     {
       "epoch": 1.9096045197740112,
-      "grad_norm": 0.6796875,
+      "grad_norm": 0.53515625,
       "learning_rate": 0.00018334476907153177,
-      "loss": 0.2663,
+      "loss": 0.2089,
       "step": 338
     },
     {
       "epoch": 1.9152542372881356,
-      "grad_norm": 1.0,
+      "grad_norm": 0.8203125,
       "learning_rate": 0.00018324599777047734,
-      "loss": 0.2295,
+      "loss": 0.2534,
       "step": 339
     },
     {
       "epoch": 1.92090395480226,
-      "grad_norm": 0.73046875,
+      "grad_norm": 1.03125,
       "learning_rate": 0.00018314696123025454,
-      "loss": 0.2068,
+      "loss": 0.2968,
       "step": 340
     },
     {
       "epoch": 1.926553672316384,
-      "grad_norm": 0.87109375,
+      "grad_norm": 0.65625,
       "learning_rate": 0.00018304765976641455,
-      "loss": 0.1881,
+      "loss": 0.1958,
       "step": 341
     },
     {
       "epoch": 1.9322033898305084,
-      "grad_norm": 0.99609375,
+      "grad_norm": 0.9765625,
       "learning_rate": 0.00018294809369535263,
-      "loss": 0.2855,
+      "loss": 0.2487,
       "step": 342
     },
     {
       "epoch": 1.9378531073446328,
-      "grad_norm": 0.94921875,
+      "grad_norm": 0.8359375,
       "learning_rate": 0.0001828482633343071,
-      "loss": 0.1964,
+      "loss": 0.1908,
       "step": 343
     },
     {
       "epoch": 1.943502824858757,
-      "grad_norm": 0.9453125,
+      "grad_norm": 0.8828125,
       "learning_rate": 0.0001827481690013584,
-      "loss": 0.2707,
+      "loss": 0.2017,
       "step": 344
     },
     {
       "epoch": 1.9491525423728815,
-      "grad_norm": 0.86328125,
+      "grad_norm": 0.9609375,
       "learning_rate": 0.000182647811015428,
-      "loss": 0.2194,
+      "loss": 0.2562,
       "step": 345
     },
     {
       "epoch": 1.9548022598870056,
-      "grad_norm": 0.734375,
+      "grad_norm": 0.68359375,
       "learning_rate": 0.0001825471896962774,
-      "loss": 0.2327,
+      "loss": 0.2456,
       "step": 346
     },
     {
       "epoch": 1.96045197740113,
-      "grad_norm": 0.89453125,
+      "grad_norm": 0.86328125,
       "learning_rate": 0.00018244630536450726,
-      "loss": 0.2434,
+      "loss": 0.1977,
       "step": 347
     },
     {
       "epoch": 1.9661016949152543,
-      "grad_norm": 0.84375,
+      "grad_norm": 0.69921875,
       "learning_rate": 0.0001823451583415561,
-      "loss": 0.1974,
+      "loss": 0.25,
       "step": 348
     },
     {
       "epoch": 1.9717514124293785,
-      "grad_norm": 0.9296875,
+      "grad_norm": 0.6875,
       "learning_rate": 0.00018224374894969958,
-      "loss": 0.1389,
+      "loss": 0.22,
       "step": 349
     },
     {
       "epoch": 1.9774011299435028,
-      "grad_norm": 0.59765625,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.00018214207751204918,
-      "loss": 0.2124,
+      "loss": 0.1882,
       "step": 350
     },
     {
       "epoch": 1.9830508474576272,
-      "grad_norm": 0.9375,
+      "grad_norm": 0.70703125,
       "learning_rate": 0.00018204014435255135,
-      "loss": 0.1816,
+      "loss": 0.2139,
       "step": 351
     },
     {
       "epoch": 1.9887005649717513,
-      "grad_norm": 0.8359375,
+      "grad_norm": 0.9375,
       "learning_rate": 0.00018193794979598656,
-      "loss": 0.2549,
+      "loss": 0.138,
       "step": 352
     },
     {
       "epoch": 1.9943502824858759,
-      "grad_norm": 0.77734375,
+      "grad_norm": 0.828125,
       "learning_rate": 0.000181835494167968,
-      "loss": 0.1852,
+      "loss": 0.1912,
       "step": 353
     },
     {
       "epoch": 2.0,
-      "grad_norm": 0.76953125,
+      "grad_norm": 0.875,
       "learning_rate": 0.0001817327777949407,
-      "loss": 0.1247,
+      "loss": 0.1552,
       "step": 354
     },
     {
       "epoch": 2.005649717514124,
-      "grad_norm": 0.6328125,
+      "grad_norm": 0.64453125,
       "learning_rate": 0.0001816298010041806,
-      "loss": 0.1333,
+      "loss": 0.1725,
       "step": 355
     },
     {
       "epoch": 2.0112994350282487,
-      "grad_norm": 0.83984375,
+      "grad_norm": 0.72265625,
       "learning_rate": 0.00018152656412379322,
-      "loss": 0.1228,
+      "loss": 0.1274,
       "step": 356
     },
     {
       "epoch": 2.016949152542373,
-      "grad_norm": 0.6328125,
+      "grad_norm": 0.68359375,
       "learning_rate": 0.00018142306748271291,
-      "loss": 0.1349,
+      "loss": 0.1981,
       "step": 357
     },
     {
       "epoch": 2.022598870056497,
-      "grad_norm": 0.9140625,
+      "grad_norm": 0.81640625,
       "learning_rate": 0.00018131931141070163,
-      "loss": 0.1186,
+      "loss": 0.1248,
       "step": 358
     },
     {
       "epoch": 2.0282485875706215,
-      "grad_norm": 0.74609375,
+      "grad_norm": 0.765625,
       "learning_rate": 0.00018121529623834793,
-      "loss": 0.0987,
+      "loss": 0.1229,
       "step": 359
     },
     {
       "epoch": 2.0338983050847457,
-      "grad_norm": 0.70703125,
+      "grad_norm": 0.77734375,
       "learning_rate": 0.0001811110222970659,
-      "loss": 0.1484,
+      "loss": 0.1367,
       "step": 360
     },
     {
       "epoch": 2.0338983050847457,
-      "eval_loss": 0.17419962584972382,
-      "eval_runtime": 5.7761,
-      "eval_samples_per_second": 25.796,
-      "eval_steps_per_second": 3.289,
+      "eval_loss": 0.16075760126113892,
+      "eval_runtime": 5.796,
+      "eval_samples_per_second": 25.88,
+      "eval_steps_per_second": 3.278,
       "step": 360
     },
     {
       "epoch": 2.0395480225988702,
-      "grad_norm": 0.72265625,
+      "grad_norm": 0.51171875,
       "learning_rate": 0.0001810064899190942,
-      "loss": 0.1561,
+      "loss": 0.154,
       "step": 361
     },
     {
       "epoch": 2.0451977401129944,
-      "grad_norm": 0.80078125,
+      "grad_norm": 0.7734375,
       "learning_rate": 0.00018090169943749476,
-      "loss": 0.129,
+      "loss": 0.1356,
       "step": 362
     },
     {
       "epoch": 2.0508474576271185,
-      "grad_norm": 0.60546875,
+      "grad_norm": 0.5625,
       "learning_rate": 0.00018079665118615203,
-      "loss": 0.1551,
+      "loss": 0.1196,
       "step": 363
     },
     {
       "epoch": 2.056497175141243,
-      "grad_norm": 0.640625,
+      "grad_norm": 0.734375,
       "learning_rate": 0.00018069134549977172,
-      "loss": 0.143,
+      "loss": 0.162,
       "step": 364
     },
     {
       "epoch": 2.062146892655367,
-      "grad_norm": 0.703125,
+      "grad_norm": 0.56640625,
       "learning_rate": 0.0001805857827138798,
-      "loss": 0.1828,
+      "loss": 0.1837,
       "step": 365
     },
     {
       "epoch": 2.0677966101694913,
-      "grad_norm": 0.80078125,
+      "grad_norm": 0.8671875,
       "learning_rate": 0.00018047996316482135,
-      "loss": 0.131,
+      "loss": 0.1639,
       "step": 366
     },
     {
       "epoch": 2.073446327683616,
-      "grad_norm": 0.73828125,
+      "grad_norm": 0.79296875,
       "learning_rate": 0.00018037388718975965,
-      "loss": 0.1576,
+      "loss": 0.0932,
       "step": 367
     },
     {
       "epoch": 2.07909604519774,
-      "grad_norm": 0.6484375,
+      "grad_norm": 0.671875,
       "learning_rate": 0.0001802675551266749,
-      "loss": 0.1625,
+      "loss": 0.131,
       "step": 368
     },
     {
       "epoch": 2.084745762711864,
-      "grad_norm": 0.62890625,
+      "grad_norm": 0.6640625,
       "learning_rate": 0.00018016096731436333,
-      "loss": 0.1069,
+      "loss": 0.2056,
       "step": 369
     },
     {
       "epoch": 2.0903954802259888,
-      "grad_norm": 0.640625,
+      "grad_norm": 0.54296875,
       "learning_rate": 0.00018005412409243606,
-      "loss": 0.1649,
+      "loss": 0.1038,
       "step": 370
     },
     {
       "epoch": 2.096045197740113,
-      "grad_norm": 0.72265625,
+      "grad_norm": 0.6796875,
       "learning_rate": 0.00017994702580131789,
-      "loss": 0.1389,
+      "loss": 0.1493,
       "step": 371
     },
     {
       "epoch": 2.1016949152542375,
-      "grad_norm": 0.71875,
+      "grad_norm": 0.4609375,
       "learning_rate": 0.00017983967278224636,
-      "loss": 0.0925,
+      "loss": 0.1628,
       "step": 372
     },
     {
       "epoch": 2.1073446327683616,
-      "grad_norm": 0.65234375,
+      "grad_norm": 0.828125,
       "learning_rate": 0.00017973206537727073,
-      "loss": 0.137,
+      "loss": 0.1577,
       "step": 373
     },
     {
       "epoch": 2.1129943502824857,
-      "grad_norm": 0.59375,
+      "grad_norm": 0.91796875,
       "learning_rate": 0.00017962420392925066,
-      "loss": 0.1588,
+      "loss": 0.1291,
       "step": 374
     },
     {
       "epoch": 2.1186440677966103,
-      "grad_norm": 0.609375,
+      "grad_norm": 0.69921875,
       "learning_rate": 0.0001795160887818553,
-      "loss": 0.151,
+      "loss": 0.0944,
       "step": 375
     },
     {
       "epoch": 2.1242937853107344,
-      "grad_norm": 0.765625,
+      "grad_norm": 0.7109375,
       "learning_rate": 0.00017940772027956218,
-      "loss": 0.0962,
+      "loss": 0.1881,
       "step": 376
     },
     {
       "epoch": 2.1299435028248586,
-      "grad_norm": 0.671875,
+      "grad_norm": 0.6875,
       "learning_rate": 0.00017929909876765592,
-      "loss": 0.1371,
+      "loss": 0.0952,
       "step": 377
     },
     {
       "epoch": 2.135593220338983,
-      "grad_norm": 0.66015625,
+      "grad_norm": 0.58984375,
       "learning_rate": 0.00017919022459222752,
-      "loss": 0.1492,
+      "loss": 0.2165,
       "step": 378
     },
     {
       "epoch": 2.1412429378531073,
-      "grad_norm": 0.66796875,
+      "grad_norm": 0.63671875,
       "learning_rate": 0.0001790810981001728,
-      "loss": 0.1586,
+      "loss": 0.1581,
       "step": 379
     },
     {
       "epoch": 2.146892655367232,
-      "grad_norm": 0.61328125,
+      "grad_norm": 0.6484375,
       "learning_rate": 0.0001789717196391916,
-      "loss": 0.1244,
+      "loss": 0.1564,
       "step": 380
     },
     {
       "epoch": 2.152542372881356,
-      "grad_norm": 0.85546875,
+      "grad_norm": 0.5390625,
       "learning_rate": 0.0001788620895577866,
-      "loss": 0.1086,
+      "loss": 0.1416,
       "step": 381
     },
     {
       "epoch": 2.15819209039548,
-      "grad_norm": 0.7578125,
+      "grad_norm": 0.66015625,
       "learning_rate": 0.00017875220820526218,
-      "loss": 0.1684,
+      "loss": 0.1663,
       "step": 382
     },
     {
       "epoch": 2.1638418079096047,
-      "grad_norm": 0.6171875,
+      "grad_norm": 0.796875,
       "learning_rate": 0.00017864207593172333,
-      "loss": 0.1451,
+      "loss": 0.1063,
       "step": 383
     },
     {
       "epoch": 2.169491525423729,
-      "grad_norm": 0.61328125,
+      "grad_norm": 0.5546875,
       "learning_rate": 0.00017853169308807448,
-      "loss": 0.1636,
+      "loss": 0.105,
       "step": 384
     },
     {
       "epoch": 2.175141242937853,
-      "grad_norm": 0.84375,
+      "grad_norm": 0.546875,
       "learning_rate": 0.00017842106002601856,
-      "loss": 0.1451,
+      "loss": 0.0333,
       "step": 385
     },
     {
       "epoch": 2.1807909604519775,
-      "grad_norm": 0.8828125,
+      "grad_norm": 0.65234375,
       "learning_rate": 0.00017831017709805556,
-      "loss": 0.1302,
+      "loss": 0.0905,
       "step": 386
     },
     {
       "epoch": 2.1864406779661016,
-      "grad_norm": 0.59765625,
+      "grad_norm": 0.53515625,
       "learning_rate": 0.0001781990446574817,
-      "loss": 0.2095,
+      "loss": 0.152,
       "step": 387
     },
     {
       "epoch": 2.1920903954802258,
-      "grad_norm": 0.80859375,
+      "grad_norm": 0.59375,
       "learning_rate": 0.00017808766305838826,
-      "loss": 0.1508,
+      "loss": 0.2006,
       "step": 388
     },
     {
       "epoch": 2.1977401129943503,
-      "grad_norm": 0.69140625,
+      "grad_norm": 0.80859375,
       "learning_rate": 0.00017797603265566026,
-      "loss": 0.1078,
+      "loss": 0.1278,
       "step": 389
     },
     {
       "epoch": 2.2033898305084745,
-      "grad_norm": 0.6484375,
+      "grad_norm": 0.64453125,
       "learning_rate": 0.00017786415380497553,
-      "loss": 0.1374,
+      "loss": 0.1234,
       "step": 390
     },
     {
       "epoch": 2.209039548022599,
-      "grad_norm": 0.65625,
+      "grad_norm": 0.7734375,
       "learning_rate": 0.0001777520268628035,
-      "loss": 0.0892,
+      "loss": 0.1115,
       "step": 391
     },
     {
       "epoch": 2.214689265536723,
-      "grad_norm": 0.75,
+      "grad_norm": 0.625,
       "learning_rate": 0.0001776396521864041,
-      "loss": 0.1491,
+      "loss": 0.1289,
       "step": 392
     },
     {
       "epoch": 2.2203389830508473,
-      "grad_norm": 0.578125,
+      "grad_norm": 0.63671875,
       "learning_rate": 0.00017752703013382656,
-      "loss": 0.1588,
+      "loss": 0.1328,
       "step": 393
     },
     {
       "epoch": 2.225988700564972,
-      "grad_norm": 0.55078125,
+      "grad_norm": 0.546875,
       "learning_rate": 0.00017741416106390826,
-      "loss": 0.1316,
+      "loss": 0.183,
       "step": 394
     },
     {
       "epoch": 2.231638418079096,
-      "grad_norm": 0.5546875,
+      "grad_norm": 0.56640625,
       "learning_rate": 0.0001773010453362737,
-      "loss": 0.1508,
+      "loss": 0.1441,
       "step": 395
     },
     {
       "epoch": 2.23728813559322,
-      "grad_norm": 0.71875,
+      "grad_norm": 0.7265625,
       "learning_rate": 0.00017718768331133328,
-      "loss": 0.1047,
+      "loss": 0.1915,
       "step": 396
     },
     {
       "epoch": 2.2429378531073447,
-      "grad_norm": 0.734375,
+      "grad_norm": 0.4375,
       "learning_rate": 0.0001770740753502821,
-      "loss": 0.1078,
+      "loss": 0.1724,
       "step": 397
     },
     {
       "epoch": 2.248587570621469,
-      "grad_norm": 0.625,
+      "grad_norm": 0.60546875,
       "learning_rate": 0.00017696022181509894,
-      "loss": 0.1303,
+      "loss": 0.1314,
       "step": 398
     },
     {
       "epoch": 2.2542372881355934,
-      "grad_norm": 0.91015625,
+      "grad_norm": 0.5625,
       "learning_rate": 0.00017684612306854495,
-      "loss": 0.0802,
+      "loss": 0.2158,
       "step": 399
     },
     {
       "epoch": 2.2598870056497176,
-      "grad_norm": 0.8359375,
+      "grad_norm": 0.6953125,
       "learning_rate": 0.00017673177947416258,
-      "loss": 0.1601,
+      "loss": 0.1099,
       "step": 400
     },
     {
       "epoch": 2.2655367231638417,
-      "grad_norm": 0.8046875,
+      "grad_norm": 0.625,
       "learning_rate": 0.00017661719139627451,
-      "loss": 0.1319,
+      "loss": 0.128,
       "step": 401
     },
     {
       "epoch": 2.2711864406779663,
-      "grad_norm": 0.5,
+      "grad_norm": 0.50390625,
       "learning_rate": 0.00017650235919998232,
-      "loss": 0.1416,
+      "loss": 0.1006,
       "step": 402
     },
     {
       "epoch": 2.2768361581920904,
-      "grad_norm": 0.58984375,
+      "grad_norm": 0.7265625,
       "learning_rate": 0.0001763872832511654,
-      "loss": 0.1187,
+      "loss": 0.1532,
       "step": 403
     },
     {
       "epoch": 2.2824858757062145,
-      "grad_norm": 0.5625,
+      "grad_norm": 0.443359375,
       "learning_rate": 0.00017627196391647982,
-      "loss": 0.1749,
+      "loss": 0.1208,
       "step": 404
     },
     {
       "epoch": 2.288135593220339,
-      "grad_norm": 0.55859375,
+      "grad_norm": 0.47265625,
       "learning_rate": 0.00017615640156335712,
-      "loss": 0.2064,
+      "loss": 0.1627,
       "step": 405
     },
     {
       "epoch": 2.288135593220339,
-      "eval_loss": 0.1260523945093155,
-      "eval_runtime": 5.7301,
-      "eval_samples_per_second": 26.003,
-      "eval_steps_per_second": 3.316,
+      "eval_loss": 0.10976649075746536,
+      "eval_runtime": 5.7508,
+      "eval_samples_per_second": 26.083,
+      "eval_steps_per_second": 3.304,
       "step": 405
     },
     {
       "epoch": 2.2937853107344632,
-      "grad_norm": 0.765625,
+      "grad_norm": 0.61328125,
       "learning_rate": 0.0001760405965600031,
-      "loss": 0.1638,
+      "loss": 0.1549,
       "step": 406
     },
     {
       "epoch": 2.2994350282485874,
-      "grad_norm": 0.66015625,
+      "grad_norm": 0.78125,
       "learning_rate": 0.00017592454927539678,
-      "loss": 0.1521,
+      "loss": 0.1451,
       "step": 407
     },
     {
       "epoch": 2.305084745762712,
-      "grad_norm": 0.466796875,
+      "grad_norm": 0.58203125,
       "learning_rate": 0.00017580826007928903,
-      "loss": 0.1506,
+      "loss": 0.1392,
       "step": 408
     },
     {
       "epoch": 2.310734463276836,
-      "grad_norm": 0.5625,
+      "grad_norm": 0.68359375,
       "learning_rate": 0.0001756917293422016,
-      "loss": 0.1553,
+      "loss": 0.1466,
       "step": 409
     },
     {
       "epoch": 2.3163841807909606,
-      "grad_norm": 0.482421875,
+      "grad_norm": 0.71484375,
       "learning_rate": 0.00017557495743542585,
-      "loss": 0.1346,
+      "loss": 0.1733,
       "step": 410
     },
     {
       "epoch": 2.3220338983050848,
-      "grad_norm": 0.67578125,
+      "grad_norm": 0.546875,
       "learning_rate": 0.00017545794473102142,
-      "loss": 0.1392,
+      "loss": 0.1663,
       "step": 411
     },
     {
       "epoch": 2.327683615819209,
-      "grad_norm": 0.6171875,
+      "grad_norm": 0.6953125,
       "learning_rate": 0.00017534069160181533,
-      "loss": 0.2126,
+      "loss": 0.0944,
       "step": 412
     },
     {
       "epoch": 2.3333333333333335,
-      "grad_norm": 0.7109375,
+      "grad_norm": 0.53125,
       "learning_rate": 0.00017522319842140055,
-      "loss": 0.1302,
+      "loss": 0.0702,
       "step": 413
     },
     {
       "epoch": 2.3389830508474576,
-      "grad_norm": 0.5546875,
+      "grad_norm": 0.7578125,
       "learning_rate": 0.00017510546556413498,
-      "loss": 0.1001,
+      "loss": 0.0995,
       "step": 414
     },
     {
       "epoch": 2.3446327683615817,
-      "grad_norm": 0.515625,
+      "grad_norm": 0.625,
       "learning_rate": 0.0001749874934051401,
-      "loss": 0.1458,
+      "loss": 0.1027,
       "step": 415
     },
     {
       "epoch": 2.3502824858757063,
-      "grad_norm": 0.80859375,
+      "grad_norm": 0.5703125,
       "learning_rate": 0.00017486928232029992,
-      "loss": 0.1641,
+      "loss": 0.1022,
       "step": 416
     },
     {
       "epoch": 2.3559322033898304,
-      "grad_norm": 0.66796875,
+      "grad_norm": 0.52734375,
       "learning_rate": 0.0001747508326862597,
-      "loss": 0.1532,
+      "loss": 0.0948,
       "step": 417
     },
     {
       "epoch": 2.361581920903955,
-      "grad_norm": 0.66796875,
+      "grad_norm": 0.73046875,
       "learning_rate": 0.00017463214488042472,
-      "loss": 0.0665,
+      "loss": 0.091,
       "step": 418
     },
     {
       "epoch": 2.367231638418079,
-      "grad_norm": 0.6953125,
+      "grad_norm": 0.6796875,
       "learning_rate": 0.0001745132192809592,
-      "loss": 0.1204,
+      "loss": 0.1356,
       "step": 419
     },
     {
       "epoch": 2.3728813559322033,
-      "grad_norm": 0.490234375,
+      "grad_norm": 0.66015625,
       "learning_rate": 0.00017439405626678496,
-      "loss": 0.1045,
+      "loss": 0.1232,
       "step": 420
     },
     {
       "epoch": 2.378531073446328,
-      "grad_norm": 0.6171875,
+      "grad_norm": 0.578125,
       "learning_rate": 0.00017427465621758033,
-      "loss": 0.077,
+      "loss": 0.0712,
       "step": 421
     },
     {
       "epoch": 2.384180790960452,
-      "grad_norm": 0.6015625,
+      "grad_norm": 0.66015625,
       "learning_rate": 0.0001741550195137788,
-      "loss": 0.1626,
+      "loss": 0.1879,
       "step": 422
     },
     {
       "epoch": 2.389830508474576,
-      "grad_norm": 0.5703125,
+      "grad_norm": 0.640625,
       "learning_rate": 0.00017403514653656802,
-      "loss": 0.1051,
+      "loss": 0.1091,
       "step": 423
     },
     {
       "epoch": 2.3954802259887007,
-      "grad_norm": 0.5390625,
+      "grad_norm": 0.5234375,
       "learning_rate": 0.00017391503766788828,
-      "loss": 0.1339,
+      "loss": 0.0693,
       "step": 424
     },
     {
       "epoch": 2.401129943502825,
-      "grad_norm": 0.5,
+      "grad_norm": 0.6015625,
       "learning_rate": 0.00017379469329043167,
-      "loss": 0.1209,
+      "loss": 0.1592,
       "step": 425
     },
     {
       "epoch": 2.406779661016949,
-      "grad_norm": 0.546875,
+      "grad_norm": 0.43359375,
       "learning_rate": 0.0001736741137876405,
-      "loss": 0.1685,
+      "loss": 0.1538,
       "step": 426
     },
     {
       "epoch": 2.4124293785310735,
-      "grad_norm": 0.71484375,
+      "grad_norm": 0.4921875,
       "learning_rate": 0.0001735532995437063,
-      "loss": 0.2087,
+      "loss": 0.1463,
       "step": 427
     },
     {
       "epoch": 2.4180790960451977,
-      "grad_norm": 0.671875,
+      "grad_norm": 0.69140625,
       "learning_rate": 0.00017343225094356855,
-      "loss": 0.1292,
+      "loss": 0.1626,
       "step": 428
     },
     {
       "epoch": 2.423728813559322,
-      "grad_norm": 0.62109375,
+      "grad_norm": 0.7890625,
       "learning_rate": 0.00017331096837291344,
-      "loss": 0.1329,
+      "loss": 0.1325,
       "step": 429
     },
     {
       "epoch": 2.4293785310734464,
-      "grad_norm": 0.8125,
+      "grad_norm": 0.5546875,
       "learning_rate": 0.00017318945221817255,
-      "loss": 0.1297,
+      "loss": 0.0322,
       "step": 430
     },
     {
       "epoch": 2.4350282485875705,
-      "grad_norm": 0.58203125,
+      "grad_norm": 0.89453125,
       "learning_rate": 0.00017306770286652182,
-      "loss": 0.1352,
+      "loss": 0.1504,
       "step": 431
     },
     {
       "epoch": 2.440677966101695,
-      "grad_norm": 0.50390625,
+      "grad_norm": 0.578125,
       "learning_rate": 0.00017294572070588018,
-      "loss": 0.062,
+      "loss": 0.1169,
       "step": 432
     },
     {
       "epoch": 2.446327683615819,
-      "grad_norm": 0.82421875,
+      "grad_norm": 0.48828125,
       "learning_rate": 0.0001728235061249083,
-      "loss": 0.171,
+      "loss": 0.1087,
       "step": 433
     },
     {
       "epoch": 2.4519774011299433,
-      "grad_norm": 0.6484375,
+      "grad_norm": 0.59765625,
       "learning_rate": 0.00017270105951300738,
-      "loss": 0.1118,
+      "loss": 0.1205,
       "step": 434
     },
     {
       "epoch": 2.457627118644068,
-      "grad_norm": 0.53125,
+      "grad_norm": 0.44140625,
       "learning_rate": 0.00017257838126031797,
-      "loss": 0.1153,
+      "loss": 0.1332,
       "step": 435
     },
     {
       "epoch": 2.463276836158192,
-      "grad_norm": 0.5234375,
+      "grad_norm": 0.57421875,
       "learning_rate": 0.0001724554717577187,
-      "loss": 0.0851,
+      "loss": 0.103,
       "step": 436
     },
     {
       "epoch": 2.4689265536723166,
-      "grad_norm": 0.53125,
+      "grad_norm": 0.466796875,
       "learning_rate": 0.0001723323313968249,
-      "loss": 0.1096,
+      "loss": 0.123,
       "step": 437
     },
     {
       "epoch": 2.4745762711864407,
-      "grad_norm": 0.609375,
+      "grad_norm": 0.51953125,
       "learning_rate": 0.0001722089605699875,
-      "loss": 0.1278,
+      "loss": 0.1735,
       "step": 438
     },
     {
       "epoch": 2.480225988700565,
-      "grad_norm": 0.5078125,
+      "grad_norm": 0.74609375,
       "learning_rate": 0.00017208535967029188,
-      "loss": 0.1185,
+      "loss": 0.0936,
       "step": 439
     },
     {
       "epoch": 2.4858757062146895,
-      "grad_norm": 0.55078125,
+      "grad_norm": 0.7734375,
       "learning_rate": 0.00017196152909155628,
-      "loss": 0.1263,
+      "loss": 0.1332,
       "step": 440
     },
     {
       "epoch": 2.4915254237288136,
-      "grad_norm": 0.392578125,
+      "grad_norm": 0.51171875,
       "learning_rate": 0.0001718374692283309,
-      "loss": 0.1545,
+      "loss": 0.0931,
       "step": 441
     },
     {
       "epoch": 2.4971751412429377,
-      "grad_norm": 0.59375,
+      "grad_norm": 0.60546875,
       "learning_rate": 0.00017171318047589637,
-      "loss": 0.122,
+      "loss": 0.0848,
       "step": 442
     },
     {
       "epoch": 2.5028248587570623,
-      "grad_norm": 0.466796875,
+      "grad_norm": 0.6953125,
       "learning_rate": 0.00017158866323026265,
-      "loss": 0.1092,
+      "loss": 0.0583,
       "step": 443
     },
     {
       "epoch": 2.5084745762711864,
-      "grad_norm": 0.71875,
+      "grad_norm": 0.388671875,
       "learning_rate": 0.0001714639178881678,
-      "loss": 0.0713,
+      "loss": 0.124,
       "step": 444
     },
     {
       "epoch": 2.5141242937853105,
-      "grad_norm": 0.53125,
+      "grad_norm": 0.640625,
       "learning_rate": 0.00017133894484707655,
-      "loss": 0.1416,
+      "loss": 0.1289,
       "step": 445
     },
     {
       "epoch": 2.519774011299435,
-      "grad_norm": 0.67578125,
+      "grad_norm": 0.60546875,
       "learning_rate": 0.00017121374450517916,
-      "loss": 0.1206,
+      "loss": 0.1119,
       "step": 446
     },
     {
       "epoch": 2.5254237288135593,
-      "grad_norm": 0.859375,
+      "grad_norm": 0.62890625,
       "learning_rate": 0.00017108831726139013,
-      "loss": 0.0892,
+      "loss": 0.0917,
       "step": 447
     },
     {
       "epoch": 2.5310734463276834,
-      "grad_norm": 0.61328125,
+      "grad_norm": 0.8125,
       "learning_rate": 0.00017096266351534692,
-      "loss": 0.1074,
+      "loss": 0.0975,
       "step": 448
     },
     {
       "epoch": 2.536723163841808,
       "grad_norm": 0.53125,
       "learning_rate": 0.0001708367836674086,
-      "loss": 0.0833,
+      "loss": 0.1363,
       "step": 449
     },
     {
       "epoch": 2.542372881355932,
-      "grad_norm": 0.416015625,
+      "grad_norm": 0.671875,
       "learning_rate": 0.00017071067811865476,
-      "loss": 0.1144,
+      "loss": 0.1465,
       "step": 450
     },
     {
       "epoch": 2.542372881355932,
-      "eval_loss": 0.06996988505125046,
-      "eval_runtime": 5.699,
-      "eval_samples_per_second": 26.145,
-      "eval_steps_per_second": 3.334,
+      "eval_loss": 0.07218772917985916,
+      "eval_runtime": 5.7234,
+      "eval_samples_per_second": 26.208,
+      "eval_steps_per_second": 3.32,
       "step": 450
     },
     {
       "epoch": 2.548022598870056,
-      "grad_norm": 0.46484375,
+      "grad_norm": 0.57421875,
       "learning_rate": 0.00017058434727088402,
-      "loss": 0.1579,
+      "loss": 0.0839,
       "step": 451
     },
     {
       "epoch": 2.553672316384181,
-      "grad_norm": 0.4765625,
+      "grad_norm": 0.75,
       "learning_rate": 0.00017045779152661297,
-      "loss": 0.0879,
+      "loss": 0.09,
       "step": 452
     },
     {
       "epoch": 2.559322033898305,
-      "grad_norm": 0.51171875,
+      "grad_norm": 0.703125,
       "learning_rate": 0.00017033101128907462,
-      "loss": 0.0885,
+      "loss": 0.1236,
       "step": 453
     },
     {
       "epoch": 2.5649717514124295,
-      "grad_norm": 0.55859375,
+      "grad_norm": 0.671875,
       "learning_rate": 0.00017020400696221737,
-      "loss": 0.1015,
+      "loss": 0.1371,
       "step": 454
     },
     {
       "epoch": 2.5706214689265536,
-      "grad_norm": 0.62109375,
+      "grad_norm": 0.63671875,
       "learning_rate": 0.00017007677895070357,
-      "loss": 0.1787,
+      "loss": 0.0876,
       "step": 455
     },
     {
       "epoch": 2.576271186440678,
-      "grad_norm": 0.65625,
+      "grad_norm": 0.33984375,
       "learning_rate": 0.0001699493276599083,
-      "loss": 0.0781,
+      "loss": 0.1452,
       "step": 456
     },
     {
       "epoch": 2.5819209039548023,
-      "grad_norm": 0.7421875,
+      "grad_norm": 0.67578125,
       "learning_rate": 0.000169821653495918,
-      "loss": 0.1154,
+      "loss": 0.1715,
       "step": 457
     },
     {
       "epoch": 2.5875706214689265,
-      "grad_norm": 0.40625,
+      "grad_norm": 0.7890625,
       "learning_rate": 0.00016969375686552937,
-      "loss": 0.114,
+      "loss": 0.1315,
       "step": 458
     },
     {
       "epoch": 2.593220338983051,
-      "grad_norm": 0.77734375,
+      "grad_norm": 0.45703125,
       "learning_rate": 0.00016956563817624773,
-      "loss": 0.0928,
+      "loss": 0.0992,
       "step": 459
     },
     {
       "epoch": 2.598870056497175,
-      "grad_norm": 0.41015625,
+      "grad_norm": 0.5078125,
       "learning_rate": 0.00016943729783628608,
-      "loss": 0.1595,
+      "loss": 0.1605,
       "step": 460
     },
     {
       "epoch": 2.6045197740112993,
-      "grad_norm": 0.52734375,
+      "grad_norm": 0.451171875,
       "learning_rate": 0.0001693087362545636,
-      "loss": 0.0771,
+      "loss": 0.144,
       "step": 461
     },
     {
       "epoch": 2.610169491525424,
-      "grad_norm": 0.625,
+      "grad_norm": 0.71484375,
       "learning_rate": 0.0001691799538407044,
-      "loss": 0.1167,
+      "loss": 0.0895,
       "step": 462
     },
     {
       "epoch": 2.615819209039548,
-      "grad_norm": 0.55078125,
+      "grad_norm": 0.6328125,
       "learning_rate": 0.00016905095100503616,
-      "loss": 0.0836,
+      "loss": 0.0858,
       "step": 463
     },
     {
       "epoch": 2.621468926553672,
-      "grad_norm": 0.380859375,
+      "grad_norm": 0.6015625,
       "learning_rate": 0.00016892172815858894,
-      "loss": 0.1129,
+      "loss": 0.1434,
       "step": 464
     },
     {
       "epoch": 2.6271186440677967,
-      "grad_norm": 0.765625,
+      "grad_norm": 0.62109375,
       "learning_rate": 0.00016879228571309377,
-      "loss": 0.114,
+      "loss": 0.1157,
       "step": 465
     },
     {
       "epoch": 2.632768361581921,
-      "grad_norm": 0.5859375,
+      "grad_norm": 0.5078125,
       "learning_rate": 0.00016866262408098134,
-      "loss": 0.039,
+      "loss": 0.1146,
       "step": 466
     },
     {
       "epoch": 2.638418079096045,
-      "grad_norm": 0.58984375,
+      "grad_norm": 0.6015625,
       "learning_rate": 0.00016853274367538082,
-      "loss": 0.1278,
+      "loss": 0.0859,
       "step": 467
     },
     {
       "epoch": 2.6440677966101696,
-      "grad_norm": 0.65625,
+      "grad_norm": 0.5390625,
       "learning_rate": 0.00016840264491011823,
-      "loss": 0.1451,
+      "loss": 0.0713,
       "step": 468
     },
     {
       "epoch": 2.6497175141242937,
-      "grad_norm": 0.515625,
+      "grad_norm": 0.388671875,
       "learning_rate": 0.00016827232819971558,
-      "loss": 0.0811,
+      "loss": 0.1833,
       "step": 469
     },
     {
       "epoch": 2.655367231638418,
-      "grad_norm": 0.6640625,
+      "grad_norm": 0.5859375,
       "learning_rate": 0.00016814179395938913,
-      "loss": 0.0762,
+      "loss": 0.0929,
       "step": 470
     },
     {
       "epoch": 2.6610169491525424,
-      "grad_norm": 0.6328125,
+      "grad_norm": 0.5859375,
       "learning_rate": 0.00016801104260504826,
-      "loss": 0.0994,
+      "loss": 0.0702,
       "step": 471
     },
     {
       "epoch": 2.6666666666666665,
-      "grad_norm": 0.65234375,
+      "grad_norm": 0.61328125,
       "learning_rate": 0.0001678800745532942,
-      "loss": 0.0825,
+      "loss": 0.0856,
       "step": 472
     },
     {
       "epoch": 2.672316384180791,
-      "grad_norm": 0.55859375,
+      "grad_norm": 0.7734375,
       "learning_rate": 0.0001677488902214185,
-      "loss": 0.0863,
+      "loss": 0.1118,
       "step": 473
     },
     {
       "epoch": 2.6779661016949152,
-      "grad_norm": 0.6328125,
+      "grad_norm": 0.6171875,
       "learning_rate": 0.00016761749002740193,
-      "loss": 0.1134,
+      "loss": 0.1031,
       "step": 474
     },
     {
       "epoch": 2.68361581920904,
-      "grad_norm": 0.5234375,
+      "grad_norm": 0.51953125,
       "learning_rate": 0.00016748587438991303,
-      "loss": 0.1288,
+      "loss": 0.1153,
       "step": 475
     },
     {
       "epoch": 2.689265536723164,
-      "grad_norm": 0.54296875,
+      "grad_norm": 0.4609375,
       "learning_rate": 0.00016735404372830674,
-      "loss": 0.0776,
+      "loss": 0.1033,
       "step": 476
     },
     {
       "epoch": 2.694915254237288,
-      "grad_norm": 0.435546875,
+      "grad_norm": 0.60546875,
       "learning_rate": 0.00016722199846262313,
-      "loss": 0.1,
+      "loss": 0.1267,
       "step": 477
     },
     {
       "epoch": 2.7005649717514126,
-      "grad_norm": 0.6171875,
+      "grad_norm": 0.46484375,
       "learning_rate": 0.00016708973901358602,
-      "loss": 0.0831,
+      "loss": 0.0816,
       "step": 478
     },
     {
       "epoch": 2.7062146892655368,
-      "grad_norm": 0.474609375,
+      "grad_norm": 0.40625,
       "learning_rate": 0.00016695726580260174,
-      "loss": 0.0949,
+      "loss": 0.1618,
       "step": 479
     },
     {
       "epoch": 2.711864406779661,
-      "grad_norm": 0.439453125,
+      "grad_norm": 0.5234375,
       "learning_rate": 0.00016682457925175763,
-      "loss": 0.1198,
+      "loss": 0.1277,
       "step": 480
     },
     {
       "epoch": 2.7175141242937855,
-      "grad_norm": 0.51171875,
+      "grad_norm": 0.62109375,
       "learning_rate": 0.00016669167978382082,
-      "loss": 0.0581,
+      "loss": 0.0733,
       "step": 481
     },
     {
       "epoch": 2.7231638418079096,
-      "grad_norm": 0.53515625,
+      "grad_norm": 0.47265625,
       "learning_rate": 0.00016655856782223682,
-      "loss": 0.0972,
+      "loss": 0.077,
       "step": 482
     },
     {
       "epoch": 2.7288135593220337,
-      "grad_norm": 0.451171875,
+      "grad_norm": 0.5078125,
       "learning_rate": 0.00016642524379112817,
-      "loss": 0.0946,
+      "loss": 0.0666,
       "step": 483
     },
     {
       "epoch": 2.7344632768361583,
-      "grad_norm": 0.6640625,
+      "grad_norm": 0.53125,
       "learning_rate": 0.00016629170811529318,
-      "loss": 0.1094,
+      "loss": 0.0665,
       "step": 484
     },
     {
       "epoch": 2.7401129943502824,
-      "grad_norm": 0.462890625,
+      "grad_norm": 0.34765625,
       "learning_rate": 0.00016615796122020443,
-      "loss": 0.0922,
+      "loss": 0.154,
       "step": 485
     },
     {
       "epoch": 2.7457627118644066,
-      "grad_norm": 0.69140625,
+      "grad_norm": 0.6171875,
       "learning_rate": 0.0001660240035320075,
-      "loss": 0.0677,
+      "loss": 0.0861,
       "step": 486
     },
     {
       "epoch": 2.751412429378531,
-      "grad_norm": 0.5390625,
+      "grad_norm": 0.447265625,
       "learning_rate": 0.00016588983547751962,
-      "loss": 0.1124,
+      "loss": 0.0939,
       "step": 487
     },
     {
       "epoch": 2.7570621468926553,
-      "grad_norm": 0.484375,
+      "grad_norm": 0.396484375,
       "learning_rate": 0.0001657554574842283,
-      "loss": 0.1058,
+      "loss": 0.0789,
       "step": 488
     },
     {
       "epoch": 2.7627118644067794,
-      "grad_norm": 0.4921875,
+      "grad_norm": 0.53515625,
       "learning_rate": 0.00016562086998028996,
-      "loss": 0.1664,
+      "loss": 0.0522,
       "step": 489
     },
     {
       "epoch": 2.768361581920904,
-      "grad_norm": 0.546875,
+      "grad_norm": 0.490234375,
       "learning_rate": 0.00016548607339452853,
-      "loss": 0.1106,
+      "loss": 0.1131,
       "step": 490
     },
     {
       "epoch": 2.774011299435028,
-      "grad_norm": 0.6015625,
+      "grad_norm": 0.443359375,
       "learning_rate": 0.0001653510681564341,
-      "loss": 0.1396,
+      "loss": 0.0911,
       "step": 491
     },
     {
       "epoch": 2.7796610169491527,
-      "grad_norm": 0.4296875,
+      "grad_norm": 0.42578125,
       "learning_rate": 0.0001652158546961617,
-      "loss": 0.0556,
+      "loss": 0.1278,
       "step": 492
     },
     {
       "epoch": 2.785310734463277,
-      "grad_norm": 0.546875,
+      "grad_norm": 0.490234375,
       "learning_rate": 0.00016508043344452963,
-      "loss": 0.1089,
+      "loss": 0.1402,
       "step": 493
     },
     {
       "epoch": 2.7909604519774014,
-      "grad_norm": 0.421875,
+      "grad_norm": 0.609375,
       "learning_rate": 0.00016494480483301836,
-      "loss": 0.1495,
+      "loss": 0.0605,
       "step": 494
     },
     {
       "epoch": 2.7966101694915255,
-      "grad_norm": 0.49609375,
+      "grad_norm": 0.63671875,
       "learning_rate": 0.00016480896929376907,
-      "loss": 0.0787,
+      "loss": 0.1019,
       "step": 495
     },
     {
       "epoch": 2.7966101694915255,
-      "eval_loss": 0.039022162556648254,
-      "eval_runtime": 5.6887,
-      "eval_samples_per_second": 26.192,
-      "eval_steps_per_second": 3.34,
+      "eval_loss": 0.04575210064649582,
+      "eval_runtime": 5.7463,
+      "eval_samples_per_second": 26.104,
+      "eval_steps_per_second": 3.306,
       "step": 495
     },
     {
       "epoch": 2.8022598870056497,
-      "grad_norm": 0.490234375,
+      "grad_norm": 0.5234375,
       "learning_rate": 0.00016467292725958212,
-      "loss": 0.0604,
+      "loss": 0.1471,
       "step": 496
     },
     {
       "epoch": 2.8079096045197742,
-      "grad_norm": 0.49609375,
+      "grad_norm": 0.70703125,
       "learning_rate": 0.00016453667916391598,
-      "loss": 0.0877,
+      "loss": 0.0417,
       "step": 497
     },
     {
       "epoch": 2.8135593220338984,
-      "grad_norm": 0.48046875,
+      "grad_norm": 0.443359375,
       "learning_rate": 0.00016440022544088553,
-      "loss": 0.1094,
+      "loss": 0.1109,
       "step": 498
     },
     {
       "epoch": 2.8192090395480225,
-      "grad_norm": 0.55859375,
+      "grad_norm": 0.455078125,
       "learning_rate": 0.00016426356652526097,
-      "loss": 0.1129,
+      "loss": 0.1472,
       "step": 499
     },
     {
       "epoch": 2.824858757062147,
-      "grad_norm": 0.54296875,
+      "grad_norm": 0.5390625,
       "learning_rate": 0.0001641267028524661,
-      "loss": 0.1056,
+      "loss": 0.121,
       "step": 500
     },
     {
       "epoch": 2.830508474576271,
-      "grad_norm": 0.37890625,
+      "grad_norm": 0.447265625,
       "learning_rate": 0.0001639896348585773,
-      "loss": 0.1371,
+      "loss": 0.0624,
       "step": 501
     },
     {
       "epoch": 2.8361581920903953,
-      "grad_norm": 0.439453125,
+      "grad_norm": 0.474609375,
       "learning_rate": 0.00016385236298032183,
-      "loss": 0.0787,
+      "loss": 0.1218,
       "step": 502
     },
     {
       "epoch": 2.84180790960452,
-      "grad_norm": 0.48046875,
+      "grad_norm": 0.482421875,
       "learning_rate": 0.00016371488765507666,
-      "loss": 0.114,
+      "loss": 0.1082,
       "step": 503
     },
     {
       "epoch": 2.847457627118644,
-      "grad_norm": 0.64453125,
+      "grad_norm": 0.546875,
       "learning_rate": 0.00016357720932086688,
-      "loss": 0.1144,
+      "loss": 0.0983,
       "step": 504
     },
     {
       "epoch": 2.853107344632768,
-      "grad_norm": 0.37890625,
+      "grad_norm": 0.50390625,
       "learning_rate": 0.00016343932841636456,
-      "loss": 0.093,
+      "loss": 0.0832,
       "step": 505
     },
     {
       "epoch": 2.8587570621468927,
-      "grad_norm": 0.42578125,
+      "grad_norm": 0.5703125,
       "learning_rate": 0.00016330124538088705,
-      "loss": 0.1173,
+      "loss": 0.1004,
       "step": 506
     },
     {
       "epoch": 2.864406779661017,
-      "grad_norm": 0.302734375,
+      "grad_norm": 0.384765625,
       "learning_rate": 0.0001631629606543958,
-      "loss": 0.1046,
+      "loss": 0.104,
       "step": 507
     },
     {
       "epoch": 2.870056497175141,
-      "grad_norm": 0.48046875,
+      "grad_norm": 0.4296875,
       "learning_rate": 0.00016302447467749493,
-      "loss": 0.0555,
+      "loss": 0.0751,
       "step": 508
     },
     {
       "epoch": 2.8757062146892656,
-      "grad_norm": 0.421875,
+      "grad_norm": 0.5234375,
       "learning_rate": 0.0001628857878914297,
-      "loss": 0.1478,
+      "loss": 0.0659,
       "step": 509
     },
     {
       "epoch": 2.8813559322033897,
-      "grad_norm": 0.40234375,
+      "grad_norm": 0.4140625,
       "learning_rate": 0.0001627469007380852,
-      "loss": 0.1031,
+      "loss": 0.14,
       "step": 510
     },
     {
       "epoch": 2.8870056497175143,
-      "grad_norm": 0.50390625,
+      "grad_norm": 0.37109375,
       "learning_rate": 0.00016260781365998503,
-      "loss": 0.0986,
+      "loss": 0.0914,
       "step": 511
     },
     {
       "epoch": 2.8926553672316384,
-      "grad_norm": 0.470703125,
+      "grad_norm": 0.3203125,
       "learning_rate": 0.00016246852710028972,
-      "loss": 0.0518,
+      "loss": 0.0694,
       "step": 512
     },
     {
       "epoch": 2.898305084745763,
-      "grad_norm": 0.416015625,
+      "grad_norm": 0.40625,
       "learning_rate": 0.00016232904150279532,
-      "loss": 0.1475,
+      "loss": 0.0528,
       "step": 513
     },
     {
       "epoch": 2.903954802259887,
-      "grad_norm": 0.4296875,
+      "grad_norm": 0.48828125,
       "learning_rate": 0.00016218935731193224,
-      "loss": 0.1669,
+      "loss": 0.0672,
       "step": 514
     },
     {
       "epoch": 2.9096045197740112,
-      "grad_norm": 0.380859375,
+      "grad_norm": 0.435546875,
       "learning_rate": 0.00016204947497276345,
-      "loss": 0.0661,
+      "loss": 0.1225,
       "step": 515
     },
     {
       "epoch": 2.915254237288136,
-      "grad_norm": 0.51171875,
+      "grad_norm": 0.4765625,
       "learning_rate": 0.00016190939493098344,
-      "loss": 0.1303,
+      "loss": 0.0717,
       "step": 516
     },
     {
       "epoch": 2.92090395480226,
-      "grad_norm": 0.5703125,
+      "grad_norm": 0.40625,
       "learning_rate": 0.0001617691176329164,
-      "loss": 0.1174,
+      "loss": 0.1019,
       "step": 517
     },
     {
       "epoch": 2.926553672316384,
-      "grad_norm": 0.416015625,
+      "grad_norm": 0.41796875,
       "learning_rate": 0.0001616286435255153,
-      "loss": 0.0957,
+      "loss": 0.0618,
       "step": 518
     },
     {
       "epoch": 2.9322033898305087,
-      "grad_norm": 0.65234375,
+      "grad_norm": 0.484375,
       "learning_rate": 0.00016148797305635997,
-      "loss": 0.0294,
+      "loss": 0.113,
       "step": 519
     },
     {
       "epoch": 2.937853107344633,
-      "grad_norm": 0.68359375,
+      "grad_norm": 0.546875,
       "learning_rate": 0.00016134710667365596,
-      "loss": 0.1508,
+      "loss": 0.0919,
       "step": 520
     },
     {
       "epoch": 2.943502824858757,
-      "grad_norm": 0.41796875,
+      "grad_norm": 0.62109375,
       "learning_rate": 0.00016120604482623306,
-      "loss": 0.1577,
+      "loss": 0.0997,
       "step": 521
     },
     {
       "epoch": 2.9491525423728815,
-      "grad_norm": 0.51171875,
+      "grad_norm": 0.373046875,
       "learning_rate": 0.00016106478796354382,
-      "loss": 0.0938,
+      "loss": 0.1475,
       "step": 522
     },
     {
       "epoch": 2.9548022598870056,
-      "grad_norm": 0.43359375,
+      "grad_norm": 0.4375,
       "learning_rate": 0.00016092333653566215,
-      "loss": 0.1115,
+      "loss": 0.0902,
       "step": 523
     },
     {
       "epoch": 2.9604519774011298,
-      "grad_norm": 0.52734375,
+      "grad_norm": 0.53515625,
       "learning_rate": 0.00016078169099328197,
-      "loss": 0.0721,
+      "loss": 0.1288,
       "step": 524
     },
     {
       "epoch": 2.9661016949152543,
-      "grad_norm": 0.59765625,
+      "grad_norm": 0.53125,
       "learning_rate": 0.00016063985178771555,
-      "loss": 0.0691,
+      "loss": 0.0964,
       "step": 525
     },
     {
       "epoch": 2.9717514124293785,
-      "grad_norm": 0.6015625,
+      "grad_norm": 0.609375,
       "learning_rate": 0.00016049781937089227,
-      "loss": 0.0971,
+      "loss": 0.0831,
       "step": 526
     },
     {
       "epoch": 2.9774011299435026,
-      "grad_norm": 0.291015625,
+      "grad_norm": 0.494140625,
       "learning_rate": 0.00016035559419535716,
-      "loss": 0.1096,
+      "loss": 0.0564,
       "step": 527
     },
     {
       "epoch": 2.983050847457627,
-      "grad_norm": 0.4609375,
+      "grad_norm": 0.44140625,
       "learning_rate": 0.0001602131767142694,
-      "loss": 0.1174,
+      "loss": 0.1105,
       "step": 528
     },
     {
       "epoch": 2.9887005649717513,
-      "grad_norm": 0.39453125,
+      "grad_norm": 0.451171875,
       "learning_rate": 0.00016007056738140085,
-      "loss": 0.1203,
+      "loss": 0.1365,
       "step": 529
     },
     {
       "epoch": 2.994350282485876,
-      "grad_norm": 0.59375,
+      "grad_norm": 0.345703125,
       "learning_rate": 0.0001599277666511347,
-      "loss": 0.1285,
+      "loss": 0.1079,
       "step": 530
     },
     {
       "epoch": 3.0,
-      "grad_norm": 0.419921875,
+      "grad_norm": 0.53515625,
       "learning_rate": 0.00015978477497846397,
-      "loss": 0.119,
+      "loss": 0.1166,
       "step": 531
     },
     {
       "epoch": 3.005649717514124,
-      "grad_norm": 0.3359375,
+      "grad_norm": 0.515625,
       "learning_rate": 0.00015964159281899008,
-      "loss": 0.0806,
+      "loss": 0.0968,
       "step": 532
     },
     {
       "epoch": 3.0112994350282487,
-      "grad_norm": 0.341796875,
+      "grad_norm": 0.30859375,
       "learning_rate": 0.00015949822062892127,
-      "loss": 0.0482,
+      "loss": 0.0953,
       "step": 533
     },
     {
       "epoch": 3.016949152542373,
-      "grad_norm": 0.55859375,
+      "grad_norm": 0.400390625,
       "learning_rate": 0.00015935465886507142,
-      "loss": 0.0903,
+      "loss": 0.0778,
       "step": 534
     },
     {
       "epoch": 3.022598870056497,
-      "grad_norm": 0.298828125,
+      "grad_norm": 0.341796875,
       "learning_rate": 0.00015921090798485832,
-      "loss": 0.0656,
+      "loss": 0.0501,
       "step": 535
     },
     {
       "epoch": 3.0282485875706215,
-      "grad_norm": 0.4453125,
+      "grad_norm": 0.408203125,
       "learning_rate": 0.0001590669684463023,
-      "loss": 0.082,
+      "loss": 0.0637,
       "step": 536
     },
     {
       "epoch": 3.0338983050847457,
       "grad_norm": 0.400390625,
       "learning_rate": 0.00015892284070802493,
-      "loss": 0.0528,
+      "loss": 0.1284,
       "step": 537
     },
     {
       "epoch": 3.0395480225988702,
-      "grad_norm": 0.423828125,
+      "grad_norm": 0.60546875,
       "learning_rate": 0.00015877852522924732,
-      "loss": 0.0677,
+      "loss": 0.091,
       "step": 538
     },
     {
       "epoch": 3.0451977401129944,
-      "grad_norm": 0.47265625,
+      "grad_norm": 0.427734375,
       "learning_rate": 0.00015863402246978878,
-      "loss": 0.0468,
+      "loss": 0.0696,
       "step": 539
     },
     {
       "epoch": 3.0508474576271185,
-      "grad_norm": 0.423828125,
+      "grad_norm": 0.4140625,
       "learning_rate": 0.0001584893328900653,
-      "loss": 0.0523,
+      "loss": 0.161,
       "step": 540
     },
     {
       "epoch": 3.0508474576271185,
-      "eval_loss": 0.026925092563033104,
-      "eval_runtime": 5.7968,
-      "eval_samples_per_second": 25.704,
-      "eval_steps_per_second": 3.278,
+      "eval_loss": 0.035421498119831085,
+      "eval_runtime": 5.726,
+      "eval_samples_per_second": 26.197,
+      "eval_steps_per_second": 3.318,
       "step": 540
     },
     {
       "epoch": 3.056497175141243,
-      "grad_norm": 0.33203125,
+      "grad_norm": 0.5234375,
       "learning_rate": 0.00015834445695108823,
-      "loss": 0.0931,
+      "loss": 0.0376,
       "step": 541
     },
     {
       "epoch": 3.062146892655367,
-      "grad_norm": 0.4296875,
+      "grad_norm": 0.3828125,
       "learning_rate": 0.0001581993951144626,
-      "loss": 0.0859,
+      "loss": 0.0768,
       "step": 542
     },
     {
       "epoch": 3.0677966101694913,
-      "grad_norm": 0.39453125,
+      "grad_norm": 0.40234375,
       "learning_rate": 0.00015805414784238571,
-      "loss": 0.0772,
+      "loss": 0.0319,
       "step": 543
     },
     {
       "epoch": 3.073446327683616,
-      "grad_norm": 0.267578125,
+      "grad_norm": 0.349609375,
       "learning_rate": 0.0001579087155976459,
-      "loss": 0.121,
+      "loss": 0.1377,
       "step": 544
     },
     {
       "epoch": 3.07909604519774,
-      "grad_norm": 0.421875,
+      "grad_norm": 0.48828125,
       "learning_rate": 0.0001577630988436206,
-      "loss": 0.0916,
+      "loss": 0.0683,
       "step": 545
     },
     {
       "epoch": 3.084745762711864,
-      "grad_norm": 0.345703125,
+      "grad_norm": 0.34765625,
       "learning_rate": 0.0001576172980442753,
-      "loss": 0.1164,
+      "loss": 0.0622,
       "step": 546
     },
     {
       "epoch": 3.0903954802259888,
-      "grad_norm": 0.375,
+      "grad_norm": 0.337890625,
       "learning_rate": 0.0001574713136641619,
-      "loss": 0.0646,
+      "loss": 0.1864,
       "step": 547
     },
     {
       "epoch": 3.096045197740113,
-      "grad_norm": 0.30078125,
+      "grad_norm": 0.361328125,
       "learning_rate": 0.00015732514616841715,
-      "loss": 0.072,
+      "loss": 0.0443,
       "step": 548
     },
     {
       "epoch": 3.1016949152542375,
-      "grad_norm": 0.4453125,
+      "grad_norm": 0.296875,
       "learning_rate": 0.00015717879602276122,
-      "loss": 0.0779,
+      "loss": 0.1397,
       "step": 549
     },
     {
       "epoch": 3.1073446327683616,
-      "grad_norm": 0.365234375,
+      "grad_norm": 0.337890625,
       "learning_rate": 0.0001570322636934964,
-      "loss": 0.1022,
+      "loss": 0.0843,
       "step": 550
     },
     {
       "epoch": 3.1129943502824857,
-      "grad_norm": 0.30078125,
+      "grad_norm": 0.4140625,
       "learning_rate": 0.00015688554964750533,
-      "loss": 0.0963,
+      "loss": 0.1233,
       "step": 551
     },
     {
       "epoch": 3.1186440677966103,
-      "grad_norm": 0.365234375,
+      "grad_norm": 0.515625,
       "learning_rate": 0.00015673865435224962,
-      "loss": 0.1167,
+      "loss": 0.0687,
       "step": 552
     },
     {
       "epoch": 3.1242937853107344,
-      "grad_norm": 0.28125,
+      "grad_norm": 0.2255859375,
       "learning_rate": 0.00015659157827576842,
-      "loss": 0.0881,
+      "loss": 0.1127,
       "step": 553
     },
     {
       "epoch": 3.1299435028248586,
-      "grad_norm": 0.404296875,
+      "grad_norm": 0.3671875,
       "learning_rate": 0.00015644432188667695,
-      "loss": 0.1178,
+      "loss": 0.069,
       "step": 554
     },
     {
       "epoch": 3.135593220338983,
-      "grad_norm": 0.48828125,
+      "grad_norm": 0.484375,
       "learning_rate": 0.00015629688565416478,
-      "loss": 0.0441,
+      "loss": 0.0406,
       "step": 555
     },
     {
       "epoch": 3.1412429378531073,
-      "grad_norm": 0.388671875,
+      "grad_norm": 0.416015625,
       "learning_rate": 0.0001561492700479947,
-      "loss": 0.1159,
+      "loss": 0.0985,
       "step": 556
     },
     {
       "epoch": 3.146892655367232,
-      "grad_norm": 0.2578125,
+      "grad_norm": 0.400390625,
       "learning_rate": 0.00015600147553850082,
-      "loss": 0.0721,
+      "loss": 0.0812,
       "step": 557
     },
     {
       "epoch": 3.152542372881356,
-      "grad_norm": 0.453125,
+      "grad_norm": 0.357421875,
       "learning_rate": 0.00015585350259658746,
-      "loss": 0.0948,
+      "loss": 0.0668,
       "step": 558
     },
     {
       "epoch": 3.15819209039548,
-      "grad_norm": 0.26953125,
+      "grad_norm": 0.212890625,
       "learning_rate": 0.00015570535169372732,
-      "loss": 0.1409,
+      "loss": 0.1309,
       "step": 559
     },
     {
       "epoch": 3.1638418079096047,
-      "grad_norm": 0.41796875,
+      "grad_norm": 0.373046875,
       "learning_rate": 0.00015555702330196023,
-      "loss": 0.1408,
+      "loss": 0.0328,
       "step": 560
     },
     {
       "epoch": 3.169491525423729,
-      "grad_norm": 0.359375,
+      "grad_norm": 0.37890625,
       "learning_rate": 0.00015540851789389145,
-      "loss": 0.0941,
+      "loss": 0.0553,
       "step": 561
     },
     {
       "epoch": 3.175141242937853,
-      "grad_norm": 0.357421875,
+      "grad_norm": 0.5,
       "learning_rate": 0.00015525983594269027,
-      "loss": 0.0609,
+      "loss": 0.0875,
       "step": 562
     },
     {
       "epoch": 3.1807909604519775,
-      "grad_norm": 0.28125,
+      "grad_norm": 0.439453125,
       "learning_rate": 0.00015511097792208852,
-      "loss": 0.148,
+      "loss": 0.0944,
       "step": 563
     },
     {
       "epoch": 3.1864406779661016,
-      "grad_norm": 0.40625,
+      "grad_norm": 0.416015625,
       "learning_rate": 0.00015496194430637903,
-      "loss": 0.1367,
+      "loss": 0.1143,
       "step": 564
     },
     {
       "epoch": 3.1920903954802258,
-      "grad_norm": 0.341796875,
+      "grad_norm": 0.359375,
       "learning_rate": 0.000154812735570414,
-      "loss": 0.0759,
+      "loss": 0.0747,
       "step": 565
     },
     {
       "epoch": 3.1977401129943503,
-      "grad_norm": 0.318359375,
+      "grad_norm": 0.302734375,
       "learning_rate": 0.0001546633521896038,
-      "loss": 0.0805,
+      "loss": 0.1003,
       "step": 566
     },
     {
       "epoch": 3.2033898305084745,
-      "grad_norm": 0.1630859375,
+      "grad_norm": 0.26171875,
       "learning_rate": 0.00015451379463991507,
-      "loss": 0.0831,
+      "loss": 0.0953,
       "step": 567
     },
     {
       "epoch": 3.209039548022599,
-      "grad_norm": 0.359375,
+      "grad_norm": 0.361328125,
       "learning_rate": 0.00015436406339786943,
-      "loss": 0.101,
+      "loss": 0.0811,
       "step": 568
     },
     {
       "epoch": 3.214689265536723,
-      "grad_norm": 0.484375,
+      "grad_norm": 0.3515625,
       "learning_rate": 0.000154214158940542,
-      "loss": 0.0826,
+      "loss": 0.0934,
       "step": 569
     },
     {
       "epoch": 3.2203389830508473,
-      "grad_norm": 0.3828125,
+      "grad_norm": 0.6015625,
       "learning_rate": 0.00015406408174555976,
-      "loss": 0.0643,
+      "loss": 0.0869,
       "step": 570
     },
     {
       "epoch": 3.225988700564972,
-      "grad_norm": 0.3203125,
+      "grad_norm": 0.48828125,
       "learning_rate": 0.00015391383229110007,
-      "loss": 0.1304,
+      "loss": 0.118,
       "step": 571
     },
     {
       "epoch": 3.231638418079096,
-      "grad_norm": 0.279296875,
+      "grad_norm": 0.142578125,
       "learning_rate": 0.00015376341105588902,
-      "loss": 0.0777,
+      "loss": 0.1527,
       "step": 572
     },
     {
       "epoch": 3.23728813559322,
-      "grad_norm": 0.439453125,
+      "grad_norm": 0.419921875,
       "learning_rate": 0.00015361281851920025,
-      "loss": 0.0663,
+      "loss": 0.1209,
       "step": 573
     },
     {
       "epoch": 3.2429378531073447,
-      "grad_norm": 0.384765625,
+      "grad_norm": 0.396484375,
       "learning_rate": 0.00015346205516085306,
-      "loss": 0.0277,
+      "loss": 0.0753,
       "step": 574
     },
     {
       "epoch": 3.248587570621469,
-      "grad_norm": 0.294921875,
+      "grad_norm": 0.390625,
       "learning_rate": 0.00015331112146121104,
-      "loss": 0.0579,
+      "loss": 0.0999,
       "step": 575
     },
     {
       "epoch": 3.2542372881355934,
-      "grad_norm": 0.275390625,
+      "grad_norm": 0.3515625,
       "learning_rate": 0.0001531600179011806,
-      "loss": 0.0714,
+      "loss": 0.0982,
       "step": 576
     },
     {
       "epoch": 3.2598870056497176,
-      "grad_norm": 0.314453125,
+      "grad_norm": 0.2431640625,
       "learning_rate": 0.00015300874496220924,
-      "loss": 0.0746,
+      "loss": 0.1202,
       "step": 577
     },
     {
       "epoch": 3.2655367231638417,
-      "grad_norm": 0.373046875,
+      "grad_norm": 0.4375,
       "learning_rate": 0.0001528573031262842,
-      "loss": 0.0596,
+      "loss": 0.074,
       "step": 578
     },
     {
       "epoch": 3.2711864406779663,
-      "grad_norm": 0.34765625,
+      "grad_norm": 0.296875,
       "learning_rate": 0.00015270569287593091,
-      "loss": 0.0461,
+      "loss": 0.0442,
       "step": 579
     },
     {
       "epoch": 3.2768361581920904,
-      "grad_norm": 0.1689453125,
+      "grad_norm": 0.53125,
       "learning_rate": 0.00015255391469421128,
-      "loss": 0.1111,
+      "loss": 0.0751,
       "step": 580
     },
     {
       "epoch": 3.2824858757062145,
-      "grad_norm": 0.404296875,
+      "grad_norm": 0.314453125,
       "learning_rate": 0.00015240196906472244,
-      "loss": 0.1048,
+      "loss": 0.0583,
       "step": 581
     },
     {
       "epoch": 3.288135593220339,
-      "grad_norm": 0.353515625,
+      "grad_norm": 0.318359375,
       "learning_rate": 0.0001522498564715949,
-      "loss": 0.0627,
+      "loss": 0.044,
       "step": 582
     },
     {
       "epoch": 3.2937853107344632,
-      "grad_norm": 0.271484375,
+      "grad_norm": 0.375,
       "learning_rate": 0.00015209757739949122,
-      "loss": 0.1162,
+      "loss": 0.0423,
       "step": 583
     },
     {
       "epoch": 3.2994350282485874,
-      "grad_norm": 0.65234375,
+      "grad_norm": 0.5078125,
       "learning_rate": 0.0001519451323336044,
-      "loss": 0.0913,
+      "loss": 0.09,
       "step": 584
     },
     {
       "epoch": 3.305084745762712,
-      "grad_norm": 0.326171875,
+      "grad_norm": 0.2353515625,
       "learning_rate": 0.00015179252175965633,
-      "loss": 0.0606,
+      "loss": 0.0597,
       "step": 585
     },
     {
       "epoch": 3.305084745762712,
-      "eval_loss": 0.0192513819783926,
-      "eval_runtime": 5.7206,
-      "eval_samples_per_second": 26.046,
-      "eval_steps_per_second": 3.321,
+      "eval_loss": 0.018856162205338478,
+      "eval_runtime": 5.7947,
+      "eval_samples_per_second": 25.886,
+      "eval_steps_per_second": 3.279,
       "step": 585
     },
     {
       "epoch": 3.310734463276836,
-      "grad_norm": 0.302734375,
+      "grad_norm": 0.486328125,
       "learning_rate": 0.0001516397461638962,
-      "loss": 0.062,
+      "loss": 0.0881,
       "step": 586
     },
     {
       "epoch": 3.3163841807909606,
-      "grad_norm": 0.4609375,
+      "grad_norm": 0.314453125,
       "learning_rate": 0.00015148680603309904,
-      "loss": 0.064,
+      "loss": 0.0397,
       "step": 587
     },
     {
       "epoch": 3.3220338983050848,
-      "grad_norm": 0.3046875,
+      "grad_norm": 0.2578125,
       "learning_rate": 0.00015133370185456415,
-      "loss": 0.0788,
+      "loss": 0.1276,
       "step": 588
     },
     {
       "epoch": 3.327683615819209,
-      "grad_norm": 0.376953125,
+      "grad_norm": 0.228515625,
       "learning_rate": 0.00015118043411611342,
-      "loss": 0.1242,
+      "loss": 0.1303,
       "step": 589
     },
     {
       "epoch": 3.3333333333333335,
-      "grad_norm": 0.5546875,
+      "grad_norm": 0.392578125,
       "learning_rate": 0.00015102700330609,
-      "loss": 0.1125,
+      "loss": 0.0953,
       "step": 590
     },
     {
       "epoch": 3.3389830508474576,
-      "grad_norm": 0.361328125,
+      "grad_norm": 0.294921875,
       "learning_rate": 0.00015087340991335645,
-      "loss": 0.0587,
+      "loss": 0.0855,
       "step": 591
     },
     {
       "epoch": 3.3446327683615817,
       "grad_norm": 0.33984375,
       "learning_rate": 0.00015071965442729356,
-      "loss": 0.1133,
+      "loss": 0.1481,
       "step": 592
     },
     {
       "epoch": 3.3502824858757063,
-      "grad_norm": 0.2578125,
+      "grad_norm": 0.265625,
       "learning_rate": 0.00015056573733779848,
-      "loss": 0.0915,
+      "loss": 0.1141,
       "step": 593
     },
     {
       "epoch": 3.3559322033898304,
-      "grad_norm": 0.3359375,
+      "grad_norm": 0.32421875,
       "learning_rate": 0.0001504116591352832,
-      "loss": 0.0978,
+      "loss": 0.0667,
       "step": 594
     },
     {
       "epoch": 3.361581920903955,
-      "grad_norm": 0.369140625,
+      "grad_norm": 0.58203125,
       "learning_rate": 0.00015025742031067317,
-      "loss": 0.066,
+      "loss": 0.0623,
       "step": 595
     },
     {
       "epoch": 3.367231638418079,
-      "grad_norm": 0.365234375,
+      "grad_norm": 0.4609375,
       "learning_rate": 0.00015010302135540552,
-      "loss": 0.0638,
+      "loss": 0.0465,
       "step": 596
     },
     {
       "epoch": 3.3728813559322033,
-      "grad_norm": 0.28515625,
+      "grad_norm": 0.3046875,
       "learning_rate": 0.00014994846276142766,
-      "loss": 0.0697,
+      "loss": 0.0463,
       "step": 597
     },
     {
       "epoch": 3.378531073446328,
-      "grad_norm": 0.328125,
+      "grad_norm": 0.359375,
       "learning_rate": 0.0001497937450211956,
-      "loss": 0.0812,
+      "loss": 0.1036,
       "step": 598
     },
     {
       "epoch": 3.384180790960452,
-      "grad_norm": 0.37109375,
+      "grad_norm": 0.345703125,
       "learning_rate": 0.00014963886862767245,
-      "loss": 0.1016,
+      "loss": 0.1135,
       "step": 599
     },
     {
       "epoch": 3.389830508474576,
-      "grad_norm": 0.18359375,
+      "grad_norm": 0.28515625,
       "learning_rate": 0.00014948383407432678,
-      "loss": 0.1218,
+      "loss": 0.1076,
       "step": 600
     },
     {
       "epoch": 3.3954802259887007,
-      "grad_norm": 0.40625,
+      "grad_norm": 0.306640625,
       "learning_rate": 0.00014932864185513114,
-      "loss": 0.0634,
+      "loss": 0.0846,
       "step": 601
     },
     {
       "epoch": 3.401129943502825,
-      "grad_norm": 0.419921875,
+      "grad_norm": 0.353515625,
       "learning_rate": 0.0001491732924645604,
-      "loss": 0.0656,
+      "loss": 0.0951,
       "step": 602
     },
     {
       "epoch": 3.406779661016949,
-      "grad_norm": 0.32421875,
+      "grad_norm": 0.353515625,
       "learning_rate": 0.00014901778639759023,
-      "loss": 0.0419,
+      "loss": 0.0412,
       "step": 603
     },
     {
       "epoch": 3.4124293785310735,
-      "grad_norm": 0.384765625,
+      "grad_norm": 0.36328125,
       "learning_rate": 0.00014886212414969553,
-      "loss": 0.0839,
+      "loss": 0.0988,
       "step": 604
     },
     {
       "epoch": 3.4180790960451977,
-      "grad_norm": 0.3359375,
+      "grad_norm": 0.296875,
       "learning_rate": 0.00014870630621684872,
-      "loss": 0.0694,
+      "loss": 0.112,
       "step": 605
     },
     {
       "epoch": 3.423728813559322,
-      "grad_norm": 0.259765625,
+      "grad_norm": 0.333984375,
       "learning_rate": 0.00014855033309551843,
-      "loss": 0.0961,
+      "loss": 0.1076,
       "step": 606
     },
     {
       "epoch": 3.4293785310734464,
-      "grad_norm": 0.291015625,
+      "grad_norm": 0.3125,
       "learning_rate": 0.0001483942052826676,
-      "loss": 0.0829,
+      "loss": 0.0585,
       "step": 607
     },
     {
       "epoch": 3.4350282485875705,
-      "grad_norm": 0.1943359375,
+      "grad_norm": 0.2890625,
       "learning_rate": 0.0001482379232757521,
-      "loss": 0.1282,
+      "loss": 0.0607,
       "step": 608
     },
     {
       "epoch": 3.440677966101695,
-      "grad_norm": 0.1640625,
+      "grad_norm": 0.2255859375,
       "learning_rate": 0.0001480814875727192,
-      "loss": 0.06,
+      "loss": 0.0432,
       "step": 609
     },
     {
       "epoch": 3.446327683615819,
-      "grad_norm": 0.34375,
+      "grad_norm": 0.275390625,
       "learning_rate": 0.0001479248986720057,
-      "loss": 0.0933,
+      "loss": 0.094,
       "step": 610
     },
     {
       "epoch": 3.4519774011299433,
-      "grad_norm": 0.443359375,
+      "grad_norm": 0.1875,
       "learning_rate": 0.00014776815707253662,
-      "loss": 0.0927,
+      "loss": 0.0705,
       "step": 611
     },
     {
       "epoch": 3.457627118644068,
-      "grad_norm": 0.400390625,
+      "grad_norm": 0.41015625,
       "learning_rate": 0.00014761126327372348,
-      "loss": 0.0504,
+      "loss": 0.0887,
       "step": 612
     },
     {
       "epoch": 3.463276836158192,
-      "grad_norm": 0.19140625,
+      "grad_norm": 0.349609375,
       "learning_rate": 0.00014745421777546283,
-      "loss": 0.1507,
+      "loss": 0.088,
       "step": 613
     },
     {
       "epoch": 3.4689265536723166,
-      "grad_norm": 0.314453125,
+      "grad_norm": 0.435546875,
       "learning_rate": 0.00014729702107813436,
-      "loss": 0.0613,
+      "loss": 0.0329,
       "step": 614
     },
     {
       "epoch": 3.4745762711864407,
-      "grad_norm": 0.423828125,
+      "grad_norm": 0.279296875,
       "learning_rate": 0.0001471396736825998,
-      "loss": 0.0827,
+      "loss": 0.0752,
       "step": 615
     },
     {
       "epoch": 3.480225988700565,
-      "grad_norm": 0.390625,
+      "grad_norm": 0.3359375,
       "learning_rate": 0.00014698217609020073,
-      "loss": 0.0631,
+      "loss": 0.1034,
       "step": 616
     },
     {
       "epoch": 3.4858757062146895,
-      "grad_norm": 0.52734375,
+      "grad_norm": 0.291015625,
       "learning_rate": 0.0001468245288027575,
-      "loss": 0.1119,
+      "loss": 0.1112,
       "step": 617
     },
     {
       "epoch": 3.4915254237288136,
-      "grad_norm": 0.306640625,
+      "grad_norm": 0.298828125,
       "learning_rate": 0.00014666673232256738,
-      "loss": 0.0653,
+      "loss": 0.0788,
       "step": 618
     },
     {
       "epoch": 3.4971751412429377,
-      "grad_norm": 0.34765625,
+      "grad_norm": 0.2578125,
       "learning_rate": 0.00014650878715240294,
-      "loss": 0.0488,
+      "loss": 0.0717,
       "step": 619
     },
     {
       "epoch": 3.5028248587570623,
-      "grad_norm": 0.263671875,
+      "grad_norm": 0.193359375,
       "learning_rate": 0.00014635069379551055,
-      "loss": 0.0818,
+      "loss": 0.0742,
       "step": 620
     },
     {
       "epoch": 3.5084745762711864,
-      "grad_norm": 0.3515625,
+      "grad_norm": 0.29296875,
       "learning_rate": 0.00014619245275560876,
-      "loss": 0.0823,
+      "loss": 0.0266,
       "step": 621
     },
     {
       "epoch": 3.5141242937853105,
-      "grad_norm": 0.375,
+      "grad_norm": 0.236328125,
       "learning_rate": 0.00014603406453688654,
-      "loss": 0.0643,
+      "loss": 0.0258,
       "step": 622
     },
     {
       "epoch": 3.519774011299435,
-      "grad_norm": 0.23046875,
+      "grad_norm": 0.77734375,
       "learning_rate": 0.000145875529644002,
-      "loss": 0.0793,
+      "loss": 0.1026,
       "step": 623
     },
     {
       "epoch": 3.5254237288135593,
-      "grad_norm": 0.26171875,
+      "grad_norm": 0.419921875,
       "learning_rate": 0.00014571684858208044,
-      "loss": 0.0723,
+      "loss": 0.0333,
       "step": 624
     },
     {
       "epoch": 3.5310734463276834,
-      "grad_norm": 0.16015625,
+      "grad_norm": 0.2177734375,
       "learning_rate": 0.00014555802185671298,
-      "loss": 0.1233,
+      "loss": 0.0876,
       "step": 625
     },
     {
       "epoch": 3.536723163841808,
-      "grad_norm": 0.302734375,
+      "grad_norm": 0.23046875,
       "learning_rate": 0.00014539904997395468,
-      "loss": 0.0386,
+      "loss": 0.0657,
       "step": 626
     },
     {
       "epoch": 3.542372881355932,
-      "grad_norm": 0.458984375,
+      "grad_norm": 0.1455078125,
       "learning_rate": 0.00014523993344032335,
-      "loss": 0.0814,
+      "loss": 0.0986,
       "step": 627
     },
     {
       "epoch": 3.548022598870056,
-      "grad_norm": 0.2314453125,
+      "grad_norm": 0.2890625,
       "learning_rate": 0.0001450806727627975,
-      "loss": 0.1336,
+      "loss": 0.1032,
       "step": 628
     },
     {
       "epoch": 3.553672316384181,
-      "grad_norm": 0.1962890625,
+      "grad_norm": 0.345703125,
       "learning_rate": 0.00014492126844881494,
-      "loss": 0.0603,
+      "loss": 0.0957,
       "step": 629
     },
     {
       "epoch": 3.559322033898305,
-      "grad_norm": 0.22265625,
+      "grad_norm": 0.31640625,
       "learning_rate": 0.00014476172100627127,
-      "loss": 0.0568,
+      "loss": 0.1038,
       "step": 630
     },
     {
       "epoch": 3.559322033898305,
-      "eval_loss": 0.013199886307120323,
-      "eval_runtime": 5.733,
-      "eval_samples_per_second": 25.99,
-      "eval_steps_per_second": 3.314,
+      "eval_loss": 0.012973220087587833,
+      "eval_runtime": 5.959,
+      "eval_samples_per_second": 25.172,
+      "eval_steps_per_second": 3.188,
       "step": 630
     },
     {
       "epoch": 3.5649717514124295,
-      "grad_norm": 0.388671875,
+      "grad_norm": 0.2353515625,
       "learning_rate": 0.00014460203094351795,
-      "loss": 0.0626,
+      "loss": 0.1189,
       "step": 631
     },
     {
       "epoch": 3.5706214689265536,
-      "grad_norm": 0.310546875,
+      "grad_norm": 0.390625,
       "learning_rate": 0.00014444219876936097,
-      "loss": 0.0637,
+      "loss": 0.0594,
       "step": 632
     },
     {
       "epoch": 3.576271186440678,
-      "grad_norm": 0.27734375,
+      "grad_norm": 0.380859375,
       "learning_rate": 0.00014428222499305904,
-      "loss": 0.0746,
+      "loss": 0.0736,
       "step": 633
     },
     {
       "epoch": 3.5819209039548023,
-      "grad_norm": 0.263671875,
+      "grad_norm": 0.44140625,
       "learning_rate": 0.00014412211012432212,
-      "loss": 0.0988,
+      "loss": 0.1071,
       "step": 634
     },
     {
       "epoch": 3.5875706214689265,
-      "grad_norm": 0.255859375,
+      "grad_norm": 0.287109375,
       "learning_rate": 0.00014396185467330973,
-      "loss": 0.1188,
+      "loss": 0.0583,
       "step": 635
     },
     {
       "epoch": 3.593220338983051,
-      "grad_norm": 0.345703125,
+      "grad_norm": 0.283203125,
       "learning_rate": 0.00014380145915062918,
-      "loss": 0.1147,
+      "loss": 0.1013,
       "step": 636
     },
     {
       "epoch": 3.598870056497175,
-      "grad_norm": 0.3046875,
+      "grad_norm": 0.2119140625,
       "learning_rate": 0.0001436409240673342,
-      "loss": 0.061,
+      "loss": 0.1301,
       "step": 637
     },
     {
       "epoch": 3.6045197740112993,
-      "grad_norm": 0.322265625,
+      "grad_norm": 0.1650390625,
       "learning_rate": 0.0001434802499349232,
-      "loss": 0.1477,
+      "loss": 0.1108,
       "step": 638
     },
     {
       "epoch": 3.610169491525424,
-      "grad_norm": 0.2890625,
+      "grad_norm": 0.302734375,
       "learning_rate": 0.00014331943726533756,
-      "loss": 0.1111,
+      "loss": 0.1154,
       "step": 639
     },
     {
       "epoch": 3.615819209039548,
-      "grad_norm": 0.2392578125,
+      "grad_norm": 0.341796875,
       "learning_rate": 0.00014315848657096004,
-      "loss": 0.0392,
+      "loss": 0.0425,
       "step": 640
     },
     {
       "epoch": 3.621468926553672,
-      "grad_norm": 0.2275390625,
+      "grad_norm": 0.435546875,
       "learning_rate": 0.00014299739836461336,
-      "loss": 0.0255,
+      "loss": 0.1122,
       "step": 641
     },
     {
       "epoch": 3.6271186440677967,
-      "grad_norm": 0.1923828125,
+      "grad_norm": 0.1572265625,
       "learning_rate": 0.00014283617315955814,
-      "loss": 0.1162,
+      "loss": 0.1051,
       "step": 642
     },
     {
       "epoch": 3.632768361581921,
-      "grad_norm": 0.2294921875,
+      "grad_norm": 0.404296875,
       "learning_rate": 0.00014267481146949167,
-      "loss": 0.0451,
+      "loss": 0.0841,
       "step": 643
     },
     {
       "epoch": 3.638418079096045,
-      "grad_norm": 0.201171875,
+      "grad_norm": 0.2314453125,
       "learning_rate": 0.00014251331380854603,
-      "loss": 0.0784,
+      "loss": 0.0537,
       "step": 644
     },
     {
       "epoch": 3.6440677966101696,
-      "grad_norm": 0.23046875,
+      "grad_norm": 0.390625,
       "learning_rate": 0.00014235168069128657,
-      "loss": 0.0858,
+      "loss": 0.027,
       "step": 645
     },
     {
       "epoch": 3.6497175141242937,
-      "grad_norm": 0.40625,
+      "grad_norm": 0.208984375,
       "learning_rate": 0.00014218991263271024,
-      "loss": 0.0614,
+      "loss": 0.0858,
       "step": 646
     },
     {
       "epoch": 3.655367231638418,
-      "grad_norm": 0.1572265625,
+      "grad_norm": 0.2021484375,
       "learning_rate": 0.00014202801014824394,
-      "loss": 0.0806,
+      "loss": 0.1138,
       "step": 647
     },
     {
       "epoch": 3.6610169491525424,
-      "grad_norm": 0.37109375,
+      "grad_norm": 0.25390625,
       "learning_rate": 0.0001418659737537428,
-      "loss": 0.0983,
+      "loss": 0.121,
       "step": 648
     },
     {
       "epoch": 3.6666666666666665,
-      "grad_norm": 0.302734375,
+      "grad_norm": 0.314453125,
       "learning_rate": 0.00014170380396548877,
-      "loss": 0.0381,
+      "loss": 0.0656,
       "step": 649
     },
     {
       "epoch": 3.672316384180791,
-      "grad_norm": 0.283203125,
+      "grad_norm": 0.171875,
       "learning_rate": 0.00014154150130018866,
-      "loss": 0.071,
+      "loss": 0.087,
       "step": 650
     },
     {
       "epoch": 3.6779661016949152,
-      "grad_norm": 0.33984375,
+      "grad_norm": 0.328125,
       "learning_rate": 0.00014137906627497272,
-      "loss": 0.0957,
+      "loss": 0.1003,
       "step": 651
     },
     {
       "epoch": 3.68361581920904,
-      "grad_norm": 0.265625,
+      "grad_norm": 0.2275390625,
       "learning_rate": 0.00014121649940739301,
-      "loss": 0.0555,
+      "loss": 0.061,
       "step": 652
     },
     {
       "epoch": 3.689265536723164,
-      "grad_norm": 0.1669921875,
+      "grad_norm": 0.275390625,
       "learning_rate": 0.0001410538012154215,
-      "loss": 0.1616,
+      "loss": 0.1431,
       "step": 653
     },
     {
       "epoch": 3.694915254237288,
-      "grad_norm": 0.28125,
+      "grad_norm": 0.2119140625,
       "learning_rate": 0.00014089097221744868,
-      "loss": 0.0231,
+      "loss": 0.1552,
       "step": 654
     },
     {
       "epoch": 3.7005649717514126,
-      "grad_norm": 0.12255859375,
+      "grad_norm": 0.2431640625,
       "learning_rate": 0.00014072801293228188,
-      "loss": 0.1052,
+      "loss": 0.078,
       "step": 655
     },
     {
       "epoch": 3.7062146892655368,
-      "grad_norm": 0.1640625,
+      "grad_norm": 0.2138671875,
       "learning_rate": 0.0001405649238791434,
-      "loss": 0.074,
+      "loss": 0.0745,
       "step": 656
     },
     {
       "epoch": 3.711864406779661,
-      "grad_norm": 0.27734375,
+      "grad_norm": 0.171875,
       "learning_rate": 0.00014040170557766917,
-      "loss": 0.0758,
+      "loss": 0.0428,
       "step": 657
     },
     {
       "epoch": 3.7175141242937855,
-      "grad_norm": 0.228515625,
+      "grad_norm": 0.267578125,
       "learning_rate": 0.0001402383585479068,
-      "loss": 0.074,
+      "loss": 0.1354,
       "step": 658
     },
     {
       "epoch": 3.7231638418079096,
-      "grad_norm": 0.32421875,
+      "grad_norm": 0.40234375,
       "learning_rate": 0.0001400748833103141,
-      "loss": 0.0456,
+      "loss": 0.0354,
       "step": 659
     },
     {
       "epoch": 3.7288135593220337,
-      "grad_norm": 0.296875,
+      "grad_norm": 0.3046875,
       "learning_rate": 0.00013991128038575741,
-      "loss": 0.0989,
+      "loss": 0.1629,
       "step": 660
     },
     {
       "epoch": 3.7344632768361583,
-      "grad_norm": 0.16796875,
+      "grad_norm": 0.306640625,
       "learning_rate": 0.0001397475502955099,
-      "loss": 0.1154,
+      "loss": 0.1283,
       "step": 661
     },
     {
       "epoch": 3.7401129943502824,
-      "grad_norm": 0.37890625,
+      "grad_norm": 0.28515625,
       "learning_rate": 0.00013958369356124986,
-      "loss": 0.0783,
+      "loss": 0.0372,
       "step": 662
     },
     {
       "epoch": 3.7457627118644066,
-      "grad_norm": 0.2275390625,
+      "grad_norm": 0.1904296875,
       "learning_rate": 0.00013941971070505915,
-      "loss": 0.1046,
+      "loss": 0.0569,
       "step": 663
     },
     {
       "epoch": 3.751412429378531,
-      "grad_norm": 0.40234375,
+      "grad_norm": 0.09814453125,
       "learning_rate": 0.00013925560224942144,
-      "loss": 0.1102,
+      "loss": 0.1149,
       "step": 664
     },
     {
       "epoch": 3.7570621468926553,
-      "grad_norm": 0.24609375,
+      "grad_norm": 0.380859375,
       "learning_rate": 0.00013909136871722067,
-      "loss": 0.1173,
+      "loss": 0.0465,
       "step": 665
     },
     {
       "epoch": 3.7627118644067794,
-      "grad_norm": 0.328125,
+      "grad_norm": 0.171875,
       "learning_rate": 0.00013892701063173918,
-      "loss": 0.1176,
+      "loss": 0.0899,
       "step": 666
     },
     {
       "epoch": 3.768361581920904,
-      "grad_norm": 0.142578125,
+      "grad_norm": 0.32421875,
       "learning_rate": 0.00013876252851665623,
-      "loss": 0.0927,
+      "loss": 0.0765,
       "step": 667
     },
     {
       "epoch": 3.774011299435028,
-      "grad_norm": 0.314453125,
+      "grad_norm": 0.32421875,
       "learning_rate": 0.00013859792289604628,
-      "loss": 0.1375,
+      "loss": 0.1139,
       "step": 668
     },
     {
       "epoch": 3.7796610169491527,
-      "grad_norm": 0.455078125,
+      "grad_norm": 0.2421875,
       "learning_rate": 0.00013843319429437728,
-      "loss": 0.0257,
+      "loss": 0.0433,
       "step": 669
     },
     {
       "epoch": 3.785310734463277,
-      "grad_norm": 0.34765625,
+      "grad_norm": 0.171875,
       "learning_rate": 0.000138268343236509,
-      "loss": 0.1077,
+      "loss": 0.0218,
       "step": 670
     },
     {
       "epoch": 3.7909604519774014,
-      "grad_norm": 0.1943359375,
+      "grad_norm": 0.2890625,
       "learning_rate": 0.0001381033702476914,
-      "loss": 0.0389,
+      "loss": 0.0523,
       "step": 671
     },
     {
       "epoch": 3.7966101694915255,
-      "grad_norm": 0.34375,
+      "grad_norm": 0.375,
       "learning_rate": 0.000137938275853563,
-      "loss": 0.0416,
+      "loss": 0.0457,
       "step": 672
     },
     {
       "epoch": 3.8022598870056497,
-      "grad_norm": 0.134765625,
+      "grad_norm": 0.1787109375,
       "learning_rate": 0.00013777306058014905,
-      "loss": 0.1873,
+      "loss": 0.1648,
       "step": 673
     },
     {
       "epoch": 3.8079096045197742,
-      "grad_norm": 0.27734375,
+      "grad_norm": 0.29296875,
       "learning_rate": 0.00013760772495385998,
-      "loss": 0.145,
+      "loss": 0.0392,
       "step": 674
     },
     {
       "epoch": 3.8135593220338984,
-      "grad_norm": 0.279296875,
+      "grad_norm": 0.216796875,
       "learning_rate": 0.00013744226950148972,
-      "loss": 0.063,
+      "loss": 0.0754,
       "step": 675
     },
     {
       "epoch": 3.8135593220338984,
-      "eval_loss": 0.006439562421292067,
-      "eval_runtime": 5.7798,
-      "eval_samples_per_second": 25.779,
-      "eval_steps_per_second": 3.287,
+      "eval_loss": 0.00783731322735548,
+      "eval_runtime": 5.7971,
+      "eval_samples_per_second": 25.875,
+      "eval_steps_per_second": 3.277,
       "step": 675
     },
     {
       "epoch": 3.8192090395480225,
-      "grad_norm": 0.154296875,
+      "grad_norm": 0.25390625,
       "learning_rate": 0.0001372766947502139,
-      "loss": 0.1658,
+      "loss": 0.0426,
       "step": 676
     },
     {
       "epoch": 3.824858757062147,
-      "grad_norm": 0.1982421875,
+      "grad_norm": 0.2578125,
       "learning_rate": 0.00013711100122758838,
-      "loss": 0.1096,
+      "loss": 0.1097,
       "step": 677
     },
     {
       "epoch": 3.830508474576271,
-      "grad_norm": 0.2890625,
+      "grad_norm": 0.2578125,
       "learning_rate": 0.0001369451894615474,
-      "loss": 0.0599,
+      "loss": 0.0961,
       "step": 678
     },
     {
       "epoch": 3.8361581920903953,
-      "grad_norm": 0.28125,
+      "grad_norm": 0.13671875,
       "learning_rate": 0.00013677925998040188,
-      "loss": 0.0801,
+      "loss": 0.0195,
       "step": 679
     },
     {
       "epoch": 3.84180790960452,
-      "grad_norm": 0.2294921875,
+      "grad_norm": 0.37890625,
       "learning_rate": 0.00013661321331283796,
-      "loss": 0.0532,
+      "loss": 0.1321,
       "step": 680
     },
     {
       "epoch": 3.847457627118644,
-      "grad_norm": 0.1669921875,
+      "grad_norm": 0.146484375,
       "learning_rate": 0.000136447049987915,
-      "loss": 0.0578,
+      "loss": 0.0208,
       "step": 681
     },
     {
       "epoch": 3.853107344632768,
-      "grad_norm": 0.310546875,
+      "grad_norm": 0.185546875,
       "learning_rate": 0.0001362807705350641,
-      "loss": 0.0439,
+      "loss": 0.0375,
       "step": 682
     },
     {
       "epoch": 3.8587570621468927,
-      "grad_norm": 0.236328125,
+      "grad_norm": 0.1962890625,
       "learning_rate": 0.00013611437548408648,
-      "loss": 0.0423,
+      "loss": 0.0656,
       "step": 683
     },
     {
       "epoch": 3.864406779661017,
-      "grad_norm": 0.376953125,
+      "grad_norm": 0.1884765625,
       "learning_rate": 0.00013594786536515153,
-      "loss": 0.1319,
+      "loss": 0.0387,
       "step": 684
     },
     {
       "epoch": 3.870056497175141,
-      "grad_norm": 0.1875,
+      "grad_norm": 0.265625,
       "learning_rate": 0.00013578124070879534,
-      "loss": 0.0633,
+      "loss": 0.0663,
       "step": 685
     },
     {
       "epoch": 3.8757062146892656,
-      "grad_norm": 0.1767578125,
+      "grad_norm": 0.28515625,
       "learning_rate": 0.00013561450204591896,
-      "loss": 0.1335,
+      "loss": 0.0773,
       "step": 686
     },
     {
       "epoch": 3.8813559322033897,
-      "grad_norm": 0.09716796875,
+      "grad_norm": 0.12060546875,
       "learning_rate": 0.00013544764990778662,
-      "loss": 0.0598,
+      "loss": 0.0973,
       "step": 687
     },
     {
       "epoch": 3.8870056497175143,
-      "grad_norm": 0.1689453125,
+      "grad_norm": 0.5078125,
       "learning_rate": 0.00013528068482602417,
-      "loss": 0.0827,
+      "loss": 0.0413,
       "step": 688
     },
     {
       "epoch": 3.8926553672316384,
-      "grad_norm": 0.2470703125,
+      "grad_norm": 0.439453125,
       "learning_rate": 0.00013511360733261732,
-      "loss": 0.096,
+      "loss": 0.0651,
       "step": 689
     },
     {
       "epoch": 3.898305084745763,
-      "grad_norm": 0.171875,
+      "grad_norm": 0.1982421875,
       "learning_rate": 0.00013494641795990986,
-      "loss": 0.0548,
+      "loss": 0.1238,
       "step": 690
     },
     {
       "epoch": 3.903954802259887,
-      "grad_norm": 0.2890625,
+      "grad_norm": 0.2353515625,
       "learning_rate": 0.00013477911724060215,
-      "loss": 0.0967,
+      "loss": 0.0882,
       "step": 691
     },
     {
       "epoch": 3.9096045197740112,
-      "grad_norm": 0.1533203125,
+      "grad_norm": 0.130859375,
       "learning_rate": 0.0001346117057077493,
-      "loss": 0.0439,
+      "loss": 0.0869,
       "step": 692
     },
     {
       "epoch": 3.915254237288136,
-      "grad_norm": 0.10986328125,
+      "grad_norm": 0.26953125,
       "learning_rate": 0.0001344441838947595,
-      "loss": 0.0987,
+      "loss": 0.0186,
       "step": 693
     },
     {
       "epoch": 3.92090395480226,
-      "grad_norm": 0.166015625,
+      "grad_norm": 0.2041015625,
       "learning_rate": 0.00013427655233539228,
-      "loss": 0.0854,
+      "loss": 0.0428,
       "step": 694
     },
     {
       "epoch": 3.926553672316384,
-      "grad_norm": 0.197265625,
+      "grad_norm": 0.1728515625,
       "learning_rate": 0.00013410881156375684,
-      "loss": 0.0575,
+      "loss": 0.1084,
       "step": 695
     },
     {
       "epoch": 3.9322033898305087,
-      "grad_norm": 0.1982421875,
+      "grad_norm": 0.26953125,
       "learning_rate": 0.00013394096211431043,
-      "loss": 0.0359,
+      "loss": 0.0992,
       "step": 696
     },
     {
       "epoch": 3.937853107344633,
-      "grad_norm": 0.376953125,
+      "grad_norm": 0.384765625,
       "learning_rate": 0.00013377300452185645,
-      "loss": 0.0347,
+      "loss": 0.0464,
       "step": 697
     },
     {
       "epoch": 3.943502824858757,
-      "grad_norm": 0.251953125,
+      "grad_norm": 0.404296875,
       "learning_rate": 0.00013360493932154302,
-      "loss": 0.0036,
+      "loss": 0.0053,
       "step": 698
     },
     {
       "epoch": 3.9491525423728815,
-      "grad_norm": 0.318359375,
+      "grad_norm": 0.3828125,
       "learning_rate": 0.000133436767048861,
-      "loss": 0.0902,
+      "loss": 0.0805,
       "step": 699
     },
     {
       "epoch": 3.9548022598870056,
-      "grad_norm": 0.2099609375,
+      "grad_norm": 0.146484375,
       "learning_rate": 0.00013326848823964243,
-      "loss": 0.0189,
+      "loss": 0.0367,
       "step": 700
     },
     {
       "epoch": 3.9604519774011298,
-      "grad_norm": 0.271484375,
+      "grad_norm": 0.2392578125,
       "learning_rate": 0.00013310010343005877,
-      "loss": 0.0921,
+      "loss": 0.0769,
       "step": 701
     },
     {
       "epoch": 3.9661016949152543,
-      "grad_norm": 0.1083984375,
+      "grad_norm": 0.09423828125,
       "learning_rate": 0.00013293161315661931,
-      "loss": 0.1019,
+      "loss": 0.1547,
       "step": 702
     },
     {
       "epoch": 3.9717514124293785,
-      "grad_norm": 0.185546875,
+      "grad_norm": 0.1376953125,
       "learning_rate": 0.00013276301795616936,
-      "loss": 0.0617,
+      "loss": 0.0952,
       "step": 703
     },
     {
       "epoch": 3.9774011299435026,
-      "grad_norm": 0.2109375,
+      "grad_norm": 0.267578125,
       "learning_rate": 0.00013259431836588843,
-      "loss": 0.1172,
+      "loss": 0.0598,
       "step": 704
     },
     {
       "epoch": 3.983050847457627,
-      "grad_norm": 0.2734375,
+      "grad_norm": 0.30078125,
       "learning_rate": 0.00013242551492328875,
-      "loss": 0.0356,
+      "loss": 0.0402,
       "step": 705
     },
     {
       "epoch": 3.9887005649717513,
-      "grad_norm": 0.1982421875,
+      "grad_norm": 0.28515625,
       "learning_rate": 0.0001322566081662134,
-      "loss": 0.0212,
+      "loss": 0.0601,
       "step": 706
     },
     {
       "epoch": 3.994350282485876,
-      "grad_norm": 0.25390625,
+      "grad_norm": 0.40625,
       "learning_rate": 0.00013208759863283465,
-      "loss": 0.0653,
+      "loss": 0.0757,
       "step": 707
     },
     {
       "epoch": 4.0,
-      "grad_norm": 0.1328125,
+      "grad_norm": 0.2197265625,
       "learning_rate": 0.00013191848686165228,
-      "loss": 0.0962,
+      "loss": 0.0977,
       "step": 708
     },
     {
       "epoch": 4.005649717514125,
-      "grad_norm": 0.060302734375,
+      "grad_norm": 0.15625,
       "learning_rate": 0.0001317492733914917,
-      "loss": 0.0397,
+      "loss": 0.0486,
       "step": 709
     },
     {
       "epoch": 4.011299435028248,
-      "grad_norm": 0.1240234375,
+      "grad_norm": 0.2158203125,
       "learning_rate": 0.0001315799587615025,
-      "loss": 0.0543,
+      "loss": 0.0962,
       "step": 710
     },
     {
       "epoch": 4.016949152542373,
-      "grad_norm": 0.08935546875,
+      "grad_norm": 0.142578125,
       "learning_rate": 0.0001314105435111565,
-      "loss": 0.1115,
+      "loss": 0.0531,
       "step": 711
     },
     {
       "epoch": 4.022598870056497,
-      "grad_norm": 0.1513671875,
+      "grad_norm": 0.1533203125,
       "learning_rate": 0.00013124102818024615,
-      "loss": 0.0708,
+      "loss": 0.0544,
       "step": 712
     },
     {
       "epoch": 4.028248587570621,
-      "grad_norm": 0.09814453125,
+      "grad_norm": 0.3515625,
       "learning_rate": 0.00013107141330888272,
-      "loss": 0.0748,
+      "loss": 0.0617,
       "step": 713
     },
     {
       "epoch": 4.033898305084746,
-      "grad_norm": 0.1748046875,
+      "grad_norm": 0.29296875,
       "learning_rate": 0.00013090169943749476,
-      "loss": 0.0638,
+      "loss": 0.0379,
       "step": 714
     },
     {
       "epoch": 4.03954802259887,
-      "grad_norm": 0.1962890625,
+      "grad_norm": 0.2060546875,
       "learning_rate": 0.0001307318871068261,
-      "loss": 0.0759,
+      "loss": 0.0853,
       "step": 715
     },
     {
       "epoch": 4.045197740112994,
-      "grad_norm": 0.1376953125,
+      "grad_norm": 0.27734375,
       "learning_rate": 0.0001305619768579344,
-      "loss": 0.0534,
+      "loss": 0.0755,
       "step": 716
     },
     {
       "epoch": 4.0508474576271185,
-      "grad_norm": 0.10400390625,
+      "grad_norm": 0.25390625,
       "learning_rate": 0.0001303919692321893,
-      "loss": 0.1504,
+      "loss": 0.0934,
       "step": 717
     },
     {
       "epoch": 4.056497175141243,
-      "grad_norm": 0.078125,
+      "grad_norm": 0.1318359375,
       "learning_rate": 0.0001302218647712706,
-      "loss": 0.0639,
+      "loss": 0.0711,
       "step": 718
     },
     {
       "epoch": 4.062146892655368,
-      "grad_norm": 0.1513671875,
+      "grad_norm": 0.15625,
       "learning_rate": 0.0001300516640171667,
-      "loss": 0.1144,
+      "loss": 0.0818,
       "step": 719
     },
     {
       "epoch": 4.067796610169491,
-      "grad_norm": 0.1875,
+      "grad_norm": 0.1416015625,
       "learning_rate": 0.00012988136751217291,
-      "loss": 0.081,
+      "loss": 0.0632,
       "step": 720
     },
     {
       "epoch": 4.067796610169491,
-      "eval_loss": 0.0039033321663737297,
-      "eval_runtime": 6.167,
-      "eval_samples_per_second": 24.161,
-      "eval_steps_per_second": 3.081,
+      "eval_loss": 0.005122837144881487,
+      "eval_runtime": 5.8351,
+      "eval_samples_per_second": 25.706,
+      "eval_steps_per_second": 3.256,
       "step": 720
     },
     {
       "epoch": 4.073446327683616,
-      "grad_norm": 0.1787109375,
+      "grad_norm": 0.10791015625,
       "learning_rate": 0.00012971097579888948,
-      "loss": 0.0394,
+      "loss": 0.039,
       "step": 721
     },
     {
       "epoch": 4.0790960451977405,
-      "grad_norm": 0.43359375,
+      "grad_norm": 0.169921875,
       "learning_rate": 0.00012954048942022002,
-      "loss": 0.0537,
+      "loss": 0.1026,
       "step": 722
     },
     {
       "epoch": 4.084745762711864,
-      "grad_norm": 0.10546875,
+      "grad_norm": 0.2138671875,
       "learning_rate": 0.0001293699089193699,
-      "loss": 0.0502,
+      "loss": 0.0827,
       "step": 723
     },
     {
       "epoch": 4.090395480225989,
-      "grad_norm": 0.14453125,
+      "grad_norm": 0.1953125,
       "learning_rate": 0.00012919923483984414,
-      "loss": 0.0204,
+      "loss": 0.0578,
       "step": 724
     },
     {
       "epoch": 4.096045197740113,
-      "grad_norm": 0.1884765625,
+      "grad_norm": 0.0712890625,
       "learning_rate": 0.00012902846772544624,
-      "loss": 0.0976,
+      "loss": 0.0412,
       "step": 725
     },
     {
       "epoch": 4.101694915254237,
-      "grad_norm": 0.158203125,
+      "grad_norm": 0.142578125,
       "learning_rate": 0.00012885760812027589,
-      "loss": 0.0969,
+      "loss": 0.1037,
       "step": 726
     },
     {
       "epoch": 4.107344632768362,
-      "grad_norm": 0.3203125,
+      "grad_norm": 0.1416015625,
       "learning_rate": 0.0001286866565687275,
-      "loss": 0.09,
+      "loss": 0.0529,
       "step": 727
     },
     {
       "epoch": 4.112994350282486,
-      "grad_norm": 0.216796875,
+      "grad_norm": 0.447265625,
       "learning_rate": 0.00012851561361548858,
-      "loss": 0.0391,
+      "loss": 0.0724,
       "step": 728
     },
     {
       "epoch": 4.11864406779661,
-      "grad_norm": 0.08349609375,
+      "grad_norm": 0.2216796875,
       "learning_rate": 0.00012834447980553768,
-      "loss": 0.1024,
+      "loss": 0.1165,
       "step": 729
     },
     {
       "epoch": 4.124293785310734,
-      "grad_norm": 0.1376953125,
+      "grad_norm": 0.126953125,
       "learning_rate": 0.00012817325568414297,
-      "loss": 0.0326,
+      "loss": 0.0561,
       "step": 730
     },
     {
       "epoch": 4.129943502824859,
-      "grad_norm": 0.1337890625,
+      "grad_norm": 0.09130859375,
       "learning_rate": 0.00012800194179686037,
-      "loss": 0.112,
+      "loss": 0.1052,
       "step": 731
     },
     {
       "epoch": 4.135593220338983,
-      "grad_norm": 0.1025390625,
+      "grad_norm": 0.09033203125,
       "learning_rate": 0.00012783053868953168,
-      "loss": 0.0913,
+      "loss": 0.0703,
       "step": 732
     },
     {
       "epoch": 4.141242937853107,
-      "grad_norm": 0.1572265625,
+      "grad_norm": 0.345703125,
       "learning_rate": 0.00012765904690828313,
-      "loss": 0.1063,
+      "loss": 0.0848,
       "step": 733
     },
     {
       "epoch": 4.146892655367232,
-      "grad_norm": 0.1640625,
+      "grad_norm": 0.1572265625,
       "learning_rate": 0.00012748746699952338,
-      "loss": 0.0962,
+      "loss": 0.0545,
       "step": 734
     },
     {
       "epoch": 4.1525423728813555,
-      "grad_norm": 0.111328125,
+      "grad_norm": 0.09423828125,
       "learning_rate": 0.000127315799509942,
-      "loss": 0.1167,
+      "loss": 0.0974,
       "step": 735
     },
     {
       "epoch": 4.15819209039548,
-      "grad_norm": 0.1357421875,
+      "grad_norm": 0.19140625,
       "learning_rate": 0.00012714404498650743,
-      "loss": 0.0566,
+      "loss": 0.075,
       "step": 736
     },
     {
       "epoch": 4.163841807909605,
-      "grad_norm": 0.10009765625,
+      "grad_norm": 0.21875,
       "learning_rate": 0.00012697220397646562,
-      "loss": 0.1305,
+      "loss": 0.0538,
       "step": 737
     },
     {
       "epoch": 4.169491525423728,
-      "grad_norm": 0.091796875,
+      "grad_norm": 0.3359375,
       "learning_rate": 0.0001268002770273379,
-      "loss": 0.0709,
+      "loss": 0.0423,
       "step": 738
     },
     {
       "epoch": 4.175141242937853,
-      "grad_norm": 0.1962890625,
+      "grad_norm": 0.134765625,
       "learning_rate": 0.00012662826468691954,
-      "loss": 0.0384,
+      "loss": 0.0018,
       "step": 739
     },
     {
       "epoch": 4.1807909604519775,
-      "grad_norm": 0.06396484375,
+      "grad_norm": 0.06982421875,
       "learning_rate": 0.0001264561675032779,
-      "loss": 0.0853,
+      "loss": 0.0595,
       "step": 740
     },
     {
       "epoch": 4.186440677966102,
-      "grad_norm": 0.03857421875,
+      "grad_norm": 0.3125,
       "learning_rate": 0.00012628398602475055,
-      "loss": 0.0499,
+      "loss": 0.0906,
       "step": 741
     },
     {
       "epoch": 4.192090395480226,
-      "grad_norm": 0.05517578125,
+      "grad_norm": 0.10546875,
       "learning_rate": 0.00012611172079994375,
-      "loss": 0.055,
+      "loss": 0.0918,
       "step": 742
     },
     {
       "epoch": 4.19774011299435,
-      "grad_norm": 0.283203125,
+      "grad_norm": 0.1748046875,
       "learning_rate": 0.00012593937237773053,
-      "loss": 0.0863,
+      "loss": 0.0762,
       "step": 743
     },
     {
       "epoch": 4.203389830508475,
-      "grad_norm": 0.1171875,
+      "grad_norm": 0.11474609375,
       "learning_rate": 0.00012576694130724905,
-      "loss": 0.0556,
+      "loss": 0.0192,
       "step": 744
     },
     {
       "epoch": 4.209039548022599,
-      "grad_norm": 0.150390625,
+      "grad_norm": 0.06884765625,
       "learning_rate": 0.00012559442813790076,
-      "loss": 0.0423,
+      "loss": 0.0324,
       "step": 745
     },
     {
       "epoch": 4.214689265536723,
-      "grad_norm": 0.072265625,
+      "grad_norm": 0.08837890625,
       "learning_rate": 0.00012542183341934872,
-      "loss": 0.0887,
+      "loss": 0.0929,
       "step": 746
     },
     {
       "epoch": 4.220338983050848,
-      "grad_norm": 0.07373046875,
+      "grad_norm": 0.08642578125,
       "learning_rate": 0.0001252491577015158,
-      "loss": 0.0881,
+      "loss": 0.0762,
       "step": 747
     },
     {
       "epoch": 4.2259887005649714,
-      "grad_norm": 0.083984375,
+      "grad_norm": 0.09326171875,
       "learning_rate": 0.00012507640153458302,
-      "loss": 0.1079,
+      "loss": 0.0356,
       "step": 748
     },
     {
       "epoch": 4.231638418079096,
-      "grad_norm": 0.1708984375,
+      "grad_norm": 0.2109375,
       "learning_rate": 0.00012490356546898768,
-      "loss": 0.0762,
+      "loss": 0.0586,
       "step": 749
     },
     {
       "epoch": 4.237288135593221,
-      "grad_norm": 0.23828125,
+      "grad_norm": 0.2109375,
       "learning_rate": 0.00012473065005542155,
-      "loss": 0.0715,
+      "loss": 0.0757,
       "step": 750
     },
     {
       "epoch": 4.242937853107344,
-      "grad_norm": 0.053466796875,
+      "grad_norm": 0.1240234375,
       "learning_rate": 0.00012455765584482945,
-      "loss": 0.0803,
+      "loss": 0.0553,
       "step": 751
     },
     {
       "epoch": 4.248587570621469,
-      "grad_norm": 0.07373046875,
+      "grad_norm": 0.1728515625,
       "learning_rate": 0.00012438458338840706,
-      "loss": 0.0395,
+      "loss": 0.1261,
       "step": 752
     },
     {
       "epoch": 4.254237288135593,
-      "grad_norm": 0.10302734375,
+      "grad_norm": 0.07421875,
       "learning_rate": 0.00012421143323759946,
-      "loss": 0.0833,
+      "loss": 0.0571,
       "step": 753
     },
     {
       "epoch": 4.259887005649717,
-      "grad_norm": 0.056640625,
+      "grad_norm": 0.1318359375,
       "learning_rate": 0.00012403820594409924,
-      "loss": 0.1022,
+      "loss": 0.0553,
       "step": 754
     },
     {
       "epoch": 4.265536723163842,
-      "grad_norm": 0.0732421875,
+      "grad_norm": 0.09521484375,
       "learning_rate": 0.00012386490205984488,
-      "loss": 0.1243,
+      "loss": 0.076,
       "step": 755
     },
     {
       "epoch": 4.271186440677966,
-      "grad_norm": 0.07275390625,
+      "grad_norm": 0.06591796875,
       "learning_rate": 0.00012369152213701873,
-      "loss": 0.0911,
+      "loss": 0.0409,
       "step": 756
     },
     {
       "epoch": 4.27683615819209,
-      "grad_norm": 0.1025390625,
+      "grad_norm": 0.130859375,
       "learning_rate": 0.00012351806672804557,
-      "loss": 0.0843,
+      "loss": 0.1105,
       "step": 757
     },
     {
       "epoch": 4.2824858757062145,
-      "grad_norm": 0.06787109375,
+      "grad_norm": 0.0634765625,
       "learning_rate": 0.00012334453638559057,
-      "loss": 0.0887,
+      "loss": 0.0743,
       "step": 758
     },
     {
       "epoch": 4.288135593220339,
-      "grad_norm": 0.05859375,
+      "grad_norm": 0.06884765625,
       "learning_rate": 0.00012317093166255773,
-      "loss": 0.0658,
+      "loss": 0.0891,
       "step": 759
     },
     {
       "epoch": 4.293785310734464,
-      "grad_norm": 0.061767578125,
+      "grad_norm": 0.0859375,
       "learning_rate": 0.00012299725311208808,
-      "loss": 0.0874,
+      "loss": 0.0495,
       "step": 760
     },
     {
       "epoch": 4.299435028248587,
-      "grad_norm": 0.1865234375,
+      "grad_norm": 0.1474609375,
       "learning_rate": 0.00012282350128755773,
-      "loss": 0.0567,
+      "loss": 0.0584,
       "step": 761
     },
     {
       "epoch": 4.305084745762712,
-      "grad_norm": 0.10498046875,
+      "grad_norm": 0.1904296875,
       "learning_rate": 0.00012264967674257646,
-      "loss": 0.1024,
+      "loss": 0.079,
       "step": 762
     },
     {
       "epoch": 4.3107344632768365,
-      "grad_norm": 0.03759765625,
+      "grad_norm": 0.1474609375,
       "learning_rate": 0.00012247578003098553,
-      "loss": 0.0343,
+      "loss": 0.0377,
       "step": 763
     },
     {
       "epoch": 4.31638418079096,
-      "grad_norm": 0.1826171875,
+      "grad_norm": 0.140625,
       "learning_rate": 0.00012230181170685636,
-      "loss": 0.0529,
+      "loss": 0.0924,
       "step": 764
     },
     {
       "epoch": 4.322033898305085,
-      "grad_norm": 0.06298828125,
+      "grad_norm": 0.10400390625,
       "learning_rate": 0.00012212777232448838,
-      "loss": 0.0748,
+      "loss": 0.0364,
       "step": 765
     },
     {
       "epoch": 4.322033898305085,
-      "eval_loss": 0.0021968409419059753,
-      "eval_runtime": 5.8458,
-      "eval_samples_per_second": 25.488,
-      "eval_steps_per_second": 3.25,
+      "eval_loss": 0.0032166026066988707,
+      "eval_runtime": 5.8145,
+      "eval_samples_per_second": 25.798,
+      "eval_steps_per_second": 3.268,
       "step": 765
     },
     {
       "epoch": 4.327683615819209,
-      "grad_norm": 0.07421875,
+      "grad_norm": 0.23046875,
       "learning_rate": 0.00012195366243840745,
-      "loss": 0.0691,
+      "loss": 0.1378,
       "step": 766
     },
     {
       "epoch": 4.333333333333333,
-      "grad_norm": 0.318359375,
+      "grad_norm": 0.28125,
       "learning_rate": 0.00012177948260336413,
-      "loss": 0.0947,
+      "loss": 0.0547,
       "step": 767
     },
     {
       "epoch": 4.338983050847458,
-      "grad_norm": 0.064453125,
+      "grad_norm": 0.1875,
       "learning_rate": 0.00012160523337433188,
-      "loss": 0.0911,
+      "loss": 0.1095,
       "step": 768
     },
     {
       "epoch": 4.344632768361582,
-      "grad_norm": 0.06640625,
+      "grad_norm": 0.1376953125,
       "learning_rate": 0.00012143091530650508,
-      "loss": 0.092,
+      "loss": 0.0533,
       "step": 769
     },
     {
       "epoch": 4.350282485875706,
-      "grad_norm": 0.06396484375,
+      "grad_norm": 0.09423828125,
       "learning_rate": 0.00012125652895529766,
-      "loss": 0.0574,
+      "loss": 0.0887,
       "step": 770
     },
     {
       "epoch": 4.3559322033898304,
-      "grad_norm": 0.07470703125,
+      "grad_norm": 0.13671875,
       "learning_rate": 0.00012108207487634104,
-      "loss": 0.0734,
+      "loss": 0.0685,
       "step": 771
     },
     {
       "epoch": 4.361581920903955,
-      "grad_norm": 0.2001953125,
+      "grad_norm": 0.0986328125,
       "learning_rate": 0.00012090755362548236,
-      "loss": 0.0762,
+      "loss": 0.0528,
       "step": 772
     },
     {
       "epoch": 4.367231638418079,
-      "grad_norm": 0.1591796875,
+      "grad_norm": 0.1328125,
       "learning_rate": 0.00012073296575878288,
-      "loss": 0.0356,
+      "loss": 0.0505,
       "step": 773
     },
     {
       "epoch": 4.372881355932203,
-      "grad_norm": 0.06982421875,
+      "grad_norm": 0.20703125,
       "learning_rate": 0.00012055831183251607,
-      "loss": 0.1282,
+      "loss": 0.0742,
       "step": 774
     },
     {
       "epoch": 4.378531073446328,
-      "grad_norm": 0.1953125,
+      "grad_norm": 0.09619140625,
       "learning_rate": 0.00012038359240316589,
-      "loss": 0.0403,
+      "loss": 0.0644,
       "step": 775
     },
     {
       "epoch": 4.3841807909604515,
-      "grad_norm": 0.04345703125,
+      "grad_norm": 0.1103515625,
       "learning_rate": 0.00012020880802742498,
-      "loss": 0.0325,
+      "loss": 0.0858,
       "step": 776
     },
     {
       "epoch": 4.389830508474576,
-      "grad_norm": 0.078125,
+      "grad_norm": 0.267578125,
       "learning_rate": 0.00012003395926219295,
-      "loss": 0.0926,
+      "loss": 0.0554,
       "step": 777
     },
     {
       "epoch": 4.395480225988701,
-      "grad_norm": 0.047607421875,
+      "grad_norm": 0.1650390625,
       "learning_rate": 0.00011985904666457455,
-      "loss": 0.0732,
+      "loss": 0.0956,
       "step": 778
     },
     {
       "epoch": 4.401129943502825,
-      "grad_norm": 0.2041015625,
+      "grad_norm": 0.08056640625,
       "learning_rate": 0.00011968407079187794,
-      "loss": 0.0898,
+      "loss": 0.1002,
       "step": 779
     },
     {
       "epoch": 4.406779661016949,
-      "grad_norm": 0.05224609375,
+      "grad_norm": 0.09619140625,
       "learning_rate": 0.00011950903220161285,
-      "loss": 0.0577,
+      "loss": 0.1037,
       "step": 780
     },
     {
       "epoch": 4.4124293785310735,
-      "grad_norm": 0.047607421875,
+      "grad_norm": 0.07861328125,
       "learning_rate": 0.00011933393145148883,
-      "loss": 0.0332,
+      "loss": 0.1006,
       "step": 781
     },
     {
       "epoch": 4.418079096045198,
-      "grad_norm": 0.0751953125,
+      "grad_norm": 0.119140625,
       "learning_rate": 0.00011915876909941355,
-      "loss": 0.1023,
+      "loss": 0.0372,
       "step": 782
     },
     {
       "epoch": 4.423728813559322,
-      "grad_norm": 0.06982421875,
+      "grad_norm": 0.1787109375,
       "learning_rate": 0.00011898354570349088,
-      "loss": 0.0715,
+      "loss": 0.1068,
       "step": 783
     },
     {
       "epoch": 4.429378531073446,
-      "grad_norm": 0.05419921875,
+      "grad_norm": 0.09716796875,
       "learning_rate": 0.00011880826182201926,
-      "loss": 0.0716,
+      "loss": 0.1484,
       "step": 784
     },
     {
       "epoch": 4.435028248587571,
-      "grad_norm": 0.130859375,
+      "grad_norm": 0.0771484375,
       "learning_rate": 0.0001186329180134898,
-      "loss": 0.0883,
+      "loss": 0.0365,
       "step": 785
     },
     {
       "epoch": 4.440677966101695,
-      "grad_norm": 0.05126953125,
+      "grad_norm": 0.1884765625,
       "learning_rate": 0.00011845751483658453,
-      "loss": 0.0892,
+      "loss": 0.0429,
       "step": 786
     },
     {
       "epoch": 4.446327683615819,
-      "grad_norm": 0.123046875,
+      "grad_norm": 0.08056640625,
       "learning_rate": 0.00011828205285017475,
-      "loss": 0.0735,
+      "loss": 0.0887,
       "step": 787
     },
     {
       "epoch": 4.451977401129944,
-      "grad_norm": 0.05712890625,
+      "grad_norm": 0.06787109375,
       "learning_rate": 0.00011810653261331903,
-      "loss": 0.0711,
+      "loss": 0.0707,
       "step": 788
     },
     {
       "epoch": 4.4576271186440675,
-      "grad_norm": 0.09033203125,
+      "grad_norm": 0.150390625,
       "learning_rate": 0.00011793095468526154,
-      "loss": 0.0762,
+      "loss": 0.054,
       "step": 789
     },
     {
       "epoch": 4.463276836158192,
-      "grad_norm": 0.0712890625,
+      "grad_norm": 0.08935546875,
       "learning_rate": 0.00011775531962543036,
-      "loss": 0.0736,
+      "loss": 0.0379,
       "step": 790
     },
     {
       "epoch": 4.468926553672317,
-      "grad_norm": 0.076171875,
+      "grad_norm": 0.059326171875,
       "learning_rate": 0.00011757962799343547,
-      "loss": 0.0867,
+      "loss": 0.0564,
       "step": 791
     },
     {
       "epoch": 4.47457627118644,
-      "grad_norm": 0.056396484375,
+      "grad_norm": 0.10498046875,
       "learning_rate": 0.00011740388034906719,
-      "loss": 0.103,
+      "loss": 0.0911,
       "step": 792
     },
     {
       "epoch": 4.480225988700565,
-      "grad_norm": 0.04443359375,
+      "grad_norm": 0.2236328125,
       "learning_rate": 0.00011722807725229439,
-      "loss": 0.0539,
+      "loss": 0.0891,
       "step": 793
     },
     {
       "epoch": 4.4858757062146895,
-      "grad_norm": 0.1455078125,
+      "grad_norm": 0.087890625,
       "learning_rate": 0.0001170522192632624,
-      "loss": 0.0925,
+      "loss": 0.0683,
       "step": 794
     },
     {
       "epoch": 4.491525423728813,
-      "grad_norm": 0.0712890625,
+      "grad_norm": 0.11474609375,
       "learning_rate": 0.0001168763069422916,
-      "loss": 0.0541,
+      "loss": 0.1057,
       "step": 795
     },
     {
       "epoch": 4.497175141242938,
-      "grad_norm": 0.1396484375,
+      "grad_norm": 0.06884765625,
       "learning_rate": 0.00011670034084987551,
-      "loss": 0.0787,
+      "loss": 0.0426,
       "step": 796
     },
     {
       "epoch": 4.502824858757062,
-      "grad_norm": 0.080078125,
+      "grad_norm": 0.1708984375,
       "learning_rate": 0.00011652432154667887,
-      "loss": 0.0571,
+      "loss": 0.0787,
       "step": 797
     },
     {
       "epoch": 4.508474576271187,
-      "grad_norm": 0.11083984375,
+      "grad_norm": 0.059326171875,
       "learning_rate": 0.000116348249593536,
-      "loss": 0.068,
+      "loss": 0.0346,
       "step": 798
     },
     {
       "epoch": 4.5141242937853105,
-      "grad_norm": 0.07177734375,
+      "grad_norm": 0.1396484375,
       "learning_rate": 0.00011617212555144905,
-      "loss": 0.1071,
+      "loss": 0.0905,
       "step": 799
     },
     {
       "epoch": 4.519774011299435,
-      "grad_norm": 0.08251953125,
+      "grad_norm": 0.0625,
       "learning_rate": 0.00011599594998158602,
-      "loss": 0.16,
+      "loss": 0.0884,
       "step": 800
     },
     {
       "epoch": 4.52542372881356,
-      "grad_norm": 0.0380859375,
+      "grad_norm": 0.09521484375,
       "learning_rate": 0.00011581972344527918,
-      "loss": 0.0364,
+      "loss": 0.051,
       "step": 801
     },
     {
       "epoch": 4.531073446327683,
-      "grad_norm": 0.140625,
+      "grad_norm": 0.08837890625,
       "learning_rate": 0.0001156434465040231,
-      "loss": 0.0853,
+      "loss": 0.1065,
       "step": 802
     },
     {
       "epoch": 4.536723163841808,
-      "grad_norm": 0.0517578125,
+      "grad_norm": 0.2177734375,
       "learning_rate": 0.00011546711971947303,
-      "loss": 0.0315,
+      "loss": 0.0789,
       "step": 803
     },
     {
       "epoch": 4.5423728813559325,
-      "grad_norm": 0.05419921875,
+      "grad_norm": 0.16796875,
       "learning_rate": 0.00011529074365344301,
-      "loss": 0.0699,
+      "loss": 0.0398,
       "step": 804
     },
     {
       "epoch": 4.548022598870056,
-      "grad_norm": 0.045166015625,
+      "grad_norm": 0.20703125,
       "learning_rate": 0.00011511431886790407,
-      "loss": 0.0321,
+      "loss": 0.0162,
       "step": 805
     },
     {
       "epoch": 4.553672316384181,
-      "grad_norm": 0.11083984375,
+      "grad_norm": 0.0595703125,
       "learning_rate": 0.00011493784592498252,
-      "loss": 0.0429,
+      "loss": 0.0719,
       "step": 806
     },
     {
       "epoch": 4.559322033898305,
-      "grad_norm": 0.07275390625,
+      "grad_norm": 0.10205078125,
       "learning_rate": 0.00011476132538695807,
-      "loss": 0.066,
+      "loss": 0.1052,
       "step": 807
     },
     {
       "epoch": 4.564971751412429,
-      "grad_norm": 0.04736328125,
+      "grad_norm": 0.046875,
       "learning_rate": 0.00011458475781626208,
-      "loss": 0.0636,
+      "loss": 0.0369,
       "step": 808
     },
     {
       "epoch": 4.570621468926554,
-      "grad_norm": 0.06005859375,
+      "grad_norm": 0.052978515625,
       "learning_rate": 0.00011440814377547581,
-      "loss": 0.0542,
+      "loss": 0.0348,
       "step": 809
     },
     {
       "epoch": 4.576271186440678,
-      "grad_norm": 0.051025390625,
+      "grad_norm": 0.07080078125,
       "learning_rate": 0.00011423148382732853,
-      "loss": 0.0812,
+      "loss": 0.1342,
       "step": 810
     },
     {
       "epoch": 4.576271186440678,
-      "eval_loss": 0.001691528595983982,
-      "eval_runtime": 6.0671,
-      "eval_samples_per_second": 24.559,
-      "eval_steps_per_second": 3.132,
+      "eval_loss": 0.001909776940010488,
+      "eval_runtime": 5.7476,
+      "eval_samples_per_second": 26.098,
+      "eval_steps_per_second": 3.306,
       "step": 810
     },
     {
       "epoch": 4.581920903954802,
-      "grad_norm": 0.0888671875,
+      "grad_norm": 0.05859375,
       "learning_rate": 0.00011405477853469577,
-      "loss": 0.0753,
+      "loss": 0.072,
       "step": 811
     },
     {
       "epoch": 4.5875706214689265,
-      "grad_norm": 0.052734375,
+      "grad_norm": 0.1025390625,
       "learning_rate": 0.00011387802846059763,
-      "loss": 0.0706,
+      "loss": 0.1122,
       "step": 812
     },
     {
       "epoch": 4.593220338983051,
-      "grad_norm": 0.0419921875,
+      "grad_norm": 0.07275390625,
       "learning_rate": 0.00011370123416819682,
-      "loss": 0.0376,
+      "loss": 0.0807,
       "step": 813
     },
     {
       "epoch": 4.598870056497175,
-      "grad_norm": 0.12060546875,
+      "grad_norm": 0.1171875,
       "learning_rate": 0.00011352439622079689,
-      "loss": 0.1045,
+      "loss": 0.0905,
       "step": 814
     },
     {
       "epoch": 4.604519774011299,
-      "grad_norm": 0.059326171875,
+      "grad_norm": 0.06591796875,
       "learning_rate": 0.00011334751518184061,
-      "loss": 0.0708,
+      "loss": 0.1056,
       "step": 815
     },
     {
       "epoch": 4.610169491525424,
-      "grad_norm": 0.0703125,
+      "grad_norm": 0.060546875,
       "learning_rate": 0.00011317059161490797,
-      "loss": 0.107,
+      "loss": 0.079,
       "step": 816
     },
     {
       "epoch": 4.6158192090395485,
-      "grad_norm": 0.0625,
+      "grad_norm": 0.1494140625,
       "learning_rate": 0.00011299362608371446,
-      "loss": 0.0581,
+      "loss": 0.0548,
       "step": 817
     },
     {
       "epoch": 4.621468926553672,
-      "grad_norm": 0.099609375,
+      "grad_norm": 0.06005859375,
       "learning_rate": 0.0001128166191521093,
-      "loss": 0.0544,
+      "loss": 0.0477,
       "step": 818
     },
     {
       "epoch": 4.627118644067797,
-      "grad_norm": 0.0869140625,
+      "grad_norm": 0.06689453125,
       "learning_rate": 0.00011263957138407359,
-      "loss": 0.0742,
+      "loss": 0.0959,
       "step": 819
     },
     {
       "epoch": 4.632768361581921,
-      "grad_norm": 0.08251953125,
+      "grad_norm": 0.05615234375,
       "learning_rate": 0.0001124624833437186,
-      "loss": 0.1252,
+      "loss": 0.0556,
       "step": 820
     },
     {
       "epoch": 4.638418079096045,
-      "grad_norm": 0.08935546875,
+      "grad_norm": 0.072265625,
       "learning_rate": 0.00011228535559528388,
-      "loss": 0.1082,
+      "loss": 0.1328,
       "step": 821
     },
     {
       "epoch": 4.6440677966101696,
-      "grad_norm": 0.051025390625,
+      "grad_norm": 0.0966796875,
       "learning_rate": 0.00011210818870313549,
-      "loss": 0.0529,
+      "loss": 0.1328,
       "step": 822
     },
     {
       "epoch": 4.649717514124294,
-      "grad_norm": 0.0908203125,
+      "grad_norm": 0.058837890625,
       "learning_rate": 0.0001119309832317642,
-      "loss": 0.1633,
+      "loss": 0.0549,
       "step": 823
     },
     {
       "epoch": 4.655367231638418,
-      "grad_norm": 0.051513671875,
+      "grad_norm": 0.047119140625,
       "learning_rate": 0.00011175373974578378,
-      "loss": 0.0528,
+      "loss": 0.0684,
       "step": 824
     },
     {
       "epoch": 4.661016949152542,
-      "grad_norm": 0.0595703125,
+      "grad_norm": 0.033203125,
       "learning_rate": 0.00011157645880992902,
-      "loss": 0.0784,
+      "loss": 0.0309,
       "step": 825
     },
     {
       "epoch": 4.666666666666667,
-      "grad_norm": 0.059814453125,
+      "grad_norm": 0.06396484375,
       "learning_rate": 0.00011139914098905406,
-      "loss": 0.057,
+      "loss": 0.0788,
       "step": 826
     },
     {
       "epoch": 4.672316384180791,
-      "grad_norm": 0.12060546875,
+      "grad_norm": 0.134765625,
       "learning_rate": 0.00011122178684813065,
-      "loss": 0.0572,
+      "loss": 0.1368,
       "step": 827
     },
     {
       "epoch": 4.677966101694915,
-      "grad_norm": 0.07177734375,
+      "grad_norm": 0.0625,
       "learning_rate": 0.00011104439695224613,
-      "loss": 0.0364,
+      "loss": 0.0669,
       "step": 828
     },
     {
       "epoch": 4.68361581920904,
-      "grad_norm": 0.047607421875,
+      "grad_norm": 0.06298828125,
       "learning_rate": 0.00011086697186660186,
-      "loss": 0.0559,
+      "loss": 0.0896,
       "step": 829
     },
     {
       "epoch": 4.6892655367231635,
-      "grad_norm": 0.056884765625,
+      "grad_norm": 0.041748046875,
       "learning_rate": 0.00011068951215651132,
-      "loss": 0.0563,
+      "loss": 0.0524,
       "step": 830
     },
     {
       "epoch": 4.694915254237288,
-      "grad_norm": 0.0546875,
+      "grad_norm": 0.1435546875,
       "learning_rate": 0.00011051201838739817,
-      "loss": 0.0728,
+      "loss": 0.1099,
       "step": 831
     },
     {
       "epoch": 4.700564971751413,
-      "grad_norm": 0.0673828125,
+      "grad_norm": 0.068359375,
       "learning_rate": 0.00011033449112479481,
-      "loss": 0.0869,
+      "loss": 0.0955,
       "step": 832
     },
     {
       "epoch": 4.706214689265536,
-      "grad_norm": 0.146484375,
+      "grad_norm": 0.061767578125,
       "learning_rate": 0.00011015693093434023,
-      "loss": 0.0935,
+      "loss": 0.0719,
       "step": 833
     },
     {
       "epoch": 4.711864406779661,
-      "grad_norm": 0.06103515625,
+      "grad_norm": 0.1103515625,
       "learning_rate": 0.00010997933838177827,
-      "loss": 0.0823,
+      "loss": 0.1035,
       "step": 834
     },
     {
       "epoch": 4.7175141242937855,
-      "grad_norm": 0.07568359375,
+      "grad_norm": 0.06884765625,
       "learning_rate": 0.0001098017140329561,
-      "loss": 0.0592,
+      "loss": 0.0749,
       "step": 835
     },
     {
       "epoch": 4.72316384180791,
-      "grad_norm": 0.045166015625,
+      "grad_norm": 0.0615234375,
       "learning_rate": 0.00010962405845382193,
-      "loss": 0.0504,
+      "loss": 0.0599,
       "step": 836
     },
     {
       "epoch": 4.728813559322034,
-      "grad_norm": 0.05810546875,
+      "grad_norm": 0.044921875,
       "learning_rate": 0.00010944637221042366,
-      "loss": 0.0854,
+      "loss": 0.05,
       "step": 837
     },
     {
       "epoch": 4.734463276836158,
-      "grad_norm": 0.051513671875,
+      "grad_norm": 0.0615234375,
       "learning_rate": 0.00010926865586890689,
-      "loss": 0.0657,
+      "loss": 0.107,
       "step": 838
     },
     {
       "epoch": 4.740112994350282,
-      "grad_norm": 0.056640625,
+      "grad_norm": 0.06494140625,
       "learning_rate": 0.00010909090999551304,
-      "loss": 0.1176,
+      "loss": 0.0927,
       "step": 839
     },
     {
       "epoch": 4.745762711864407,
-      "grad_norm": 0.037353515625,
+      "grad_norm": 0.078125,
       "learning_rate": 0.0001089131351565776,
-      "loss": 0.0341,
+      "loss": 0.0706,
       "step": 840
     },
     {
       "epoch": 4.751412429378531,
-      "grad_norm": 0.0751953125,
+      "grad_norm": 0.0791015625,
       "learning_rate": 0.00010873533191852849,
-      "loss": 0.1303,
+      "loss": 0.0578,
       "step": 841
     },
     {
       "epoch": 4.757062146892656,
-      "grad_norm": 0.057373046875,
+      "grad_norm": 0.052490234375,
       "learning_rate": 0.00010855750084788398,
-      "loss": 0.0696,
+      "loss": 0.0377,
       "step": 842
     },
     {
       "epoch": 4.762711864406779,
-      "grad_norm": 0.11669921875,
+      "grad_norm": 0.052490234375,
       "learning_rate": 0.0001083796425112511,
-      "loss": 0.0949,
+      "loss": 0.0742,
       "step": 843
     },
     {
       "epoch": 4.768361581920904,
-      "grad_norm": 0.01251220703125,
+      "grad_norm": 0.042236328125,
       "learning_rate": 0.00010820175747532373,
-      "loss": 0.0005,
+      "loss": 0.0339,
       "step": 844
     },
     {
       "epoch": 4.7740112994350286,
-      "grad_norm": 0.119140625,
+      "grad_norm": 0.054443359375,
       "learning_rate": 0.00010802384630688078,
-      "loss": 0.0902,
+      "loss": 0.0857,
       "step": 845
     },
     {
       "epoch": 4.779661016949152,
-      "grad_norm": 0.0966796875,
+      "grad_norm": 0.072265625,
       "learning_rate": 0.0001078459095727845,
-      "loss": 0.0805,
+      "loss": 0.097,
       "step": 846
     },
     {
       "epoch": 4.785310734463277,
-      "grad_norm": 0.052734375,
+      "grad_norm": 0.09814453125,
       "learning_rate": 0.00010766794783997855,
-      "loss": 0.0563,
+      "loss": 0.0359,
       "step": 847
     },
     {
       "epoch": 4.790960451977401,
-      "grad_norm": 0.0380859375,
+      "grad_norm": 0.07666015625,
       "learning_rate": 0.00010748996167548623,
-      "loss": 0.047,
+      "loss": 0.0697,
       "step": 848
     },
     {
       "epoch": 4.796610169491525,
-      "grad_norm": 0.353515625,
+      "grad_norm": 0.051025390625,
       "learning_rate": 0.00010731195164640873,
-      "loss": 0.0901,
+      "loss": 0.064,
       "step": 849
     },
     {
       "epoch": 4.80225988700565,
-      "grad_norm": 0.0498046875,
+      "grad_norm": 0.08935546875,
       "learning_rate": 0.00010713391831992323,
-      "loss": 0.0529,
+      "loss": 0.1196,
       "step": 850
     },
     {
       "epoch": 4.807909604519774,
-      "grad_norm": 0.0751953125,
+      "grad_norm": 0.173828125,
       "learning_rate": 0.00010695586226328118,
-      "loss": 0.0723,
+      "loss": 0.0554,
       "step": 851
     },
     {
       "epoch": 4.813559322033898,
-      "grad_norm": 0.045654296875,
+      "grad_norm": 0.052490234375,
       "learning_rate": 0.00010677778404380643,
-      "loss": 0.0515,
+      "loss": 0.0817,
       "step": 852
     },
     {
       "epoch": 4.8192090395480225,
-      "grad_norm": 0.053466796875,
+      "grad_norm": 0.083984375,
       "learning_rate": 0.00010659968422889346,
-      "loss": 0.0808,
+      "loss": 0.1326,
       "step": 853
     },
     {
       "epoch": 4.824858757062147,
-      "grad_norm": 0.08203125,
+      "grad_norm": 0.1845703125,
       "learning_rate": 0.00010642156338600551,
-      "loss": 0.0379,
+      "loss": 0.0511,
       "step": 854
     },
     {
       "epoch": 4.830508474576272,
-      "grad_norm": 0.028564453125,
+      "grad_norm": 0.0595703125,
       "learning_rate": 0.00010624342208267292,
-      "loss": 0.0313,
+      "loss": 0.0776,
       "step": 855
     },
     {
       "epoch": 4.830508474576272,
-      "eval_loss": 0.0015072287060320377,
-      "eval_runtime": 5.9324,
-      "eval_samples_per_second": 25.116,
-      "eval_steps_per_second": 3.203,
+      "eval_loss": 0.0014068098971620202,
+      "eval_runtime": 5.7264,
+      "eval_samples_per_second": 26.194,
+      "eval_steps_per_second": 3.318,
       "step": 855
     },
     {
       "epoch": 4.836158192090395,
-      "grad_norm": 0.07177734375,
+      "grad_norm": 0.061767578125,
       "learning_rate": 0.00010606526088649112,
-      "loss": 0.0949,
+      "loss": 0.1094,
       "step": 856
     },
     {
       "epoch": 4.84180790960452,
-      "grad_norm": 0.04833984375,
+      "grad_norm": 0.0634765625,
       "learning_rate": 0.0001058870803651189,
-      "loss": 0.0402,
+      "loss": 0.0867,
       "step": 857
     },
     {
       "epoch": 4.847457627118644,
-      "grad_norm": 0.05908203125,
+      "grad_norm": 0.050048828125,
       "learning_rate": 0.00010570888108627681,
-      "loss": 0.0842,
+      "loss": 0.0694,
       "step": 858
     },
     {
       "epoch": 4.853107344632768,
-      "grad_norm": 0.056884765625,
+      "grad_norm": 0.07275390625,
       "learning_rate": 0.00010553066361774497,
-      "loss": 0.0691,
+      "loss": 0.1105,
       "step": 859
     },
     {
       "epoch": 4.858757062146893,
-      "grad_norm": 0.04541015625,
+      "grad_norm": 0.1435546875,
       "learning_rate": 0.00010535242852736151,
-      "loss": 0.0528,
+      "loss": 0.0521,
       "step": 860
     },
     {
       "epoch": 4.864406779661017,
-      "grad_norm": 0.07861328125,
+      "grad_norm": 0.04248046875,
       "learning_rate": 0.00010517417638302081,
-      "loss": 0.0863,
+      "loss": 0.0528,
       "step": 861
     },
     {
       "epoch": 4.870056497175141,
-      "grad_norm": 0.10693359375,
+      "grad_norm": 0.1513671875,
       "learning_rate": 0.0001049959077526714,
-      "loss": 0.0674,
+      "loss": 0.1105,
       "step": 862
     },
     {
       "epoch": 4.875706214689266,
-      "grad_norm": 0.07373046875,
+      "grad_norm": 0.06494140625,
       "learning_rate": 0.0001048176232043145,
-      "loss": 0.0831,
+      "loss": 0.0937,
       "step": 863
     },
     {
       "epoch": 4.88135593220339,
-      "grad_norm": 0.07568359375,
+      "grad_norm": 0.044921875,
       "learning_rate": 0.00010463932330600196,
-      "loss": 0.1288,
+      "loss": 0.0531,
       "step": 864
     },
     {
       "epoch": 4.887005649717514,
-      "grad_norm": 0.04833984375,
+      "grad_norm": 0.045654296875,
       "learning_rate": 0.00010446100862583459,
-      "loss": 0.0697,
+      "loss": 0.0482,
       "step": 865
     },
     {
       "epoch": 4.892655367231638,
-      "grad_norm": 0.05810546875,
+      "grad_norm": 0.08203125,
       "learning_rate": 0.00010428267973196027,
-      "loss": 0.0968,
+      "loss": 0.1131,
       "step": 866
     },
     {
       "epoch": 4.898305084745763,
-      "grad_norm": 0.053955078125,
+      "grad_norm": 0.07861328125,
       "learning_rate": 0.00010410433719257213,
-      "loss": 0.0642,
+      "loss": 0.0932,
       "step": 867
     },
     {
       "epoch": 4.903954802259887,
-      "grad_norm": 0.07763671875,
+      "grad_norm": 0.04638671875,
       "learning_rate": 0.00010392598157590688,
-      "loss": 0.136,
+      "loss": 0.0629,
       "step": 868
     },
     {
       "epoch": 4.909604519774011,
-      "grad_norm": 0.09423828125,
+      "grad_norm": 0.07421875,
       "learning_rate": 0.0001037476134502428,
-      "loss": 0.1351,
+      "loss": 0.0858,
       "step": 869
     },
     {
       "epoch": 4.915254237288136,
-      "grad_norm": 0.043701171875,
+      "grad_norm": 0.08447265625,
       "learning_rate": 0.00010356923338389806,
-      "loss": 0.0525,
+      "loss": 0.1058,
       "step": 870
     },
     {
       "epoch": 4.9209039548022595,
-      "grad_norm": 0.056396484375,
+      "grad_norm": 0.193359375,
       "learning_rate": 0.00010339084194522888,
-      "loss": 0.0908,
+      "loss": 0.0736,
       "step": 871
     },
     {
       "epoch": 4.926553672316384,
-      "grad_norm": 0.119140625,
+      "grad_norm": 0.061279296875,
       "learning_rate": 0.00010321243970262777,
-      "loss": 0.0429,
+      "loss": 0.0962,
       "step": 872
     },
     {
       "epoch": 4.932203389830509,
-      "grad_norm": 0.060302734375,
+      "grad_norm": 0.0712890625,
       "learning_rate": 0.0001030340272245215,
-      "loss": 0.0922,
+      "loss": 0.0901,
       "step": 873
     },
     {
       "epoch": 4.937853107344633,
-      "grad_norm": 0.076171875,
+      "grad_norm": 0.062255859375,
       "learning_rate": 0.00010285560507936961,
-      "loss": 0.1181,
+      "loss": 0.0861,
       "step": 874
     },
     {
       "epoch": 4.943502824858757,
-      "grad_norm": 0.009033203125,
+      "grad_norm": 0.055419921875,
       "learning_rate": 0.00010267717383566246,
-      "loss": 0.0005,
+      "loss": 0.0527,
       "step": 875
     },
     {
       "epoch": 4.9491525423728815,
-      "grad_norm": 0.0791015625,
+      "grad_norm": 0.044921875,
       "learning_rate": 0.00010249873406191921,
-      "loss": 0.0532,
+      "loss": 0.0477,
       "step": 876
     },
     {
       "epoch": 4.954802259887005,
-      "grad_norm": 0.041259765625,
+      "grad_norm": 0.06494140625,
       "learning_rate": 0.00010232028632668638,
-      "loss": 0.0383,
+      "loss": 0.0953,
       "step": 877
     },
     {
       "epoch": 4.96045197740113,
-      "grad_norm": 0.06640625,
+      "grad_norm": 0.056640625,
       "learning_rate": 0.00010214183119853582,
-      "loss": 0.1118,
+      "loss": 0.0721,
       "step": 878
     },
     {
       "epoch": 4.966101694915254,
-      "grad_norm": 0.062255859375,
+      "grad_norm": 0.0703125,
       "learning_rate": 0.00010196336924606283,
-      "loss": 0.1089,
+      "loss": 0.0984,
       "step": 879
     },
     {
       "epoch": 4.971751412429379,
-      "grad_norm": 0.041748046875,
+      "grad_norm": 0.052490234375,
       "learning_rate": 0.0001017849010378846,
-      "loss": 0.0488,
+      "loss": 0.0695,
       "step": 880
     },
     {
       "epoch": 4.977401129943503,
-      "grad_norm": 0.043701171875,
+      "grad_norm": 0.060791015625,
       "learning_rate": 0.0001016064271426382,
-      "loss": 0.056,
+      "loss": 0.0958,
       "step": 881
     },
     {
       "epoch": 4.983050847457627,
-      "grad_norm": 0.0458984375,
+      "grad_norm": 0.09375,
       "learning_rate": 0.00010142794812897873,
-      "loss": 0.0656,
+      "loss": 0.076,
       "step": 882
     },
     {
       "epoch": 4.988700564971752,
-      "grad_norm": 0.037353515625,
+      "grad_norm": 0.0634765625,
       "learning_rate": 0.00010124946456557776,
-      "loss": 0.0533,
+      "loss": 0.077,
       "step": 883
     },
     {
       "epoch": 4.994350282485875,
-      "grad_norm": 0.04443359375,
+      "grad_norm": 0.08544921875,
       "learning_rate": 0.0001010709770211212,
-      "loss": 0.0209,
+      "loss": 0.1083,
       "step": 884
     },
     {
       "epoch": 5.0,
-      "grad_norm": 0.0654296875,
+      "grad_norm": 0.042724609375,
       "learning_rate": 0.00010089248606430775,
-      "loss": 0.1099,
+      "loss": 0.0379,
       "step": 885
     },
     {
       "epoch": 5.005649717514125,
-      "grad_norm": 0.059326171875,
+      "grad_norm": 0.0576171875,
       "learning_rate": 0.00010071399226384694,
-      "loss": 0.1115,
+      "loss": 0.1181,
       "step": 886
     },
     {
       "epoch": 5.011299435028248,
-      "grad_norm": 0.04150390625,
+      "grad_norm": 0.0615234375,
       "learning_rate": 0.00010053549618845739,
-      "loss": 0.0561,
+      "loss": 0.0759,
       "step": 887
     },
     {
       "epoch": 5.016949152542373,
-      "grad_norm": 0.056884765625,
+      "grad_norm": 0.0390625,
       "learning_rate": 0.00010035699840686493,
-      "loss": 0.0895,
+      "loss": 0.0338,
       "step": 888
     },
     {
       "epoch": 5.022598870056497,
-      "grad_norm": 0.035400390625,
+      "grad_norm": 0.0537109375,
       "learning_rate": 0.00010017849948780083,
-      "loss": 0.048,
+      "loss": 0.0871,
       "step": 889
     },
     {
       "epoch": 5.028248587570621,
-      "grad_norm": 0.06591796875,
+      "grad_norm": 0.049072265625,
       "learning_rate": 0.0001,
-      "loss": 0.0946,
+      "loss": 0.0676,
       "step": 890
     },
     {
       "epoch": 5.033898305084746,
-      "grad_norm": 0.0380859375,
+      "grad_norm": 0.048828125,
       "learning_rate": 9.98215005121992e-05,
-      "loss": 0.0474,
+      "loss": 0.0744,
       "step": 891
     },
     {
       "epoch": 5.03954802259887,
-      "grad_norm": 0.07958984375,
+      "grad_norm": 0.039794921875,
       "learning_rate": 9.964300159313512e-05,
-      "loss": 0.0897,
+      "loss": 0.0393,
       "step": 892
     },
     {
       "epoch": 5.045197740112994,
-      "grad_norm": 0.0576171875,
+      "grad_norm": 0.05419921875,
       "learning_rate": 9.946450381154264e-05,
-      "loss": 0.0809,
+      "loss": 0.0686,
       "step": 893
     },
     {
       "epoch": 5.0508474576271185,
-      "grad_norm": 0.043701171875,
+      "grad_norm": 0.060546875,
       "learning_rate": 9.928600773615307e-05,
-      "loss": 0.0547,
+      "loss": 0.0743,
       "step": 894
     },
     {
       "epoch": 5.056497175141243,
-      "grad_norm": 0.037841796875,
+      "grad_norm": 0.06494140625,
       "learning_rate": 9.910751393569229e-05,
-      "loss": 0.0186,
+      "loss": 0.0922,
       "step": 895
     },
     {
       "epoch": 5.062146892655368,
-      "grad_norm": 0.07275390625,
+      "grad_norm": 0.042236328125,
       "learning_rate": 9.892902297887881e-05,
-      "loss": 0.1233,
+      "loss": 0.0364,
       "step": 896
     },
     {
       "epoch": 5.067796610169491,
-      "grad_norm": 0.042724609375,
+      "grad_norm": 0.047119140625,
       "learning_rate": 9.875053543442227e-05,
-      "loss": 0.0482,
+      "loss": 0.0541,
       "step": 897
     },
     {
       "epoch": 5.073446327683616,
-      "grad_norm": 0.045654296875,
+      "grad_norm": 0.0791015625,
       "learning_rate": 9.85720518710213e-05,
-      "loss": 0.0357,
+      "loss": 0.1149,
       "step": 898
     },
     {
       "epoch": 5.0790960451977405,
-      "grad_norm": 0.0654296875,
+      "grad_norm": 0.052978515625,
       "learning_rate": 9.839357285736183e-05,
-      "loss": 0.0904,
+      "loss": 0.05,
       "step": 899
     },
     {
       "epoch": 5.084745762711864,
-      "grad_norm": 0.11328125,
+      "grad_norm": 0.04052734375,
       "learning_rate": 9.821509896211539e-05,
-      "loss": 0.0229,
+      "loss": 0.0337,
       "step": 900
     },
     {
       "epoch": 5.084745762711864,
-      "eval_loss": 0.001152906334027648,
-      "eval_runtime": 5.7876,
-      "eval_samples_per_second": 25.745,
-      "eval_steps_per_second": 3.283,
+      "eval_loss": 0.0011893091723322868,
+      "eval_runtime": 5.7843,
+      "eval_samples_per_second": 25.932,
+      "eval_steps_per_second": 3.285,
       "step": 900
     },
     {
       "epoch": 5.090395480225989,
-      "grad_norm": 0.07275390625,
+      "grad_norm": 0.064453125,
       "learning_rate": 9.803663075393718e-05,
-      "loss": 0.1145,
+      "loss": 0.0651,
       "step": 901
     },
     {
       "epoch": 5.096045197740113,
-      "grad_norm": 0.0712890625,
+      "grad_norm": 0.04833984375,
       "learning_rate": 9.78581688014642e-05,
-      "loss": 0.0919,
+      "loss": 0.0488,
       "step": 902
     },
     {
       "epoch": 5.101694915254237,
       "grad_norm": 0.05615234375,
       "learning_rate": 9.767971367331363e-05,
-      "loss": 0.0847,
+      "loss": 0.0567,
       "step": 903
     },
     {
       "epoch": 5.107344632768362,
-      "grad_norm": 0.037353515625,
+      "grad_norm": 0.06005859375,
       "learning_rate": 9.750126593808082e-05,
-      "loss": 0.0353,
+      "loss": 0.0853,
       "step": 904
     },
     {
       "epoch": 5.112994350282486,
-      "grad_norm": 0.06884765625,
+      "grad_norm": 0.028564453125,
       "learning_rate": 9.732282616433756e-05,
-      "loss": 0.101,
+      "loss": 0.0013,
       "step": 905
     },
     {
       "epoch": 5.11864406779661,
-      "grad_norm": 0.060302734375,
+      "grad_norm": 0.060546875,
       "learning_rate": 9.71443949206304e-05,
-      "loss": 0.0739,
+      "loss": 0.0787,
       "step": 906
     },
     {
       "epoch": 5.124293785310734,
-      "grad_norm": 0.04931640625,
+      "grad_norm": 0.0732421875,
       "learning_rate": 9.696597277547853e-05,
-      "loss": 0.0502,
+      "loss": 0.1145,
       "step": 907
     },
     {
       "epoch": 5.129943502824859,
-      "grad_norm": 0.0537109375,
+      "grad_norm": 0.042724609375,
       "learning_rate": 9.678756029737225e-05,
-      "loss": 0.0824,
+      "loss": 0.0492,
       "step": 908
     },
     {
       "epoch": 5.135593220338983,
-      "grad_norm": 0.050048828125,
+      "grad_norm": 0.04541015625,
       "learning_rate": 9.660915805477116e-05,
-      "loss": 0.0537,
+      "loss": 0.0484,
       "step": 909
     },
     {
       "epoch": 5.141242937853107,
-      "grad_norm": 0.07421875,
+      "grad_norm": 0.041015625,
       "learning_rate": 9.643076661610196e-05,
-      "loss": 0.1025,
+      "loss": 0.0251,
       "step": 910
     },
     {
       "epoch": 5.146892655367232,
-      "grad_norm": 0.05078125,
+      "grad_norm": 0.04150390625,
       "learning_rate": 9.625238654975724e-05,
-      "loss": 0.0532,
+      "loss": 0.0511,
       "step": 911
     },
     {
       "epoch": 5.1525423728813555,
-      "grad_norm": 0.052490234375,
+      "grad_norm": 0.0419921875,
       "learning_rate": 9.607401842409317e-05,
-      "loss": 0.085,
+      "loss": 0.0533,
       "step": 912
     },
     {
       "epoch": 5.15819209039548,
-      "grad_norm": 0.0634765625,
+      "grad_norm": 0.078125,
       "learning_rate": 9.589566280742788e-05,
-      "loss": 0.1039,
+      "loss": 0.1093,
       "step": 913
     },
     {
       "epoch": 5.163841807909605,
-      "grad_norm": 0.052490234375,
+      "grad_norm": 0.05029296875,
       "learning_rate": 9.571732026803977e-05,
-      "loss": 0.0521,
+      "loss": 0.0522,
       "step": 914
     },
     {
       "epoch": 5.169491525423728,
-      "grad_norm": 0.0625,
+      "grad_norm": 0.0225830078125,
       "learning_rate": 9.553899137416545e-05,
-      "loss": 0.0878,
+      "loss": 0.0151,
       "step": 915
     },
     {
       "epoch": 5.175141242937853,
-      "grad_norm": 0.06884765625,
+      "grad_norm": 0.06591796875,
       "learning_rate": 9.536067669399806e-05,
-      "loss": 0.1012,
+      "loss": 0.0874,
       "step": 916
     },
     {
       "epoch": 5.1807909604519775,
-      "grad_norm": 0.0308837890625,
+      "grad_norm": 0.0458984375,
       "learning_rate": 9.518237679568553e-05,
-      "loss": 0.0208,
+      "loss": 0.048,
       "step": 917
     },
     {
       "epoch": 5.186440677966102,
-      "grad_norm": 0.0615234375,
+      "grad_norm": 0.06591796875,
       "learning_rate": 9.500409224732863e-05,
-      "loss": 0.0724,
+      "loss": 0.0622,
       "step": 918
     },
     {
       "epoch": 5.192090395480226,
-      "grad_norm": 0.0751953125,
+      "grad_norm": 0.06689453125,
       "learning_rate": 9.482582361697921e-05,
-      "loss": 0.122,
+      "loss": 0.1027,
       "step": 919
     },
     {
       "epoch": 5.19774011299435,
-      "grad_norm": 0.0576171875,
+      "grad_norm": 0.045654296875,
       "learning_rate": 9.464757147263849e-05,
-      "loss": 0.0728,
+      "loss": 0.0529,
       "step": 920
     },
     {
       "epoch": 5.203389830508475,
-      "grad_norm": 0.06103515625,
+      "grad_norm": 0.07080078125,
       "learning_rate": 9.446933638225506e-05,
-      "loss": 0.082,
+      "loss": 0.0905,
       "step": 921
     },
     {
       "epoch": 5.209039548022599,
-      "grad_norm": 0.056640625,
+      "grad_norm": 0.052490234375,
       "learning_rate": 9.42911189137232e-05,
-      "loss": 0.0604,
+      "loss": 0.0679,
       "step": 922
     },
     {
       "epoch": 5.214689265536723,
-      "grad_norm": 0.052490234375,
+      "grad_norm": 0.0712890625,
       "learning_rate": 9.411291963488109e-05,
-      "loss": 0.0714,
+      "loss": 0.099,
       "step": 923
     },
     {
       "epoch": 5.220338983050848,
-      "grad_norm": 0.0419921875,
+      "grad_norm": 0.062255859375,
       "learning_rate": 9.393473911350893e-05,
-      "loss": 0.0345,
+      "loss": 0.0713,
       "step": 924
     },
     {
       "epoch": 5.2259887005649714,
-      "grad_norm": 0.06591796875,
+      "grad_norm": 0.050537109375,
       "learning_rate": 9.37565779173271e-05,
-      "loss": 0.1178,
+      "loss": 0.0516,
       "step": 925
     },
     {
       "epoch": 5.231638418079096,
-      "grad_norm": 0.0341796875,
+      "grad_norm": 0.06201171875,
       "learning_rate": 9.357843661399447e-05,
-      "loss": 0.0304,
+      "loss": 0.1037,
       "step": 926
     },
     {
       "epoch": 5.237288135593221,
-      "grad_norm": 0.06787109375,
+      "grad_norm": 0.060791015625,
       "learning_rate": 9.340031577110658e-05,
-      "loss": 0.1097,
+      "loss": 0.0751,
       "step": 927
     },
     {
       "epoch": 5.242937853107344,
-      "grad_norm": 0.0634765625,
+      "grad_norm": 0.07080078125,
       "learning_rate": 9.322221595619358e-05,
-      "loss": 0.0671,
+      "loss": 0.1077,
       "step": 928
     },
     {
       "epoch": 5.248587570621469,
-      "grad_norm": 0.039794921875,
+      "grad_norm": 0.064453125,
       "learning_rate": 9.304413773671881e-05,
-      "loss": 0.0368,
+      "loss": 0.0865,
       "step": 929
     },
     {
       "epoch": 5.254237288135593,
-      "grad_norm": 0.06591796875,
+      "grad_norm": 0.03759765625,
       "learning_rate": 9.286608168007678e-05,
-      "loss": 0.1071,
+      "loss": 0.034,
       "step": 930
     },
     {
       "epoch": 5.259887005649717,
-      "grad_norm": 0.06640625,
+      "grad_norm": 0.0576171875,
       "learning_rate": 9.268804835359128e-05,
-      "loss": 0.0878,
+      "loss": 0.0675,
       "step": 931
     },
     {
       "epoch": 5.265536723163842,
-      "grad_norm": 0.036865234375,
+      "grad_norm": 0.0289306640625,
       "learning_rate": 9.25100383245138e-05,
-      "loss": 0.0338,
+      "loss": 0.0188,
       "step": 932
     },
     {
       "epoch": 5.271186440677966,
-      "grad_norm": 0.07177734375,
+      "grad_norm": 0.037109375,
       "learning_rate": 9.233205216002147e-05,
-      "loss": 0.1112,
+      "loss": 0.0332,
       "step": 933
     },
     {
       "epoch": 5.27683615819209,
-      "grad_norm": 0.072265625,
+      "grad_norm": 0.049560546875,
       "learning_rate": 9.215409042721552e-05,
-      "loss": 0.1101,
+      "loss": 0.0496,
       "step": 934
     },
     {
       "epoch": 5.2824858757062145,
-      "grad_norm": 0.0615234375,
+      "grad_norm": 0.056396484375,
       "learning_rate": 9.197615369311925e-05,
-      "loss": 0.0784,
+      "loss": 0.0671,
       "step": 935
     },
     {
       "epoch": 5.288135593220339,
-      "grad_norm": 0.0390625,
+      "grad_norm": 0.08447265625,
       "learning_rate": 9.179824252467629e-05,
-      "loss": 0.034,
+      "loss": 0.1107,
       "step": 936
     },
     {
       "epoch": 5.293785310734464,
-      "grad_norm": 0.025634765625,
+      "grad_norm": 0.060302734375,
       "learning_rate": 9.162035748874891e-05,
-      "loss": 0.0184,
+      "loss": 0.0692,
       "step": 937
     },
     {
       "epoch": 5.299435028248587,
-      "grad_norm": 0.06689453125,
+      "grad_norm": 0.048583984375,
       "learning_rate": 9.144249915211605e-05,
-      "loss": 0.0762,
+      "loss": 0.0497,
       "step": 938
     },
     {
       "epoch": 5.305084745762712,
-      "grad_norm": 0.05712890625,
+      "grad_norm": 0.05322265625,
       "learning_rate": 9.126466808147153e-05,
-      "loss": 0.0723,
+      "loss": 0.0364,
       "step": 939
     },
     {
       "epoch": 5.3107344632768365,
-      "grad_norm": 0.0546875,
+      "grad_norm": 0.06884765625,
       "learning_rate": 9.108686484342241e-05,
-      "loss": 0.0557,
+      "loss": 0.0908,
       "step": 940
     },
     {
       "epoch": 5.31638418079096,
-      "grad_norm": 0.059814453125,
+      "grad_norm": 0.10546875,
       "learning_rate": 9.090909000448702e-05,
-      "loss": 0.075,
+      "loss": 0.1612,
       "step": 941
     },
     {
       "epoch": 5.322033898305085,
-      "grad_norm": 0.0703125,
+      "grad_norm": 0.072265625,
       "learning_rate": 9.073134413109312e-05,
-      "loss": 0.0782,
+      "loss": 0.0906,
       "step": 942
     },
     {
       "epoch": 5.327683615819209,
-      "grad_norm": 0.06298828125,
+      "grad_norm": 0.051025390625,
       "learning_rate": 9.055362778957634e-05,
-      "loss": 0.0987,
+      "loss": 0.0369,
       "step": 943
     },
     {
       "epoch": 5.333333333333333,
-      "grad_norm": 0.0380859375,
+      "grad_norm": 0.060302734375,
       "learning_rate": 9.037594154617812e-05,
-      "loss": 0.0343,
+      "loss": 0.0724,
       "step": 944
     },
     {
       "epoch": 5.338983050847458,
-      "grad_norm": 0.04736328125,
+      "grad_norm": 0.05908203125,
       "learning_rate": 9.019828596704394e-05,
-      "loss": 0.0518,
+      "loss": 0.0591,
       "step": 945
     },
     {
       "epoch": 5.338983050847458,
-      "eval_loss": 0.0010936502367258072,
-      "eval_runtime": 5.8508,
-      "eval_samples_per_second": 25.467,
-      "eval_steps_per_second": 3.247,
+      "eval_loss": 0.0011234070407226682,
+      "eval_runtime": 5.79,
+      "eval_samples_per_second": 25.907,
+      "eval_steps_per_second": 3.282,
       "step": 945
     },
     {
       "epoch": 5.344632768361582,
-      "grad_norm": 0.055908203125,
+      "grad_norm": 0.08447265625,
       "learning_rate": 9.002066161822172e-05,
-      "loss": 0.0628,
+      "loss": 0.1037,
       "step": 946
     },
     {
       "epoch": 5.350282485875706,
-      "grad_norm": 0.06396484375,
+      "grad_norm": 0.0546875,
       "learning_rate": 8.984306906565982e-05,
-      "loss": 0.0683,
+      "loss": 0.0498,
       "step": 947
     },
     {
       "epoch": 5.3559322033898304,
-      "grad_norm": 0.04833984375,
+      "grad_norm": 0.0634765625,
       "learning_rate": 8.966550887520521e-05,
-      "loss": 0.0522,
+      "loss": 0.0571,
       "step": 948
     },
     {
       "epoch": 5.361581920903955,
-      "grad_norm": 0.05224609375,
+      "grad_norm": 0.06494140625,
       "learning_rate": 8.948798161260184e-05,
-      "loss": 0.0648,
+      "loss": 0.0746,
       "step": 949
     },
     {
       "epoch": 5.367231638418079,
-      "grad_norm": 0.03955078125,
+      "grad_norm": 0.0751953125,
       "learning_rate": 8.931048784348875e-05,
-      "loss": 0.037,
+      "loss": 0.1047,
       "step": 950
     },
     {
       "epoch": 5.372881355932203,
-      "grad_norm": 0.06396484375,
+      "grad_norm": 0.036376953125,
       "learning_rate": 8.913302813339816e-05,
-      "loss": 0.1068,
+      "loss": 0.0466,
       "step": 951
     },
     {
       "epoch": 5.378531073446328,
-      "grad_norm": 0.0556640625,
+      "grad_norm": 0.0712890625,
       "learning_rate": 8.895560304775387e-05,
-      "loss": 0.0619,
+      "loss": 0.109,
       "step": 952
     },
     {
       "epoch": 5.3841807909604515,
-      "grad_norm": 0.040771484375,
+      "grad_norm": 0.0810546875,
       "learning_rate": 8.877821315186939e-05,
-      "loss": 0.0402,
+      "loss": 0.0878,
       "step": 953
     },
     {
       "epoch": 5.389830508474576,
-      "grad_norm": 0.037109375,
+      "grad_norm": 0.06884765625,
       "learning_rate": 8.860085901094595e-05,
-      "loss": 0.0351,
+      "loss": 0.0919,
       "step": 954
     },
     {
       "epoch": 5.395480225988701,
-      "grad_norm": 0.047119140625,
+      "grad_norm": 0.05419921875,
       "learning_rate": 8.842354119007099e-05,
-      "loss": 0.0655,
+      "loss": 0.0521,
       "step": 955
     },
     {
       "epoch": 5.401129943502825,
-      "grad_norm": 0.049072265625,
+      "grad_norm": 0.0625,
       "learning_rate": 8.824626025421626e-05,
-      "loss": 0.0644,
+      "loss": 0.098,
       "step": 956
     },
     {
       "epoch": 5.406779661016949,
-      "grad_norm": 0.044189453125,
+      "grad_norm": 0.05029296875,
       "learning_rate": 8.806901676823581e-05,
-      "loss": 0.0486,
+      "loss": 0.0572,
       "step": 957
     },
     {
       "epoch": 5.4124293785310735,
-      "grad_norm": 0.037841796875,
+      "grad_norm": 0.061767578125,
       "learning_rate": 8.789181129686452e-05,
-      "loss": 0.0337,
+      "loss": 0.0992,
       "step": 958
     },
     {
       "epoch": 5.418079096045198,
-      "grad_norm": 0.050537109375,
+      "grad_norm": 0.043701171875,
       "learning_rate": 8.771464440471615e-05,
-      "loss": 0.0653,
+      "loss": 0.0377,
       "step": 959
     },
     {
       "epoch": 5.423728813559322,
-      "grad_norm": 0.041259765625,
+      "grad_norm": 0.062255859375,
       "learning_rate": 8.753751665628141e-05,
-      "loss": 0.0521,
+      "loss": 0.0715,
       "step": 960
     },
     {
       "epoch": 5.429378531073446,
-      "grad_norm": 0.059326171875,
+      "grad_norm": 0.0947265625,
       "learning_rate": 8.736042861592645e-05,
-      "loss": 0.0763,
+      "loss": 0.0939,
       "step": 961
     },
     {
       "epoch": 5.435028248587571,
-      "grad_norm": 0.043212890625,
+      "grad_norm": 0.06005859375,
       "learning_rate": 8.718338084789072e-05,
-      "loss": 0.0356,
+      "loss": 0.0756,
       "step": 962
     },
     {
       "epoch": 5.440677966101695,
-      "grad_norm": 0.07373046875,
+      "grad_norm": 0.052978515625,
       "learning_rate": 8.700637391628556e-05,
-      "loss": 0.1104,
+      "loss": 0.0521,
       "step": 963
     },
     {
       "epoch": 5.446327683615819,
-      "grad_norm": 0.0419921875,
+      "grad_norm": 0.05810546875,
       "learning_rate": 8.682940838509207e-05,
-      "loss": 0.0399,
+      "loss": 0.0558,
       "step": 964
     },
     {
       "epoch": 5.451977401129944,
-      "grad_norm": 0.036865234375,
+      "grad_norm": 0.07177734375,
       "learning_rate": 8.665248481815941e-05,
-      "loss": 0.0349,
+      "loss": 0.0853,
       "step": 965
     },
     {
       "epoch": 5.4576271186440675,
-      "grad_norm": 0.0673828125,
+      "grad_norm": 0.08056640625,
       "learning_rate": 8.647560377920311e-05,
-      "loss": 0.1138,
+      "loss": 0.1004,
       "step": 966
     },
     {
       "epoch": 5.463276836158192,
-      "grad_norm": 0.072265625,
+      "grad_norm": 0.049072265625,
       "learning_rate": 8.629876583180321e-05,
-      "loss": 0.1022,
+      "loss": 0.0374,
       "step": 967
     },
     {
       "epoch": 5.468926553672317,
-      "grad_norm": 0.07080078125,
+      "grad_norm": 0.0634765625,
       "learning_rate": 8.612197153940238e-05,
-      "loss": 0.1012,
+      "loss": 0.0732,
       "step": 968
     },
     {
       "epoch": 5.47457627118644,
-      "grad_norm": 0.0595703125,
+      "grad_norm": 0.054931640625,
       "learning_rate": 8.594522146530422e-05,
-      "loss": 0.0707,
+      "loss": 0.0656,
       "step": 969
     },
     {
       "epoch": 5.480225988700565,
-      "grad_norm": 0.054443359375,
+      "grad_norm": 0.045654296875,
       "learning_rate": 8.57685161726715e-05,
-      "loss": 0.0775,
+      "loss": 0.0472,
       "step": 970
     },
     {
       "epoch": 5.4858757062146895,
-      "grad_norm": 0.046142578125,
+      "grad_norm": 0.068359375,
       "learning_rate": 8.559185622452421e-05,
-      "loss": 0.0522,
+      "loss": 0.0898,
       "step": 971
     },
     {
       "epoch": 5.491525423728813,
-      "grad_norm": 0.045654296875,
+      "grad_norm": 0.10693359375,
       "learning_rate": 8.54152421837379e-05,
-      "loss": 0.0392,
+      "loss": 0.1492,
       "step": 972
     },
     {
       "epoch": 5.497175141242938,
-      "grad_norm": 0.0693359375,
+      "grad_norm": 0.06640625,
       "learning_rate": 8.523867461304196e-05,
-      "loss": 0.1182,
+      "loss": 0.0631,
       "step": 973
     },
     {
       "epoch": 5.502824858757062,
-      "grad_norm": 0.0771484375,
+      "grad_norm": 0.083984375,
       "learning_rate": 8.50621540750175e-05,
-      "loss": 0.1138,
+      "loss": 0.1225,
       "step": 974
     },
     {
       "epoch": 5.508474576271187,
-      "grad_norm": 0.08203125,
+      "grad_norm": 0.07177734375,
       "learning_rate": 8.488568113209593e-05,
-      "loss": 0.1231,
+      "loss": 0.0919,
       "step": 975
     },
     {
       "epoch": 5.5141242937853105,
-      "grad_norm": 0.0081787109375,
+      "grad_norm": 0.068359375,
       "learning_rate": 8.470925634655702e-05,
-      "loss": 0.0004,
+      "loss": 0.0702,
       "step": 976
     },
     {
       "epoch": 5.519774011299435,
-      "grad_norm": 0.059814453125,
+      "grad_norm": 0.07568359375,
       "learning_rate": 8.453288028052699e-05,
-      "loss": 0.0713,
+      "loss": 0.1103,
       "step": 977
     },
     {
       "epoch": 5.52542372881356,
-      "grad_norm": 0.045654296875,
+      "grad_norm": 0.06640625,
       "learning_rate": 8.435655349597689e-05,
-      "loss": 0.049,
+      "loss": 0.0829,
       "step": 978
     },
     {
       "epoch": 5.531073446327683,
-      "grad_norm": 0.06884765625,
+      "grad_norm": 0.05859375,
       "learning_rate": 8.418027655472085e-05,
-      "loss": 0.0826,
+      "loss": 0.0547,
       "step": 979
     },
     {
       "epoch": 5.536723163841808,
-      "grad_norm": 0.0123291015625,
+      "grad_norm": 0.06640625,
       "learning_rate": 8.400405001841399e-05,
-      "loss": 0.0007,
+      "loss": 0.0856,
       "step": 980
     },
     {
       "epoch": 5.5423728813559325,
-      "grad_norm": 0.07958984375,
+      "grad_norm": 0.08837890625,
       "learning_rate": 8.382787444855093e-05,
-      "loss": 0.0938,
+      "loss": 0.1497,
       "step": 981
     },
     {
       "epoch": 5.548022598870056,
-      "grad_norm": 0.0546875,
+      "grad_norm": 0.040771484375,
       "learning_rate": 8.365175040646402e-05,
-      "loss": 0.0521,
+      "loss": 0.0382,
       "step": 982
     },
     {
       "epoch": 5.553672316384181,
-      "grad_norm": 0.10107421875,
+      "grad_norm": 0.054443359375,
       "learning_rate": 8.347567845332115e-05,
-      "loss": 0.0868,
+      "loss": 0.0533,
       "step": 983
     },
     {
       "epoch": 5.559322033898305,
-      "grad_norm": 0.054443359375,
+      "grad_norm": 0.049072265625,
       "learning_rate": 8.329965915012451e-05,
-      "loss": 0.0692,
+      "loss": 0.0517,
       "step": 984
     },
     {
       "epoch": 5.564971751412429,
-      "grad_norm": 0.04736328125,
+      "grad_norm": 0.06298828125,
       "learning_rate": 8.312369305770843e-05,
-      "loss": 0.0358,
+      "loss": 0.0937,
       "step": 985
     },
     {
       "epoch": 5.570621468926554,
-      "grad_norm": 0.053466796875,
+      "grad_norm": 0.04638671875,
       "learning_rate": 8.294778073673762e-05,
-      "loss": 0.0519,
+      "loss": 0.0505,
       "step": 986
     },
     {
       "epoch": 5.576271186440678,
-      "grad_norm": 0.07177734375,
+      "grad_norm": 0.06298828125,
       "learning_rate": 8.277192274770566e-05,
-      "loss": 0.0934,
+      "loss": 0.084,
       "step": 987
     },
     {
       "epoch": 5.581920903954802,
-      "grad_norm": 0.04833984375,
+      "grad_norm": 0.07373046875,
       "learning_rate": 8.259611965093282e-05,
-      "loss": 0.0495,
+      "loss": 0.1135,
       "step": 988
     },
     {
       "epoch": 5.5875706214689265,
-      "grad_norm": 0.0576171875,
+      "grad_norm": 0.04931640625,
       "learning_rate": 8.242037200656455e-05,
-      "loss": 0.0534,
+      "loss": 0.0505,
       "step": 989
     },
     {
       "epoch": 5.593220338983051,
-      "grad_norm": 0.034423828125,
+      "grad_norm": 0.035888671875,
       "learning_rate": 8.224468037456969e-05,
-      "loss": 0.019,
+      "loss": 0.0171,
       "step": 990
     },
     {
       "epoch": 5.593220338983051,
-      "eval_loss": 0.001057442044839263,
-      "eval_runtime": 5.7483,
-      "eval_samples_per_second": 25.921,
-      "eval_steps_per_second": 3.305,
+      "eval_loss": 0.001048309844918549,
+      "eval_runtime": 5.8181,
+      "eval_samples_per_second": 25.782,
+      "eval_steps_per_second": 3.266,
       "step": 990
     },
     {
       "epoch": 5.598870056497175,
-      "grad_norm": 0.0869140625,
+      "grad_norm": 0.052001953125,
       "learning_rate": 8.20690453147385e-05,
-      "loss": 0.1213,
+      "loss": 0.0695,
       "step": 991
     },
     {
       "epoch": 5.604519774011299,
-      "grad_norm": 0.06689453125,
+      "grad_norm": 0.03857421875,
       "learning_rate": 8.189346738668099e-05,
-      "loss": 0.1004,
+      "loss": 0.0349,
       "step": 992
     },
     {
       "epoch": 5.610169491525424,
-      "grad_norm": 0.07763671875,
+      "grad_norm": 0.016357421875,
       "learning_rate": 8.171794714982528e-05,
-      "loss": 0.1002,
+      "loss": 0.0006,
       "step": 993
     },
     {
       "epoch": 5.6158192090395485,
-      "grad_norm": 0.0634765625,
+      "grad_norm": 0.059814453125,
       "learning_rate": 8.154248516341548e-05,
-      "loss": 0.0729,
+      "loss": 0.0734,
       "step": 994
     },
     {
       "epoch": 5.621468926553672,
-      "grad_norm": 0.07177734375,
+      "grad_norm": 0.052490234375,
       "learning_rate": 8.136708198651021e-05,
-      "loss": 0.109,
+      "loss": 0.0528,
       "step": 995
     },
     {
       "epoch": 5.627118644067797,
-      "grad_norm": 0.0751953125,
+      "grad_norm": 0.06494140625,
       "learning_rate": 8.119173817798077e-05,
-      "loss": 0.0905,
+      "loss": 0.0766,
       "step": 996
     },
     {
       "epoch": 5.632768361581921,
-      "grad_norm": 0.062255859375,
+      "grad_norm": 0.0791015625,
       "learning_rate": 8.101645429650914e-05,
-      "loss": 0.0684,
+      "loss": 0.0917,
       "step": 997
     },
     {
       "epoch": 5.638418079096045,
-      "grad_norm": 0.07763671875,
+      "grad_norm": 0.09423828125,
       "learning_rate": 8.084123090058646e-05,
-      "loss": 0.1102,
+      "loss": 0.1787,
       "step": 998
     },
     {
       "epoch": 5.6440677966101696,
-      "grad_norm": 0.08447265625,
+      "grad_norm": 0.08349609375,
       "learning_rate": 8.06660685485112e-05,
-      "loss": 0.1394,
+      "loss": 0.1206,
       "step": 999
     },
     {
       "epoch": 5.649717514124294,
-      "grad_norm": 0.05712890625,
+      "grad_norm": 0.055419921875,
       "learning_rate": 8.049096779838719e-05,
-      "loss": 0.0815,
+      "loss": 0.0645,
       "step": 1000
     },
     {
       "epoch": 5.655367231638418,
-      "grad_norm": 0.06201171875,
+      "grad_norm": 0.09716796875,
       "learning_rate": 8.031592920812207e-05,
-      "loss": 0.0931,
+      "loss": 0.1413,
       "step": 1001
     },
     {
       "epoch": 5.661016949152542,
-      "grad_norm": 0.033203125,
+      "grad_norm": 0.076171875,
       "learning_rate": 8.014095333542548e-05,
-      "loss": 0.033,
+      "loss": 0.0857,
       "step": 1002
     },
     {
       "epoch": 5.666666666666667,
-      "grad_norm": 0.0673828125,
+      "grad_norm": 0.05810546875,
       "learning_rate": 7.996604073780706e-05,
-      "loss": 0.104,
+      "loss": 0.069,
       "step": 1003
     },
     {
       "epoch": 5.672316384180791,
-      "grad_norm": 0.05908203125,
+      "grad_norm": 0.052001953125,
       "learning_rate": 7.979119197257505e-05,
-      "loss": 0.0905,
+      "loss": 0.0532,
       "step": 1004
     },
     {
       "epoch": 5.677966101694915,
-      "grad_norm": 0.05859375,
+      "grad_norm": 0.0654296875,
       "learning_rate": 7.961640759683416e-05,
-      "loss": 0.0688,
+      "loss": 0.0738,
       "step": 1005
     },
     {
       "epoch": 5.68361581920904,
-      "grad_norm": 0.05615234375,
+      "grad_norm": 0.058837890625,
       "learning_rate": 7.944168816748395e-05,
-      "loss": 0.0724,
+      "loss": 0.0698,
       "step": 1006
     },
     {
       "epoch": 5.6892655367231635,
-      "grad_norm": 0.0113525390625,
+      "grad_norm": 0.0615234375,
       "learning_rate": 7.926703424121713e-05,
-      "loss": 0.0006,
+      "loss": 0.0668,
       "step": 1007
     },
     {
       "epoch": 5.694915254237288,
-      "grad_norm": 0.08056640625,
+      "grad_norm": 0.064453125,
       "learning_rate": 7.909244637451767e-05,
-      "loss": 0.0985,
+      "loss": 0.1004,
       "step": 1008
     },
     {
       "epoch": 5.700564971751413,
-      "grad_norm": 0.052001953125,
+      "grad_norm": 0.060302734375,
       "learning_rate": 7.891792512365897e-05,
-      "loss": 0.0415,
+      "loss": 0.0863,
       "step": 1009
     },
     {
       "epoch": 5.706214689265536,
-      "grad_norm": 0.06591796875,
+      "grad_norm": 0.0693359375,
       "learning_rate": 7.874347104470234e-05,
-      "loss": 0.0892,
+      "loss": 0.1059,
       "step": 1010
     },
     {
       "epoch": 5.711864406779661,
-      "grad_norm": 0.07958984375,
+      "grad_norm": 0.07763671875,
       "learning_rate": 7.856908469349495e-05,
-      "loss": 0.1116,
+      "loss": 0.1318,
       "step": 1011
     },
     {
       "epoch": 5.7175141242937855,
-      "grad_norm": 0.068359375,
+      "grad_norm": 0.0439453125,
       "learning_rate": 7.839476662566816e-05,
-      "loss": 0.123,
+      "loss": 0.0477,
       "step": 1012
     },
     {
       "epoch": 5.72316384180791,
-      "grad_norm": 0.046875,
+      "grad_norm": 0.06494140625,
       "learning_rate": 7.822051739663588e-05,
-      "loss": 0.0626,
+      "loss": 0.0855,
       "step": 1013
     },
     {
       "epoch": 5.728813559322034,
-      "grad_norm": 0.0361328125,
+      "grad_norm": 0.07373046875,
       "learning_rate": 7.804633756159259e-05,
-      "loss": 0.0312,
+      "loss": 0.0874,
       "step": 1014
     },
     {
       "epoch": 5.734463276836158,
-      "grad_norm": 0.062255859375,
+      "grad_norm": 0.07275390625,
       "learning_rate": 7.787222767551164e-05,
-      "loss": 0.0702,
+      "loss": 0.0877,
       "step": 1015
     },
     {
       "epoch": 5.740112994350282,
-      "grad_norm": 0.061279296875,
+      "grad_norm": 0.053955078125,
       "learning_rate": 7.769818829314366e-05,
-      "loss": 0.072,
+      "loss": 0.0571,
       "step": 1016
     },
     {
       "epoch": 5.745762711864407,
-      "grad_norm": 0.0703125,
+      "grad_norm": 0.03955078125,
       "learning_rate": 7.752421996901448e-05,
-      "loss": 0.1181,
+      "loss": 0.0339,
       "step": 1017
     },
     {
       "epoch": 5.751412429378531,
-      "grad_norm": 0.0615234375,
+      "grad_norm": 0.080078125,
       "learning_rate": 7.735032325742355e-05,
-      "loss": 0.0617,
+      "loss": 0.1182,
       "step": 1018
     },
     {
       "epoch": 5.757062146892656,
-      "grad_norm": 0.0693359375,
+      "grad_norm": 0.05615234375,
       "learning_rate": 7.717649871244231e-05,
-      "loss": 0.0805,
+      "loss": 0.0729,
       "step": 1019
     },
     {
       "epoch": 5.762711864406779,
-      "grad_norm": 0.0888671875,
+      "grad_norm": 0.034912109375,
       "learning_rate": 7.700274688791196e-05,
-      "loss": 0.1372,
+      "loss": 0.0305,
       "step": 1020
     },
     {
       "epoch": 5.768361581920904,
-      "grad_norm": 0.0634765625,
+      "grad_norm": 0.0478515625,
       "learning_rate": 7.682906833744229e-05,
-      "loss": 0.0585,
+      "loss": 0.0376,
       "step": 1021
     },
     {
       "epoch": 5.7740112994350286,
-      "grad_norm": 0.05126953125,
+      "grad_norm": 0.06201171875,
       "learning_rate": 7.66554636144095e-05,
-      "loss": 0.066,
+      "loss": 0.1,
       "step": 1022
     },
     {
       "epoch": 5.779661016949152,
-      "grad_norm": 0.046875,
+      "grad_norm": 0.044677734375,
       "learning_rate": 7.648193327195447e-05,
-      "loss": 0.0375,
+      "loss": 0.0342,
       "step": 1023
     },
     {
       "epoch": 5.785310734463277,
-      "grad_norm": 0.07080078125,
+      "grad_norm": 0.045654296875,
       "learning_rate": 7.63084778629813e-05,
-      "loss": 0.0951,
+      "loss": 0.0609,
       "step": 1024
     },
     {
       "epoch": 5.790960451977401,
-      "grad_norm": 0.0595703125,
+      "grad_norm": 0.0791015625,
       "learning_rate": 7.613509794015517e-05,
-      "loss": 0.066,
+      "loss": 0.1191,
       "step": 1025
     },
     {
       "epoch": 5.796610169491525,
-      "grad_norm": 0.07177734375,
+      "grad_norm": 0.0810546875,
       "learning_rate": 7.596179405590076e-05,
-      "loss": 0.0946,
+      "loss": 0.1208,
       "step": 1026
     },
     {
       "epoch": 5.80225988700565,
-      "grad_norm": 0.07080078125,
+      "grad_norm": 0.053955078125,
       "learning_rate": 7.578856676240056e-05,
-      "loss": 0.0694,
+      "loss": 0.0514,
       "step": 1027
     },
     {
       "epoch": 5.807909604519774,
-      "grad_norm": 0.047607421875,
+      "grad_norm": 0.06689453125,
       "learning_rate": 7.561541661159298e-05,
-      "loss": 0.0485,
+      "loss": 0.0901,
       "step": 1028
     },
     {
       "epoch": 5.813559322033898,
-      "grad_norm": 0.051025390625,
+      "grad_norm": 0.057373046875,
       "learning_rate": 7.544234415517058e-05,
-      "loss": 0.0482,
+      "loss": 0.038,
       "step": 1029
     },
     {
       "epoch": 5.8192090395480225,
-      "grad_norm": 0.0693359375,
+      "grad_norm": 0.06591796875,
       "learning_rate": 7.526934994457844e-05,
-      "loss": 0.1074,
+      "loss": 0.084,
       "step": 1030
     },
     {
       "epoch": 5.824858757062147,
-      "grad_norm": 0.044677734375,
+      "grad_norm": 0.07177734375,
       "learning_rate": 7.509643453101238e-05,
-      "loss": 0.0523,
+      "loss": 0.0967,
       "step": 1031
     },
     {
       "epoch": 5.830508474576272,
-      "grad_norm": 0.061767578125,
+      "grad_norm": 0.07275390625,
       "learning_rate": 7.4923598465417e-05,
-      "loss": 0.0537,
+      "loss": 0.1224,
       "step": 1032
     },
     {
       "epoch": 5.836158192090395,
-      "grad_norm": 0.06640625,
+      "grad_norm": 0.049560546875,
       "learning_rate": 7.47508422984842e-05,
-      "loss": 0.1016,
+      "loss": 0.0544,
       "step": 1033
     },
     {
       "epoch": 5.84180790960452,
-      "grad_norm": 0.06884765625,
+      "grad_norm": 0.033203125,
       "learning_rate": 7.457816658065134e-05,
-      "loss": 0.0904,
+      "loss": 0.0201,
       "step": 1034
     },
     {
       "epoch": 5.847457627118644,
-      "grad_norm": 0.07568359375,
+      "grad_norm": 0.06591796875,
       "learning_rate": 7.440557186209926e-05,
-      "loss": 0.09,
+      "loss": 0.0732,
       "step": 1035
     },
     {
       "epoch": 5.847457627118644,
-      "eval_loss": 0.0010066000977531075,
-      "eval_runtime": 5.7122,
-      "eval_samples_per_second": 26.084,
-      "eval_steps_per_second": 3.326,
+      "eval_loss": 0.0010012869024649262,
+      "eval_runtime": 5.7155,
+      "eval_samples_per_second": 26.245,
+      "eval_steps_per_second": 3.324,
       "step": 1035
     },
     {
       "epoch": 5.853107344632768,
-      "grad_norm": 0.06982421875,
+      "grad_norm": 0.06103515625,
       "learning_rate": 7.423305869275097e-05,
-      "loss": 0.1062,
+      "loss": 0.0678,
       "step": 1036
     },
     {
       "epoch": 5.858757062146893,
-      "grad_norm": 0.07373046875,
+      "grad_norm": 0.056640625,
       "learning_rate": 7.40606276222695e-05,
-      "loss": 0.1066,
+      "loss": 0.0686,
       "step": 1037
     },
     {
       "epoch": 5.864406779661017,
-      "grad_norm": 0.0458984375,
+      "grad_norm": 0.046630859375,
       "learning_rate": 7.388827920005627e-05,
-      "loss": 0.0349,
+      "loss": 0.0558,
       "step": 1038
     },
     {
       "epoch": 5.870056497175141,
-      "grad_norm": 0.06005859375,
+      "grad_norm": 0.04833984375,
       "learning_rate": 7.371601397524945e-05,
-      "loss": 0.0768,
+      "loss": 0.0474,
       "step": 1039
     },
     {
       "epoch": 5.875706214689266,
-      "grad_norm": 0.055419921875,
+      "grad_norm": 0.06103515625,
       "learning_rate": 7.354383249672212e-05,
-      "loss": 0.0682,
+      "loss": 0.1006,
       "step": 1040
     },
     {
       "epoch": 5.88135593220339,
-      "grad_norm": 0.03125,
+      "grad_norm": 0.068359375,
       "learning_rate": 7.337173531308046e-05,
-      "loss": 0.0295,
+      "loss": 0.0997,
       "step": 1041
     },
     {
       "epoch": 5.887005649717514,
-      "grad_norm": 0.0576171875,
+      "grad_norm": 0.052978515625,
       "learning_rate": 7.319972297266214e-05,
-      "loss": 0.07,
+      "loss": 0.0654,
       "step": 1042
     },
     {
       "epoch": 5.892655367231638,
-      "grad_norm": 0.08447265625,
+      "grad_norm": 0.07568359375,
       "learning_rate": 7.302779602353442e-05,
-      "loss": 0.1392,
+      "loss": 0.0938,
       "step": 1043
     },
     {
       "epoch": 5.898305084745763,
-      "grad_norm": 0.06689453125,
+      "grad_norm": 0.06396484375,
       "learning_rate": 7.285595501349258e-05,
-      "loss": 0.0775,
+      "loss": 0.0848,
       "step": 1044
     },
     {
       "epoch": 5.903954802259887,
-      "grad_norm": 0.0703125,
+      "grad_norm": 0.039306640625,
       "learning_rate": 7.268420049005807e-05,
-      "loss": 0.094,
+      "loss": 0.0328,
       "step": 1045
     },
     {
       "epoch": 5.909604519774011,
-      "grad_norm": 0.0810546875,
+      "grad_norm": 0.047119140625,
       "learning_rate": 7.251253300047663e-05,
-      "loss": 0.0529,
+      "loss": 0.0505,
       "step": 1046
     },
     {
       "epoch": 5.915254237288136,
-      "grad_norm": 0.0517578125,
+      "grad_norm": 0.0654296875,
       "learning_rate": 7.23409530917169e-05,
-      "loss": 0.051,
+      "loss": 0.0762,
       "step": 1047
     },
     {
       "epoch": 5.9209039548022595,
-      "grad_norm": 0.06982421875,
+      "grad_norm": 0.033447265625,
       "learning_rate": 7.216946131046837e-05,
-      "loss": 0.0849,
+      "loss": 0.0313,
       "step": 1048
     },
     {
       "epoch": 5.926553672316384,
-      "grad_norm": 0.099609375,
+      "grad_norm": 0.052734375,
       "learning_rate": 7.199805820313968e-05,
-      "loss": 0.1465,
+      "loss": 0.0552,
       "step": 1049
     },
     {
       "epoch": 5.932203389830509,
-      "grad_norm": 0.043212890625,
+      "grad_norm": 0.060546875,
       "learning_rate": 7.182674431585704e-05,
-      "loss": 0.034,
+      "loss": 0.0714,
       "step": 1050
     },
     {
       "epoch": 5.937853107344633,
-      "grad_norm": 0.0830078125,
+      "grad_norm": 0.06591796875,
       "learning_rate": 7.165552019446235e-05,
-      "loss": 0.1395,
+      "loss": 0.0744,
       "step": 1051
     },
     {
       "epoch": 5.943502824858757,
-      "grad_norm": 0.044677734375,
+      "grad_norm": 0.06298828125,
       "learning_rate": 7.148438638451144e-05,
-      "loss": 0.036,
+      "loss": 0.0848,
       "step": 1052
     },
     {
       "epoch": 5.9491525423728815,
-      "grad_norm": 0.0062255859375,
+      "grad_norm": 0.0751953125,
       "learning_rate": 7.131334343127249e-05,
-      "loss": 0.0003,
+      "loss": 0.1139,
       "step": 1053
     },
     {
       "epoch": 5.954802259887005,
-      "grad_norm": 0.026123046875,
+      "grad_norm": 0.041015625,
       "learning_rate": 7.114239187972416e-05,
-      "loss": 0.0162,
+      "loss": 0.033,
       "step": 1054
     },
     {
       "epoch": 5.96045197740113,
-      "grad_norm": 0.07666015625,
+      "grad_norm": 0.0615234375,
       "learning_rate": 7.097153227455379e-05,
-      "loss": 0.1096,
+      "loss": 0.0663,
       "step": 1055
     },
     {
       "epoch": 5.966101694915254,
-      "grad_norm": 0.0390625,
+      "grad_norm": 0.07470703125,
       "learning_rate": 7.080076516015584e-05,
-      "loss": 0.0318,
+      "loss": 0.0889,
       "step": 1056
     },
     {
       "epoch": 5.971751412429379,
-      "grad_norm": 0.078125,
+      "grad_norm": 0.055908203125,
       "learning_rate": 7.063009108063017e-05,
-      "loss": 0.1249,
+      "loss": 0.0507,
       "step": 1057
     },
     {
       "epoch": 5.977401129943503,
-      "grad_norm": 0.07421875,
+      "grad_norm": 0.04833984375,
       "learning_rate": 7.045951057978e-05,
-      "loss": 0.0886,
+      "loss": 0.0519,
       "step": 1058
     },
     {
       "epoch": 5.983050847457627,
-      "grad_norm": 0.056884765625,
+      "grad_norm": 0.07763671875,
       "learning_rate": 7.028902420111055e-05,
-      "loss": 0.0817,
+      "loss": 0.0875,
       "step": 1059
     },
     {
       "epoch": 5.988700564971752,
-      "grad_norm": 0.076171875,
+      "grad_norm": 0.07177734375,
       "learning_rate": 7.011863248782711e-05,
-      "loss": 0.1085,
+      "loss": 0.079,
       "step": 1060
     },
     {
       "epoch": 5.994350282485875,
-      "grad_norm": 0.059326171875,
+      "grad_norm": 0.076171875,
       "learning_rate": 6.99483359828333e-05,
-      "loss": 0.0676,
+      "loss": 0.0926,
       "step": 1061
     },
     {
       "epoch": 6.0,
-      "grad_norm": 0.0751953125,
+      "grad_norm": 0.068359375,
       "learning_rate": 6.977813522872943e-05,
-      "loss": 0.098,
+      "loss": 0.0883,
       "step": 1062
     },
     {
       "epoch": 6.005649717514125,
-      "grad_norm": 0.06494140625,
+      "grad_norm": 0.061767578125,
       "learning_rate": 6.960803076781075e-05,
-      "loss": 0.0739,
+      "loss": 0.0833,
       "step": 1063
     },
     {
       "epoch": 6.011299435028248,
-      "grad_norm": 0.0654296875,
+      "grad_norm": 0.056884765625,
       "learning_rate": 6.94380231420656e-05,
-      "loss": 0.0743,
+      "loss": 0.0712,
       "step": 1064
     },
     {
       "epoch": 6.016949152542373,
-      "grad_norm": 0.0069580078125,
+      "grad_norm": 0.055419921875,
       "learning_rate": 6.92681128931739e-05,
-      "loss": 0.0004,
+      "loss": 0.0493,
       "step": 1065
     },
     {
       "epoch": 6.022598870056497,
-      "grad_norm": 0.0517578125,
+      "grad_norm": 0.076171875,
       "learning_rate": 6.909830056250527e-05,
-      "loss": 0.0703,
+      "loss": 0.0932,
       "step": 1066
     },
     {
       "epoch": 6.028248587570621,
-      "grad_norm": 0.007781982421875,
+      "grad_norm": 0.07275390625,
       "learning_rate": 6.892858669111728e-05,
-      "loss": 0.0004,
+      "loss": 0.0933,
       "step": 1067
     },
     {
       "epoch": 6.033898305084746,
-      "grad_norm": 0.04931640625,
+      "grad_norm": 0.0625,
       "learning_rate": 6.875897181975386e-05,
-      "loss": 0.0521,
+      "loss": 0.07,
       "step": 1068
     },
     {
       "epoch": 6.03954802259887,
-      "grad_norm": 0.0751953125,
+      "grad_norm": 0.064453125,
       "learning_rate": 6.858945648884351e-05,
-      "loss": 0.0884,
+      "loss": 0.0855,
       "step": 1069
     },
     {
       "epoch": 6.045197740112994,
-      "grad_norm": 0.0810546875,
+      "grad_norm": 0.05419921875,
       "learning_rate": 6.842004123849752e-05,
-      "loss": 0.123,
+      "loss": 0.056,
       "step": 1070
     },
     {
       "epoch": 6.0508474576271185,
-      "grad_norm": 0.06103515625,
+      "grad_norm": 0.07080078125,
       "learning_rate": 6.825072660850833e-05,
-      "loss": 0.0762,
+      "loss": 0.1177,
       "step": 1071
     },
     {
       "epoch": 6.056497175141243,
-      "grad_norm": 0.03857421875,
+      "grad_norm": 0.0625,
       "learning_rate": 6.808151313834776e-05,
-      "loss": 0.0321,
+      "loss": 0.0832,
       "step": 1072
     },
     {
       "epoch": 6.062146892655368,
-      "grad_norm": 0.06396484375,
+      "grad_norm": 0.06982421875,
       "learning_rate": 6.791240136716534e-05,
-      "loss": 0.0742,
+      "loss": 0.0859,
       "step": 1073
     },
     {
       "epoch": 6.067796610169491,
-      "grad_norm": 0.042724609375,
+      "grad_norm": 0.053466796875,
       "learning_rate": 6.774339183378663e-05,
-      "loss": 0.0359,
+      "loss": 0.0623,
       "step": 1074
     },
     {
       "epoch": 6.073446327683616,
-      "grad_norm": 0.0625,
+      "grad_norm": 0.05712890625,
       "learning_rate": 6.757448507671128e-05,
-      "loss": 0.0717,
+      "loss": 0.0388,
       "step": 1075
     },
     {
       "epoch": 6.0790960451977405,
-      "grad_norm": 0.0810546875,
+      "grad_norm": 0.06689453125,
       "learning_rate": 6.740568163411158e-05,
-      "loss": 0.1111,
+      "loss": 0.087,
       "step": 1076
     },
     {
       "epoch": 6.084745762711864,
-      "grad_norm": 0.08935546875,
+      "grad_norm": 0.0478515625,
       "learning_rate": 6.723698204383066e-05,
-      "loss": 0.1571,
+      "loss": 0.0331,
       "step": 1077
     },
     {
       "epoch": 6.090395480225989,
-      "grad_norm": 0.072265625,
+      "grad_norm": 0.0849609375,
       "learning_rate": 6.70683868433807e-05,
-      "loss": 0.0915,
+      "loss": 0.1339,
       "step": 1078
     },
     {
       "epoch": 6.096045197740113,
-      "grad_norm": 0.06298828125,
+      "grad_norm": 0.083984375,
       "learning_rate": 6.689989656994124e-05,
-      "loss": 0.0737,
+      "loss": 0.1228,
       "step": 1079
     },
     {
       "epoch": 6.101694915254237,
-      "grad_norm": 0.07177734375,
+      "grad_norm": 0.056640625,
       "learning_rate": 6.673151176035762e-05,
-      "loss": 0.0907,
+      "loss": 0.0538,
       "step": 1080
     },
     {
       "epoch": 6.101694915254237,
-      "eval_loss": 0.0009795993100851774,
-      "eval_runtime": 5.7777,
-      "eval_samples_per_second": 25.789,
-      "eval_steps_per_second": 3.288,
+      "eval_loss": 0.0009557561133988202,
+      "eval_runtime": 5.7397,
+      "eval_samples_per_second": 26.134,
+      "eval_steps_per_second": 3.31,
       "step": 1080
     },
     {
       "epoch": 6.107344632768362,
-      "grad_norm": 0.054931640625,
+      "grad_norm": 0.083984375,
       "learning_rate": 6.656323295113904e-05,
-      "loss": 0.0705,
+      "loss": 0.0992,
       "step": 1081
     },
     {
       "epoch": 6.112994350282486,
-      "grad_norm": 0.06884765625,
+      "grad_norm": 0.06787109375,
       "learning_rate": 6.639506067845697e-05,
-      "loss": 0.0835,
+      "loss": 0.0685,
       "step": 1082
     },
     {
       "epoch": 6.11864406779661,
-      "grad_norm": 0.058349609375,
+      "grad_norm": 0.05078125,
       "learning_rate": 6.622699547814356e-05,
-      "loss": 0.0709,
+      "loss": 0.0351,
       "step": 1083
     },
     {
       "epoch": 6.124293785310734,
-      "grad_norm": 0.0712890625,
+      "grad_norm": 0.06787109375,
       "learning_rate": 6.60590378856896e-05,
-      "loss": 0.0904,
+      "loss": 0.0834,
       "step": 1084
     },
     {
       "epoch": 6.129943502824859,
-      "grad_norm": 0.07666015625,
+      "grad_norm": 0.0546875,
       "learning_rate": 6.589118843624315e-05,
-      "loss": 0.1063,
+      "loss": 0.053,
       "step": 1085
     },
     {
       "epoch": 6.135593220338983,
-      "grad_norm": 0.0308837890625,
+      "grad_norm": 0.0517578125,
       "learning_rate": 6.572344766460776e-05,
-      "loss": 0.018,
+      "loss": 0.0346,
       "step": 1086
     },
     {
       "epoch": 6.141242937853107,
-      "grad_norm": 0.07373046875,
+      "grad_norm": 0.06982421875,
       "learning_rate": 6.555581610524053e-05,
-      "loss": 0.0996,
+      "loss": 0.0816,
       "step": 1087
     },
     {
       "epoch": 6.146892655367232,
-      "grad_norm": 0.07421875,
+      "grad_norm": 0.057861328125,
       "learning_rate": 6.538829429225069e-05,
-      "loss": 0.0904,
+      "loss": 0.0398,
       "step": 1088
     },
     {
       "epoch": 6.1525423728813555,
-      "grad_norm": 0.0634765625,
+      "grad_norm": 0.05810546875,
       "learning_rate": 6.522088275939788e-05,
-      "loss": 0.0696,
+      "loss": 0.061,
       "step": 1089
     },
     {
       "epoch": 6.15819209039548,
-      "grad_norm": 0.07421875,
+      "grad_norm": 0.048828125,
       "learning_rate": 6.505358204009017e-05,
-      "loss": 0.1048,
+      "loss": 0.056,
       "step": 1090
     },
     {
       "epoch": 6.163841807909605,
-      "grad_norm": 0.06640625,
+      "grad_norm": 0.044189453125,
       "learning_rate": 6.48863926673827e-05,
-      "loss": 0.0757,
+      "loss": 0.0354,
       "step": 1091
     },
     {
       "epoch": 6.169491525423728,
-      "grad_norm": 0.03369140625,
+      "grad_norm": 0.052490234375,
       "learning_rate": 6.471931517397587e-05,
-      "loss": 0.0193,
+      "loss": 0.0507,
       "step": 1092
     },
     {
       "epoch": 6.175141242937853,
-      "grad_norm": 0.0673828125,
+      "grad_norm": 0.0294189453125,
       "learning_rate": 6.455235009221339e-05,
-      "loss": 0.0702,
+      "loss": 0.016,
       "step": 1093
     },
     {
       "epoch": 6.1807909604519775,
-      "grad_norm": 0.09765625,
+      "grad_norm": 0.048095703125,
       "learning_rate": 6.438549795408106e-05,
-      "loss": 0.1602,
+      "loss": 0.0348,
       "step": 1094
     },
     {
       "epoch": 6.186440677966102,
-      "grad_norm": 0.052490234375,
+      "grad_norm": 0.0458984375,
       "learning_rate": 6.421875929120469e-05,
-      "loss": 0.0501,
+      "loss": 0.0347,
       "step": 1095
     },
     {
       "epoch": 6.192090395480226,
-      "grad_norm": 0.0537109375,
+      "grad_norm": 0.08251953125,
       "learning_rate": 6.405213463484848e-05,
-      "loss": 0.0658,
+      "loss": 0.086,
       "step": 1096
     },
     {
       "epoch": 6.19774011299435,
-      "grad_norm": 0.06787109375,
+      "grad_norm": 0.08642578125,
       "learning_rate": 6.388562451591355e-05,
-      "loss": 0.0728,
+      "loss": 0.1056,
       "step": 1097
     },
     {
       "epoch": 6.203389830508475,
-      "grad_norm": 0.068359375,
+      "grad_norm": 0.0947265625,
       "learning_rate": 6.371922946493591e-05,
-      "loss": 0.0849,
+      "loss": 0.1101,
       "step": 1098
     },
     {
       "epoch": 6.209039548022599,
-      "grad_norm": 0.06396484375,
+      "grad_norm": 0.0830078125,
       "learning_rate": 6.355295001208504e-05,
-      "loss": 0.0587,
+      "loss": 0.1057,
       "step": 1099
     },
     {
       "epoch": 6.214689265536723,
-      "grad_norm": 0.078125,
+      "grad_norm": 0.076171875,
       "learning_rate": 6.338678668716209e-05,
-      "loss": 0.1016,
+      "loss": 0.0921,
       "step": 1100
     },
     {
       "epoch": 6.220338983050848,
-      "grad_norm": 0.061279296875,
+      "grad_norm": 0.05859375,
       "learning_rate": 6.322074001959815e-05,
-      "loss": 0.0812,
+      "loss": 0.0679,
       "step": 1101
     },
     {
       "epoch": 6.2259887005649714,
-      "grad_norm": 0.06494140625,
+      "grad_norm": 0.0712890625,
       "learning_rate": 6.305481053845261e-05,
-      "loss": 0.0898,
+      "loss": 0.0907,
       "step": 1102
     },
     {
       "epoch": 6.231638418079096,
-      "grad_norm": 0.04736328125,
+      "grad_norm": 0.07275390625,
       "learning_rate": 6.288899877241163e-05,
-      "loss": 0.0494,
+      "loss": 0.0902,
       "step": 1103
     },
     {
       "epoch": 6.237288135593221,
-      "grad_norm": 0.038818359375,
+      "grad_norm": 0.0810546875,
       "learning_rate": 6.272330524978613e-05,
-      "loss": 0.0212,
+      "loss": 0.0861,
       "step": 1104
     },
     {
       "epoch": 6.242937853107344,
-      "grad_norm": 0.07373046875,
+      "grad_norm": 0.0791015625,
       "learning_rate": 6.25577304985103e-05,
-      "loss": 0.1021,
+      "loss": 0.0813,
       "step": 1105
     },
     {
       "epoch": 6.248587570621469,
-      "grad_norm": 0.06884765625,
+      "grad_norm": 0.0341796875,
       "learning_rate": 6.239227504614003e-05,
-      "loss": 0.0829,
+      "loss": 0.018,
       "step": 1106
     },
     {
       "epoch": 6.254237288135593,
-      "grad_norm": 0.058837890625,
+      "grad_norm": 0.07763671875,
       "learning_rate": 6.222693941985096e-05,
-      "loss": 0.0557,
+      "loss": 0.0885,
       "step": 1107
     },
     {
       "epoch": 6.259887005649717,
-      "grad_norm": 0.0673828125,
+      "grad_norm": 0.026611328125,
       "learning_rate": 6.206172414643699e-05,
-      "loss": 0.0729,
+      "loss": 0.0152,
       "step": 1108
     },
     {
       "epoch": 6.265536723163842,
-      "grad_norm": 0.06689453125,
+      "grad_norm": 0.055908203125,
       "learning_rate": 6.189662975230861e-05,
-      "loss": 0.0669,
+      "loss": 0.0659,
       "step": 1109
     },
     {
       "epoch": 6.271186440677966,
-      "grad_norm": 0.0751953125,
+      "grad_norm": 0.0634765625,
       "learning_rate": 6.173165676349103e-05,
-      "loss": 0.1013,
+      "loss": 0.0798,
       "step": 1110
     },
     {
       "epoch": 6.27683615819209,
-      "grad_norm": 0.06298828125,
+      "grad_norm": 0.08203125,
       "learning_rate": 6.156680570562277e-05,
-      "loss": 0.0668,
+      "loss": 0.1024,
       "step": 1111
     },
     {
       "epoch": 6.2824858757062145,
-      "grad_norm": 0.06982421875,
+      "grad_norm": 0.06396484375,
       "learning_rate": 6.140207710395378e-05,
-      "loss": 0.0905,
+      "loss": 0.0676,
       "step": 1112
     },
     {
       "epoch": 6.288135593220339,
-      "grad_norm": 0.053466796875,
+      "grad_norm": 0.043701171875,
       "learning_rate": 6.12374714833438e-05,
-      "loss": 0.0387,
+      "loss": 0.0361,
       "step": 1113
     },
     {
       "epoch": 6.293785310734464,
-      "grad_norm": 0.0927734375,
+      "grad_norm": 0.0478515625,
       "learning_rate": 6.107298936826086e-05,
-      "loss": 0.1038,
+      "loss": 0.0475,
       "step": 1114
     },
     {
       "epoch": 6.299435028248587,
-      "grad_norm": 0.064453125,
+      "grad_norm": 0.0732421875,
       "learning_rate": 6.090863128277938e-05,
-      "loss": 0.0631,
+      "loss": 0.0852,
       "step": 1115
     },
     {
       "epoch": 6.305084745762712,
-      "grad_norm": 0.07470703125,
+      "grad_norm": 0.09033203125,
       "learning_rate": 6.074439775057856e-05,
-      "loss": 0.1037,
+      "loss": 0.0929,
       "step": 1116
     },
     {
       "epoch": 6.3107344632768365,
-      "grad_norm": 0.083984375,
+      "grad_norm": 0.0703125,
       "learning_rate": 6.0580289294940874e-05,
-      "loss": 0.0984,
+      "loss": 0.057,
       "step": 1117
     },
     {
       "epoch": 6.31638418079096,
-      "grad_norm": 0.0634765625,
+      "grad_norm": 0.03564453125,
       "learning_rate": 6.041630643875018e-05,
-      "loss": 0.0706,
+      "loss": 0.0198,
       "step": 1118
     },
     {
       "epoch": 6.322033898305085,
-      "grad_norm": 0.05712890625,
+      "grad_norm": 0.05615234375,
       "learning_rate": 6.025244970449011e-05,
-      "loss": 0.0468,
+      "loss": 0.0531,
       "step": 1119
     },
     {
       "epoch": 6.327683615819209,
-      "grad_norm": 0.07470703125,
+      "grad_norm": 0.052734375,
       "learning_rate": 6.008871961424258e-05,
-      "loss": 0.0846,
+      "loss": 0.0359,
       "step": 1120
     },
     {
       "epoch": 6.333333333333333,
-      "grad_norm": 0.08154296875,
+      "grad_norm": 0.0732421875,
       "learning_rate": 5.992511668968592e-05,
-      "loss": 0.0697,
+      "loss": 0.0911,
       "step": 1121
     },
     {
       "epoch": 6.338983050847458,
-      "grad_norm": 0.060791015625,
+      "grad_norm": 0.052734375,
       "learning_rate": 5.976164145209322e-05,
-      "loss": 0.0569,
+      "loss": 0.0506,
       "step": 1122
     },
     {
       "epoch": 6.344632768361582,
-      "grad_norm": 0.08935546875,
+      "grad_norm": 0.0927734375,
       "learning_rate": 5.959829442233085e-05,
-      "loss": 0.116,
+      "loss": 0.1108,
       "step": 1123
     },
     {
       "epoch": 6.350282485875706,
-      "grad_norm": 0.050537109375,
+      "grad_norm": 0.0927734375,
       "learning_rate": 5.9435076120856616e-05,
-      "loss": 0.037,
+      "loss": 0.1338,
       "step": 1124
     },
     {
       "epoch": 6.3559322033898304,
-      "grad_norm": 0.0908203125,
+      "grad_norm": 0.0439453125,
       "learning_rate": 5.9271987067718125e-05,
-      "loss": 0.0876,
+      "loss": 0.0234,
       "step": 1125
     },
     {
       "epoch": 6.3559322033898304,
-      "eval_loss": 0.0009804321452975273,
-      "eval_runtime": 5.8499,
-      "eval_samples_per_second": 25.47,
-      "eval_steps_per_second": 3.248,
+      "eval_loss": 0.0009568367386236787,
+      "eval_runtime": 5.8411,
+      "eval_samples_per_second": 25.68,
+      "eval_steps_per_second": 3.253,
       "step": 1125
     },
     {
       "epoch": 6.361581920903955,
-      "grad_norm": 0.052978515625,
+      "grad_norm": 0.083984375,
       "learning_rate": 5.910902778255134e-05,
-      "loss": 0.0358,
+      "loss": 0.107,
       "step": 1126
     },
     {
       "epoch": 6.367231638418079,
       "grad_norm": 0.06787109375,
       "learning_rate": 5.894619878457854e-05,
-      "loss": 0.0696,
+      "loss": 0.0653,
       "step": 1127
     },
     {
       "epoch": 6.372881355932203,
-      "grad_norm": 0.08837890625,
+      "grad_norm": 0.0576171875,
       "learning_rate": 5.878350059260701e-05,
-      "loss": 0.1188,
+      "loss": 0.0483,
       "step": 1128
     },
     {
       "epoch": 6.378531073446328,
-      "grad_norm": 0.06298828125,
+      "grad_norm": 0.0458984375,
       "learning_rate": 5.862093372502731e-05,
-      "loss": 0.0593,
+      "loss": 0.0343,
       "step": 1129
     },
     {
       "epoch": 6.3841807909604515,
-      "grad_norm": 0.08154296875,
+      "grad_norm": 0.08056640625,
       "learning_rate": 5.845849869981137e-05,
-      "loss": 0.0814,
+      "loss": 0.106,
       "step": 1130
     },
     {
       "epoch": 6.389830508474576,
-      "grad_norm": 0.06396484375,
+      "grad_norm": 0.06298828125,
       "learning_rate": 5.829619603451124e-05,
-      "loss": 0.0537,
+      "loss": 0.0667,
       "step": 1131
     },
     {
       "epoch": 6.395480225988701,
-      "grad_norm": 0.08642578125,
+      "grad_norm": 0.06494140625,
       "learning_rate": 5.8134026246257225e-05,
-      "loss": 0.1032,
+      "loss": 0.0649,
       "step": 1132
     },
     {
       "epoch": 6.401129943502825,
-      "grad_norm": 0.0673828125,
+      "grad_norm": 0.0927734375,
       "learning_rate": 5.79719898517561e-05,
-      "loss": 0.0703,
+      "loss": 0.1072,
       "step": 1133
     },
     {
       "epoch": 6.406779661016949,
-      "grad_norm": 0.0771484375,
+      "grad_norm": 0.04296875,
       "learning_rate": 5.7810087367289744e-05,
-      "loss": 0.0771,
+      "loss": 0.0345,
       "step": 1134
     },
     {
       "epoch": 6.4124293785310735,
-      "grad_norm": 0.044921875,
+      "grad_norm": 0.0537109375,
       "learning_rate": 5.7648319308713464e-05,
-      "loss": 0.0331,
+      "loss": 0.0335,
       "step": 1135
     },
     {
       "epoch": 6.418079096045198,
-      "grad_norm": 0.043212890625,
+      "grad_norm": 0.061767578125,
       "learning_rate": 5.748668619145401e-05,
-      "loss": 0.0332,
+      "loss": 0.0439,
       "step": 1136
     },
     {
       "epoch": 6.423728813559322,
-      "grad_norm": 0.048828125,
+      "grad_norm": 0.03515625,
       "learning_rate": 5.732518853050835e-05,
-      "loss": 0.045,
+      "loss": 0.0181,
       "step": 1137
     },
     {
       "epoch": 6.429378531073446,
-      "grad_norm": 0.06201171875,
+      "grad_norm": 0.0927734375,
       "learning_rate": 5.71638268404419e-05,
-      "loss": 0.0843,
+      "loss": 0.1217,
       "step": 1138
     },
     {
       "epoch": 6.435028248587571,
-      "grad_norm": 0.0810546875,
+      "grad_norm": 0.07958984375,
       "learning_rate": 5.700260163538668e-05,
-      "loss": 0.0803,
+      "loss": 0.0846,
       "step": 1139
     },
     {
       "epoch": 6.440677966101695,
-      "grad_norm": 0.0703125,
+      "grad_norm": 0.10400390625,
       "learning_rate": 5.684151342903992e-05,
-      "loss": 0.0954,
+      "loss": 0.1456,
       "step": 1140
     },
     {
       "epoch": 6.446327683615819,
-      "grad_norm": 0.095703125,
+      "grad_norm": 0.09326171875,
       "learning_rate": 5.668056273466248e-05,
-      "loss": 0.1076,
+      "loss": 0.1248,
       "step": 1141
     },
     {
       "epoch": 6.451977401129944,
-      "grad_norm": 0.0654296875,
+      "grad_norm": 0.078125,
       "learning_rate": 5.651975006507681e-05,
-      "loss": 0.0816,
+      "loss": 0.0706,
       "step": 1142
     },
     {
       "epoch": 6.4576271186440675,
-      "grad_norm": 0.0751953125,
+      "grad_norm": 0.07080078125,
       "learning_rate": 5.6359075932665775e-05,
-      "loss": 0.089,
+      "loss": 0.0809,
       "step": 1143
     },
     {
       "epoch": 6.463276836158192,
-      "grad_norm": 0.060302734375,
+      "grad_norm": 0.052734375,
       "learning_rate": 5.619854084937085e-05,
-      "loss": 0.0664,
+      "loss": 0.0378,
       "step": 1144
     },
     {
       "epoch": 6.468926553672317,
-      "grad_norm": 0.060546875,
+      "grad_norm": 0.0517578125,
       "learning_rate": 5.603814532669032e-05,
-      "loss": 0.0517,
+      "loss": 0.0537,
       "step": 1145
     },
     {
       "epoch": 6.47457627118644,
-      "grad_norm": 0.07861328125,
+      "grad_norm": 0.099609375,
       "learning_rate": 5.5877889875677845e-05,
-      "loss": 0.1066,
+      "loss": 0.1067,
       "step": 1146
     },
     {
       "epoch": 6.480225988700565,
-      "grad_norm": 0.068359375,
+      "grad_norm": 0.0830078125,
       "learning_rate": 5.5717775006940976e-05,
-      "loss": 0.089,
+      "loss": 0.0938,
       "step": 1147
     },
     {
       "epoch": 6.4858757062146895,
-      "grad_norm": 0.059326171875,
+      "grad_norm": 0.06005859375,
       "learning_rate": 5.555780123063907e-05,
-      "loss": 0.0696,
+      "loss": 0.0644,
       "step": 1148
     },
     {
       "epoch": 6.491525423728813,
-      "grad_norm": 0.057373046875,
+      "grad_norm": 0.064453125,
       "learning_rate": 5.539796905648208e-05,
-      "loss": 0.0511,
+      "loss": 0.0543,
       "step": 1149
     },
     {
       "epoch": 6.497175141242938,
-      "grad_norm": 0.0966796875,
+      "grad_norm": 0.033447265625,
       "learning_rate": 5.5238278993728756e-05,
-      "loss": 0.1457,
+      "loss": 0.018,
       "step": 1150
     },
     {
       "epoch": 6.502824858757062,
-      "grad_norm": 0.09765625,
+      "grad_norm": 0.06982421875,
       "learning_rate": 5.507873155118506e-05,
-      "loss": 0.1197,
+      "loss": 0.054,
       "step": 1151
     },
     {
       "epoch": 6.508474576271187,
-      "grad_norm": 0.07177734375,
+      "grad_norm": 0.06298828125,
       "learning_rate": 5.4919327237202546e-05,
-      "loss": 0.1056,
+      "loss": 0.0637,
       "step": 1152
     },
     {
       "epoch": 6.5141242937853105,
-      "grad_norm": 0.05126953125,
+      "grad_norm": 0.053466796875,
       "learning_rate": 5.476006655967668e-05,
-      "loss": 0.0523,
+      "loss": 0.0457,
       "step": 1153
     },
     {
       "epoch": 6.519774011299435,
-      "grad_norm": 0.07568359375,
+      "grad_norm": 0.07763671875,
       "learning_rate": 5.4600950026045326e-05,
-      "loss": 0.1045,
+      "loss": 0.1021,
       "step": 1154
     },
     {
       "epoch": 6.52542372881356,
-      "grad_norm": 0.06640625,
+      "grad_norm": 0.0751953125,
       "learning_rate": 5.4441978143287066e-05,
-      "loss": 0.0693,
+      "loss": 0.0574,
       "step": 1155
     },
     {
       "epoch": 6.531073446327683,
-      "grad_norm": 0.053466796875,
+      "grad_norm": 0.08544921875,
       "learning_rate": 5.428315141791955e-05,
-      "loss": 0.0523,
+      "loss": 0.0976,
       "step": 1156
     },
     {
       "epoch": 6.536723163841808,
-      "grad_norm": 0.05224609375,
+      "grad_norm": 0.046630859375,
       "learning_rate": 5.4124470355998005e-05,
-      "loss": 0.0501,
+      "loss": 0.0315,
       "step": 1157
     },
     {
       "epoch": 6.5423728813559325,
-      "grad_norm": 0.036865234375,
+      "grad_norm": 0.0625,
       "learning_rate": 5.396593546311346e-05,
-      "loss": 0.0312,
+      "loss": 0.056,
       "step": 1158
     },
     {
       "epoch": 6.548022598870056,
-      "grad_norm": 0.051513671875,
+      "grad_norm": 0.087890625,
       "learning_rate": 5.380754724439128e-05,
-      "loss": 0.0341,
+      "loss": 0.1129,
       "step": 1159
     },
     {
       "epoch": 6.553672316384181,
-      "grad_norm": 0.09033203125,
+      "grad_norm": 0.0966796875,
       "learning_rate": 5.364930620448946e-05,
-      "loss": 0.1368,
+      "loss": 0.1235,
       "step": 1160
     },
     {
       "epoch": 6.559322033898305,
-      "grad_norm": 0.05517578125,
+      "grad_norm": 0.05908203125,
       "learning_rate": 5.349121284759707e-05,
-      "loss": 0.0391,
+      "loss": 0.0639,
       "step": 1161
     },
     {
       "epoch": 6.564971751412429,
-      "grad_norm": 0.0751953125,
+      "grad_norm": 0.0908203125,
       "learning_rate": 5.333326767743263e-05,
       "loss": 0.0867,
       "step": 1162
     },
     {
       "epoch": 6.570621468926554,
-      "grad_norm": 0.045654296875,
+      "grad_norm": 0.06787109375,
       "learning_rate": 5.317547119724251e-05,
-      "loss": 0.0359,
+      "loss": 0.0703,
       "step": 1163
     },
     {
       "epoch": 6.576271186440678,
-      "grad_norm": 0.095703125,
+      "grad_norm": 0.07373046875,
       "learning_rate": 5.301782390979929e-05,
-      "loss": 0.1265,
+      "loss": 0.0784,
       "step": 1164
     },
     {
       "epoch": 6.581920903954802,
-      "grad_norm": 0.06103515625,
+      "grad_norm": 0.052490234375,
       "learning_rate": 5.286032631740023e-05,
-      "loss": 0.0693,
+      "loss": 0.0503,
       "step": 1165
     },
     {
       "epoch": 6.5875706214689265,
-      "grad_norm": 0.056640625,
+      "grad_norm": 0.07861328125,
       "learning_rate": 5.270297892186563e-05,
-      "loss": 0.0514,
+      "loss": 0.0689,
       "step": 1166
     },
     {
       "epoch": 6.593220338983051,
-      "grad_norm": 0.0556640625,
+      "grad_norm": 0.0625,
       "learning_rate": 5.254578222453723e-05,
-      "loss": 0.0591,
+      "loss": 0.0693,
       "step": 1167
     },
     {
       "epoch": 6.598870056497175,
-      "grad_norm": 0.060546875,
+      "grad_norm": 0.07421875,
       "learning_rate": 5.238873672627651e-05,
-      "loss": 0.0579,
+      "loss": 0.0876,
       "step": 1168
     },
     {
       "epoch": 6.604519774011299,
-      "grad_norm": 0.0625,
+      "grad_norm": 0.0712890625,
       "learning_rate": 5.223184292746339e-05,
-      "loss": 0.0629,
+      "loss": 0.073,
       "step": 1169
     },
     {
       "epoch": 6.610169491525424,
-      "grad_norm": 0.068359375,
+      "grad_norm": 0.09326171875,
       "learning_rate": 5.207510132799436e-05,
-      "loss": 0.0716,
+      "loss": 0.1259,
       "step": 1170
     },
     {
       "epoch": 6.610169491525424,
-      "eval_loss": 0.0009521428146399558,
-      "eval_runtime": 5.7429,
-      "eval_samples_per_second": 25.945,
-      "eval_steps_per_second": 3.308,
+      "eval_loss": 0.0009285201667807996,
+      "eval_runtime": 5.7572,
+      "eval_samples_per_second": 26.054,
+      "eval_steps_per_second": 3.3,
       "step": 1170
     },
     {
       "epoch": 6.6158192090395485,
-      "grad_norm": 0.0673828125,
+      "grad_norm": 0.09375,
       "learning_rate": 5.191851242728081e-05,
-      "loss": 0.0727,
+      "loss": 0.1239,
       "step": 1171
     },
     {
       "epoch": 6.621468926553672,
-      "grad_norm": 0.06298828125,
+      "grad_norm": 0.078125,
       "learning_rate": 5.1762076724247876e-05,
-      "loss": 0.0767,
+      "loss": 0.0874,
       "step": 1172
     },
     {
       "epoch": 6.627118644067797,
-      "grad_norm": 0.046630859375,
+      "grad_norm": 0.064453125,
       "learning_rate": 5.160579471733245e-05,
-      "loss": 0.0192,
+      "loss": 0.0688,
       "step": 1173
     },
     {
       "epoch": 6.632768361581921,
-      "grad_norm": 0.08203125,
+      "grad_norm": 0.083984375,
       "learning_rate": 5.1449666904481585e-05,
-      "loss": 0.0931,
+      "loss": 0.079,
       "step": 1174
     },
     {
       "epoch": 6.638418079096045,
-      "grad_norm": 0.06591796875,
+      "grad_norm": 0.08349609375,
       "learning_rate": 5.1293693783151275e-05,
-      "loss": 0.0821,
+      "loss": 0.0927,
       "step": 1175
     },
     {
       "epoch": 6.6440677966101696,
-      "grad_norm": 0.0458984375,
+      "grad_norm": 0.07421875,
       "learning_rate": 5.113787585030454e-05,
-      "loss": 0.045,
+      "loss": 0.067,
       "step": 1176
     },
     {
       "epoch": 6.649717514124294,
-      "grad_norm": 0.0654296875,
+      "grad_norm": 0.060791015625,
       "learning_rate": 5.098221360240978e-05,
-      "loss": 0.0725,
+      "loss": 0.0391,
       "step": 1177
     },
     {
       "epoch": 6.655367231638418,
-      "grad_norm": 0.06787109375,
+      "grad_norm": 0.08837890625,
       "learning_rate": 5.082670753543961e-05,
-      "loss": 0.0899,
+      "loss": 0.1295,
       "step": 1178
     },
     {
       "epoch": 6.661016949152542,
-      "grad_norm": 0.07275390625,
+      "grad_norm": 0.0654296875,
       "learning_rate": 5.067135814486892e-05,
-      "loss": 0.0865,
+      "loss": 0.0703,
       "step": 1179
     },
     {
       "epoch": 6.666666666666667,
-      "grad_norm": 0.07373046875,
+      "grad_norm": 0.0771484375,
       "learning_rate": 5.051616592567323e-05,
-      "loss": 0.1041,
+      "loss": 0.0838,
       "step": 1180
     },
     {
       "epoch": 6.672316384180791,
-      "grad_norm": 0.043701171875,
+      "grad_norm": 0.076171875,
       "learning_rate": 5.0361131372327595e-05,
-      "loss": 0.0363,
+      "loss": 0.0729,
       "step": 1181
     },
     {
       "epoch": 6.677966101694915,
-      "grad_norm": 0.0869140625,
+      "grad_norm": 0.059814453125,
       "learning_rate": 5.0206254978804435e-05,
-      "loss": 0.1191,
+      "loss": 0.0532,
       "step": 1182
     },
     {
       "epoch": 6.68361581920904,
-      "grad_norm": 0.072265625,
+      "grad_norm": 0.060302734375,
       "learning_rate": 5.005153723857235e-05,
-      "loss": 0.0906,
+      "loss": 0.0646,
       "step": 1183
     },
     {
       "epoch": 6.6892655367231635,
-      "grad_norm": 0.072265625,
+      "grad_norm": 0.07861328125,
       "learning_rate": 4.989697864459452e-05,
-      "loss": 0.0703,
+      "loss": 0.0778,
       "step": 1184
     },
     {
       "epoch": 6.694915254237288,
-      "grad_norm": 0.053466796875,
+      "grad_norm": 0.0615234375,
       "learning_rate": 4.9742579689326874e-05,
-      "loss": 0.0523,
+      "loss": 0.0613,
       "step": 1185
     },
     {
       "epoch": 6.700564971751413,
-      "grad_norm": 0.05126953125,
+      "grad_norm": 0.07275390625,
       "learning_rate": 4.958834086471683e-05,
-      "loss": 0.0527,
+      "loss": 0.0705,
       "step": 1186
     },
     {
       "epoch": 6.706214689265536,
-      "grad_norm": 0.07275390625,
+      "grad_norm": 0.06103515625,
       "learning_rate": 4.943426266220156e-05,
-      "loss": 0.0719,
+      "loss": 0.0514,
       "step": 1187
     },
     {
       "epoch": 6.711864406779661,
-      "grad_norm": 0.0859375,
+      "grad_norm": 0.061767578125,
       "learning_rate": 4.928034557270645e-05,
-      "loss": 0.1142,
+      "loss": 0.0563,
       "step": 1188
     },
     {
       "epoch": 6.7175141242937855,
-      "grad_norm": 0.078125,
+      "grad_norm": 0.06298828125,
       "learning_rate": 4.912659008664357e-05,
-      "loss": 0.1024,
+      "loss": 0.0513,
       "step": 1189
     },
     {
       "epoch": 6.72316384180791,
-      "grad_norm": 0.06103515625,
+      "grad_norm": 0.049072265625,
       "learning_rate": 4.8972996693910054e-05,
-      "loss": 0.0577,
+      "loss": 0.0371,
       "step": 1190
     },
     {
       "epoch": 6.728813559322034,
-      "grad_norm": 0.047119140625,
+      "grad_norm": 0.07958984375,
       "learning_rate": 4.88195658838866e-05,
-      "loss": 0.0503,
+      "loss": 0.1025,
       "step": 1191
     },
     {
       "epoch": 6.734463276836158,
-      "grad_norm": 0.095703125,
+      "grad_norm": 0.080078125,
       "learning_rate": 4.8666298145435875e-05,
-      "loss": 0.1513,
+      "loss": 0.0727,
       "step": 1192
     },
     {
       "epoch": 6.740112994350282,
-      "grad_norm": 0.048828125,
+      "grad_norm": 0.07275390625,
       "learning_rate": 4.851319396690097e-05,
-      "loss": 0.0488,
+      "loss": 0.0689,
       "step": 1193
     },
     {
       "epoch": 6.745762711864407,
-      "grad_norm": 0.0732421875,
+      "grad_norm": 0.0888671875,
       "learning_rate": 4.836025383610382e-05,
-      "loss": 0.0699,
+      "loss": 0.1173,
       "step": 1194
     },
     {
       "epoch": 6.751412429378531,
-      "grad_norm": 0.08740234375,
+      "grad_norm": 0.0693359375,
       "learning_rate": 4.82074782403437e-05,
-      "loss": 0.1061,
+      "loss": 0.0824,
       "step": 1195
     },
     {
       "epoch": 6.757062146892656,
-      "grad_norm": 0.08349609375,
+      "grad_norm": 0.0380859375,
       "learning_rate": 4.805486766639562e-05,
-      "loss": 0.093,
+      "loss": 0.031,
       "step": 1196
     },
     {
       "epoch": 6.762711864406779,
-      "grad_norm": 0.054931640625,
+      "grad_norm": 0.031494140625,
       "learning_rate": 4.79024226005088e-05,
-      "loss": 0.0536,
+      "loss": 0.0166,
       "step": 1197
     },
     {
       "epoch": 6.768361581920904,
-      "grad_norm": 0.048828125,
+      "grad_norm": 0.07470703125,
       "learning_rate": 4.7750143528405126e-05,
-      "loss": 0.0341,
+      "loss": 0.0698,
       "step": 1198
     },
     {
       "epoch": 6.7740112994350286,
-      "grad_norm": 0.059814453125,
+      "grad_norm": 0.0634765625,
       "learning_rate": 4.759803093527758e-05,
-      "loss": 0.0712,
+      "loss": 0.0701,
       "step": 1199
     },
     {
       "epoch": 6.779661016949152,
-      "grad_norm": 0.0208740234375,
+      "grad_norm": 0.06494140625,
       "learning_rate": 4.7446085305788725e-05,
-      "loss": 0.0137,
+      "loss": 0.0762,
       "step": 1200
     },
     {
       "epoch": 6.785310734463277,
-      "grad_norm": 0.042236328125,
+      "grad_norm": 0.050537109375,
       "learning_rate": 4.7294307124069115e-05,
-      "loss": 0.0184,
+      "loss": 0.0361,
       "step": 1201
     },
     {
       "epoch": 6.790960451977401,
-      "grad_norm": 0.038818359375,
+      "grad_norm": 0.0751953125,
       "learning_rate": 4.714269687371581e-05,
-      "loss": 0.0328,
+      "loss": 0.0616,
       "step": 1202
     },
     {
       "epoch": 6.796610169491525,
-      "grad_norm": 0.05322265625,
+      "grad_norm": 0.059814453125,
       "learning_rate": 4.6991255037790784e-05,
-      "loss": 0.0499,
+      "loss": 0.057,
       "step": 1203
     },
     {
       "epoch": 6.80225988700565,
-      "grad_norm": 0.0849609375,
+      "grad_norm": 0.07958984375,
       "learning_rate": 4.683998209881943e-05,
-      "loss": 0.1086,
+      "loss": 0.0956,
       "step": 1204
     },
     {
       "epoch": 6.807909604519774,
-      "grad_norm": 0.08251953125,
+      "grad_norm": 0.0654296875,
       "learning_rate": 4.668887853878896e-05,
-      "loss": 0.1072,
+      "loss": 0.0771,
       "step": 1205
     },
     {
       "epoch": 6.813559322033898,
-      "grad_norm": 0.076171875,
+      "grad_norm": 0.08056640625,
       "learning_rate": 4.653794483914696e-05,
-      "loss": 0.0901,
+      "loss": 0.0793,
       "step": 1206
     },
     {
       "epoch": 6.8192090395480225,
-      "grad_norm": 0.04638671875,
+      "grad_norm": 0.11083984375,
       "learning_rate": 4.6387181480799804e-05,
-      "loss": 0.0497,
+      "loss": 0.1599,
       "step": 1207
     },
     {
       "epoch": 6.824858757062147,
-      "grad_norm": 0.0693359375,
+      "grad_norm": 0.07958984375,
       "learning_rate": 4.623658894411099e-05,
-      "loss": 0.0705,
+      "loss": 0.0735,
       "step": 1208
     },
     {
       "epoch": 6.830508474576272,
-      "grad_norm": 0.054443359375,
+      "grad_norm": 0.034912109375,
       "learning_rate": 4.6086167708899975e-05,
-      "loss": 0.0488,
+      "loss": 0.017,
       "step": 1209
     },
     {
       "epoch": 6.836158192090395,
-      "grad_norm": 0.08642578125,
+      "grad_norm": 0.076171875,
       "learning_rate": 4.593591825444028e-05,
-      "loss": 0.1129,
+      "loss": 0.0708,
       "step": 1210
     },
     {
       "epoch": 6.84180790960452,
-      "grad_norm": 0.04736328125,
+      "grad_norm": 0.0771484375,
       "learning_rate": 4.578584105945799e-05,
-      "loss": 0.0334,
+      "loss": 0.1072,
       "step": 1211
     },
     {
       "epoch": 6.847457627118644,
-      "grad_norm": 0.06591796875,
+      "grad_norm": 0.07763671875,
       "learning_rate": 4.563593660213057e-05,
-      "loss": 0.0718,
+      "loss": 0.0897,
       "step": 1212
     },
     {
       "epoch": 6.853107344632768,
-      "grad_norm": 0.06298828125,
+      "grad_norm": 0.07763671875,
       "learning_rate": 4.5486205360085e-05,
-      "loss": 0.0533,
+      "loss": 0.0857,
       "step": 1213
     },
     {
       "epoch": 6.858757062146893,
-      "grad_norm": 0.0732421875,
+      "grad_norm": 0.09228515625,
       "learning_rate": 4.5336647810396215e-05,
-      "loss": 0.088,
+      "loss": 0.1312,
       "step": 1214
     },
     {
       "epoch": 6.864406779661017,
-      "grad_norm": 0.0712890625,
+      "grad_norm": 0.09423828125,
       "learning_rate": 4.518726442958599e-05,
-      "loss": 0.0728,
+      "loss": 0.1216,
       "step": 1215
     },
     {
       "epoch": 6.864406779661017,
-      "eval_loss": 0.000947348540648818,
-      "eval_runtime": 5.794,
-      "eval_samples_per_second": 25.716,
-      "eval_steps_per_second": 3.279,
+      "eval_loss": 0.0008980315760709345,
+      "eval_runtime": 5.7213,
+      "eval_samples_per_second": 26.218,
+      "eval_steps_per_second": 3.321,
       "step": 1215
     },
     {
       "epoch": 6.870056497175141,
-      "grad_norm": 0.033203125,
+      "grad_norm": 0.0732421875,
       "learning_rate": 4.5038055693621025e-05,
-      "loss": 0.0176,
+      "loss": 0.0732,
       "step": 1216
     },
     {
       "epoch": 6.875706214689266,
-      "grad_norm": 0.0634765625,
+      "grad_norm": 0.08837890625,
       "learning_rate": 4.488902207791147e-05,
-      "loss": 0.0687,
+      "loss": 0.1035,
       "step": 1217
     },
     {
       "epoch": 6.88135593220339,
-      "grad_norm": 0.0693359375,
+      "grad_norm": 0.056884765625,
       "learning_rate": 4.474016405730973e-05,
-      "loss": 0.0877,
+      "loss": 0.0504,
       "step": 1218
     },
     {
       "epoch": 6.887005649717514,
-      "grad_norm": 0.04638671875,
+      "grad_norm": 0.061767578125,
       "learning_rate": 4.4591482106108596e-05,
-      "loss": 0.0332,
+      "loss": 0.064,
       "step": 1219
     },
     {
       "epoch": 6.892655367231638,
-      "grad_norm": 0.01116943359375,
+      "grad_norm": 0.0390625,
       "learning_rate": 4.444297669803981e-05,
-      "loss": 0.0006,
+      "loss": 0.0321,
       "step": 1220
     },
     {
       "epoch": 6.898305084745763,
-      "grad_norm": 0.043701171875,
+      "grad_norm": 0.0634765625,
       "learning_rate": 4.4294648306272666e-05,
-      "loss": 0.033,
+      "loss": 0.0679,
       "step": 1221
     },
     {
       "epoch": 6.903954802259887,
-      "grad_norm": 0.06396484375,
+      "grad_norm": 0.059326171875,
       "learning_rate": 4.414649740341258e-05,
-      "loss": 0.0689,
+      "loss": 0.0553,
       "step": 1222
     },
     {
       "epoch": 6.909604519774011,
-      "grad_norm": 0.08251953125,
+      "grad_norm": 0.0732421875,
       "learning_rate": 4.3998524461499214e-05,
-      "loss": 0.0948,
+      "loss": 0.0689,
       "step": 1223
     },
     {
       "epoch": 6.915254237288136,
-      "grad_norm": 0.0625,
+      "grad_norm": 0.06787109375,
       "learning_rate": 4.385072995200532e-05,
-      "loss": 0.0581,
+      "loss": 0.0654,
       "step": 1224
     },
     {
       "epoch": 6.9209039548022595,
-      "grad_norm": 0.0517578125,
+      "grad_norm": 0.056396484375,
       "learning_rate": 4.370311434583525e-05,
-      "loss": 0.0467,
+      "loss": 0.0669,
       "step": 1225
     },
     {
       "epoch": 6.926553672316384,
-      "grad_norm": 0.08740234375,
+      "grad_norm": 0.055419921875,
       "learning_rate": 4.355567811332311e-05,
-      "loss": 0.1039,
+      "loss": 0.0541,
       "step": 1226
     },
     {
       "epoch": 6.932203389830509,
-      "grad_norm": 0.0634765625,
+      "grad_norm": 0.08447265625,
       "learning_rate": 4.340842172423156e-05,
-      "loss": 0.0633,
+      "loss": 0.0904,
       "step": 1227
     },
     {
       "epoch": 6.937853107344633,
-      "grad_norm": 0.07666015625,
+      "grad_norm": 0.08837890625,
       "learning_rate": 4.3261345647750417e-05,
-      "loss": 0.0767,
+      "loss": 0.0934,
       "step": 1228
     },
     {
       "epoch": 6.943502824858757,
-      "grad_norm": 0.060302734375,
+      "grad_norm": 0.052978515625,
       "learning_rate": 4.3114450352494704e-05,
-      "loss": 0.066,
+      "loss": 0.0499,
       "step": 1229
     },
     {
       "epoch": 6.9491525423728815,
-      "grad_norm": 0.07373046875,
+      "grad_norm": 0.07421875,
       "learning_rate": 4.296773630650358e-05,
-      "loss": 0.0807,
+      "loss": 0.0729,
       "step": 1230
     },
     {
       "epoch": 6.954802259887005,
-      "grad_norm": 0.055908203125,
+      "grad_norm": 0.0576171875,
       "learning_rate": 4.282120397723879e-05,
-      "loss": 0.0536,
+      "loss": 0.053,
       "step": 1231
     },
     {
       "epoch": 6.96045197740113,
-      "grad_norm": 0.07861328125,
+      "grad_norm": 0.061767578125,
       "learning_rate": 4.267485383158291e-05,
-      "loss": 0.0856,
+      "loss": 0.0584,
       "step": 1232
     },
     {
       "epoch": 6.966101694915254,
-      "grad_norm": 0.0830078125,
+      "grad_norm": 0.08203125,
       "learning_rate": 4.2528686335838094e-05,
-      "loss": 0.1034,
+      "loss": 0.1064,
       "step": 1233
     },
     {
       "epoch": 6.971751412429379,
-      "grad_norm": 0.058349609375,
+      "grad_norm": 0.06787109375,
       "learning_rate": 4.238270195572472e-05,
-      "loss": 0.0618,
+      "loss": 0.0728,
       "step": 1234
     },
     {
       "epoch": 6.977401129943503,
-      "grad_norm": 0.0673828125,
+      "grad_norm": 0.06396484375,
       "learning_rate": 4.223690115637944e-05,
-      "loss": 0.0855,
+      "loss": 0.0711,
       "step": 1235
     },
     {
       "epoch": 6.983050847457627,
-      "grad_norm": 0.060302734375,
+      "grad_norm": 0.0654296875,
       "learning_rate": 4.2091284402354146e-05,
-      "loss": 0.067,
+      "loss": 0.0725,
       "step": 1236
     },
     {
       "epoch": 6.988700564971752,
-      "grad_norm": 0.04296875,
+      "grad_norm": 0.053955078125,
       "learning_rate": 4.1945852157614295e-05,
-      "loss": 0.0334,
+      "loss": 0.0463,
       "step": 1237
     },
     {
       "epoch": 6.994350282485875,
-      "grad_norm": 0.07861328125,
+      "grad_norm": 0.08203125,
       "learning_rate": 4.180060488553743e-05,
-      "loss": 0.0892,
+      "loss": 0.1111,
       "step": 1238
     },
     {
       "epoch": 7.0,
-      "grad_norm": 0.0634765625,
+      "grad_norm": 0.046875,
       "learning_rate": 4.165554304891178e-05,
-      "loss": 0.0566,
+      "loss": 0.0343,
       "step": 1239
     },
     {
       "epoch": 7.005649717514125,
-      "grad_norm": 0.064453125,
+      "grad_norm": 0.06884765625,
       "learning_rate": 4.15106671099347e-05,
-      "loss": 0.0677,
+      "loss": 0.0669,
       "step": 1240
     },
     {
       "epoch": 7.011299435028248,
-      "grad_norm": 0.06884765625,
+      "grad_norm": 0.053955078125,
       "learning_rate": 4.1365977530211245e-05,
-      "loss": 0.081,
+      "loss": 0.0523,
       "step": 1241
     },
     {
       "epoch": 7.016949152542373,
-      "grad_norm": 0.047119140625,
+      "grad_norm": 0.07373046875,
       "learning_rate": 4.12214747707527e-05,
-      "loss": 0.0316,
+      "loss": 0.0863,
       "step": 1242
     },
     {
       "epoch": 7.022598870056497,
-      "grad_norm": 0.0595703125,
+      "grad_norm": 0.07470703125,
       "learning_rate": 4.107715929197508e-05,
-      "loss": 0.0524,
+      "loss": 0.0812,
       "step": 1243
     },
     {
       "epoch": 7.028248587570621,
-      "grad_norm": 0.052734375,
+      "grad_norm": 0.07861328125,
       "learning_rate": 4.093303155369771e-05,
-      "loss": 0.0516,
+      "loss": 0.0767,
       "step": 1244
     },
     {
       "epoch": 7.033898305084746,
-      "grad_norm": 0.07958984375,
+      "grad_norm": 0.07275390625,
       "learning_rate": 4.0789092015141714e-05,
-      "loss": 0.0986,
+      "loss": 0.0641,
       "step": 1245
     },
     {
       "epoch": 7.03954802259887,
-      "grad_norm": 0.0654296875,
+      "grad_norm": 0.045654296875,
       "learning_rate": 4.06453411349286e-05,
-      "loss": 0.0742,
+      "loss": 0.0461,
       "step": 1246
     },
     {
       "epoch": 7.045197740112994,
-      "grad_norm": 0.07275390625,
+      "grad_norm": 0.08837890625,
       "learning_rate": 4.050177937107874e-05,
-      "loss": 0.0756,
+      "loss": 0.1207,
       "step": 1247
     },
     {
       "epoch": 7.0508474576271185,
-      "grad_norm": 0.06103515625,
+      "grad_norm": 0.068359375,
       "learning_rate": 4.035840718100995e-05,
-      "loss": 0.0707,
+      "loss": 0.0687,
       "step": 1248
     },
     {
       "epoch": 7.056497175141243,
-      "grad_norm": 0.06494140625,
+      "grad_norm": 0.091796875,
       "learning_rate": 4.021522502153603e-05,
-      "loss": 0.0854,
+      "loss": 0.1126,
       "step": 1249
     },
     {
       "epoch": 7.062146892655368,
-      "grad_norm": 0.078125,
+      "grad_norm": 0.06689453125,
       "learning_rate": 4.007223334886531e-05,
-      "loss": 0.082,
+      "loss": 0.0556,
       "step": 1250
     },
     {
       "epoch": 7.067796610169491,
-      "grad_norm": 0.04931640625,
+      "grad_norm": 0.04248046875,
       "learning_rate": 3.992943261859917e-05,
-      "loss": 0.0366,
+      "loss": 0.0327,
       "step": 1251
     },
     {
       "epoch": 7.073446327683616,
-      "grad_norm": 0.042724609375,
+      "grad_norm": 0.08251953125,
       "learning_rate": 3.978682328573063e-05,
-      "loss": 0.0303,
+      "loss": 0.1048,
       "step": 1252
     },
     {
       "epoch": 7.0790960451977405,
-      "grad_norm": 0.04541015625,
+      "grad_norm": 0.053955078125,
       "learning_rate": 3.964440580464286e-05,
-      "loss": 0.043,
+      "loss": 0.0507,
       "step": 1253
     },
     {
       "epoch": 7.084745762711864,
-      "grad_norm": 0.0693359375,
+      "grad_norm": 0.0625,
       "learning_rate": 3.9502180629107756e-05,
-      "loss": 0.0887,
+      "loss": 0.0546,
       "step": 1254
     },
     {
       "epoch": 7.090395480225989,
-      "grad_norm": 0.08642578125,
+      "grad_norm": 0.060546875,
       "learning_rate": 3.9360148212284475e-05,
-      "loss": 0.103,
+      "loss": 0.0437,
       "step": 1255
     },
     {
       "epoch": 7.096045197740113,
-      "grad_norm": 0.07080078125,
+      "grad_norm": 0.09228515625,
       "learning_rate": 3.9218309006718043e-05,
-      "loss": 0.0779,
+      "loss": 0.1066,
       "step": 1256
     },
     {
       "epoch": 7.101694915254237,
-      "grad_norm": 0.07470703125,
+      "grad_norm": 0.0849609375,
       "learning_rate": 3.907666346433787e-05,
-      "loss": 0.0839,
+      "loss": 0.0722,
       "step": 1257
     },
     {
       "epoch": 7.107344632768362,
-      "grad_norm": 0.06005859375,
+      "grad_norm": 0.09912109375,
       "learning_rate": 3.893521203645618e-05,
-      "loss": 0.0583,
+      "loss": 0.1072,
       "step": 1258
     },
     {
       "epoch": 7.112994350282486,
-      "grad_norm": 0.048583984375,
+      "grad_norm": 0.0908203125,
       "learning_rate": 3.879395517376695e-05,
-      "loss": 0.0486,
+      "loss": 0.0899,
       "step": 1259
     },
     {
       "epoch": 7.11864406779661,
-      "grad_norm": 0.044921875,
+      "grad_norm": 0.080078125,
       "learning_rate": 3.865289332634407e-05,
-      "loss": 0.0338,
+      "loss": 0.0687,
       "step": 1260
     },
     {
       "epoch": 7.11864406779661,
-      "eval_loss": 0.0009293659240938723,
-      "eval_runtime": 5.7166,
-      "eval_samples_per_second": 26.064,
-      "eval_steps_per_second": 3.324,
+      "eval_loss": 0.0008977354736998677,
+      "eval_runtime": 5.8377,
+      "eval_samples_per_second": 25.695,
+      "eval_steps_per_second": 3.255,
       "step": 1260
     },
     {
       "epoch": 7.124293785310734,
-      "grad_norm": 0.06689453125,
+      "grad_norm": 0.0556640625,
       "learning_rate": 3.8512026943640034e-05,
-      "loss": 0.0557,
+      "loss": 0.0346,
       "step": 1261
     },
     {
       "epoch": 7.129943502824859,
-      "grad_norm": 0.072265625,
+      "grad_norm": 0.06787109375,
       "learning_rate": 3.8371356474484746e-05,
-      "loss": 0.0863,
+      "loss": 0.07,
       "step": 1262
     },
     {
       "epoch": 7.135593220338983,
-      "grad_norm": 0.07763671875,
+      "grad_norm": 0.06640625,
       "learning_rate": 3.823088236708362e-05,
-      "loss": 0.092,
+      "loss": 0.0535,
       "step": 1263
     },
     {
       "epoch": 7.141242937853107,
-      "grad_norm": 0.078125,
+      "grad_norm": 0.0693359375,
       "learning_rate": 3.8090605069016595e-05,
-      "loss": 0.1279,
+      "loss": 0.0672,
       "step": 1264
     },
     {
       "epoch": 7.146892655367232,
-      "grad_norm": 0.05810546875,
+      "grad_norm": 0.08544921875,
       "learning_rate": 3.7950525027236585e-05,
-      "loss": 0.0582,
+      "loss": 0.0877,
       "step": 1265
     },
     {
       "epoch": 7.1525423728813555,
-      "grad_norm": 0.09228515625,
+      "grad_norm": 0.07080078125,
       "learning_rate": 3.7810642688067796e-05,
-      "loss": 0.1074,
+      "loss": 0.0576,
       "step": 1266
     },
     {
       "epoch": 7.15819209039548,
-      "grad_norm": 0.045654296875,
+      "grad_norm": 0.047607421875,
       "learning_rate": 3.767095849720466e-05,
-      "loss": 0.0404,
+      "loss": 0.0365,
       "step": 1267
     },
     {
       "epoch": 7.163841807909605,
-      "grad_norm": 0.05078125,
+      "grad_norm": 0.059326171875,
       "learning_rate": 3.753147289971033e-05,
-      "loss": 0.0367,
+      "loss": 0.0477,
       "step": 1268
     },
     {
       "epoch": 7.169491525423728,
-      "grad_norm": 0.05517578125,
+      "grad_norm": 0.0595703125,
       "learning_rate": 3.739218634001499e-05,
-      "loss": 0.0544,
+      "loss": 0.0471,
       "step": 1269
     },
     {
       "epoch": 7.175141242937853,
-      "grad_norm": 0.0703125,
+      "grad_norm": 0.0859375,
       "learning_rate": 3.725309926191479e-05,
       "loss": 0.0748,
       "step": 1270
     },
     {
       "epoch": 7.1807909604519775,
-      "grad_norm": 0.08056640625,
+      "grad_norm": 0.08251953125,
       "learning_rate": 3.7114212108570356e-05,
-      "loss": 0.0979,
+      "loss": 0.0752,
       "step": 1271
     },
     {
       "epoch": 7.186440677966102,
-      "grad_norm": 0.0517578125,
+      "grad_norm": 0.07763671875,
       "learning_rate": 3.697552532250511e-05,
-      "loss": 0.0366,
+      "loss": 0.0765,
       "step": 1272
     },
     {
       "epoch": 7.192090395480226,
-      "grad_norm": 0.039794921875,
+      "grad_norm": 0.080078125,
       "learning_rate": 3.683703934560419e-05,
-      "loss": 0.0298,
+      "loss": 0.0671,
       "step": 1273
     },
     {
       "epoch": 7.19774011299435,
-      "grad_norm": 0.0771484375,
+      "grad_norm": 0.09716796875,
       "learning_rate": 3.669875461911297e-05,
-      "loss": 0.0879,
+      "loss": 0.1181,
       "step": 1274
     },
     {
       "epoch": 7.203389830508475,
-      "grad_norm": 0.06787109375,
+      "grad_norm": 0.0615234375,
       "learning_rate": 3.6560671583635467e-05,
-      "loss": 0.0563,
+      "loss": 0.0409,
       "step": 1275
     },
     {
       "epoch": 7.209039548022599,
-      "grad_norm": 0.04345703125,
+      "grad_norm": 0.07470703125,
       "learning_rate": 3.642279067913312e-05,
-      "loss": 0.033,
+      "loss": 0.0674,
       "step": 1276
     },
     {
       "epoch": 7.214689265536723,
-      "grad_norm": 0.076171875,
+      "grad_norm": 0.0390625,
       "learning_rate": 3.628511234492338e-05,
-      "loss": 0.0876,
+      "loss": 0.018,
       "step": 1277
     },
     {
       "epoch": 7.220338983050848,
-      "grad_norm": 0.062255859375,
+      "grad_norm": 0.037841796875,
       "learning_rate": 3.6147637019678195e-05,
-      "loss": 0.0539,
+      "loss": 0.0185,
       "step": 1278
     },
     {
       "epoch": 7.2259887005649714,
-      "grad_norm": 0.09228515625,
+      "grad_norm": 0.050537109375,
       "learning_rate": 3.601036514142273e-05,
-      "loss": 0.1165,
+      "loss": 0.0353,
       "step": 1279
     },
     {
       "epoch": 7.231638418079096,
-      "grad_norm": 0.060302734375,
+      "grad_norm": 0.0908203125,
       "learning_rate": 3.5873297147533915e-05,
-      "loss": 0.0673,
+      "loss": 0.0835,
       "step": 1280
     },
     {
       "epoch": 7.237288135593221,
-      "grad_norm": 0.10791015625,
+      "grad_norm": 0.07421875,
       "learning_rate": 3.573643347473906e-05,
-      "loss": 0.1615,
+      "loss": 0.0676,
       "step": 1281
     },
     {
       "epoch": 7.242937853107344,
-      "grad_norm": 0.05908203125,
+      "grad_norm": 0.054931640625,
       "learning_rate": 3.5599774559114475e-05,
-      "loss": 0.0648,
+      "loss": 0.035,
       "step": 1282
     },
     {
       "epoch": 7.248587570621469,
-      "grad_norm": 0.055419921875,
+      "grad_norm": 0.076171875,
       "learning_rate": 3.546332083608405e-05,
-      "loss": 0.0502,
+      "loss": 0.0743,
       "step": 1283
     },
     {
       "epoch": 7.254237288135593,
-      "grad_norm": 0.0693359375,
+      "grad_norm": 0.0859375,
       "learning_rate": 3.53270727404179e-05,
-      "loss": 0.0657,
+      "loss": 0.0888,
       "step": 1284
     },
     {
       "epoch": 7.259887005649717,
-      "grad_norm": 0.1005859375,
+      "grad_norm": 0.06201171875,
       "learning_rate": 3.519103070623096e-05,
-      "loss": 0.0972,
+      "loss": 0.0531,
       "step": 1285
     },
     {
       "epoch": 7.265536723163842,
-      "grad_norm": 0.05029296875,
+      "grad_norm": 0.072265625,
       "learning_rate": 3.5055195166981645e-05,
-      "loss": 0.0352,
+      "loss": 0.0754,
       "step": 1286
     },
     {
       "epoch": 7.271186440677966,
-      "grad_norm": 0.080078125,
+      "grad_norm": 0.0966796875,
       "learning_rate": 3.491956655547039e-05,
-      "loss": 0.0764,
+      "loss": 0.0946,
       "step": 1287
     },
     {
       "epoch": 7.27683615819209,
-      "grad_norm": 0.07421875,
+      "grad_norm": 0.0869140625,
       "learning_rate": 3.478414530383832e-05,
-      "loss": 0.0836,
+      "loss": 0.0764,
       "step": 1288
     },
     {
       "epoch": 7.2824858757062145,
-      "grad_norm": 0.06298828125,
+      "grad_norm": 0.06689453125,
       "learning_rate": 3.4648931843565904e-05,
-      "loss": 0.0384,
+      "loss": 0.0589,
       "step": 1289
     },
     {
       "epoch": 7.288135593220339,
-      "grad_norm": 0.10107421875,
+      "grad_norm": 0.0986328125,
       "learning_rate": 3.45139266054715e-05,
-      "loss": 0.1541,
+      "loss": 0.125,
       "step": 1290
     },
     {
       "epoch": 7.293785310734464,
-      "grad_norm": 0.09521484375,
+      "grad_norm": 0.051025390625,
       "learning_rate": 3.4379130019710056e-05,
-      "loss": 0.12,
+      "loss": 0.0176,
       "step": 1291
     },
     {
       "epoch": 7.299435028248587,
-      "grad_norm": 0.0634765625,
+      "grad_norm": 0.08447265625,
       "learning_rate": 3.4244542515771706e-05,
-      "loss": 0.0518,
+      "loss": 0.0854,
       "step": 1292
     },
     {
       "epoch": 7.305084745762712,
-      "grad_norm": 0.058349609375,
+      "grad_norm": 0.06689453125,
       "learning_rate": 3.4110164522480394e-05,
-      "loss": 0.0647,
+      "loss": 0.0477,
       "step": 1293
     },
     {
       "epoch": 7.3107344632768365,
-      "grad_norm": 0.033203125,
+      "grad_norm": 0.0703125,
       "learning_rate": 3.397599646799256e-05,
-      "loss": 0.0186,
+      "loss": 0.067,
       "step": 1294
     },
     {
       "epoch": 7.31638418079096,
-      "grad_norm": 0.076171875,
+      "grad_norm": 0.05859375,
       "learning_rate": 3.384203877979559e-05,
-      "loss": 0.0837,
+      "loss": 0.0325,
       "step": 1295
     },
     {
       "epoch": 7.322033898305085,
-      "grad_norm": 0.06201171875,
+      "grad_norm": 0.0732421875,
       "learning_rate": 3.370829188470683e-05,
-      "loss": 0.0524,
+      "loss": 0.0849,
       "step": 1296
     },
     {
       "epoch": 7.327683615819209,
-      "grad_norm": 0.072265625,
+      "grad_norm": 0.07958984375,
       "learning_rate": 3.357475620887186e-05,
-      "loss": 0.0839,
+      "loss": 0.0708,
       "step": 1297
     },
     {
       "epoch": 7.333333333333333,
-      "grad_norm": 0.1025390625,
+      "grad_norm": 0.0947265625,
       "learning_rate": 3.344143217776319e-05,
-      "loss": 0.1385,
+      "loss": 0.1016,
       "step": 1298
     },
     {
       "epoch": 7.338983050847458,
-      "grad_norm": 0.06298828125,
+      "grad_norm": 0.055419921875,
       "learning_rate": 3.330832021617919e-05,
-      "loss": 0.051,
+      "loss": 0.0362,
       "step": 1299
     },
     {
       "epoch": 7.344632768361582,
-      "grad_norm": 0.08251953125,
+      "grad_norm": 0.05712890625,
       "learning_rate": 3.3175420748242406e-05,
-      "loss": 0.0936,
+      "loss": 0.047,
       "step": 1300
     },
     {
       "epoch": 7.350282485875706,
-      "grad_norm": 0.09765625,
+      "grad_norm": 0.0859375,
       "learning_rate": 3.3042734197398265e-05,
-      "loss": 0.14,
+      "loss": 0.07,
       "step": 1301
     },
     {
       "epoch": 7.3559322033898304,
-      "grad_norm": 0.10693359375,
+      "grad_norm": 0.06787109375,
       "learning_rate": 3.291026098641397e-05,
-      "loss": 0.1377,
+      "loss": 0.0673,
       "step": 1302
     },
     {
       "epoch": 7.361581920903955,
-      "grad_norm": 0.062255859375,
+      "grad_norm": 0.0888671875,
       "learning_rate": 3.277800153737691e-05,
-      "loss": 0.0555,
+      "loss": 0.0852,
       "step": 1303
     },
     {
       "epoch": 7.367231638418079,
-      "grad_norm": 0.0859375,
+      "grad_norm": 0.05419921875,
       "learning_rate": 3.2645956271693257e-05,
-      "loss": 0.0763,
+      "loss": 0.037,
       "step": 1304
     },
     {
       "epoch": 7.372881355932203,
-      "grad_norm": 0.047607421875,
+      "grad_norm": 0.0986328125,
       "learning_rate": 3.2514125610086955e-05,
-      "loss": 0.032,
+      "loss": 0.1172,
       "step": 1305
     },
     {
       "epoch": 7.372881355932203,
-      "eval_loss": 0.0009304272825829685,
-      "eval_runtime": 5.7052,
-      "eval_samples_per_second": 26.117,
-      "eval_steps_per_second": 3.33,
+      "eval_loss": 0.0008774159359745681,
+      "eval_runtime": 5.7651,
+      "eval_samples_per_second": 26.019,
+      "eval_steps_per_second": 3.296,
       "step": 1305
     },
     {
       "epoch": 7.378531073446328,
-      "grad_norm": 0.05517578125,
+      "grad_norm": 0.06396484375,
       "learning_rate": 3.238250997259808e-05,
-      "loss": 0.0367,
+      "loss": 0.0507,
       "step": 1306
     },
     {
       "epoch": 7.3841807909604515,
-      "grad_norm": 0.09765625,
+      "grad_norm": 0.08349609375,
       "learning_rate": 3.22511097785815e-05,
-      "loss": 0.1235,
+      "loss": 0.0857,
       "step": 1307
     },
     {
       "epoch": 7.389830508474576,
-      "grad_norm": 0.0791015625,
+      "grad_norm": 0.09130859375,
       "learning_rate": 3.211992544670582e-05,
-      "loss": 0.0707,
+      "loss": 0.0953,
       "step": 1308
     },
     {
       "epoch": 7.395480225988701,
-      "grad_norm": 0.0908203125,
+      "grad_norm": 0.06787109375,
       "learning_rate": 3.198895739495177e-05,
-      "loss": 0.0947,
+      "loss": 0.0685,
       "step": 1309
     },
     {
       "epoch": 7.401129943502825,
-      "grad_norm": 0.078125,
+      "grad_norm": 0.1015625,
       "learning_rate": 3.185820604061088e-05,
-      "loss": 0.0653,
+      "loss": 0.1113,
       "step": 1310
     },
     {
       "epoch": 7.406779661016949,
-      "grad_norm": 0.08642578125,
+      "grad_norm": 0.068359375,
       "learning_rate": 3.172767180028442e-05,
-      "loss": 0.0876,
+      "loss": 0.0516,
       "step": 1311
     },
     {
       "epoch": 7.4124293785310735,
-      "grad_norm": 0.060302734375,
+      "grad_norm": 0.10302734375,
       "learning_rate": 3.1597355089881783e-05,
-      "loss": 0.0507,
+      "loss": 0.1048,
       "step": 1312
     },
     {
       "epoch": 7.418079096045198,
-      "grad_norm": 0.09521484375,
+      "grad_norm": 0.078125,
       "learning_rate": 3.146725632461923e-05,
-      "loss": 0.0885,
+      "loss": 0.0709,
       "step": 1313
     },
     {
       "epoch": 7.423728813559322,
-      "grad_norm": 0.068359375,
+      "grad_norm": 0.0625,
       "learning_rate": 3.133737591901864e-05,
-      "loss": 0.0377,
+      "loss": 0.048,
       "step": 1314
     },
     {
       "epoch": 7.429378531073446,
-      "grad_norm": 0.07421875,
+      "grad_norm": 0.09130859375,
       "learning_rate": 3.1207714286906256e-05,
-      "loss": 0.0736,
+      "loss": 0.0939,
       "step": 1315
     },
     {
       "epoch": 7.435028248587571,
-      "grad_norm": 0.083984375,
+      "grad_norm": 0.047607421875,
       "learning_rate": 3.1078271841411085e-05,
-      "loss": 0.1061,
+      "loss": 0.0336,
       "step": 1316
     },
     {
       "epoch": 7.440677966101695,
-      "grad_norm": 0.08447265625,
+      "grad_norm": 0.06982421875,
       "learning_rate": 3.094904899496387e-05,
-      "loss": 0.0989,
+      "loss": 0.0691,
       "step": 1317
     },
     {
       "epoch": 7.446327683615819,
-      "grad_norm": 0.05322265625,
+      "grad_norm": 0.0703125,
       "learning_rate": 3.0820046159295644e-05,
-      "loss": 0.0462,
+      "loss": 0.0508,
       "step": 1318
     },
     {
       "epoch": 7.451977401129944,
-      "grad_norm": 0.0546875,
+      "grad_norm": 0.11474609375,
       "learning_rate": 3.069126374543643e-05,
-      "loss": 0.0477,
+      "loss": 0.1409,
       "step": 1319
     },
     {
       "epoch": 7.4576271186440675,
-      "grad_norm": 0.0634765625,
+      "grad_norm": 0.08984375,
       "learning_rate": 3.056270216371395e-05,
-      "loss": 0.0511,
+      "loss": 0.0902,
       "step": 1320
     },
     {
       "epoch": 7.463276836158192,
-      "grad_norm": 0.0771484375,
+      "grad_norm": 0.032958984375,
       "learning_rate": 3.04343618237523e-05,
-      "loss": 0.0799,
+      "loss": 0.017,
       "step": 1321
     },
     {
       "epoch": 7.468926553672317,
-      "grad_norm": 0.040771484375,
+      "grad_norm": 0.08056640625,
       "learning_rate": 3.030624313447067e-05,
-      "loss": 0.0309,
+      "loss": 0.0702,
       "step": 1322
     },
     {
       "epoch": 7.47457627118644,
-      "grad_norm": 0.08740234375,
+      "grad_norm": 0.08203125,
       "learning_rate": 3.0178346504082e-05,
-      "loss": 0.1023,
+      "loss": 0.0832,
       "step": 1323
     },
     {
       "epoch": 7.480225988700565,
-      "grad_norm": 0.07470703125,
+      "grad_norm": 0.09130859375,
       "learning_rate": 3.0050672340091725e-05,
-      "loss": 0.0449,
+      "loss": 0.0883,
       "step": 1324
     },
     {
       "epoch": 7.4858757062146895,
-      "grad_norm": 0.07470703125,
+      "grad_norm": 0.083984375,
       "learning_rate": 2.9923221049296446e-05,
-      "loss": 0.0743,
+      "loss": 0.0703,
       "step": 1325
     },
     {
       "epoch": 7.491525423728813,
-      "grad_norm": 0.07275390625,
+      "grad_norm": 0.07861328125,
       "learning_rate": 2.9795993037782655e-05,
-      "loss": 0.0723,
+      "loss": 0.0868,
       "step": 1326
     },
     {
       "epoch": 7.497175141242938,
-      "grad_norm": 0.0869140625,
+      "grad_norm": 0.0732421875,
       "learning_rate": 2.9668988710925395e-05,
-      "loss": 0.0879,
+      "loss": 0.0785,
       "step": 1327
     },
     {
       "epoch": 7.502824858757062,
-      "grad_norm": 0.055419921875,
+      "grad_norm": 0.08984375,
       "learning_rate": 2.9542208473387044e-05,
-      "loss": 0.0457,
+      "loss": 0.0599,
       "step": 1328
     },
     {
       "epoch": 7.508474576271187,
-      "grad_norm": 0.08935546875,
+      "grad_norm": 0.1123046875,
       "learning_rate": 2.9415652729115972e-05,
-      "loss": 0.0734,
+      "loss": 0.1403,
       "step": 1329
     },
     {
       "epoch": 7.5141242937853105,
-      "grad_norm": 0.07080078125,
+      "grad_norm": 0.06298828125,
       "learning_rate": 2.9289321881345254e-05,
-      "loss": 0.0725,
+      "loss": 0.0518,
       "step": 1330
     },
     {
       "epoch": 7.519774011299435,
-      "grad_norm": 0.08447265625,
+      "grad_norm": 0.08203125,
       "learning_rate": 2.9163216332591403e-05,
-      "loss": 0.0852,
+      "loss": 0.0745,
       "step": 1331
     },
     {
       "epoch": 7.52542372881356,
-      "grad_norm": 0.072265625,
+      "grad_norm": 0.08984375,
       "learning_rate": 2.90373364846531e-05,
-      "loss": 0.0799,
+      "loss": 0.087,
       "step": 1332
     },
     {
       "epoch": 7.531073446327683,
-      "grad_norm": 0.07080078125,
+      "grad_norm": 0.07861328125,
       "learning_rate": 2.8911682738609867e-05,
-      "loss": 0.0712,
+      "loss": 0.05,
       "step": 1333
     },
     {
       "epoch": 7.536723163841808,
-      "grad_norm": 0.07373046875,
+      "grad_norm": 0.0634765625,
       "learning_rate": 2.8786255494820835e-05,
-      "loss": 0.0697,
+      "loss": 0.0509,
       "step": 1334
     },
     {
       "epoch": 7.5423728813559325,
-      "grad_norm": 0.095703125,
+      "grad_norm": 0.0693359375,
       "learning_rate": 2.8661055152923456e-05,
-      "loss": 0.1083,
+      "loss": 0.0651,
       "step": 1335
     },
     {
       "epoch": 7.548022598870056,
-      "grad_norm": 0.07470703125,
+      "grad_norm": 0.07958984375,
       "learning_rate": 2.853608211183221e-05,
-      "loss": 0.07,
+      "loss": 0.0855,
       "step": 1336
     },
     {
       "epoch": 7.553672316384181,
-      "grad_norm": 0.1015625,
+      "grad_norm": 0.0791015625,
       "learning_rate": 2.841133676973735e-05,
-      "loss": 0.1136,
+      "loss": 0.0697,
       "step": 1337
     },
     {
       "epoch": 7.559322033898305,
-      "grad_norm": 0.06396484375,
+      "grad_norm": 0.080078125,
       "learning_rate": 2.828681952410366e-05,
-      "loss": 0.0374,
+      "loss": 0.0551,
       "step": 1338
     },
     {
       "epoch": 7.564971751412429,
-      "grad_norm": 0.048583984375,
+      "grad_norm": 0.06494140625,
       "learning_rate": 2.8162530771669115e-05,
-      "loss": 0.0356,
+      "loss": 0.0514,
       "step": 1339
     },
     {
       "epoch": 7.570621468926554,
-      "grad_norm": 0.1103515625,
+      "grad_norm": 0.0771484375,
       "learning_rate": 2.8038470908443714e-05,
-      "loss": 0.1335,
+      "loss": 0.0707,
       "step": 1340
     },
     {
       "epoch": 7.576271186440678,
-      "grad_norm": 0.0634765625,
+      "grad_norm": 0.1005859375,
       "learning_rate": 2.7914640329708118e-05,
-      "loss": 0.0519,
+      "loss": 0.1219,
       "step": 1341
     },
     {
       "epoch": 7.581920903954802,
-      "grad_norm": 0.091796875,
+      "grad_norm": 0.08251953125,
       "learning_rate": 2.7791039430012476e-05,
-      "loss": 0.0868,
+      "loss": 0.0759,
       "step": 1342
     },
     {
       "epoch": 7.5875706214689265,
-      "grad_norm": 0.01080322265625,
+      "grad_norm": 0.08203125,
       "learning_rate": 2.766766860317512e-05,
-      "loss": 0.0006,
+      "loss": 0.0847,
       "step": 1343
     },
     {
       "epoch": 7.593220338983051,
-      "grad_norm": 0.034423828125,
+      "grad_norm": 0.055419921875,
       "learning_rate": 2.754452824228132e-05,
-      "loss": 0.0174,
+      "loss": 0.0372,
       "step": 1344
     },
     {
       "epoch": 7.598870056497175,
-      "grad_norm": 0.09814453125,
+      "grad_norm": 0.09912109375,
       "learning_rate": 2.742161873968202e-05,
-      "loss": 0.1061,
+      "loss": 0.1391,
       "step": 1345
     },
     {
       "epoch": 7.604519774011299,
-      "grad_norm": 0.05029296875,
+      "grad_norm": 0.078125,
       "learning_rate": 2.729894048699265e-05,
-      "loss": 0.0488,
+      "loss": 0.0861,
       "step": 1346
     },
     {
       "epoch": 7.610169491525424,
-      "grad_norm": 0.0458984375,
+      "grad_norm": 0.0859375,
       "learning_rate": 2.7176493875091714e-05,
-      "loss": 0.0299,
+      "loss": 0.0854,
       "step": 1347
     },
     {
       "epoch": 7.6158192090395485,
-      "grad_norm": 0.06591796875,
+      "grad_norm": 0.08349609375,
       "learning_rate": 2.7054279294119823e-05,
-      "loss": 0.0799,
+      "loss": 0.1013,
       "step": 1348
     },
     {
       "epoch": 7.621468926553672,
-      "grad_norm": 0.11669921875,
+      "grad_norm": 0.08935546875,
       "learning_rate": 2.69322971334782e-05,
-      "loss": 0.1557,
+      "loss": 0.0827,
       "step": 1349
     },
     {
       "epoch": 7.627118644067797,
-      "grad_norm": 0.04443359375,
+      "grad_norm": 0.09228515625,
       "learning_rate": 2.681054778182748e-05,
-      "loss": 0.0304,
+      "loss": 0.1007,
       "step": 1350
     },
     {
       "epoch": 7.627118644067797,
-      "eval_loss": 0.0009185844101011753,
-      "eval_runtime": 5.8176,
-      "eval_samples_per_second": 25.612,
-      "eval_steps_per_second": 3.266,
+      "eval_loss": 0.0008750773267820477,
+      "eval_runtime": 5.7503,
+      "eval_samples_per_second": 26.086,
+      "eval_steps_per_second": 3.304,
       "step": 1350
     },
     {
       "epoch": 7.632768361581921,
-      "grad_norm": 0.0810546875,
+      "grad_norm": 0.07568359375,
       "learning_rate": 2.668903162708658e-05,
-      "loss": 0.0979,
+      "loss": 0.0567,
       "step": 1351
     },
     {
       "epoch": 7.638418079096045,
-      "grad_norm": 0.044189453125,
+      "grad_norm": 0.0693359375,
       "learning_rate": 2.6567749056431467e-05,
-      "loss": 0.0339,
+      "loss": 0.0627,
       "step": 1352
     },
     {
       "epoch": 7.6440677966101696,
-      "grad_norm": 0.09619140625,
+      "grad_norm": 0.05078125,
       "learning_rate": 2.6446700456293717e-05,
-      "loss": 0.12,
+      "loss": 0.0376,
       "step": 1353
     },
     {
       "epoch": 7.649717514124294,
-      "grad_norm": 0.0966796875,
+      "grad_norm": 0.07421875,
       "learning_rate": 2.6325886212359498e-05,
-      "loss": 0.1019,
+      "loss": 0.0484,
       "step": 1354
     },
     {
       "epoch": 7.655367231638418,
-      "grad_norm": 0.057373046875,
+      "grad_norm": 0.07421875,
       "learning_rate": 2.6205306709568354e-05,
-      "loss": 0.0516,
+      "loss": 0.056,
       "step": 1355
     },
     {
       "epoch": 7.661016949152542,
-      "grad_norm": 0.048583984375,
+      "grad_norm": 0.10791015625,
       "learning_rate": 2.608496233211174e-05,
-      "loss": 0.0339,
+      "loss": 0.1524,
       "step": 1356
     },
     {
       "epoch": 7.666666666666667,
-      "grad_norm": 0.0859375,
+      "grad_norm": 0.10595703125,
       "learning_rate": 2.5964853463431994e-05,
-      "loss": 0.1021,
+      "loss": 0.1372,
       "step": 1357
     },
     {
       "epoch": 7.672316384180791,
-      "grad_norm": 0.0654296875,
+      "grad_norm": 0.058349609375,
       "learning_rate": 2.5844980486221228e-05,
-      "loss": 0.0547,
+      "loss": 0.042,
       "step": 1358
     },
     {
       "epoch": 7.677966101694915,
-      "grad_norm": 0.0615234375,
+      "grad_norm": 0.09228515625,
       "learning_rate": 2.572534378241972e-05,
-      "loss": 0.0542,
+      "loss": 0.0999,
       "step": 1359
     },
     {
       "epoch": 7.68361581920904,
-      "grad_norm": 0.07861328125,
+      "grad_norm": 0.07763671875,
       "learning_rate": 2.5605943733215042e-05,
-      "loss": 0.0701,
+      "loss": 0.0839,
       "step": 1360
     },
     {
       "epoch": 7.6892655367231635,
-      "grad_norm": 0.07177734375,
+      "grad_norm": 0.08642578125,
       "learning_rate": 2.5486780719040838e-05,
-      "loss": 0.0799,
+      "loss": 0.0862,
       "step": 1361
     },
     {
       "epoch": 7.694915254237288,
-      "grad_norm": 0.10595703125,
+      "grad_norm": 0.076171875,
       "learning_rate": 2.536785511957531e-05,
-      "loss": 0.1214,
+      "loss": 0.0708,
       "step": 1362
     },
     {
       "epoch": 7.700564971751413,
-      "grad_norm": 0.0419921875,
+      "grad_norm": 0.052490234375,
       "learning_rate": 2.5249167313740308e-05,
-      "loss": 0.0319,
+      "loss": 0.0331,
       "step": 1363
     },
     {
       "epoch": 7.706214689265536,
-      "grad_norm": 0.0517578125,
+      "grad_norm": 0.034912109375,
       "learning_rate": 2.51307176797001e-05,
-      "loss": 0.0389,
+      "loss": 0.0173,
       "step": 1364
     },
     {
       "epoch": 7.711864406779661,
-      "grad_norm": 0.07763671875,
+      "grad_norm": 0.033203125,
       "learning_rate": 2.501250659485992e-05,
-      "loss": 0.0786,
+      "loss": 0.017,
       "step": 1365
     },
     {
       "epoch": 7.7175141242937855,
-      "grad_norm": 0.10498046875,
+      "grad_norm": 0.06298828125,
       "learning_rate": 2.4894534435865015e-05,
-      "loss": 0.1258,
+      "loss": 0.0632,
       "step": 1366
     },
     {
       "epoch": 7.72316384180791,
-      "grad_norm": 0.06298828125,
+      "grad_norm": 0.06884765625,
       "learning_rate": 2.4776801578599464e-05,
-      "loss": 0.0548,
+      "loss": 0.056,
       "step": 1367
     },
     {
       "epoch": 7.728813559322034,
-      "grad_norm": 0.0634765625,
+      "grad_norm": 0.07568359375,
       "learning_rate": 2.46593083981847e-05,
-      "loss": 0.0694,
+      "loss": 0.0803,
       "step": 1368
     },
     {
       "epoch": 7.734463276836158,
-      "grad_norm": 0.10107421875,
+      "grad_norm": 0.08544921875,
       "learning_rate": 2.4542055268978603e-05,
-      "loss": 0.1398,
+      "loss": 0.0867,
       "step": 1369
     },
     {
       "epoch": 7.740112994350282,
-      "grad_norm": 0.05615234375,
+      "grad_norm": 0.0849609375,
       "learning_rate": 2.4425042564574184e-05,
-      "loss": 0.0398,
+      "loss": 0.0961,
       "step": 1370
     },
     {
       "epoch": 7.745762711864407,
-      "grad_norm": 0.062255859375,
+      "grad_norm": 0.0654296875,
       "learning_rate": 2.4308270657798404e-05,
-      "loss": 0.0524,
+      "loss": 0.0532,
       "step": 1371
     },
     {
       "epoch": 7.751412429378531,
-      "grad_norm": 0.10205078125,
+      "grad_norm": 0.08984375,
       "learning_rate": 2.4191739920710988e-05,
-      "loss": 0.1392,
+      "loss": 0.1032,
       "step": 1372
     },
     {
       "epoch": 7.757062146892656,
-      "grad_norm": 0.0771484375,
+      "grad_norm": 0.08203125,
       "learning_rate": 2.4075450724603256e-05,
-      "loss": 0.0875,
+      "loss": 0.0734,
       "step": 1373
     },
     {
       "epoch": 7.762711864406779,
-      "grad_norm": 0.09765625,
+      "grad_norm": 0.064453125,
       "learning_rate": 2.3959403439996907e-05,
-      "loss": 0.1223,
+      "loss": 0.0619,
       "step": 1374
     },
     {
       "epoch": 7.768361581920904,
-      "grad_norm": 0.08056640625,
+      "grad_norm": 0.08740234375,
       "learning_rate": 2.38435984366429e-05,
-      "loss": 0.0878,
+      "loss": 0.1038,
       "step": 1375
     },
     {
       "epoch": 7.7740112994350286,
-      "grad_norm": 0.00732421875,
+      "grad_norm": 0.08203125,
       "learning_rate": 2.372803608352018e-05,
-      "loss": 0.0004,
+      "loss": 0.0901,
       "step": 1376
     },
     {
       "epoch": 7.779661016949152,
-      "grad_norm": 0.06591796875,
+      "grad_norm": 0.0712890625,
       "learning_rate": 2.3612716748834597e-05,
-      "loss": 0.0685,
+      "loss": 0.0677,
       "step": 1377
     },
     {
       "epoch": 7.785310734463277,
-      "grad_norm": 0.099609375,
+      "grad_norm": 0.0732421875,
       "learning_rate": 2.3497640800017685e-05,
-      "loss": 0.1253,
+      "loss": 0.0725,
       "step": 1378
     },
     {
       "epoch": 7.790960451977401,
-      "grad_norm": 0.05224609375,
+      "grad_norm": 0.045654296875,
       "learning_rate": 2.3382808603725492e-05,
-      "loss": 0.0381,
+      "loss": 0.0316,
       "step": 1379
     },
     {
       "epoch": 7.796610169491525,
-      "grad_norm": 0.07958984375,
+      "grad_norm": 0.058349609375,
       "learning_rate": 2.3268220525837437e-05,
-      "loss": 0.0833,
+      "loss": 0.0654,
       "step": 1380
     },
     {
       "epoch": 7.80225988700565,
-      "grad_norm": 0.08203125,
+      "grad_norm": 0.10498046875,
       "learning_rate": 2.3153876931455088e-05,
-      "loss": 0.0874,
+      "loss": 0.1084,
       "step": 1381
     },
     {
       "epoch": 7.807909604519774,
-      "grad_norm": 0.0859375,
+      "grad_norm": 0.08447265625,
       "learning_rate": 2.3039778184901084e-05,
-      "loss": 0.0883,
+      "loss": 0.0699,
       "step": 1382
     },
     {
       "epoch": 7.813559322033898,
-      "grad_norm": 0.06982421875,
+      "grad_norm": 0.068359375,
       "learning_rate": 2.29259246497179e-05,
-      "loss": 0.0674,
+      "loss": 0.0713,
       "step": 1383
     },
     {
       "epoch": 7.8192090395480225,
-      "grad_norm": 0.048095703125,
+      "grad_norm": 0.06591796875,
       "learning_rate": 2.2812316688666737e-05,
-      "loss": 0.0493,
+      "loss": 0.0544,
       "step": 1384
     },
     {
       "epoch": 7.824858757062147,
-      "grad_norm": 0.046875,
+      "grad_norm": 0.048095703125,
       "learning_rate": 2.26989546637263e-05,
-      "loss": 0.0353,
+      "loss": 0.0245,
       "step": 1385
     },
     {
       "epoch": 7.830508474576272,
-      "grad_norm": 0.06787109375,
+      "grad_norm": 0.06982421875,
       "learning_rate": 2.2585838936091754e-05,
-      "loss": 0.0667,
+      "loss": 0.0557,
       "step": 1386
     },
     {
       "epoch": 7.836158192090395,
-      "grad_norm": 0.06201171875,
+      "grad_norm": 0.08642578125,
       "learning_rate": 2.2472969866173488e-05,
-      "loss": 0.0619,
+      "loss": 0.1046,
       "step": 1387
     },
     {
       "epoch": 7.84180790960452,
-      "grad_norm": 0.041259765625,
+      "grad_norm": 0.095703125,
       "learning_rate": 2.2360347813595893e-05,
-      "loss": 0.0193,
+      "loss": 0.1221,
       "step": 1388
     },
     {
       "epoch": 7.847457627118644,
-      "grad_norm": 0.0966796875,
+      "grad_norm": 0.0615234375,
       "learning_rate": 2.22479731371965e-05,
-      "loss": 0.1129,
+      "loss": 0.0496,
       "step": 1389
     },
     {
       "epoch": 7.853107344632768,
-      "grad_norm": 0.0947265625,
+      "grad_norm": 0.0830078125,
       "learning_rate": 2.2135846195024513e-05,
-      "loss": 0.0924,
+      "loss": 0.0617,
       "step": 1390
     },
     {
       "epoch": 7.858757062146893,
-      "grad_norm": 0.0478515625,
+      "grad_norm": 0.07666015625,
       "learning_rate": 2.2023967344339758e-05,
-      "loss": 0.0349,
+      "loss": 0.0542,
       "step": 1391
     },
     {
       "epoch": 7.864406779661017,
-      "grad_norm": 0.08544921875,
+      "grad_norm": 0.080078125,
       "learning_rate": 2.1912336941611754e-05,
-      "loss": 0.0735,
+      "loss": 0.0862,
       "step": 1392
     },
     {
       "epoch": 7.870056497175141,
-      "grad_norm": 0.052001953125,
+      "grad_norm": 0.06884765625,
       "learning_rate": 2.1800955342518325e-05,
-      "loss": 0.0374,
+      "loss": 0.0628,
       "step": 1393
     },
     {
       "epoch": 7.875706214689266,
-      "grad_norm": 0.06884765625,
+      "grad_norm": 0.07470703125,
       "learning_rate": 2.1689822901944457e-05,
-      "loss": 0.065,
+      "loss": 0.072,
       "step": 1394
     },
     {
       "epoch": 7.88135593220339,
-      "grad_norm": 0.060302734375,
+      "grad_norm": 0.09912109375,
       "learning_rate": 2.157893997398146e-05,
-      "loss": 0.0508,
+      "loss": 0.1372,
       "step": 1395
     },
     {
       "epoch": 7.88135593220339,
-      "eval_loss": 0.0009222907829098403,
-      "eval_runtime": 5.6916,
-      "eval_samples_per_second": 26.179,
-      "eval_steps_per_second": 3.338,
+      "eval_loss": 0.0008622172754257917,
+      "eval_runtime": 5.7698,
+      "eval_samples_per_second": 25.997,
+      "eval_steps_per_second": 3.293,
       "step": 1395
     },
     {
       "epoch": 7.887005649717514,
-      "grad_norm": 0.10205078125,
+      "grad_norm": 0.080078125,
       "learning_rate": 2.146830691192553e-05,
-      "loss": 0.1047,
+      "loss": 0.0846,
       "step": 1396
     },
     {
       "epoch": 7.892655367231638,
-      "grad_norm": 0.0341796875,
+      "grad_norm": 0.05029296875,
       "learning_rate": 2.135792406827668e-05,
-      "loss": 0.0175,
+      "loss": 0.029,
       "step": 1397
     },
     {
       "epoch": 7.898305084745763,
-      "grad_norm": 0.07763671875,
+      "grad_norm": 0.052978515625,
       "learning_rate": 2.1247791794737826e-05,
-      "loss": 0.0661,
+      "loss": 0.0505,
       "step": 1398
     },
     {
       "epoch": 7.903954802259887,
-      "grad_norm": 0.08203125,
+      "grad_norm": 0.059814453125,
       "learning_rate": 2.1137910442213425e-05,
-      "loss": 0.0702,
+      "loss": 0.0368,
       "step": 1399
     },
     {
       "epoch": 7.909604519774011,
-      "grad_norm": 0.09326171875,
+      "grad_norm": 0.059326171875,
       "learning_rate": 2.1028280360808407e-05,
-      "loss": 0.1186,
+      "loss": 0.0326,
       "step": 1400
     },
     {
       "epoch": 7.915254237288136,
-      "grad_norm": 0.076171875,
+      "grad_norm": 0.083984375,
       "learning_rate": 2.0918901899827237e-05,
-      "loss": 0.0816,
+      "loss": 0.0675,
       "step": 1401
     },
     {
       "epoch": 7.9209039548022595,
-      "grad_norm": 0.05810546875,
+      "grad_norm": 0.041748046875,
       "learning_rate": 2.0809775407772503e-05,
-      "loss": 0.0518,
+      "loss": 0.0313,
       "step": 1402
     },
     {
       "epoch": 7.926553672316384,
-      "grad_norm": 0.072265625,
+      "grad_norm": 0.0947265625,
       "learning_rate": 2.070090123234405e-05,
-      "loss": 0.0734,
+      "loss": 0.1132,
       "step": 1403
     },
     {
       "epoch": 7.932203389830509,
-      "grad_norm": 0.103515625,
+      "grad_norm": 0.11669921875,
       "learning_rate": 2.0592279720437858e-05,
-      "loss": 0.1408,
+      "loss": 0.1236,
       "step": 1404
     },
     {
       "epoch": 7.937853107344633,
-      "grad_norm": 0.037109375,
+      "grad_norm": 0.0390625,
       "learning_rate": 2.0483911218144715e-05,
-      "loss": 0.0207,
+      "loss": 0.0299,
       "step": 1405
     },
     {
       "epoch": 7.943502824858757,
-      "grad_norm": 0.060791015625,
+      "grad_norm": 0.09521484375,
       "learning_rate": 2.0375796070749363e-05,
-      "loss": 0.0365,
+      "loss": 0.1013,
       "step": 1406
     },
     {
       "epoch": 7.9491525423728815,
-      "grad_norm": 0.035888671875,
+      "grad_norm": 0.099609375,
       "learning_rate": 2.02679346227293e-05,
-      "loss": 0.0163,
+      "loss": 0.1107,
       "step": 1407
     },
     {
       "epoch": 7.954802259887005,
-      "grad_norm": 0.04736328125,
+      "grad_norm": 0.064453125,
       "learning_rate": 2.016032721775366e-05,
-      "loss": 0.0342,
+      "loss": 0.0556,
       "step": 1408
     },
     {
       "epoch": 7.96045197740113,
-      "grad_norm": 0.07275390625,
+      "grad_norm": 0.09423828125,
       "learning_rate": 2.005297419868216e-05,
-      "loss": 0.0534,
+      "loss": 0.0934,
       "step": 1409
     },
     {
       "epoch": 7.966101694915254,
-      "grad_norm": 0.06005859375,
+      "grad_norm": 0.044677734375,
       "learning_rate": 1.994587590756397e-05,
-      "loss": 0.0606,
+      "loss": 0.0311,
       "step": 1410
     },
     {
       "epoch": 7.971751412429379,
-      "grad_norm": 0.06591796875,
+      "grad_norm": 0.10107421875,
       "learning_rate": 1.9839032685636672e-05,
-      "loss": 0.0672,
+      "loss": 0.0862,
       "step": 1411
     },
     {
       "epoch": 7.977401129943503,
-      "grad_norm": 0.09716796875,
+      "grad_norm": 0.07861328125,
       "learning_rate": 1.973244487332512e-05,
-      "loss": 0.0945,
+      "loss": 0.0568,
       "step": 1412
     },
     {
       "epoch": 7.983050847457627,
-      "grad_norm": 0.06396484375,
+      "grad_norm": 0.11279296875,
       "learning_rate": 1.9626112810240382e-05,
-      "loss": 0.0497,
+      "loss": 0.1499,
       "step": 1413
     },
     {
       "epoch": 7.988700564971752,
-      "grad_norm": 0.09375,
+      "grad_norm": 0.080078125,
       "learning_rate": 1.9520036835178668e-05,
-      "loss": 0.1042,
+      "loss": 0.0694,
       "step": 1414
     },
     {
       "epoch": 7.994350282485875,
-      "grad_norm": 0.09326171875,
+      "grad_norm": 0.061767578125,
       "learning_rate": 1.941421728612023e-05,
-      "loss": 0.089,
+      "loss": 0.0487,
       "step": 1415
     },
     {
       "epoch": 8.0,
-      "grad_norm": 0.078125,
+      "grad_norm": 0.06396484375,
       "learning_rate": 1.9308654500228295e-05,
-      "loss": 0.1019,
+      "loss": 0.0479,
       "step": 1416
     },
     {
       "epoch": 8.005649717514125,
-      "grad_norm": 0.049072265625,
+      "grad_norm": 0.08837890625,
       "learning_rate": 1.9203348813847986e-05,
-      "loss": 0.0353,
+      "loss": 0.0903,
       "step": 1417
     },
     {
       "epoch": 8.01129943502825,
-      "grad_norm": 0.06591796875,
+      "grad_norm": 0.07666015625,
       "learning_rate": 1.9098300562505266e-05,
-      "loss": 0.0708,
+      "loss": 0.074,
       "step": 1418
     },
     {
       "epoch": 8.016949152542374,
-      "grad_norm": 0.07861328125,
+      "grad_norm": 0.09130859375,
       "learning_rate": 1.899351008090583e-05,
-      "loss": 0.071,
+      "loss": 0.0861,
       "step": 1419
     },
     {
       "epoch": 8.022598870056497,
-      "grad_norm": 0.034423828125,
+      "grad_norm": 0.07470703125,
       "learning_rate": 1.8888977702934085e-05,
-      "loss": 0.0163,
+      "loss": 0.0571,
       "step": 1420
     },
     {
       "epoch": 8.028248587570621,
-      "grad_norm": 0.05322265625,
+      "grad_norm": 0.08740234375,
       "learning_rate": 1.8784703761652068e-05,
-      "loss": 0.0361,
+      "loss": 0.1015,
       "step": 1421
     },
     {
       "epoch": 8.033898305084746,
-      "grad_norm": 0.076171875,
+      "grad_norm": 0.0791015625,
       "learning_rate": 1.8680688589298368e-05,
-      "loss": 0.0697,
+      "loss": 0.0572,
       "step": 1422
     },
     {
       "epoch": 8.03954802259887,
-      "grad_norm": 0.06982421875,
+      "grad_norm": 0.03369140625,
       "learning_rate": 1.8576932517287093e-05,
-      "loss": 0.0768,
+      "loss": 0.0169,
       "step": 1423
     },
     {
       "epoch": 8.045197740112995,
-      "grad_norm": 0.041259765625,
+      "grad_norm": 0.072265625,
       "learning_rate": 1.847343587620679e-05,
-      "loss": 0.031,
+      "loss": 0.0674,
       "step": 1424
     },
     {
       "epoch": 8.05084745762712,
-      "grad_norm": 0.08447265625,
+      "grad_norm": 0.0771484375,
       "learning_rate": 1.837019899581943e-05,
-      "loss": 0.068,
+      "loss": 0.0689,
       "step": 1425
     },
     {
       "epoch": 8.056497175141242,
-      "grad_norm": 0.061279296875,
+      "grad_norm": 0.07373046875,
       "learning_rate": 1.8267222205059308e-05,
-      "loss": 0.035,
+      "loss": 0.0623,
       "step": 1426
     },
     {
       "epoch": 8.062146892655367,
-      "grad_norm": 0.0693359375,
+      "grad_norm": 0.1044921875,
       "learning_rate": 1.8164505832032054e-05,
-      "loss": 0.0617,
+      "loss": 0.1092,
       "step": 1427
     },
     {
       "epoch": 8.067796610169491,
-      "grad_norm": 0.052978515625,
+      "grad_norm": 0.049560546875,
       "learning_rate": 1.8062050204013446e-05,
-      "loss": 0.034,
+      "loss": 0.033,
       "step": 1428
     },
     {
       "epoch": 8.073446327683616,
-      "grad_norm": 0.08447265625,
+      "grad_norm": 0.08837890625,
       "learning_rate": 1.795985564744864e-05,
-      "loss": 0.0879,
+      "loss": 0.1,
       "step": 1429
     },
     {
       "epoch": 8.07909604519774,
-      "grad_norm": 0.061767578125,
+      "grad_norm": 0.07275390625,
       "learning_rate": 1.7857922487950874e-05,
-      "loss": 0.0524,
+      "loss": 0.0706,
       "step": 1430
     },
     {
       "epoch": 8.084745762711865,
-      "grad_norm": 0.0615234375,
+      "grad_norm": 0.072265625,
       "learning_rate": 1.7756251050300432e-05,
-      "loss": 0.0535,
+      "loss": 0.0528,
       "step": 1431
     },
     {
       "epoch": 8.090395480225988,
-      "grad_norm": 0.0615234375,
+      "grad_norm": 0.0634765625,
       "learning_rate": 1.765484165844389e-05,
       "loss": 0.0542,
       "step": 1432
     },
     {
       "epoch": 8.096045197740112,
-      "grad_norm": 0.119140625,
+      "grad_norm": 0.0712890625,
       "learning_rate": 1.755369463549278e-05,
-      "loss": 0.1637,
+      "loss": 0.0536,
       "step": 1433
     },
     {
       "epoch": 8.101694915254237,
-      "grad_norm": 0.06640625,
+      "grad_norm": 0.051513671875,
       "learning_rate": 1.74528103037226e-05,
-      "loss": 0.0786,
+      "loss": 0.0305,
       "step": 1434
     },
     {
       "epoch": 8.107344632768362,
-      "grad_norm": 0.09423828125,
+      "grad_norm": 0.087890625,
       "learning_rate": 1.7352188984572026e-05,
-      "loss": 0.135,
+      "loss": 0.0806,
       "step": 1435
     },
     {
       "epoch": 8.112994350282486,
-      "grad_norm": 0.1005859375,
+      "grad_norm": 0.0751953125,
       "learning_rate": 1.7251830998641638e-05,
-      "loss": 0.1469,
+      "loss": 0.0706,
       "step": 1436
     },
     {
       "epoch": 8.11864406779661,
-      "grad_norm": 0.10400390625,
+      "grad_norm": 0.099609375,
       "learning_rate": 1.7151736665692896e-05,
-      "loss": 0.1372,
+      "loss": 0.118,
       "step": 1437
     },
     {
       "epoch": 8.124293785310735,
-      "grad_norm": 0.041748046875,
+      "grad_norm": 0.08837890625,
       "learning_rate": 1.7051906304647368e-05,
-      "loss": 0.0178,
+      "loss": 0.1044,
       "step": 1438
     },
     {
       "epoch": 8.129943502824858,
-      "grad_norm": 0.08056640625,
+      "grad_norm": 0.08349609375,
       "learning_rate": 1.6952340233585473e-05,
-      "loss": 0.0958,
+      "loss": 0.0682,
       "step": 1439
     },
     {
       "epoch": 8.135593220338983,
-      "grad_norm": 0.036865234375,
+      "grad_norm": 0.09130859375,
       "learning_rate": 1.6853038769745467e-05,
-      "loss": 0.0196,
+      "loss": 0.0925,
       "step": 1440
     },
     {
       "epoch": 8.135593220338983,
-      "eval_loss": 0.0009117891313508153,
-      "eval_runtime": 5.7882,
-      "eval_samples_per_second": 25.742,
-      "eval_steps_per_second": 3.283,
+      "eval_loss": 0.0008618851425126195,
+      "eval_runtime": 5.8148,
+      "eval_samples_per_second": 25.796,
+      "eval_steps_per_second": 3.268,
       "step": 1440
     },
     {
       "epoch": 8.141242937853107,
-      "grad_norm": 0.060791015625,
+      "grad_norm": 0.076171875,
       "learning_rate": 1.675400222952268e-05,
-      "loss": 0.0516,
+      "loss": 0.0346,
       "step": 1441
     },
     {
       "epoch": 8.146892655367232,
-      "grad_norm": 0.0927734375,
+      "grad_norm": 0.07568359375,
       "learning_rate": 1.6655230928468258e-05,
-      "loss": 0.111,
+      "loss": 0.0705,
       "step": 1442
     },
     {
       "epoch": 8.152542372881356,
-      "grad_norm": 0.07763671875,
+      "grad_norm": 0.09716796875,
       "learning_rate": 1.6556725181288156e-05,
-      "loss": 0.0871,
+      "loss": 0.124,
       "step": 1443
     },
     {
       "epoch": 8.158192090395481,
-      "grad_norm": 0.054443359375,
+      "grad_norm": 0.052490234375,
       "learning_rate": 1.645848530184233e-05,
-      "loss": 0.0345,
+      "loss": 0.0331,
       "step": 1444
     },
     {
       "epoch": 8.163841807909604,
-      "grad_norm": 0.04443359375,
+      "grad_norm": 0.1005859375,
       "learning_rate": 1.6360511603143648e-05,
-      "loss": 0.0301,
+      "loss": 0.1247,
       "step": 1445
     },
     {
       "epoch": 8.169491525423728,
-      "grad_norm": 0.07275390625,
+      "grad_norm": 0.08056640625,
       "learning_rate": 1.6262804397356746e-05,
-      "loss": 0.0826,
+      "loss": 0.0703,
       "step": 1446
     },
     {
       "epoch": 8.175141242937853,
-      "grad_norm": 0.04345703125,
+      "grad_norm": 0.099609375,
       "learning_rate": 1.6165363995797223e-05,
-      "loss": 0.03,
+      "loss": 0.1215,
       "step": 1447
     },
     {
       "epoch": 8.180790960451978,
-      "grad_norm": 0.05517578125,
+      "grad_norm": 0.08837890625,
       "learning_rate": 1.6068190708930675e-05,
-      "loss": 0.0522,
+      "loss": 0.0759,
       "step": 1448
     },
     {
       "epoch": 8.186440677966102,
-      "grad_norm": 0.07177734375,
+      "grad_norm": 0.0439453125,
       "learning_rate": 1.597128484637144e-05,
-      "loss": 0.0711,
+      "loss": 0.0168,
       "step": 1449
     },
     {
       "epoch": 8.192090395480227,
-      "grad_norm": 0.0947265625,
+      "grad_norm": 0.06982421875,
       "learning_rate": 1.587464671688187e-05,
-      "loss": 0.1074,
+      "loss": 0.0605,
       "step": 1450
     },
     {
       "epoch": 8.19774011299435,
-      "grad_norm": 0.08154296875,
+      "grad_norm": 0.1005859375,
       "learning_rate": 1.5778276628371357e-05,
-      "loss": 0.0741,
+      "loss": 0.1049,
       "step": 1451
     },
     {
       "epoch": 8.203389830508474,
-      "grad_norm": 0.08349609375,
+      "grad_norm": 0.0712890625,
       "learning_rate": 1.5682174887895083e-05,
-      "loss": 0.0875,
+      "loss": 0.0649,
       "step": 1452
     },
     {
       "epoch": 8.209039548022599,
-      "grad_norm": 0.0810546875,
+      "grad_norm": 0.0908203125,
       "learning_rate": 1.5586341801653315e-05,
-      "loss": 0.0869,
+      "loss": 0.071,
       "step": 1453
     },
     {
       "epoch": 8.214689265536723,
-      "grad_norm": 0.078125,
+      "grad_norm": 0.060546875,
       "learning_rate": 1.5490777674990376e-05,
-      "loss": 0.0664,
+      "loss": 0.0316,
       "step": 1454
     },
     {
       "epoch": 8.220338983050848,
-      "grad_norm": 0.055908203125,
+      "grad_norm": 0.09228515625,
       "learning_rate": 1.5395482812393514e-05,
-      "loss": 0.0328,
+      "loss": 0.0939,
       "step": 1455
     },
     {
       "epoch": 8.225988700564972,
-      "grad_norm": 0.07763671875,
+      "grad_norm": 0.049072265625,
       "learning_rate": 1.530045751749212e-05,
-      "loss": 0.0853,
+      "loss": 0.0316,
       "step": 1456
     },
     {
       "epoch": 8.231638418079097,
-      "grad_norm": 0.0986328125,
+      "grad_norm": 0.08935546875,
       "learning_rate": 1.520570209305665e-05,
-      "loss": 0.1274,
+      "loss": 0.0849,
       "step": 1457
     },
     {
       "epoch": 8.23728813559322,
-      "grad_norm": 0.0986328125,
+      "grad_norm": 0.0927734375,
       "learning_rate": 1.5111216840997743e-05,
-      "loss": 0.1227,
+      "loss": 0.1051,
       "step": 1458
     },
     {
       "epoch": 8.242937853107344,
-      "grad_norm": 0.08154296875,
+      "grad_norm": 0.04833984375,
       "learning_rate": 1.5017002062365172e-05,
-      "loss": 0.0809,
+      "loss": 0.0289,
       "step": 1459
     },
     {
       "epoch": 8.248587570621469,
-      "grad_norm": 0.055419921875,
+      "grad_norm": 0.080078125,
       "learning_rate": 1.4923058057346929e-05,
-      "loss": 0.0502,
+      "loss": 0.0534,
       "step": 1460
     },
     {
       "epoch": 8.254237288135593,
-      "grad_norm": 0.06005859375,
+      "grad_norm": 0.0869140625,
       "learning_rate": 1.482938512526828e-05,
-      "loss": 0.0508,
+      "loss": 0.0859,
       "step": 1461
     },
     {
       "epoch": 8.259887005649718,
-      "grad_norm": 0.06201171875,
+      "grad_norm": 0.0849609375,
       "learning_rate": 1.4735983564590783e-05,
-      "loss": 0.0536,
+      "loss": 0.0708,
       "step": 1462
     },
     {
       "epoch": 8.265536723163843,
-      "grad_norm": 0.10107421875,
+      "grad_norm": 0.0771484375,
       "learning_rate": 1.464285367291136e-05,
-      "loss": 0.1049,
+      "loss": 0.0682,
       "step": 1463
     },
     {
       "epoch": 8.271186440677965,
-      "grad_norm": 0.07421875,
+      "grad_norm": 0.0732421875,
       "learning_rate": 1.4549995746961332e-05,
-      "loss": 0.0684,
+      "loss": 0.0674,
       "step": 1464
     },
     {
       "epoch": 8.27683615819209,
-      "grad_norm": 0.0673828125,
+      "grad_norm": 0.072265625,
       "learning_rate": 1.4457410082605483e-05,
-      "loss": 0.0529,
+      "loss": 0.0624,
       "step": 1465
     },
     {
       "epoch": 8.282485875706215,
-      "grad_norm": 0.09033203125,
+      "grad_norm": 0.0732421875,
       "learning_rate": 1.4365096974841108e-05,
-      "loss": 0.1037,
+      "loss": 0.0548,
       "step": 1466
     },
     {
       "epoch": 8.288135593220339,
-      "grad_norm": 0.0703125,
+      "grad_norm": 0.06005859375,
       "learning_rate": 1.4273056717797084e-05,
-      "loss": 0.0407,
+      "loss": 0.0488,
       "step": 1467
     },
     {
       "epoch": 8.293785310734464,
-      "grad_norm": 0.06298828125,
+      "grad_norm": 0.0830078125,
       "learning_rate": 1.418128960473294e-05,
-      "loss": 0.0366,
+      "loss": 0.0708,
       "step": 1468
     },
     {
       "epoch": 8.299435028248588,
-      "grad_norm": 0.0732421875,
+      "grad_norm": 0.06689453125,
       "learning_rate": 1.4089795928037896e-05,
-      "loss": 0.0674,
+      "loss": 0.0525,
       "step": 1469
     },
     {
       "epoch": 8.305084745762711,
-      "grad_norm": 0.08837890625,
+      "grad_norm": 0.0869140625,
       "learning_rate": 1.3998575979229944e-05,
-      "loss": 0.099,
+      "loss": 0.0696,
       "step": 1470
     },
     {
       "epoch": 8.310734463276836,
-      "grad_norm": 0.06640625,
+      "grad_norm": 0.0771484375,
       "learning_rate": 1.3907630048954944e-05,
-      "loss": 0.0648,
+      "loss": 0.0814,
       "step": 1471
     },
     {
       "epoch": 8.31638418079096,
-      "grad_norm": 0.06396484375,
+      "grad_norm": 0.06640625,
       "learning_rate": 1.381695842698565e-05,
-      "loss": 0.0654,
+      "loss": 0.0529,
       "step": 1472
     },
     {
       "epoch": 8.322033898305085,
-      "grad_norm": 0.08935546875,
+      "grad_norm": 0.04345703125,
       "learning_rate": 1.3726561402220817e-05,
-      "loss": 0.0887,
+      "loss": 0.0303,
       "step": 1473
     },
     {
       "epoch": 8.32768361581921,
-      "grad_norm": 0.0673828125,
+      "grad_norm": 0.059326171875,
       "learning_rate": 1.3636439262684298e-05,
-      "loss": 0.0619,
+      "loss": 0.0439,
       "step": 1474
     },
     {
       "epoch": 8.333333333333334,
-      "grad_norm": 0.087890625,
+      "grad_norm": 0.033447265625,
       "learning_rate": 1.3546592295524074e-05,
-      "loss": 0.1031,
+      "loss": 0.017,
       "step": 1475
     },
     {
       "epoch": 8.338983050847457,
-      "grad_norm": 0.10009765625,
+      "grad_norm": 0.06640625,
       "learning_rate": 1.3457020787011388e-05,
-      "loss": 0.0966,
+      "loss": 0.0509,
       "step": 1476
     },
     {
       "epoch": 8.344632768361581,
-      "grad_norm": 0.04296875,
+      "grad_norm": 0.0830078125,
       "learning_rate": 1.3367725022539824e-05,
-      "loss": 0.0207,
+      "loss": 0.0871,
       "step": 1477
     },
     {
       "epoch": 8.350282485875706,
-      "grad_norm": 0.0615234375,
+      "grad_norm": 0.076171875,
       "learning_rate": 1.3278705286624327e-05,
-      "loss": 0.0498,
+      "loss": 0.0666,
       "step": 1478
     },
     {
       "epoch": 8.35593220338983,
-      "grad_norm": 0.08544921875,
+      "grad_norm": 0.061279296875,
       "learning_rate": 1.3189961862900447e-05,
-      "loss": 0.0826,
+      "loss": 0.0544,
       "step": 1479
     },
     {
       "epoch": 8.361581920903955,
-      "grad_norm": 0.09228515625,
+      "grad_norm": 0.08642578125,
       "learning_rate": 1.3101495034123313e-05,
-      "loss": 0.079,
+      "loss": 0.084,
       "step": 1480
     },
     {
       "epoch": 8.36723163841808,
-      "grad_norm": 0.07958984375,
+      "grad_norm": 0.059814453125,
       "learning_rate": 1.3013305082166694e-05,
-      "loss": 0.0698,
+      "loss": 0.0368,
       "step": 1481
     },
     {
       "epoch": 8.372881355932204,
-      "grad_norm": 0.07373046875,
+      "grad_norm": 0.0458984375,
       "learning_rate": 1.2925392288022298e-05,
-      "loss": 0.0714,
+      "loss": 0.034,
       "step": 1482
     },
     {
       "epoch": 8.378531073446327,
-      "grad_norm": 0.0927734375,
+      "grad_norm": 0.09326171875,
       "learning_rate": 1.2837756931798638e-05,
-      "loss": 0.1044,
+      "loss": 0.1084,
       "step": 1483
     },
     {
       "epoch": 8.384180790960452,
-      "grad_norm": 0.1162109375,
+      "grad_norm": 0.06494140625,
       "learning_rate": 1.2750399292720283e-05,
-      "loss": 0.1321,
+      "loss": 0.0502,
       "step": 1484
     },
     {
       "epoch": 8.389830508474576,
-      "grad_norm": 0.078125,
+      "grad_norm": 0.0537109375,
       "learning_rate": 1.2663319649127026e-05,
-      "loss": 0.0709,
+      "loss": 0.0342,
       "step": 1485
     },
     {
       "epoch": 8.389830508474576,
-      "eval_loss": 0.0009126444347202778,
-      "eval_runtime": 5.7247,
-      "eval_samples_per_second": 26.027,
-      "eval_steps_per_second": 3.319,
+      "eval_loss": 0.000868115748744458,
+      "eval_runtime": 5.9704,
+      "eval_samples_per_second": 25.124,
+      "eval_steps_per_second": 3.182,
       "step": 1485
     },
     {
       "epoch": 8.3954802259887,
-      "grad_norm": 0.08837890625,
+      "grad_norm": 0.07861328125,
       "learning_rate": 1.2576518278472759e-05,
-      "loss": 0.1146,
+      "loss": 0.0676,
       "step": 1486
     },
     {
       "epoch": 8.401129943502825,
-      "grad_norm": 0.07373046875,
+      "grad_norm": 0.08203125,
       "learning_rate": 1.2489995457324822e-05,
-      "loss": 0.0531,
+      "loss": 0.0915,
       "step": 1487
     },
     {
       "epoch": 8.40677966101695,
-      "grad_norm": 0.0810546875,
+      "grad_norm": 0.051025390625,
       "learning_rate": 1.2403751461363089e-05,
-      "loss": 0.0894,
+      "loss": 0.0311,
       "step": 1488
     },
     {
       "epoch": 8.412429378531073,
-      "grad_norm": 0.0751953125,
+      "grad_norm": 0.08056640625,
       "learning_rate": 1.2317786565378908e-05,
-      "loss": 0.0525,
+      "loss": 0.0726,
       "step": 1489
     },
     {
       "epoch": 8.418079096045197,
-      "grad_norm": 0.06640625,
+      "grad_norm": 0.09912109375,
       "learning_rate": 1.2232101043274436e-05,
-      "loss": 0.0501,
+      "loss": 0.0869,
       "step": 1490
     },
     {
       "epoch": 8.423728813559322,
-      "grad_norm": 0.07666015625,
+      "grad_norm": 0.10302734375,
       "learning_rate": 1.2146695168061728e-05,
-      "loss": 0.0739,
+      "loss": 0.126,
       "step": 1491
     },
     {
       "epoch": 8.429378531073446,
-      "grad_norm": 0.0849609375,
+      "grad_norm": 0.06591796875,
       "learning_rate": 1.2061569211861712e-05,
-      "loss": 0.1014,
+      "loss": 0.0521,
       "step": 1492
     },
     {
       "epoch": 8.435028248587571,
-      "grad_norm": 0.103515625,
+      "grad_norm": 0.0810546875,
       "learning_rate": 1.1976723445903515e-05,
-      "loss": 0.1249,
+      "loss": 0.0738,
       "step": 1493
     },
     {
       "epoch": 8.440677966101696,
-      "grad_norm": 0.061279296875,
+      "grad_norm": 0.07470703125,
       "learning_rate": 1.1892158140523546e-05,
-      "loss": 0.0379,
+      "loss": 0.0824,
       "step": 1494
     },
     {
       "epoch": 8.44632768361582,
-      "grad_norm": 0.0908203125,
+      "grad_norm": 0.08251953125,
       "learning_rate": 1.1807873565164506e-05,
-      "loss": 0.0996,
+      "loss": 0.0831,
       "step": 1495
     },
     {
       "epoch": 8.451977401129943,
-      "grad_norm": 0.048583984375,
+      "grad_norm": 0.064453125,
       "learning_rate": 1.1723869988374714e-05,
-      "loss": 0.0356,
+      "loss": 0.0643,
       "step": 1496
     },
     {
       "epoch": 8.457627118644067,
-      "grad_norm": 0.053955078125,
+      "grad_norm": 0.0673828125,
       "learning_rate": 1.1640147677807134e-05,
-      "loss": 0.0356,
+      "loss": 0.035,
       "step": 1497
     },
     {
       "epoch": 8.463276836158192,
-      "grad_norm": 0.056640625,
+      "grad_norm": 0.06103515625,
       "learning_rate": 1.1556706900218572e-05,
-      "loss": 0.0486,
+      "loss": 0.0621,
       "step": 1498
     },
     {
       "epoch": 8.468926553672317,
-      "grad_norm": 0.08349609375,
+      "grad_norm": 0.09619140625,
       "learning_rate": 1.1473547921468818e-05,
-      "loss": 0.0759,
+      "loss": 0.1157,
       "step": 1499
     },
     {
       "epoch": 8.474576271186441,
-      "grad_norm": 0.09326171875,
+      "grad_norm": 0.07275390625,
       "learning_rate": 1.139067100651976e-05,
-      "loss": 0.1201,
+      "loss": 0.0605,
       "step": 1500
     },
     {
       "epoch": 8.480225988700566,
-      "grad_norm": 0.053466796875,
+      "grad_norm": 0.06884765625,
       "learning_rate": 1.1308076419434588e-05,
-      "loss": 0.0471,
+      "loss": 0.0538,
       "step": 1501
     },
     {
       "epoch": 8.485875706214689,
-      "grad_norm": 0.058837890625,
+      "grad_norm": 0.09716796875,
       "learning_rate": 1.122576442337696e-05,
-      "loss": 0.0499,
+      "loss": 0.0859,
       "step": 1502
     },
     {
       "epoch": 8.491525423728813,
-      "grad_norm": 0.057861328125,
+      "grad_norm": 0.042724609375,
       "learning_rate": 1.1143735280610112e-05,
-      "loss": 0.0394,
+      "loss": 0.0173,
       "step": 1503
     },
     {
       "epoch": 8.497175141242938,
-      "grad_norm": 0.09130859375,
+      "grad_norm": 0.0830078125,
       "learning_rate": 1.1061989252496053e-05,
-      "loss": 0.1163,
+      "loss": 0.0709,
       "step": 1504
     },
     {
       "epoch": 8.502824858757062,
-      "grad_norm": 0.06884765625,
+      "grad_norm": 0.05859375,
       "learning_rate": 1.0980526599494733e-05,
-      "loss": 0.0658,
+      "loss": 0.0464,
       "step": 1505
     },
     {
       "epoch": 8.508474576271187,
-      "grad_norm": 0.08935546875,
+      "grad_norm": 0.064453125,
       "learning_rate": 1.0899347581163221e-05,
-      "loss": 0.0839,
+      "loss": 0.0478,
       "step": 1506
     },
     {
       "epoch": 8.514124293785311,
-      "grad_norm": 0.08203125,
+      "grad_norm": 0.08837890625,
       "learning_rate": 1.0818452456154849e-05,
-      "loss": 0.0761,
+      "loss": 0.0876,
       "step": 1507
     },
     {
       "epoch": 8.519774011299434,
-      "grad_norm": 0.09521484375,
+      "grad_norm": 0.09912109375,
       "learning_rate": 1.0737841482218403e-05,
-      "loss": 0.1206,
+      "loss": 0.1187,
       "step": 1508
     },
     {
       "epoch": 8.525423728813559,
-      "grad_norm": 0.08203125,
+      "grad_norm": 0.078125,
       "learning_rate": 1.065751491619733e-05,
-      "loss": 0.0849,
+      "loss": 0.0846,
       "step": 1509
     },
     {
       "epoch": 8.531073446327683,
-      "grad_norm": 0.0859375,
+      "grad_norm": 0.050537109375,
       "learning_rate": 1.057747301402887e-05,
-      "loss": 0.1002,
+      "loss": 0.02,
       "step": 1510
     },
     {
       "epoch": 8.536723163841808,
-      "grad_norm": 0.10986328125,
+      "grad_norm": 0.09326171875,
       "learning_rate": 1.0497716030743288e-05,
-      "loss": 0.1391,
+      "loss": 0.0849,
       "step": 1511
     },
     {
       "epoch": 8.542372881355933,
-      "grad_norm": 0.0810546875,
+      "grad_norm": 0.0634765625,
       "learning_rate": 1.0418244220462991e-05,
-      "loss": 0.0705,
+      "loss": 0.0397,
       "step": 1512
     },
     {
       "epoch": 8.548022598870057,
-      "grad_norm": 0.07373046875,
+      "grad_norm": 0.08740234375,
       "learning_rate": 1.0339057836401821e-05,
-      "loss": 0.0629,
+      "loss": 0.0679,
       "step": 1513
     },
     {
       "epoch": 8.55367231638418,
-      "grad_norm": 0.0654296875,
+      "grad_norm": 0.09326171875,
       "learning_rate": 1.026015713086418e-05,
-      "loss": 0.0534,
+      "loss": 0.1103,
       "step": 1514
     },
     {
       "epoch": 8.559322033898304,
-      "grad_norm": 0.053955078125,
+      "grad_norm": 0.068359375,
       "learning_rate": 1.0181542355244167e-05,
-      "loss": 0.0361,
+      "loss": 0.0671,
       "step": 1515
     },
     {
       "epoch": 8.564971751412429,
-      "grad_norm": 0.047119140625,
+      "grad_norm": 0.09423828125,
       "learning_rate": 1.0103213760024965e-05,
-      "loss": 0.0344,
+      "loss": 0.1074,
       "step": 1516
     },
     {
       "epoch": 8.570621468926554,
-      "grad_norm": 0.056884765625,
+      "grad_norm": 0.0712890625,
       "learning_rate": 1.0025171594777871e-05,
-      "loss": 0.0344,
+      "loss": 0.0519,
       "step": 1517
     },
     {
       "epoch": 8.576271186440678,
-      "grad_norm": 0.05419921875,
+      "grad_norm": 0.10546875,
       "learning_rate": 9.9474161081615e-06,
-      "loss": 0.0459,
+      "loss": 0.1244,
       "step": 1518
     },
     {
       "epoch": 8.581920903954803,
-      "grad_norm": 0.0556640625,
+      "grad_norm": 0.08349609375,
       "learning_rate": 9.869947547921143e-06,
-      "loss": 0.0354,
+      "loss": 0.0682,
       "step": 1519
     },
     {
       "epoch": 8.587570621468927,
-      "grad_norm": 0.09130859375,
+      "grad_norm": 0.10009765625,
       "learning_rate": 9.792766160887868e-06,
-      "loss": 0.0889,
+      "loss": 0.1056,
       "step": 1520
     },
     {
       "epoch": 8.59322033898305,
-      "grad_norm": 0.052001953125,
+      "grad_norm": 0.1142578125,
       "learning_rate": 9.71587219297765e-06,
-      "loss": 0.0323,
+      "loss": 0.1198,
       "step": 1521
     },
     {
       "epoch": 8.598870056497175,
-      "grad_norm": 0.061279296875,
+      "grad_norm": 0.06787109375,
       "learning_rate": 9.63926588919083e-06,
-      "loss": 0.0348,
+      "loss": 0.0641,
       "step": 1522
     },
     {
       "epoch": 8.6045197740113,
-      "grad_norm": 0.07763671875,
+      "grad_norm": 0.10009765625,
       "learning_rate": 9.562947493611108e-06,
-      "loss": 0.0667,
+      "loss": 0.0995,
       "step": 1523
     },
     {
       "epoch": 8.610169491525424,
-      "grad_norm": 0.08056640625,
+      "grad_norm": 0.0693359375,
       "learning_rate": 9.486917249404815e-06,
-      "loss": 0.0773,
+      "loss": 0.048,
       "step": 1524
     },
     {
       "epoch": 8.615819209039548,
-      "grad_norm": 0.056640625,
+      "grad_norm": 0.08203125,
       "learning_rate": 9.411175398820271e-06,
-      "loss": 0.0498,
+      "loss": 0.082,
       "step": 1525
     },
     {
       "epoch": 8.621468926553673,
-      "grad_norm": 0.07421875,
+      "grad_norm": 0.03955078125,
       "learning_rate": 9.335722183186868e-06,
-      "loss": 0.0695,
+      "loss": 0.019,
       "step": 1526
     },
     {
       "epoch": 8.627118644067796,
-      "grad_norm": 0.04443359375,
+      "grad_norm": 0.1025390625,
       "learning_rate": 9.260557842914275e-06,
-      "loss": 0.0329,
+      "loss": 0.1005,
       "step": 1527
     },
     {
       "epoch": 8.63276836158192,
-      "grad_norm": 0.08544921875,
+      "grad_norm": 0.0830078125,
       "learning_rate": 9.185682617491863e-06,
-      "loss": 0.0891,
+      "loss": 0.0749,
       "step": 1528
     },
     {
       "epoch": 8.638418079096045,
-      "grad_norm": 0.078125,
+      "grad_norm": 0.083984375,
       "learning_rate": 9.111096745487779e-06,
-      "loss": 0.0852,
+      "loss": 0.0711,
       "step": 1529
     },
     {
       "epoch": 8.64406779661017,
-      "grad_norm": 0.08056640625,
+      "grad_norm": 0.08984375,
       "learning_rate": 9.036800464548157e-06,
-      "loss": 0.0852,
+      "loss": 0.0688,
       "step": 1530
     },
     {
       "epoch": 8.64406779661017,
-      "eval_loss": 0.0009017279953695834,
-      "eval_runtime": 5.7363,
-      "eval_samples_per_second": 25.975,
+      "eval_loss": 0.0008601490408182144,
+      "eval_runtime": 5.7359,
+      "eval_samples_per_second": 26.151,
       "eval_steps_per_second": 3.312,
       "step": 1530
     },
     {
       "epoch": 8.649717514124294,
-      "grad_norm": 0.1044921875,
+      "grad_norm": 0.10302734375,
       "learning_rate": 8.962794011396547e-06,
-      "loss": 0.1131,
+      "loss": 0.105,
       "step": 1531
     },
     {
       "epoch": 8.655367231638419,
       "grad_norm": 0.0791015625,
       "learning_rate": 8.889077621832997e-06,
-      "loss": 0.0664,
+      "loss": 0.0672,
       "step": 1532
     },
     {
       "epoch": 8.661016949152543,
-      "grad_norm": 0.09228515625,
+      "grad_norm": 0.07666015625,
       "learning_rate": 8.815651530733327e-06,
-      "loss": 0.0903,
+      "loss": 0.0527,
       "step": 1533
     },
     {
       "epoch": 8.666666666666666,
-      "grad_norm": 0.10302734375,
+      "grad_norm": 0.060302734375,
       "learning_rate": 8.742515972048404e-06,
-      "loss": 0.1156,
+      "loss": 0.0501,
       "step": 1534
     },
     {
       "epoch": 8.67231638418079,
-      "grad_norm": 0.046142578125,
+      "grad_norm": 0.061767578125,
       "learning_rate": 8.669671178803485e-06,
-      "loss": 0.0327,
+      "loss": 0.0489,
       "step": 1535
     },
     {
       "epoch": 8.677966101694915,
-      "grad_norm": 0.040283203125,
+      "grad_norm": 0.09375,
       "learning_rate": 8.597117383097275e-06,
-      "loss": 0.0321,
+      "loss": 0.0891,
       "step": 1536
     },
     {
       "epoch": 8.68361581920904,
-      "grad_norm": 0.0947265625,
+      "grad_norm": 0.11572265625,
       "learning_rate": 8.524854816101357e-06,
-      "loss": 0.094,
+      "loss": 0.1246,
       "step": 1537
     },
     {
       "epoch": 8.689265536723164,
-      "grad_norm": 0.087890625,
+      "grad_norm": 0.08203125,
       "learning_rate": 8.4528837080594e-06,
-      "loss": 0.0933,
+      "loss": 0.0899,
       "step": 1538
     },
     {
       "epoch": 8.694915254237289,
-      "grad_norm": 0.0517578125,
+      "grad_norm": 0.035888671875,
       "learning_rate": 8.381204288286415e-06,
-      "loss": 0.05,
+      "loss": 0.0152,
       "step": 1539
     },
     {
       "epoch": 8.700564971751412,
-      "grad_norm": 0.056884765625,
+      "grad_norm": 0.06591796875,
       "learning_rate": 8.309816785168034e-06,
-      "loss": 0.0369,
+      "loss": 0.036,
       "step": 1540
     },
     {
       "epoch": 8.706214689265536,
-      "grad_norm": 0.0810546875,
+      "grad_norm": 0.0791015625,
       "learning_rate": 8.238721426159767e-06,
-      "loss": 0.0725,
+      "loss": 0.0687,
       "step": 1541
     },
     {
       "epoch": 8.711864406779661,
-      "grad_norm": 0.0537109375,
+      "grad_norm": 0.034912109375,
       "learning_rate": 8.167918437786315e-06,
-      "loss": 0.0439,
+      "loss": 0.015,
       "step": 1542
     },
     {
       "epoch": 8.717514124293785,
-      "grad_norm": 0.0791015625,
+      "grad_norm": 0.08837890625,
       "learning_rate": 8.09740804564082e-06,
-      "loss": 0.0571,
+      "loss": 0.0687,
       "step": 1543
     },
     {
       "epoch": 8.72316384180791,
-      "grad_norm": 0.0654296875,
+      "grad_norm": 0.0712890625,
       "learning_rate": 8.027190474384127e-06,
-      "loss": 0.0513,
+      "loss": 0.0551,
       "step": 1544
     },
     {
       "epoch": 8.728813559322035,
-      "grad_norm": 0.09521484375,
+      "grad_norm": 0.07568359375,
       "learning_rate": 7.95726594774413e-06,
-      "loss": 0.1026,
+      "loss": 0.0662,
       "step": 1545
     },
     {
       "epoch": 8.734463276836157,
-      "grad_norm": 0.07861328125,
+      "grad_norm": 0.06787109375,
       "learning_rate": 7.887634688515e-06,
-      "loss": 0.0696,
+      "loss": 0.0536,
       "step": 1546
     },
     {
       "epoch": 8.740112994350282,
-      "grad_norm": 0.0498046875,
+      "grad_norm": 0.087890625,
       "learning_rate": 7.818296918556479e-06,
-      "loss": 0.033,
+      "loss": 0.0833,
       "step": 1547
     },
     {
       "epoch": 8.745762711864407,
-      "grad_norm": 0.076171875,
+      "grad_norm": 0.05908203125,
       "learning_rate": 7.749252858793243e-06,
-      "loss": 0.0835,
+      "loss": 0.0303,
       "step": 1548
     },
     {
       "epoch": 8.751412429378531,
-      "grad_norm": 0.080078125,
+      "grad_norm": 0.0986328125,
       "learning_rate": 7.680502729214089e-06,
-      "loss": 0.07,
+      "loss": 0.0985,
       "step": 1549
     },
     {
       "epoch": 8.757062146892656,
-      "grad_norm": 0.10400390625,
+      "grad_norm": 0.06884765625,
       "learning_rate": 7.612046748871327e-06,
-      "loss": 0.1057,
+      "loss": 0.0645,
       "step": 1550
     },
     {
       "epoch": 8.76271186440678,
-      "grad_norm": 0.07080078125,
+      "grad_norm": 0.08154296875,
       "learning_rate": 7.543885135880036e-06,
-      "loss": 0.0525,
+      "loss": 0.0886,
       "step": 1551
     },
     {
       "epoch": 8.768361581920903,
-      "grad_norm": 0.0634765625,
+      "grad_norm": 0.080078125,
       "learning_rate": 7.476018107417371e-06,
-      "loss": 0.0687,
+      "loss": 0.0834,
       "step": 1552
     },
     {
       "epoch": 8.774011299435028,
-      "grad_norm": 0.09375,
+      "grad_norm": 0.08203125,
       "learning_rate": 7.4084458797218925e-06,
-      "loss": 0.1049,
+      "loss": 0.0638,
       "step": 1553
     },
     {
       "epoch": 8.779661016949152,
-      "grad_norm": 0.0673828125,
+      "grad_norm": 0.07080078125,
       "learning_rate": 7.341168668092857e-06,
-      "loss": 0.0537,
+      "loss": 0.0495,
       "step": 1554
     },
     {
       "epoch": 8.785310734463277,
-      "grad_norm": 0.08837890625,
+      "grad_norm": 0.0888671875,
       "learning_rate": 7.2741866868895395e-06,
-      "loss": 0.099,
+      "loss": 0.0787,
       "step": 1555
     },
     {
       "epoch": 8.790960451977401,
-      "grad_norm": 0.08447265625,
+      "grad_norm": 0.07080078125,
       "learning_rate": 7.207500149530533e-06,
-      "loss": 0.0857,
+      "loss": 0.0539,
       "step": 1556
     },
     {
       "epoch": 8.796610169491526,
-      "grad_norm": 0.08203125,
+      "grad_norm": 0.11279296875,
       "learning_rate": 7.141109268493118e-06,
-      "loss": 0.0818,
+      "loss": 0.1359,
       "step": 1557
     },
     {
       "epoch": 8.80225988700565,
-      "grad_norm": 0.061767578125,
+      "grad_norm": 0.04345703125,
       "learning_rate": 7.07501425531254e-06,
-      "loss": 0.0509,
+      "loss": 0.0295,
       "step": 1558
     },
     {
       "epoch": 8.807909604519773,
-      "grad_norm": 0.06640625,
+      "grad_norm": 0.0947265625,
       "learning_rate": 7.009215320581319e-06,
-      "loss": 0.0521,
+      "loss": 0.091,
       "step": 1559
     },
     {
       "epoch": 8.813559322033898,
-      "grad_norm": 0.072265625,
+      "grad_norm": 0.08837890625,
       "learning_rate": 6.943712673948644e-06,
-      "loss": 0.0668,
+      "loss": 0.0735,
       "step": 1560
     },
     {
       "epoch": 8.819209039548022,
-      "grad_norm": 0.0615234375,
+      "grad_norm": 0.0771484375,
       "learning_rate": 6.878506524119643e-06,
-      "loss": 0.0509,
+      "loss": 0.0808,
       "step": 1561
     },
     {
       "epoch": 8.824858757062147,
-      "grad_norm": 0.07861328125,
+      "grad_norm": 0.078125,
       "learning_rate": 6.813597078854772e-06,
-      "loss": 0.0728,
+      "loss": 0.0668,
       "step": 1562
     },
     {
       "epoch": 8.830508474576272,
-      "grad_norm": 0.09521484375,
+      "grad_norm": 0.06396484375,
       "learning_rate": 6.7489845449690836e-06,
-      "loss": 0.1085,
+      "loss": 0.0372,
       "step": 1563
     },
     {
       "epoch": 8.836158192090396,
-      "grad_norm": 0.058349609375,
+      "grad_norm": 0.06201171875,
       "learning_rate": 6.684669128331655e-06,
-      "loss": 0.0493,
+      "loss": 0.051,
       "step": 1564
     },
     {
       "epoch": 8.841807909604519,
-      "grad_norm": 0.0830078125,
+      "grad_norm": 0.0966796875,
       "learning_rate": 6.620651033864844e-06,
-      "loss": 0.074,
+      "loss": 0.0927,
       "step": 1565
     },
     {
       "epoch": 8.847457627118644,
-      "grad_norm": 0.107421875,
+      "grad_norm": 0.08984375,
       "learning_rate": 6.556930465543709e-06,
-      "loss": 0.1337,
+      "loss": 0.0707,
       "step": 1566
     },
     {
       "epoch": 8.853107344632768,
-      "grad_norm": 0.06982421875,
+      "grad_norm": 0.07421875,
       "learning_rate": 6.49350762639529e-06,
-      "loss": 0.0684,
+      "loss": 0.0546,
       "step": 1567
     },
     {
       "epoch": 8.858757062146893,
-      "grad_norm": 0.0634765625,
+      "grad_norm": 0.06640625,
       "learning_rate": 6.430382718498052e-06,
-      "loss": 0.0688,
+      "loss": 0.061,
       "step": 1568
     },
     {
       "epoch": 8.864406779661017,
-      "grad_norm": 0.06689453125,
+      "grad_norm": 0.08154296875,
       "learning_rate": 6.367555942981152e-06,
-      "loss": 0.0624,
+      "loss": 0.0749,
       "step": 1569
     },
     {
       "epoch": 8.870056497175142,
-      "grad_norm": 0.1171875,
+      "grad_norm": 0.061279296875,
       "learning_rate": 6.3050275000238414e-06,
-      "loss": 0.1366,
+      "loss": 0.0467,
       "step": 1570
     },
     {
       "epoch": 8.875706214689266,
-      "grad_norm": 0.08154296875,
+      "grad_norm": 0.083984375,
       "learning_rate": 6.2427975888547984e-06,
-      "loss": 0.0752,
+      "loss": 0.0854,
       "step": 1571
     },
     {
       "epoch": 8.88135593220339,
-      "grad_norm": 0.076171875,
+      "grad_norm": 0.1259765625,
       "learning_rate": 6.180866407751595e-06,
-      "loss": 0.0668,
+      "loss": 0.1264,
       "step": 1572
     },
     {
       "epoch": 8.887005649717514,
-      "grad_norm": 0.06982421875,
+      "grad_norm": 0.08544921875,
       "learning_rate": 6.119234154039899e-06,
-      "loss": 0.0493,
+      "loss": 0.098,
       "step": 1573
     },
     {
       "epoch": 8.892655367231638,
-      "grad_norm": 0.09423828125,
+      "grad_norm": 0.07373046875,
       "learning_rate": 6.057901024092949e-06,
-      "loss": 0.1016,
+      "loss": 0.0639,
       "step": 1574
     },
     {
       "epoch": 8.898305084745763,
-      "grad_norm": 0.072265625,
+      "grad_norm": 0.08154296875,
       "learning_rate": 5.996867213330992e-06,
-      "loss": 0.0803,
+      "loss": 0.0576,
       "step": 1575
     },
     {
       "epoch": 8.898305084745763,
-      "eval_loss": 0.0009092978434637189,
-      "eval_runtime": 5.7184,
-      "eval_samples_per_second": 26.056,
-      "eval_steps_per_second": 3.323,
+      "eval_loss": 0.0008571520447731018,
+      "eval_runtime": 5.7274,
+      "eval_samples_per_second": 26.19,
+      "eval_steps_per_second": 3.317,
       "step": 1575
     },
     {
       "epoch": 8.903954802259888,
-      "grad_norm": 0.046142578125,
+      "grad_norm": 0.11962890625,
       "learning_rate": 5.93613291622046e-06,
-      "loss": 0.0319,
+      "loss": 0.143,
       "step": 1576
     },
     {
       "epoch": 8.90960451977401,
-      "grad_norm": 0.06689453125,
+      "grad_norm": 0.0771484375,
       "learning_rate": 5.875698326273549e-06,
-      "loss": 0.0548,
+      "loss": 0.0667,
       "step": 1577
     },
     {
       "epoch": 8.915254237288135,
-      "grad_norm": 0.057373046875,
+      "grad_norm": 0.06689453125,
       "learning_rate": 5.8155636360475385e-06,
-      "loss": 0.0495,
+      "loss": 0.0488,
       "step": 1578
     },
     {
       "epoch": 8.92090395480226,
-      "grad_norm": 0.09130859375,
+      "grad_norm": 0.0693359375,
       "learning_rate": 5.755729037144097e-06,
-      "loss": 0.0855,
+      "loss": 0.0344,
       "step": 1579
     },
     {
       "epoch": 8.926553672316384,
       "grad_norm": 0.07861328125,
       "learning_rate": 5.696194720208792e-06,
-      "loss": 0.0714,
+      "loss": 0.0702,
       "step": 1580
     },
     {
       "epoch": 8.932203389830509,
       "grad_norm": 0.08154296875,
       "learning_rate": 5.636960874930431e-06,
-      "loss": 0.0815,
+      "loss": 0.0692,
       "step": 1581
     },
     {
       "epoch": 8.937853107344633,
-      "grad_norm": 0.09326171875,
+      "grad_norm": 0.125,
       "learning_rate": 5.57802769004041e-06,
-      "loss": 0.1009,
+      "loss": 0.1461,
       "step": 1582
     },
     {
       "epoch": 8.943502824858758,
-      "grad_norm": 0.09912109375,
+      "grad_norm": 0.09130859375,
       "learning_rate": 5.519395353312196e-06,
-      "loss": 0.1062,
+      "loss": 0.0875,
       "step": 1583
     },
     {
       "epoch": 8.94915254237288,
-      "grad_norm": 0.0810546875,
+      "grad_norm": 0.09521484375,
       "learning_rate": 5.461064051560705e-06,
-      "loss": 0.0803,
+      "loss": 0.1315,
       "step": 1584
     },
     {
       "epoch": 8.954802259887005,
-      "grad_norm": 0.08642578125,
+      "grad_norm": 0.10986328125,
       "learning_rate": 5.403033970641647e-06,
-      "loss": 0.0831,
+      "loss": 0.1333,
       "step": 1585
     },
     {
       "epoch": 8.96045197740113,
-      "grad_norm": 0.08447265625,
+      "grad_norm": 0.06884765625,
       "learning_rate": 5.345305295450997e-06,
-      "loss": 0.085,
+      "loss": 0.0565,
       "step": 1586
     },
     {
       "epoch": 8.966101694915254,
-      "grad_norm": 0.05419921875,
+      "grad_norm": 0.078125,
       "learning_rate": 5.287878209924435e-06,
-      "loss": 0.0477,
+      "loss": 0.0683,
       "step": 1587
     },
     {
       "epoch": 8.971751412429379,
-      "grad_norm": 0.0654296875,
+      "grad_norm": 0.09521484375,
       "learning_rate": 5.230752897036628e-06,
-      "loss": 0.0556,
+      "loss": 0.0927,
       "step": 1588
     },
     {
       "epoch": 8.977401129943503,
-      "grad_norm": 0.08349609375,
+      "grad_norm": 0.09619140625,
       "learning_rate": 5.1739295388007925e-06,
-      "loss": 0.0891,
+      "loss": 0.1064,
       "step": 1589
     },
     {
       "epoch": 8.983050847457626,
-      "grad_norm": 0.035888671875,
+      "grad_norm": 0.09130859375,
       "learning_rate": 5.1174083162680465e-06,
-      "loss": 0.0181,
+      "loss": 0.0729,
       "step": 1590
     },
     {
       "epoch": 8.98870056497175,
-      "grad_norm": 0.052978515625,
+      "grad_norm": 0.08642578125,
       "learning_rate": 5.061189409526812e-06,
-      "loss": 0.0338,
+      "loss": 0.0845,
       "step": 1591
     },
     {
       "epoch": 8.994350282485875,
-      "grad_norm": 0.0869140625,
+      "grad_norm": 0.07763671875,
       "learning_rate": 5.005272997702304e-06,
-      "loss": 0.0806,
+      "loss": 0.0683,
       "step": 1592
     },
     {
       "epoch": 9.0,
-      "grad_norm": 0.0654296875,
+      "grad_norm": 0.0703125,
       "learning_rate": 4.949659258955919e-06,
-      "loss": 0.0558,
+      "loss": 0.0714,
       "step": 1593
     }
   ],