Training in progress, step 410, checkpoint

last-checkpoint/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f9c4f278247affc7de359c904a1b8dce7c0b13b5f05f9e84953d32477ee89bea
+oid sha256:6ca1520a3649298ab68b04f841144d4338b54caa5d22eb48cd6328f2972c319a
 size 373077376

last-checkpoint/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:061d8b81bf5d5bf384b54a5ae2667084850ad01b8f621ce9740fb24aaa8e8fd6
-size 129979595
+oid sha256:0179f804a9e5120a28a9f1b84949552c7698afcd9d6950ebc6554aab994c1222
+size 373225035

last-checkpoint/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:afa414c61f15afda9f56c15a8085d690966f41ac14b2bf137cdb676429126c8c
+size 14645

last-checkpoint/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ab3f6792b2e15368d892c2ae81463f5bf7fbdb393cdbf9f22ce0395e2971286f
+oid sha256:cc9336dce323fabcb70d013e4b04bfc4fd14283d2730e1b1f6d35b089d2154dc
 size 1401

last-checkpoint/trainer_state.json

@@ -1,307 +1,2889 @@
 {
-  "best_global_step": 1000,
-  "best_metric": 8.754563331604004,
-  "best_model_checkpoint": "./artifacts/models/webtext-8x-250801/checkpoint-1000",
-  "epoch": 0.6321112515802781,
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 10.0,
   "eval_steps": 1000,
-  "global_step": 1000,
+  "global_step": 410,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
-      "epoch": 0.01580278128950695,
-      "grad_norm": 1.1875,
-      "learning_rate": 8.999999999999999e-05,
-      "loss": 10.8705,
+      "epoch": 0.024691358024691357,
+      "grad_norm": 3.625,
+      "learning_rate": 0.0,
+      "loss": 10.914,
+      "step": 1
+    },
+    {
+      "epoch": 0.04938271604938271,
+      "grad_norm": 2.90625,
+      "learning_rate": 4.761904761904762e-05,
+      "loss": 10.9199,
+      "step": 2
+    },
+    {
+      "epoch": 0.07407407407407407,
+      "grad_norm": 2.734375,
+      "learning_rate": 9.523809523809524e-05,
+      "loss": 10.8854,
+      "step": 3
+    },
+    {
+      "epoch": 0.09876543209876543,
+      "grad_norm": 3.15625,
+      "learning_rate": 0.00014285714285714284,
+      "loss": 10.8935,
+      "step": 4
+    },
+    {
+      "epoch": 0.12345679012345678,
+      "grad_norm": 2.953125,
+      "learning_rate": 0.00019047619047619048,
+      "loss": 10.8767,
+      "step": 5
+    },
+    {
+      "epoch": 0.14814814814814814,
+      "grad_norm": 2.953125,
+      "learning_rate": 0.0002380952380952381,
+      "loss": 10.8796,
+      "step": 6
+    },
+    {
+      "epoch": 0.1728395061728395,
+      "grad_norm": 2.796875,
+      "learning_rate": 0.0002857142857142857,
+      "loss": 10.8683,
+      "step": 7
+    },
+    {
+      "epoch": 0.19753086419753085,
+      "grad_norm": 2.65625,
+      "learning_rate": 0.0003333333333333333,
+      "loss": 10.854,
+      "step": 8
+    },
+    {
+      "epoch": 0.2222222222222222,
+      "grad_norm": 3.4375,
+      "learning_rate": 0.00038095238095238096,
+      "loss": 10.8677,
+      "step": 9
+    },
+    {
+      "epoch": 0.24691358024691357,
+      "grad_norm": 2.75,
+      "learning_rate": 0.00042857142857142855,
+      "loss": 10.8461,
+      "step": 10
+    },
+    {
+      "epoch": 0.2716049382716049,
+      "grad_norm": 2.515625,
+      "learning_rate": 0.0004761904761904762,
+      "loss": 10.8452,
+      "step": 11
+    },
+    {
+      "epoch": 0.2962962962962963,
+      "grad_norm": 2.953125,
+      "learning_rate": 0.0005238095238095238,
+      "loss": 10.8138,
+      "step": 12
+    },
+    {
+      "epoch": 0.32098765432098764,
+      "grad_norm": 2.328125,
+      "learning_rate": 0.0005714285714285714,
+      "loss": 10.8397,
+      "step": 13
+    },
+    {
+      "epoch": 0.345679012345679,
+      "grad_norm": 3.171875,
+      "learning_rate": 0.0006190476190476191,
+      "loss": 10.6607,
+      "step": 14
+    },
+    {
+      "epoch": 0.37037037037037035,
+      "grad_norm": 2.46875,
+      "learning_rate": 0.0006666666666666666,
+      "loss": 10.7965,
+      "step": 15
+    },
+    {
+      "epoch": 0.3950617283950617,
+      "grad_norm": 3.046875,
+      "learning_rate": 0.0007142857142857143,
+      "loss": 10.7092,
+      "step": 16
+    },
+    {
+      "epoch": 0.41975308641975306,
+      "grad_norm": 3.640625,
+      "learning_rate": 0.0007619047619047619,
+      "loss": 10.5584,
+      "step": 17
+    },
+    {
+      "epoch": 0.4444444444444444,
+      "grad_norm": 3.390625,
+      "learning_rate": 0.0008095238095238096,
+      "loss": 10.5369,
+      "step": 18
+    },
+    {
+      "epoch": 0.4691358024691358,
+      "grad_norm": 2.515625,
+      "learning_rate": 0.0008571428571428571,
+      "loss": 10.7799,
+      "step": 19
+    },
+    {
+      "epoch": 0.49382716049382713,
+      "grad_norm": 2.921875,
+      "learning_rate": 0.0009047619047619047,
+      "loss": 10.6937,
+      "step": 20
+    },
+    {
+      "epoch": 0.5185185185185185,
+      "grad_norm": 3.59375,
+      "learning_rate": 0.0009523809523809524,
+      "loss": 10.5466,
+      "step": 21
+    },
+    {
+      "epoch": 0.5432098765432098,
+      "grad_norm": 3.34375,
+      "learning_rate": 0.001,
+      "loss": 10.5925,
+      "step": 22
+    },
+    {
+      "epoch": 0.5679012345679012,
+      "grad_norm": 3.125,
+      "learning_rate": 0.0009999836943471868,
+      "loss": 10.6907,
+      "step": 23
+    },
+    {
+      "epoch": 0.5925925925925926,
+      "grad_norm": 2.703125,
+      "learning_rate": 0.000999934778452244,
+      "loss": 10.7284,
+      "step": 24
+    },
+    {
+      "epoch": 0.6172839506172839,
+      "grad_norm": 3.59375,
+      "learning_rate": 0.000999853255505594,
+      "loss": 10.4334,
       "step": 25
     },
     {
-      "epoch": 0.0316055625790139,
-      "grad_norm": 1.2265625,
-      "learning_rate": 0.00018375,
-      "loss": 10.7952,
-      "step": 50
+      "epoch": 0.6419753086419753,
+      "grad_norm": 3.625,
+      "learning_rate": 0.0009997391308243769,
+      "loss": 10.4547,
+      "step": 26
+    },
+    {
+      "epoch": 0.6666666666666666,
+      "grad_norm": 2.921875,
+      "learning_rate": 0.0009995924118521018,
+      "loss": 10.6062,
+      "step": 27
+    },
+    {
+      "epoch": 0.691358024691358,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.0009994131081581632,
+      "loss": 10.4063,
+      "step": 28
+    },
+    {
+      "epoch": 0.7160493827160493,
+      "grad_norm": 4.90625,
+      "learning_rate": 0.0009992012314372165,
+      "loss": 10.3801,
+      "step": 29
+    },
+    {
+      "epoch": 0.7407407407407407,
+      "grad_norm": 4.90625,
+      "learning_rate": 0.0009989567955084144,
+      "loss": 10.2523,
+      "step": 30
+    },
+    {
+      "epoch": 0.7654320987654321,
+      "grad_norm": 2.84375,
+      "learning_rate": 0.0009986798163145068,
+      "loss": 10.6218,
+      "step": 31
+    },
+    {
+      "epoch": 0.7901234567901234,
+      "grad_norm": 4.28125,
+      "learning_rate": 0.0009983703119208,
+      "loss": 10.3131,
+      "step": 32
+    },
+    {
+      "epoch": 0.8148148148148148,
+      "grad_norm": 5.5,
+      "learning_rate": 0.0009980283025139783,
+      "loss": 10.2454,
+      "step": 33
+    },
+    {
+      "epoch": 0.8395061728395061,
+      "grad_norm": 3.3125,
+      "learning_rate": 0.0009976538104007886,
+      "loss": 10.5066,
+      "step": 34
+    },
+    {
+      "epoch": 0.8641975308641975,
+      "grad_norm": 3.96875,
+      "learning_rate": 0.0009972468600065844,
+      "loss": 10.5809,
+      "step": 35
+    },
+    {
+      "epoch": 0.8888888888888888,
+      "grad_norm": 5.0625,
+      "learning_rate": 0.0009968074778737334,
+      "loss": 10.062,
+      "step": 36
+    },
+    {
+      "epoch": 0.9135802469135802,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.0009963356926598848,
+      "loss": 10.2692,
+      "step": 37
+    },
+    {
+      "epoch": 0.9382716049382716,
+      "grad_norm": 4.375,
+      "learning_rate": 0.000995831535136103,
+      "loss": 10.2122,
+      "step": 38
+    },
+    {
+      "epoch": 0.9629629629629629,
+      "grad_norm": 4.09375,
+      "learning_rate": 0.0009952950381848576,
+      "loss": 10.1825,
+      "step": 39
+    },
+    {
+      "epoch": 0.9876543209876543,
+      "grad_norm": 5.375,
+      "learning_rate": 0.000994726236797881,
+      "loss": 9.7581,
+      "step": 40
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 4.4375,
+      "learning_rate": 0.000994125168073885,
+      "loss": 10.368,
+      "step": 41
+    },
+    {
+      "epoch": 1.0246913580246915,
+      "grad_norm": 3.59375,
+      "learning_rate": 0.0009934918712161413,
+      "loss": 10.3786,
+      "step": 42
+    },
+    {
+      "epoch": 1.0493827160493827,
+      "grad_norm": 2.96875,
+      "learning_rate": 0.0009928263875299246,
+      "loss": 10.3233,
+      "step": 43
+    },
+    {
+      "epoch": 1.074074074074074,
+      "grad_norm": 4.34375,
+      "learning_rate": 0.0009921287604198186,
+      "loss": 10.0873,
+      "step": 44
+    },
+    {
+      "epoch": 1.0987654320987654,
+      "grad_norm": 4.5,
+      "learning_rate": 0.000991399035386885,
+      "loss": 9.8532,
+      "step": 45
+    },
+    {
+      "epoch": 1.123456790123457,
+      "grad_norm": 5.1875,
+      "learning_rate": 0.0009906372600256962,
+      "loss": 9.9951,
+      "step": 46
+    },
+    {
+      "epoch": 1.1481481481481481,
+      "grad_norm": 6.75,
+      "learning_rate": 0.0009898434840212304,
+      "loss": 9.1483,
+      "step": 47
+    },
+    {
+      "epoch": 1.1728395061728394,
+      "grad_norm": 6.0,
+      "learning_rate": 0.0009890177591456314,
+      "loss": 9.3933,
+      "step": 48
+    },
+    {
+      "epoch": 1.1975308641975309,
+      "grad_norm": 3.265625,
+      "learning_rate": 0.0009881601392548315,
+      "loss": 10.3489,
+      "step": 49
+    },
+    {
+      "epoch": 1.2222222222222223,
+      "grad_norm": 3.5625,
+      "learning_rate": 0.0009872706802850396,
+      "loss": 10.122,
+      "step": 50
+    },
+    {
+      "epoch": 1.2469135802469136,
+      "grad_norm": 3.578125,
+      "learning_rate": 0.0009863494402490923,
+      "loss": 10.3155,
+      "step": 51
+    },
+    {
+      "epoch": 1.2716049382716048,
+      "grad_norm": 3.203125,
+      "learning_rate": 0.0009853964792326704,
+      "loss": 10.364,
+      "step": 52
+    },
+    {
+      "epoch": 1.2962962962962963,
+      "grad_norm": 4.15625,
+      "learning_rate": 0.0009844118593903797,
+      "loss": 10.1379,
+      "step": 53
+    },
+    {
+      "epoch": 1.3209876543209877,
+      "grad_norm": 4.65625,
+      "learning_rate": 0.0009833956449416977,
+      "loss": 9.7886,
+      "step": 54
+    },
+    {
+      "epoch": 1.345679012345679,
+      "grad_norm": 4.8125,
+      "learning_rate": 0.0009823479021667837,
+      "loss": 9.7699,
+      "step": 55
+    },
+    {
+      "epoch": 1.3703703703703702,
+      "grad_norm": 6.90625,
+      "learning_rate": 0.0009812686994021581,
+      "loss": 8.9695,
+      "step": 56
+    },
+    {
+      "epoch": 1.3950617283950617,
+      "grad_norm": 3.71875,
+      "learning_rate": 0.000980158107036243,
+      "loss": 10.182,
+      "step": 57
+    },
+    {
+      "epoch": 1.4197530864197532,
+      "grad_norm": 4.21875,
+      "learning_rate": 0.0009790161975047724,
+      "loss": 10.3364,
+      "step": 58
+    },
+    {
+      "epoch": 1.4444444444444444,
+      "grad_norm": 4.03125,
+      "learning_rate": 0.000977843045286068,
+      "loss": 9.9323,
+      "step": 59
+    },
+    {
+      "epoch": 1.4691358024691357,
+      "grad_norm": 4.59375,
+      "learning_rate": 0.0009766387268961808,
+      "loss": 9.7666,
+      "step": 60
+    },
+    {
+      "epoch": 1.4938271604938271,
+      "grad_norm": 4.4375,
+      "learning_rate": 0.000975403320883901,
+      "loss": 9.7933,
+      "step": 61
+    },
+    {
+      "epoch": 1.5185185185185186,
+      "grad_norm": 4.375,
+      "learning_rate": 0.0009741369078256346,
+      "loss": 9.9836,
+      "step": 62
+    },
+    {
+      "epoch": 1.5432098765432098,
+      "grad_norm": 4.1875,
+      "learning_rate": 0.0009728395703201482,
+      "loss": 10.0322,
+      "step": 63
+    },
+    {
+      "epoch": 1.567901234567901,
+      "grad_norm": 6.34375,
+      "learning_rate": 0.0009715113929831816,
+      "loss": 9.5231,
+      "step": 64
+    },
+    {
+      "epoch": 1.5925925925925926,
+      "grad_norm": 5.53125,
+      "learning_rate": 0.0009701524624419289,
+      "loss": 9.4926,
+      "step": 65
+    },
+    {
+      "epoch": 1.617283950617284,
+      "grad_norm": 3.828125,
+      "learning_rate": 0.0009687628673293887,
+      "loss": 10.2509,
+      "step": 66
+    },
+    {
+      "epoch": 1.6419753086419753,
+      "grad_norm": 5.375,
+      "learning_rate": 0.0009673426982785825,
+      "loss": 9.537,
+      "step": 67
+    },
+    {
+      "epoch": 1.6666666666666665,
+      "grad_norm": 5.0,
+      "learning_rate": 0.0009658920479166444,
+      "loss": 9.4252,
+      "step": 68
+    },
+    {
+      "epoch": 1.691358024691358,
+      "grad_norm": 5.875,
+      "learning_rate": 0.0009644110108587791,
+      "loss": 9.6164,
+      "step": 69
+    },
+    {
+      "epoch": 1.7160493827160495,
+      "grad_norm": 4.5,
+      "learning_rate": 0.0009628996837020907,
+      "loss": 9.9444,
+      "step": 70
+    },
+    {
+      "epoch": 1.7407407407407407,
+      "grad_norm": 7.5,
+      "learning_rate": 0.0009613581650192831,
+      "loss": 8.9878,
+      "step": 71
+    },
+    {
+      "epoch": 1.765432098765432,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.0009597865553522297,
+      "loss": 9.7977,
+      "step": 72
+    },
+    {
+      "epoch": 1.7901234567901234,
+      "grad_norm": 5.84375,
+      "learning_rate": 0.0009581849572054171,
+      "loss": 8.9699,
+      "step": 73
+    },
+    {
+      "epoch": 1.8148148148148149,
+      "grad_norm": 5.28125,
+      "learning_rate": 0.0009565534750392586,
+      "loss": 9.2648,
+      "step": 74
+    },
+    {
+      "epoch": 1.8395061728395061,
+      "grad_norm": 5.59375,
+      "learning_rate": 0.0009548922152632812,
+      "loss": 9.4919,
+      "step": 75
+    },
+    {
+      "epoch": 1.8641975308641974,
+      "grad_norm": 5.125,
+      "learning_rate": 0.0009532012862291854,
+      "loss": 9.5113,
+      "step": 76
+    },
+    {
+      "epoch": 1.8888888888888888,
+      "grad_norm": 6.46875,
+      "learning_rate": 0.0009514807982237785,
+      "loss": 9.288,
+      "step": 77
+    },
+    {
+      "epoch": 1.9135802469135803,
+      "grad_norm": 3.875,
+      "learning_rate": 0.0009497308634617807,
+      "loss": 10.2324,
+      "step": 78
+    },
+    {
+      "epoch": 1.9382716049382716,
+      "grad_norm": 7.0,
+      "learning_rate": 0.0009479515960785068,
+      "loss": 8.892,
+      "step": 79
+    },
+    {
+      "epoch": 1.9629629629629628,
+      "grad_norm": 6.71875,
+      "learning_rate": 0.0009461431121224215,
+      "loss": 8.8531,
+      "step": 80
+    },
+    {
+      "epoch": 1.9876543209876543,
+      "grad_norm": 3.578125,
+      "learning_rate": 0.0009443055295475707,
+      "loss": 10.2284,
+      "step": 81
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 3.75,
+      "learning_rate": 0.0009424389682058887,
+      "loss": 10.6826,
+      "step": 82
+    },
+    {
+      "epoch": 2.0246913580246915,
+      "grad_norm": 6.25,
+      "learning_rate": 0.0009405435498393799,
+      "loss": 8.4895,
+      "step": 83
+    },
+    {
+      "epoch": 2.049382716049383,
+      "grad_norm": 5.84375,
+      "learning_rate": 0.00093861939807218,
+      "loss": 8.713,
+      "step": 84
+    },
+    {
+      "epoch": 2.074074074074074,
+      "grad_norm": 4.28125,
+      "learning_rate": 0.0009366666384024913,
+      "loss": 9.3152,
+      "step": 85
+    },
+    {
+      "epoch": 2.0987654320987654,
+      "grad_norm": 5.21875,
+      "learning_rate": 0.0009346853981943989,
+      "loss": 8.6951,
+      "step": 86
+    },
+    {
+      "epoch": 2.123456790123457,
+      "grad_norm": 4.46875,
+      "learning_rate": 0.0009326758066695624,
+      "loss": 9.3574,
+      "step": 87
+    },
+    {
+      "epoch": 2.148148148148148,
+      "grad_norm": 4.5625,
+      "learning_rate": 0.0009306379948987889,
+      "loss": 8.9596,
+      "step": 88
+    },
+    {
+      "epoch": 2.1728395061728394,
+      "grad_norm": 4.8125,
+      "learning_rate": 0.0009285720957934831,
+      "loss": 9.4665,
+      "step": 89
+    },
+    {
+      "epoch": 2.197530864197531,
+      "grad_norm": 4.1875,
+      "learning_rate": 0.0009264782440969795,
+      "loss": 9.2718,
+      "step": 90
+    },
+    {
+      "epoch": 2.2222222222222223,
+      "grad_norm": 7.21875,
+      "learning_rate": 0.000924356576375753,
+      "loss": 8.1262,
+      "step": 91
+    },
+    {
+      "epoch": 2.246913580246914,
+      "grad_norm": 3.84375,
+      "learning_rate": 0.0009222072310105128,
+      "loss": 9.9563,
+      "step": 92
+    },
+    {
+      "epoch": 2.271604938271605,
+      "grad_norm": 5.46875,
+      "learning_rate": 0.0009200303481871759,
+      "loss": 8.6349,
+      "step": 93
+    },
+    {
+      "epoch": 2.2962962962962963,
+      "grad_norm": 4.75,
+      "learning_rate": 0.0009178260698877248,
+      "loss": 9.2621,
+      "step": 94
+    },
+    {
+      "epoch": 2.3209876543209877,
+      "grad_norm": 6.59375,
+      "learning_rate": 0.0009155945398809458,
+      "loss": 7.6972,
+      "step": 95
+    },
+    {
+      "epoch": 2.3456790123456788,
+      "grad_norm": 6.59375,
+      "learning_rate": 0.0009133359037130531,
+      "loss": 8.3546,
+      "step": 96
+    },
+    {
+      "epoch": 2.3703703703703702,
+      "grad_norm": 4.78125,
+      "learning_rate": 0.0009110503086981956,
+      "loss": 9.5228,
+      "step": 97
+    },
+    {
+      "epoch": 2.3950617283950617,
+      "grad_norm": 3.890625,
+      "learning_rate": 0.0009087379039088481,
+      "loss": 9.8113,
+      "step": 98
+    },
+    {
+      "epoch": 2.419753086419753,
+      "grad_norm": 4.59375,
+      "learning_rate": 0.0009063988401660897,
+      "loss": 9.3982,
+      "step": 99
+    },
+    {
+      "epoch": 2.4444444444444446,
+      "grad_norm": 4.75,
+      "learning_rate": 0.0009040332700297651,
+      "loss": 9.4069,
+      "step": 100
+    },
+    {
+      "epoch": 2.4691358024691357,
+      "grad_norm": 3.3125,
+      "learning_rate": 0.0009016413477885362,
+      "loss": 10.4287,
+      "step": 101
+    },
+    {
+      "epoch": 2.493827160493827,
+      "grad_norm": 4.875,
+      "learning_rate": 0.000899223229449817,
+      "loss": 9.2696,
+      "step": 102
+    },
+    {
+      "epoch": 2.5185185185185186,
+      "grad_norm": 6.9375,
+      "learning_rate": 0.0008967790727296001,
+      "loss": 7.8461,
+      "step": 103
+    },
+    {
+      "epoch": 2.5432098765432096,
+      "grad_norm": 6.875,
+      "learning_rate": 0.0008943090370421692,
+      "loss": 8.1242,
+      "step": 104
+    },
+    {
+      "epoch": 2.567901234567901,
+      "grad_norm": 4.0,
+      "learning_rate": 0.0008918132834897016,
+      "loss": 9.7601,
+      "step": 105
+    },
+    {
+      "epoch": 2.5925925925925926,
+      "grad_norm": 3.703125,
+      "learning_rate": 0.000889291974851761,
+      "loss": 9.7302,
+      "step": 106
+    },
+    {
+      "epoch": 2.617283950617284,
+      "grad_norm": 6.5,
+      "learning_rate": 0.0008867452755746807,
+      "loss": 8.1455,
+      "step": 107
+    },
+    {
+      "epoch": 2.6419753086419755,
+      "grad_norm": 3.1875,
+      "learning_rate": 0.0008841733517608375,
+      "loss": 10.4135,
+      "step": 108
+    },
+    {
+      "epoch": 2.6666666666666665,
+      "grad_norm": 5.8125,
+      "learning_rate": 0.0008815763711578184,
+      "loss": 8.9868,
+      "step": 109
+    },
+    {
+      "epoch": 2.691358024691358,
+      "grad_norm": 2.875,
+      "learning_rate": 0.0008789545031474799,
+      "loss": 10.3719,
+      "step": 110
+    },
+    {
+      "epoch": 2.7160493827160495,
+      "grad_norm": 6.375,
+      "learning_rate": 0.0008763079187348999,
+      "loss": 8.0962,
+      "step": 111
+    },
+    {
+      "epoch": 2.7407407407407405,
+      "grad_norm": 5.28125,
+      "learning_rate": 0.0008736367905372247,
+      "loss": 8.6631,
+      "step": 112
+    },
+    {
+      "epoch": 2.765432098765432,
+      "grad_norm": 7.28125,
+      "learning_rate": 0.0008709412927724103,
+      "loss": 8.7969,
+      "step": 113
+    },
+    {
+      "epoch": 2.7901234567901234,
+      "grad_norm": 5.4375,
+      "learning_rate": 0.0008682216012478596,
+      "loss": 8.7167,
+      "step": 114
+    },
+    {
+      "epoch": 2.814814814814815,
+      "grad_norm": 4.96875,
+      "learning_rate": 0.0008654778933489557,
+      "loss": 9.2549,
+      "step": 115
+    },
+    {
+      "epoch": 2.8395061728395063,
+      "grad_norm": 5.875,
+      "learning_rate": 0.0008627103480274922,
+      "loss": 8.9388,
+      "step": 116
+    },
+    {
+      "epoch": 2.8641975308641974,
+      "grad_norm": 8.625,
+      "learning_rate": 0.0008599191457900016,
+      "loss": 7.2287,
+      "step": 117
+    },
+    {
+      "epoch": 2.888888888888889,
+      "grad_norm": 4.53125,
+      "learning_rate": 0.0008571044686859826,
+      "loss": 9.4793,
+      "step": 118
+    },
+    {
+      "epoch": 2.9135802469135803,
+      "grad_norm": 6.1875,
+      "learning_rate": 0.0008542665002960258,
+      "loss": 8.6185,
+      "step": 119
+    },
+    {
+      "epoch": 2.9382716049382713,
+      "grad_norm": 6.46875,
+      "learning_rate": 0.0008514054257198399,
+      "loss": 8.6557,
+      "step": 120
+    },
+    {
+      "epoch": 2.962962962962963,
+      "grad_norm": 6.0625,
+      "learning_rate": 0.0008485214315641801,
+      "loss": 8.2225,
+      "step": 121
+    },
+    {
+      "epoch": 2.9876543209876543,
+      "grad_norm": 5.25,
+      "learning_rate": 0.0008456147059306759,
+      "loss": 8.8138,
+      "step": 122
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 6.9375,
+      "learning_rate": 0.0008426854384035632,
+      "loss": 9.211,
+      "step": 123
+    },
+    {
+      "epoch": 3.0246913580246915,
+      "grad_norm": 5.4375,
+      "learning_rate": 0.0008397338200373195,
+      "loss": 7.7824,
+      "step": 124
+    },
+    {
+      "epoch": 3.049382716049383,
+      "grad_norm": 4.71875,
+      "learning_rate": 0.0008367600433442019,
+      "loss": 8.7634,
+      "step": 125
+    },
+    {
+      "epoch": 3.074074074074074,
+      "grad_norm": 3.25,
+      "learning_rate": 0.0008337643022816914,
+      "loss": 9.4349,
+      "step": 126
+    },
+    {
+      "epoch": 3.0987654320987654,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.0008307467922398432,
+      "loss": 9.0294,
+      "step": 127
+    },
+    {
+      "epoch": 3.123456790123457,
+      "grad_norm": 6.21875,
+      "learning_rate": 0.0008277077100285413,
+      "loss": 8.2013,
+      "step": 128
+    },
+    {
+      "epoch": 3.148148148148148,
+      "grad_norm": 6.0625,
+      "learning_rate": 0.0008246472538646634,
+      "loss": 7.1298,
+      "step": 129
+    },
+    {
+      "epoch": 3.1728395061728394,
+      "grad_norm": 6.5,
+      "learning_rate": 0.0008215656233591524,
+      "loss": 7.4791,
+      "step": 130
+    },
+    {
+      "epoch": 3.197530864197531,
+      "grad_norm": 5.75,
+      "learning_rate": 0.0008184630195039966,
+      "loss": 8.1978,
+      "step": 131
+    },
+    {
+      "epoch": 3.2222222222222223,
+      "grad_norm": 5.5,
+      "learning_rate": 0.000815339644659121,
+      "loss": 7.4613,
+      "step": 132
+    },
+    {
+      "epoch": 3.246913580246914,
+      "grad_norm": 4.40625,
+      "learning_rate": 0.0008121957025391892,
+      "loss": 8.4131,
+      "step": 133
+    },
+    {
+      "epoch": 3.271604938271605,
+      "grad_norm": 4.90625,
+      "learning_rate": 0.0008090313982003156,
+      "loss": 9.0265,
+      "step": 134
+    },
+    {
+      "epoch": 3.2962962962962963,
+      "grad_norm": 5.5,
+      "learning_rate": 0.0008058469380266922,
+      "loss": 7.4426,
+      "step": 135
+    },
+    {
+      "epoch": 3.3209876543209877,
+      "grad_norm": 4.625,
+      "learning_rate": 0.0008026425297171267,
+      "loss": 7.3086,
+      "step": 136
+    },
+    {
+      "epoch": 3.3456790123456788,
+      "grad_norm": 4.53125,
+      "learning_rate": 0.0007994183822714969,
+      "loss": 8.658,
+      "step": 137
+    },
+    {
+      "epoch": 3.3703703703703702,
+      "grad_norm": 5.25,
+      "learning_rate": 0.000796174705977118,
+      "loss": 8.8896,
+      "step": 138
+    },
+    {
+      "epoch": 3.3950617283950617,
+      "grad_norm": 4.96875,
+      "learning_rate": 0.000792911712395028,
+      "loss": 8.1086,
+      "step": 139
+    },
+    {
+      "epoch": 3.419753086419753,
+      "grad_norm": 4.96875,
+      "learning_rate": 0.000789629614346189,
+      "loss": 7.9749,
+      "step": 140
+    },
+    {
+      "epoch": 3.4444444444444446,
+      "grad_norm": 3.40625,
+      "learning_rate": 0.0007863286258976061,
+      "loss": 9.4685,
+      "step": 141
+    },
+    {
+      "epoch": 3.4691358024691357,
+      "grad_norm": 5.71875,
+      "learning_rate": 0.0007830089623483656,
+      "loss": 8.5942,
+      "step": 142
+    },
+    {
+      "epoch": 3.493827160493827,
+      "grad_norm": 4.46875,
+      "learning_rate": 0.0007796708402155925,
+      "loss": 8.0942,
+      "step": 143
+    },
+    {
+      "epoch": 3.5185185185185186,
+      "grad_norm": 4.3125,
+      "learning_rate": 0.0007763144772203292,
+      "loss": 8.7863,
+      "step": 144
+    },
+    {
+      "epoch": 3.5432098765432096,
+      "grad_norm": 5.71875,
+      "learning_rate": 0.0007729400922733345,
+      "loss": 8.1041,
+      "step": 145
+    },
+    {
+      "epoch": 3.567901234567901,
+      "grad_norm": 5.65625,
+      "learning_rate": 0.000769547905460806,
+      "loss": 8.2273,
+      "step": 146
+    },
+    {
+      "epoch": 3.5925925925925926,
+      "grad_norm": 4.78125,
+      "learning_rate": 0.0007661381380300254,
+      "loss": 8.113,
+      "step": 147
+    },
+    {
+      "epoch": 3.617283950617284,
+      "grad_norm": 4.5,
+      "learning_rate": 0.0007627110123749286,
+      "loss": 8.5411,
+      "step": 148
+    },
+    {
+      "epoch": 3.6419753086419755,
+      "grad_norm": 4.46875,
+      "learning_rate": 0.0007592667520215995,
+      "loss": 8.0202,
+      "step": 149
+    },
+    {
+      "epoch": 3.6666666666666665,
+      "grad_norm": 4.6875,
+      "learning_rate": 0.0007558055816136924,
+      "loss": 8.4248,
+      "step": 150
+    },
+    {
+      "epoch": 3.691358024691358,
+      "grad_norm": 5.125,
+      "learning_rate": 0.0007523277268977793,
+      "loss": 8.4345,
+      "step": 151
+    },
+    {
+      "epoch": 3.7160493827160495,
+      "grad_norm": 4.90625,
+      "learning_rate": 0.0007488334147086263,
+      "loss": 8.045,
+      "step": 152
+    },
+    {
+      "epoch": 3.7407407407407405,
+      "grad_norm": 3.21875,
+      "learning_rate": 0.0007453228729543988,
+      "loss": 10.2951,
+      "step": 153
+    },
+    {
+      "epoch": 3.765432098765432,
+      "grad_norm": 4.4375,
+      "learning_rate": 0.0007417963306017972,
+      "loss": 8.0272,
+      "step": 154
+    },
+    {
+      "epoch": 3.7901234567901234,
+      "grad_norm": 4.625,
+      "learning_rate": 0.0007382540176611223,
+      "loss": 8.7022,
+      "step": 155
+    },
+    {
+      "epoch": 3.814814814814815,
+      "grad_norm": 6.59375,
+      "learning_rate": 0.0007346961651712739,
+      "loss": 7.4862,
+      "step": 156
+    },
+    {
+      "epoch": 3.8395061728395063,
+      "grad_norm": 3.90625,
+      "learning_rate": 0.0007311230051846819,
+      "loss": 8.721,
+      "step": 157
+    },
+    {
+      "epoch": 3.8641975308641974,
+      "grad_norm": 4.375,
+      "learning_rate": 0.000727534770752171,
+      "loss": 7.8652,
+      "step": 158
+    },
+    {
+      "epoch": 3.888888888888889,
+      "grad_norm": 4.71875,
+      "learning_rate": 0.0007239316959077607,
+      "loss": 7.2138,
+      "step": 159
+    },
+    {
+      "epoch": 3.9135802469135803,
+      "grad_norm": 6.21875,
+      "learning_rate": 0.0007203140156534009,
+      "loss": 8.3145,
+      "step": 160
+    },
+    {
+      "epoch": 3.9382716049382713,
+      "grad_norm": 3.953125,
+      "learning_rate": 0.0007166819659436446,
+      "loss": 8.522,
+      "step": 161
+    },
+    {
+      "epoch": 3.962962962962963,
+      "grad_norm": 4.59375,
+      "learning_rate": 0.0007130357836702578,
+      "loss": 9.5293,
+      "step": 162
+    },
+    {
+      "epoch": 3.9876543209876543,
+      "grad_norm": 4.09375,
+      "learning_rate": 0.0007093757066467697,
+      "loss": 8.7476,
+      "step": 163
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 5.375,
+      "learning_rate": 0.0007057019735929611,
+      "loss": 8.7095,
+      "step": 164
+    },
+    {
+      "epoch": 4.0246913580246915,
+      "grad_norm": 5.1875,
+      "learning_rate": 0.0007020148241192947,
+      "loss": 7.6467,
+      "step": 165
+    },
+    {
+      "epoch": 4.049382716049383,
+      "grad_norm": 3.734375,
+      "learning_rate": 0.0006983144987112874,
+      "loss": 9.2119,
+      "step": 166
+    },
+    {
+      "epoch": 4.074074074074074,
+      "grad_norm": 3.8125,
+      "learning_rate": 0.0006946012387138247,
+      "loss": 8.5777,
+      "step": 167
+    },
+    {
+      "epoch": 4.098765432098766,
+      "grad_norm": 3.390625,
+      "learning_rate": 0.00069087528631542,
+      "loss": 9.8133,
+      "step": 168
+    },
+    {
+      "epoch": 4.1234567901234565,
+      "grad_norm": 4.21875,
+      "learning_rate": 0.0006871368845324181,
+      "loss": 8.4865,
+      "step": 169
+    },
+    {
+      "epoch": 4.148148148148148,
+      "grad_norm": 5.3125,
+      "learning_rate": 0.0006833862771931453,
+      "loss": 7.639,
+      "step": 170
+    },
+    {
+      "epoch": 4.172839506172839,
+      "grad_norm": 5.9375,
+      "learning_rate": 0.0006796237089220058,
+      "loss": 7.7151,
+      "step": 171
+    },
+    {
+      "epoch": 4.197530864197531,
+      "grad_norm": 5.125,
+      "learning_rate": 0.0006758494251235274,
+      "loss": 7.464,
+      "step": 172
+    },
+    {
+      "epoch": 4.222222222222222,
+      "grad_norm": 5.5,
+      "learning_rate": 0.0006720636719663549,
+      "loss": 7.2003,
+      "step": 173
+    },
+    {
+      "epoch": 4.246913580246914,
+      "grad_norm": 3.265625,
+      "learning_rate": 0.0006682666963671953,
+      "loss": 9.1617,
+      "step": 174
+    },
+    {
+      "epoch": 4.271604938271605,
+      "grad_norm": 4.6875,
+      "learning_rate": 0.0006644587459747113,
+      "loss": 7.5148,
+      "step": 175
+    },
+    {
+      "epoch": 4.296296296296296,
+      "grad_norm": 3.875,
+      "learning_rate": 0.0006606400691533715,
+      "loss": 8.4538,
+      "step": 176
+    },
+    {
+      "epoch": 4.320987654320987,
+      "grad_norm": 4.6875,
+      "learning_rate": 0.0006568109149672496,
+      "loss": 6.7651,
+      "step": 177
+    },
+    {
+      "epoch": 4.345679012345679,
+      "grad_norm": 3.21875,
+      "learning_rate": 0.0006529715331637804,
+      "loss": 9.1294,
+      "step": 178
+    },
+    {
+      "epoch": 4.37037037037037,
+      "grad_norm": 3.9375,
+      "learning_rate": 0.0006491221741574711,
+      "loss": 8.4035,
+      "step": 179
+    },
+    {
+      "epoch": 4.395061728395062,
+      "grad_norm": 4.34375,
+      "learning_rate": 0.0006452630890135672,
+      "loss": 8.3929,
+      "step": 180
+    },
+    {
+      "epoch": 4.419753086419753,
+      "grad_norm": 5.0,
+      "learning_rate": 0.0006413945294316794,
+      "loss": 7.6681,
+      "step": 181
+    },
+    {
+      "epoch": 4.444444444444445,
+      "grad_norm": 4.78125,
+      "learning_rate": 0.0006375167477293648,
+      "loss": 6.4986,
+      "step": 182
+    },
+    {
+      "epoch": 4.469135802469136,
+      "grad_norm": 4.625,
+      "learning_rate": 0.0006336299968256725,
+      "loss": 7.0889,
+      "step": 183
+    },
+    {
+      "epoch": 4.493827160493828,
+      "grad_norm": 4.375,
+      "learning_rate": 0.0006297345302246452,
+      "loss": 6.872,
+      "step": 184
+    },
+    {
+      "epoch": 4.518518518518518,
+      "grad_norm": 4.59375,
+      "learning_rate": 0.0006258306019987872,
+      "loss": 6.8268,
+      "step": 185
+    },
+    {
+      "epoch": 4.54320987654321,
+      "grad_norm": 4.53125,
+      "learning_rate": 0.000621918466772491,
+      "loss": 6.73,
+      "step": 186
+    },
+    {
+      "epoch": 4.567901234567901,
+      "grad_norm": 4.375,
+      "learning_rate": 0.0006179983797054322,
+      "loss": 6.6998,
+      "step": 187
+    },
+    {
+      "epoch": 4.592592592592593,
+      "grad_norm": 6.875,
+      "learning_rate": 0.0006140705964759251,
+      "loss": 8.6932,
+      "step": 188
+    },
+    {
+      "epoch": 4.617283950617284,
+      "grad_norm": 3.71875,
+      "learning_rate": 0.0006101353732642487,
+      "loss": 7.9493,
+      "step": 189
+    },
+    {
+      "epoch": 4.6419753086419755,
+      "grad_norm": 4.71875,
+      "learning_rate": 0.0006061929667359366,
+      "loss": 7.7456,
+      "step": 190
+    },
+    {
+      "epoch": 4.666666666666667,
+      "grad_norm": 3.890625,
+      "learning_rate": 0.0006022436340250371,
+      "loss": 8.4177,
+      "step": 191
+    },
+    {
+      "epoch": 4.6913580246913575,
+      "grad_norm": 4.0,
+      "learning_rate": 0.0005982876327173427,
+      "loss": 7.8851,
+      "step": 192
+    },
+    {
+      "epoch": 4.716049382716049,
+      "grad_norm": 4.0,
+      "learning_rate": 0.0005943252208335884,
+      "loss": 8.453,
+      "step": 193
+    },
+    {
+      "epoch": 4.7407407407407405,
+      "grad_norm": 3.59375,
+      "learning_rate": 0.0005903566568126245,
+      "loss": 9.9931,
+      "step": 194
+    },
+    {
+      "epoch": 4.765432098765432,
+      "grad_norm": 3.4375,
+      "learning_rate": 0.000586382199494559,
+      "loss": 9.2428,
+      "step": 195
+    },
+    {
+      "epoch": 4.790123456790123,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.0005824021081038767,
+      "loss": 7.5143,
+      "step": 196
+    },
+    {
+      "epoch": 4.814814814814815,
+      "grad_norm": 3.734375,
+      "learning_rate": 0.000578416642232531,
+      "loss": 9.3458,
+      "step": 197
+    },
+    {
+      "epoch": 4.839506172839506,
+      "grad_norm": 5.0,
+      "learning_rate": 0.0005744260618230132,
+      "loss": 6.0694,
+      "step": 198
+    },
+    {
+      "epoch": 4.864197530864198,
+      "grad_norm": 4.9375,
+      "learning_rate": 0.0005704306271513981,
+      "loss": 6.6276,
+      "step": 199
+    },
+    {
+      "epoch": 4.888888888888889,
+      "grad_norm": 4.3125,
+      "learning_rate": 0.0005664305988103679,
+      "loss": 7.6546,
+      "step": 200
+    },
+    {
+      "epoch": 4.91358024691358,
+      "grad_norm": 5.40625,
+      "learning_rate": 0.0005624262376922163,
+      "loss": 6.8278,
+      "step": 201
+    },
+    {
+      "epoch": 4.938271604938271,
+      "grad_norm": 5.59375,
+      "learning_rate": 0.0005584178049718315,
+      "loss": 6.1303,
+      "step": 202
+    },
+    {
+      "epoch": 4.962962962962963,
+      "grad_norm": 3.75,
+      "learning_rate": 0.0005544055620896629,
+      "loss": 9.2713,
+      "step": 203
+    },
+    {
+      "epoch": 4.987654320987654,
+      "grad_norm": 4.34375,
+      "learning_rate": 0.000550389770734668,
+      "loss": 7.6376,
+      "step": 204
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 4.9375,
+      "learning_rate": 0.0005463706928272452,
+      "loss": 8.5016,
+      "step": 205
+    },
+    {
+      "epoch": 5.0246913580246915,
+      "grad_norm": 3.765625,
+      "learning_rate": 0.0005423485905021508,
+      "loss": 7.1441,
+      "step": 206
+    },
+    {
+      "epoch": 5.049382716049383,
+      "grad_norm": 3.9375,
+      "learning_rate": 0.000538323726091401,
+      "loss": 6.4224,
+      "step": 207
+    },
+    {
+      "epoch": 5.074074074074074,
+      "grad_norm": 3.984375,
+      "learning_rate": 0.0005342963621071624,
+      "loss": 5.6542,
+      "step": 208
+    },
+    {
+      "epoch": 5.098765432098766,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.0005302667612246308,
+      "loss": 8.4893,
+      "step": 209
+    },
+    {
+      "epoch": 5.1234567901234565,
+      "grad_norm": 3.53125,
+      "learning_rate": 0.0005262351862648978,
+      "loss": 7.3327,
+      "step": 210
+    },
+    {
+      "epoch": 5.148148148148148,
+      "grad_norm": 4.40625,
+      "learning_rate": 0.00052220190017781,
+      "loss": 6.0517,
+      "step": 211
+    },
+    {
+      "epoch": 5.172839506172839,
+      "grad_norm": 5.84375,
+      "learning_rate": 0.0005181671660248179,
+      "loss": 7.6619,
+      "step": 212
+    },
+    {
+      "epoch": 5.197530864197531,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.0005141312469618183,
+      "loss": 8.5013,
+      "step": 213
+    },
+    {
+      "epoch": 5.222222222222222,
+      "grad_norm": 3.546875,
+      "learning_rate": 0.0005100944062219918,
+      "loss": 7.2512,
+      "step": 214
+    },
+    {
+      "epoch": 5.246913580246914,
+      "grad_norm": 3.75,
+      "learning_rate": 0.0005060569070986324,
+      "loss": 8.8302,
+      "step": 215
+    },
+    {
+      "epoch": 5.271604938271605,
+      "grad_norm": 4.40625,
+      "learning_rate": 0.0005020190129279759,
+      "loss": 8.1647,
+      "step": 216
+    },
+    {
+      "epoch": 5.296296296296296,
+      "grad_norm": 5.125,
+      "learning_rate": 0.0004979809870720242,
+      "loss": 4.7292,
+      "step": 217
+    },
+    {
+      "epoch": 5.320987654320987,
+      "grad_norm": 3.46875,
+      "learning_rate": 0.0004939430929013677,
+      "loss": 8.1476,
+      "step": 218
+    },
+    {
+      "epoch": 5.345679012345679,
+      "grad_norm": 3.953125,
+      "learning_rate": 0.0004899055937780083,
+      "loss": 8.0866,
+      "step": 219
+    },
+    {
+      "epoch": 5.37037037037037,
+      "grad_norm": 3.828125,
+      "learning_rate": 0.00048586875303818176,
+      "loss": 8.235,
+      "step": 220
+    },
+    {
+      "epoch": 5.395061728395062,
+      "grad_norm": 4.25,
+      "learning_rate": 0.00048183283397518233,
+      "loss": 8.2042,
+      "step": 221
+    },
+    {
+      "epoch": 5.419753086419753,
+      "grad_norm": 3.921875,
+      "learning_rate": 0.0004777980998221901,
+      "loss": 7.2502,
+      "step": 222
+    },
+    {
+      "epoch": 5.444444444444445,
+      "grad_norm": 3.671875,
+      "learning_rate": 0.0004737648137351022,
+      "loss": 8.9586,
+      "step": 223
+    },
+    {
+      "epoch": 5.469135802469136,
+      "grad_norm": 3.9375,
+      "learning_rate": 0.0004697332387753693,
+      "loss": 7.2468,
+      "step": 224
+    },
+    {
+      "epoch": 5.493827160493828,
+      "grad_norm": 3.53125,
+      "learning_rate": 0.0004657036378928376,
+      "loss": 7.3585,
+      "step": 225
+    },
+    {
+      "epoch": 5.518518518518518,
+      "grad_norm": 4.46875,
+      "learning_rate": 0.00046167627390859926,
+      "loss": 6.9372,
+      "step": 226
+    },
+    {
+      "epoch": 5.54320987654321,
+      "grad_norm": 3.546875,
+      "learning_rate": 0.00045765140949784926,
+      "loss": 8.0293,
+      "step": 227
+    },
+    {
+      "epoch": 5.567901234567901,
+      "grad_norm": 3.484375,
+      "learning_rate": 0.0004536293071727548,
+      "loss": 9.1196,
+      "step": 228
+    },
+    {
+      "epoch": 5.592592592592593,
+      "grad_norm": 4.09375,
+      "learning_rate": 0.000449610229265332,
+      "loss": 7.2503,
+      "step": 229
+    },
+    {
+      "epoch": 5.617283950617284,
+      "grad_norm": 4.75,
+      "learning_rate": 0.00044559443791033734,
+      "loss": 5.6725,
+      "step": 230
+    },
+    {
+      "epoch": 5.6419753086419755,
+      "grad_norm": 4.1875,
+      "learning_rate": 0.0004415821950281686,
+      "loss": 7.2648,
+      "step": 231
+    },
+    {
+      "epoch": 5.666666666666667,
+      "grad_norm": 4.21875,
+      "learning_rate": 0.00043757376230778387,
+      "loss": 7.2724,
+      "step": 232
+    },
+    {
+      "epoch": 5.6913580246913575,
+      "grad_norm": 3.828125,
+      "learning_rate": 0.00043356940118963205,
+      "loss": 8.7947,
+      "step": 233
+    },
+    {
+      "epoch": 5.716049382716049,
+      "grad_norm": 3.75,
+      "learning_rate": 0.00042956937284860196,
+      "loss": 8.0334,
+      "step": 234
+    },
+    {
+      "epoch": 5.7407407407407405,
+      "grad_norm": 4.8125,
+      "learning_rate": 0.0004255739381769868,
+      "loss": 7.9606,
+      "step": 235
+    },
+    {
+      "epoch": 5.765432098765432,
+      "grad_norm": 4.09375,
+      "learning_rate": 0.00042158335776746903,
+      "loss": 6.513,
+      "step": 236
+    },
+    {
+      "epoch": 5.790123456790123,
+      "grad_norm": 3.59375,
+      "learning_rate": 0.00041759789189612333,
+      "loss": 8.1531,
+      "step": 237
+    },
+    {
+      "epoch": 5.814814814814815,
+      "grad_norm": 4.78125,
+      "learning_rate": 0.00041361780050544104,
+      "loss": 7.1614,
+      "step": 238
+    },
+    {
+      "epoch": 5.839506172839506,
+      "grad_norm": 3.71875,
+      "learning_rate": 0.0004096433431873757,
+      "loss": 8.6922,
+      "step": 239
+    },
+    {
+      "epoch": 5.864197530864198,
+      "grad_norm": 3.40625,
+      "learning_rate": 0.00040567477916641164,
+      "loss": 8.9935,
+      "step": 240
+    },
+    {
+      "epoch": 5.888888888888889,
+      "grad_norm": 4.5,
+      "learning_rate": 0.0004017123672826574,
+      "loss": 8.3437,
+      "step": 241
+    },
+    {
+      "epoch": 5.91358024691358,
+      "grad_norm": 3.453125,
+      "learning_rate": 0.0003977563659749629,
+      "loss": 8.1862,
+      "step": 242
+    },
+    {
+      "epoch": 5.938271604938271,
+      "grad_norm": 3.84375,
+      "learning_rate": 0.0003938070332640636,
+      "loss": 7.3413,
+      "step": 243
+    },
+    {
+      "epoch": 5.962962962962963,
+      "grad_norm": 4.0,
+      "learning_rate": 0.00038986462673575153,
+      "loss": 6.4921,
+      "step": 244
+    },
+    {
+      "epoch": 5.987654320987654,
+      "grad_norm": 3.890625,
+      "learning_rate": 0.000385929403524075,
+      "loss": 8.07,
+      "step": 245
+    },
+    {
+      "epoch": 6.0,
+      "grad_norm": 5.84375,
+      "learning_rate": 0.00038200162029456797,
+      "loss": 6.4634,
+      "step": 246
+    },
+    {
+      "epoch": 6.0246913580246915,
+      "grad_norm": 3.78125,
+      "learning_rate": 0.0003780815332275089,
+      "loss": 6.1432,
+      "step": 247
+    },
+    {
+      "epoch": 6.049382716049383,
+      "grad_norm": 3.734375,
+      "learning_rate": 0.00037416939800121303,
+      "loss": 6.2527,
+      "step": 248
+    },
+    {
+      "epoch": 6.074074074074074,
+      "grad_norm": 3.453125,
+      "learning_rate": 0.00037026546977535484,
+      "loss": 8.8918,
+      "step": 249
+    },
+    {
+      "epoch": 6.098765432098766,
+      "grad_norm": 3.65625,
+      "learning_rate": 0.0003663700031743277,
+      "loss": 7.2045,
+      "step": 250
     },
     {
-      "epoch": 0.04740834386852086,
-      "grad_norm": 1.6875,
-      "learning_rate": 0.00027749999999999997,
-      "loss": 10.6128,
-      "step": 75
+      "epoch": 6.1234567901234565,
+      "grad_norm": 3.796875,
+      "learning_rate": 0.0003624832522706352,
+      "loss": 5.8116,
+      "step": 251
     },
     {
-      "epoch": 0.0632111251580278,
-      "grad_norm": 1.4375,
-      "learning_rate": 0.0002998815674207605,
-      "loss": 10.3251,
-      "step": 100
+      "epoch": 6.148148148148148,
+      "grad_norm": 3.34375,
+      "learning_rate": 0.0003586054705683208,
+      "loss": 8.6758,
+      "step": 252
     },
     {
-      "epoch": 0.07901390644753477,
-      "grad_norm": 1.296875,
-      "learning_rate": 0.0002993652248042369,
-      "loss": 10.0564,
-      "step": 125
+      "epoch": 6.172839506172839,
+      "grad_norm": 3.671875,
+      "learning_rate": 0.0003547369109864328,
+      "loss": 7.0771,
+      "step": 253
     },
     {
-      "epoch": 0.09481668773704172,
-      "grad_norm": 1.203125,
-      "learning_rate": 0.00029844057179344356,
-      "loss": 9.8386,
-      "step": 150
+      "epoch": 6.197530864197531,
+      "grad_norm": 3.71875,
+      "learning_rate": 0.00035087782584252905,
+      "loss": 6.3009,
+      "step": 254
     },
     {
-      "epoch": 0.11061946902654868,
-      "grad_norm": 1.1484375,
-      "learning_rate": 0.0002971101360546153,
-      "loss": 9.645,
-      "step": 175
+      "epoch": 6.222222222222222,
+      "grad_norm": 3.65625,
+      "learning_rate": 0.00034702846683621946,
+      "loss": 7.8653,
+      "step": 255
     },
     {
-      "epoch": 0.1264222503160556,
-      "grad_norm": 1.0859375,
-      "learning_rate": 0.00029537755451690867,
-      "loss": 9.4859,
-      "step": 200
+      "epoch": 6.246913580246914,
+      "grad_norm": 3.4375,
+      "learning_rate": 0.00034318908503275057,
+      "loss": 5.3887,
+      "step": 256
     },
     {
-      "epoch": 0.14222503160556257,
-      "grad_norm": 1.125,
-      "learning_rate": 0.00029324756343035576,
-      "loss": 9.3421,
-      "step": 225
+      "epoch": 6.271604938271605,
+      "grad_norm": 3.265625,
+      "learning_rate": 0.0003393599308466285,
+      "loss": 7.9237,
+      "step": 257
     },
     {
-      "epoch": 0.15802781289506954,
-      "grad_norm": 1.15625,
-      "learning_rate": 0.00029072598541867296,
-      "loss": 9.2273,
-      "step": 250
+      "epoch": 6.296296296296296,
+      "grad_norm": 3.40625,
+      "learning_rate": 0.0003355412540252887,
+      "loss": 8.8507,
+      "step": 258
+    },
+    {
+      "epoch": 6.320987654320987,
+      "grad_norm": 4.21875,
+      "learning_rate": 0.00033173330363280484,
+      "loss": 6.5917,
+      "step": 259
+    },
+    {
+      "epoch": 6.345679012345679,
+      "grad_norm": 3.21875,
+      "learning_rate": 0.0003279363280336449,
+      "loss": 8.7578,
+      "step": 260
+    },
+    {
+      "epoch": 6.37037037037037,
+      "grad_norm": 3.546875,
+      "learning_rate": 0.0003241505748764728,
+      "loss": 6.149,
+      "step": 261
+    },
+    {
+      "epoch": 6.395061728395062,
+      "grad_norm": 3.90625,
+      "learning_rate": 0.0003203762910779944,
+      "loss": 7.1028,
+      "step": 262
+    },
+    {
+      "epoch": 6.419753086419753,
+      "grad_norm": 3.53125,
+      "learning_rate": 0.00031661372280685483,
+      "loss": 6.1741,
+      "step": 263
+    },
+    {
+      "epoch": 6.444444444444445,
+      "grad_norm": 3.265625,
+      "learning_rate": 0.0003128631154675818,
+      "loss": 7.9666,
+      "step": 264
+    },
+    {
+      "epoch": 6.469135802469136,
+      "grad_norm": 3.328125,
+      "learning_rate": 0.00030912471368458005,
+      "loss": 8.0334,
+      "step": 265
+    },
+    {
+      "epoch": 6.493827160493828,
+      "grad_norm": 3.390625,
+      "learning_rate": 0.0003053987612861753,
+      "loss": 9.5293,
+      "step": 266
+    },
+    {
+      "epoch": 6.518518518518518,
+      "grad_norm": 4.09375,
+      "learning_rate": 0.00030168550128871263,
+      "loss": 6.234,
+      "step": 267
+    },
+    {
+      "epoch": 6.54320987654321,
+      "grad_norm": 3.390625,
+      "learning_rate": 0.00029798517588070536,
+      "loss": 7.8696,
+      "step": 268
+    },
+    {
+      "epoch": 6.567901234567901,
+      "grad_norm": 3.578125,
+      "learning_rate": 0.0002942980264070392,
+      "loss": 7.3372,
+      "step": 269
+    },
+    {
+      "epoch": 6.592592592592593,
+      "grad_norm": 3.515625,
+      "learning_rate": 0.0002906242933532304,
+      "loss": 7.0152,
+      "step": 270
+    },
+    {
+      "epoch": 6.617283950617284,
+      "grad_norm": 3.953125,
+      "learning_rate": 0.0002869642163297422,
+      "loss": 7.0756,
+      "step": 271
+    },
+    {
+      "epoch": 6.6419753086419755,
+      "grad_norm": 3.328125,
+      "learning_rate": 0.0002833180340563554,
+      "loss": 7.9405,
+      "step": 272
+    },
+    {
+      "epoch": 6.666666666666667,
+      "grad_norm": 3.5,
+      "learning_rate": 0.00027968598434659897,
+      "loss": 8.7255,
+      "step": 273
+    },
+    {
+      "epoch": 6.6913580246913575,
+      "grad_norm": 3.359375,
+      "learning_rate": 0.00027606830409223937,
+      "loss": 8.9699,
+      "step": 274
     },
     {
-      "epoch": 0.1738305941845765,
-      "grad_norm": 1.1796875,
-      "learning_rate": 0.0002878197135623177,
-      "loss": 9.1152,
+      "epoch": 6.716049382716049,
+      "grad_norm": 3.9375,
+      "learning_rate": 0.00027246522924782905,
+      "loss": 8.8373,
       "step": 275
     },
     {
-      "epoch": 0.18963337547408343,
-      "grad_norm": 1.1796875,
-      "learning_rate": 0.0002845366925553049,
-      "loss": 9.0362,
+      "epoch": 6.7407407407407405,
+      "grad_norm": 3.421875,
+      "learning_rate": 0.00026887699481531815,
+      "loss": 7.9459,
+      "step": 276
+    },
+    {
+      "epoch": 6.765432098765432,
+      "grad_norm": 4.03125,
+      "learning_rate": 0.0002653038348287261,
+      "loss": 6.8735,
+      "step": 277
+    },
+    {
+      "epoch": 6.790123456790123,
+      "grad_norm": 3.65625,
+      "learning_rate": 0.0002617459823388778,
+      "loss": 6.9867,
+      "step": 278
+    },
+    {
+      "epoch": 6.814814814814815,
+      "grad_norm": 3.796875,
+      "learning_rate": 0.0002582036693982028,
+      "loss": 7.8286,
+      "step": 279
+    },
+    {
+      "epoch": 6.839506172839506,
+      "grad_norm": 3.9375,
+      "learning_rate": 0.00025467712704560117,
+      "loss": 6.0116,
+      "step": 280
+    },
+    {
+      "epoch": 6.864197530864198,
+      "grad_norm": 4.0,
+      "learning_rate": 0.00025116658529137384,
+      "loss": 8.3119,
+      "step": 281
+    },
+    {
+      "epoch": 6.888888888888889,
+      "grad_norm": 3.609375,
+      "learning_rate": 0.00024767227310222074,
+      "loss": 7.1145,
+      "step": 282
+    },
+    {
+      "epoch": 6.91358024691358,
+      "grad_norm": 3.921875,
+      "learning_rate": 0.00024419441838630763,
+      "loss": 6.5803,
+      "step": 283
+    },
+    {
+      "epoch": 6.938271604938271,
+      "grad_norm": 4.0,
+      "learning_rate": 0.0002407332479784005,
+      "loss": 6.6936,
+      "step": 284
+    },
+    {
+      "epoch": 6.962962962962963,
+      "grad_norm": 3.890625,
+      "learning_rate": 0.00023728898762507146,
+      "loss": 5.3851,
+      "step": 285
+    },
+    {
+      "epoch": 6.987654320987654,
+      "grad_norm": 3.4375,
+      "learning_rate": 0.00023386186196997443,
+      "loss": 7.0678,
+      "step": 286
+    },
+    {
+      "epoch": 7.0,
+      "grad_norm": 4.25,
+      "learning_rate": 0.00023045209453919407,
+      "loss": 9.7393,
+      "step": 287
+    },
+    {
+      "epoch": 7.0246913580246915,
+      "grad_norm": 3.40625,
+      "learning_rate": 0.00022705990772666558,
+      "loss": 8.4961,
+      "step": 288
+    },
+    {
+      "epoch": 7.049382716049383,
+      "grad_norm": 3.34375,
+      "learning_rate": 0.00022368552277967081,
+      "loss": 6.9223,
+      "step": 289
+    },
+    {
+      "epoch": 7.074074074074074,
+      "grad_norm": 3.234375,
+      "learning_rate": 0.00022032915978440748,
+      "loss": 7.0141,
+      "step": 290
+    },
+    {
+      "epoch": 7.098765432098766,
+      "grad_norm": 3.125,
+      "learning_rate": 0.0002169910376516346,
+      "loss": 7.7613,
+      "step": 291
+    },
+    {
+      "epoch": 7.1234567901234565,
+      "grad_norm": 3.109375,
+      "learning_rate": 0.00021367137410239402,
+      "loss": 8.7319,
+      "step": 292
+    },
+    {
+      "epoch": 7.148148148148148,
+      "grad_norm": 3.578125,
+      "learning_rate": 0.00021037038565381106,
+      "loss": 8.2093,
+      "step": 293
+    },
+    {
+      "epoch": 7.172839506172839,
+      "grad_norm": 3.125,
+      "learning_rate": 0.000207088287604972,
+      "loss": 7.2345,
+      "step": 294
+    },
+    {
+      "epoch": 7.197530864197531,
+      "grad_norm": 3.640625,
+      "learning_rate": 0.00020382529402288209,
+      "loss": 6.2056,
+      "step": 295
+    },
+    {
+      "epoch": 7.222222222222222,
+      "grad_norm": 3.421875,
+      "learning_rate": 0.00020058161772850336,
+      "loss": 7.035,
+      "step": 296
+    },
+    {
+      "epoch": 7.246913580246914,
+      "grad_norm": 3.078125,
+      "learning_rate": 0.00019735747028287344,
+      "loss": 8.6249,
+      "step": 297
+    },
+    {
+      "epoch": 7.271604938271605,
+      "grad_norm": 3.859375,
+      "learning_rate": 0.00019415306197330802,
+      "loss": 7.2435,
+      "step": 298
+    },
+    {
+      "epoch": 7.296296296296296,
+      "grad_norm": 4.6875,
+      "learning_rate": 0.0001909686017996845,
+      "loss": 3.2569,
+      "step": 299
+    },
+    {
+      "epoch": 7.320987654320987,
+      "grad_norm": 3.734375,
+      "learning_rate": 0.0001878042974608109,
+      "loss": 7.8251,
       "step": 300
     },
     {
-      "epoch": 0.2054361567635904,
-      "grad_norm": 1.3515625,
-      "learning_rate": 0.0002808858969872923,
-      "loss": 8.9605,
+      "epoch": 7.345679012345679,
+      "grad_norm": 3.921875,
+      "learning_rate": 0.000184660355340879,
+      "loss": 5.5533,
+      "step": 301
+    },
+    {
+      "epoch": 7.37037037037037,
+      "grad_norm": 3.421875,
+      "learning_rate": 0.00018153698049600342,
+      "loss": 6.109,
+      "step": 302
+    },
+    {
+      "epoch": 7.395061728395062,
+      "grad_norm": 3.25,
+      "learning_rate": 0.00017843437664084754,
+      "loss": 7.0764,
+      "step": 303
+    },
+    {
+      "epoch": 7.419753086419753,
+      "grad_norm": 3.34375,
+      "learning_rate": 0.00017535274613533663,
+      "loss": 7.7965,
+      "step": 304
+    },
+    {
+      "epoch": 7.444444444444445,
+      "grad_norm": 3.03125,
+      "learning_rate": 0.0001722922899714588,
+      "loss": 8.0387,
+      "step": 305
+    },
+    {
+      "epoch": 7.469135802469136,
+      "grad_norm": 3.609375,
+      "learning_rate": 0.0001692532077601569,
+      "loss": 5.9805,
+      "step": 306
+    },
+    {
+      "epoch": 7.493827160493828,
+      "grad_norm": 3.328125,
+      "learning_rate": 0.00016623569771830854,
+      "loss": 6.0428,
+      "step": 307
+    },
+    {
+      "epoch": 7.518518518518518,
+      "grad_norm": 3.171875,
+      "learning_rate": 0.00016323995665579817,
+      "loss": 6.9212,
+      "step": 308
+    },
+    {
+      "epoch": 7.54320987654321,
+      "grad_norm": 3.484375,
+      "learning_rate": 0.00016026617996268066,
+      "loss": 8.4905,
+      "step": 309
+    },
+    {
+      "epoch": 7.567901234567901,
+      "grad_norm": 3.109375,
+      "learning_rate": 0.00015731456159643687,
+      "loss": 6.9324,
+      "step": 310
+    },
+    {
+      "epoch": 7.592592592592593,
+      "grad_norm": 3.375,
+      "learning_rate": 0.00015438529406932428,
+      "loss": 7.8715,
+      "step": 311
+    },
+    {
+      "epoch": 7.617283950617284,
+      "grad_norm": 3.265625,
+      "learning_rate": 0.00015147856843582004,
+      "loss": 6.2495,
+      "step": 312
+    },
+    {
+      "epoch": 7.6419753086419755,
+      "grad_norm": 3.515625,
+      "learning_rate": 0.00014859457428016015,
+      "loss": 6.6685,
+      "step": 313
+    },
+    {
+      "epoch": 7.666666666666667,
+      "grad_norm": 3.34375,
+      "learning_rate": 0.00014573349970397432,
+      "loss": 6.9555,
+      "step": 314
+    },
+    {
+      "epoch": 7.6913580246913575,
+      "grad_norm": 3.328125,
+      "learning_rate": 0.00014289553131401734,
+      "loss": 7.19,
+      "step": 315
+    },
+    {
+      "epoch": 7.716049382716049,
+      "grad_norm": 3.328125,
+      "learning_rate": 0.00014008085420999834,
+      "loss": 6.6798,
+      "step": 316
+    },
+    {
+      "epoch": 7.7407407407407405,
+      "grad_norm": 3.09375,
+      "learning_rate": 0.00013728965197250782,
+      "loss": 9.4656,
+      "step": 317
+    },
+    {
+      "epoch": 7.765432098765432,
+      "grad_norm": 3.328125,
+      "learning_rate": 0.00013452210665104442,
+      "loss": 7.7878,
+      "step": 318
+    },
+    {
+      "epoch": 7.790123456790123,
+      "grad_norm": 3.484375,
+      "learning_rate": 0.00013177839875214042,
+      "loss": 6.3782,
+      "step": 319
+    },
+    {
+      "epoch": 7.814814814814815,
+      "grad_norm": 3.109375,
+      "learning_rate": 0.00012905870722758973,
+      "loss": 7.0306,
+      "step": 320
+    },
+    {
+      "epoch": 7.839506172839506,
+      "grad_norm": 3.046875,
+      "learning_rate": 0.00012636320946277535,
+      "loss": 8.773,
+      "step": 321
+    },
+    {
+      "epoch": 7.864197530864198,
+      "grad_norm": 3.15625,
+      "learning_rate": 0.0001236920812651003,
+      "loss": 7.7132,
+      "step": 322
+    },
+    {
+      "epoch": 7.888888888888889,
+      "grad_norm": 3.078125,
+      "learning_rate": 0.00012104549685252026,
+      "loss": 7.8435,
+      "step": 323
+    },
+    {
+      "epoch": 7.91358024691358,
+      "grad_norm": 3.125,
+      "learning_rate": 0.00011842362884218167,
+      "loss": 8.5793,
+      "step": 324
+    },
+    {
+      "epoch": 7.938271604938271,
+      "grad_norm": 3.453125,
+      "learning_rate": 0.00011582664823916261,
+      "loss": 6.9349,
       "step": 325
     },
     {
-      "epoch": 0.22123893805309736,
-      "grad_norm": 1.375,
-      "learning_rate": 0.0002768773068103051,
-      "loss": 8.9049,
+      "epoch": 7.962962962962963,
+      "grad_norm": 3.359375,
+      "learning_rate": 0.00011325472442531948,
+      "loss": 7.377,
+      "step": 326
+    },
+    {
+      "epoch": 7.987654320987654,
+      "grad_norm": 3.046875,
+      "learning_rate": 0.00011070802514823913,
+      "loss": 6.9828,
+      "step": 327
+    },
+    {
+      "epoch": 8.0,
+      "grad_norm": 4.8125,
+      "learning_rate": 0.00010818671651029855,
+      "loss": 6.7613,
+      "step": 328
+    },
+    {
+      "epoch": 8.024691358024691,
+      "grad_norm": 3.046875,
+      "learning_rate": 0.00010569096295783088,
+      "loss": 6.0772,
+      "step": 329
+    },
+    {
+      "epoch": 8.049382716049383,
+      "grad_norm": 3.65625,
+      "learning_rate": 0.00010322092727039989,
+      "loss": 3.4719,
+      "step": 330
+    },
+    {
+      "epoch": 8.074074074074074,
+      "grad_norm": 2.859375,
+      "learning_rate": 0.00010077677055018303,
+      "loss": 7.9138,
+      "step": 331
+    },
+    {
+      "epoch": 8.098765432098766,
+      "grad_norm": 3.21875,
+      "learning_rate": 9.83586522114639e-05,
+      "loss": 7.2392,
+      "step": 332
+    },
+    {
+      "epoch": 8.123456790123457,
+      "grad_norm": 3.078125,
+      "learning_rate": 9.596672997023481e-05,
+      "loss": 7.6866,
+      "step": 333
+    },
+    {
+      "epoch": 8.148148148148149,
+      "grad_norm": 3.046875,
+      "learning_rate": 9.360115983391032e-05,
+      "loss": 7.9637,
+      "step": 334
+    },
+    {
+      "epoch": 8.17283950617284,
+      "grad_norm": 3.15625,
+      "learning_rate": 9.126209609115193e-05,
+      "loss": 5.9937,
+      "step": 335
+    },
+    {
+      "epoch": 8.197530864197532,
+      "grad_norm": 3.34375,
+      "learning_rate": 8.894969130180453e-05,
+      "loss": 6.9129,
+      "step": 336
+    },
+    {
+      "epoch": 8.222222222222221,
+      "grad_norm": 3.25,
+      "learning_rate": 8.666409628694693e-05,
+      "loss": 9.3835,
+      "step": 337
+    },
+    {
+      "epoch": 8.246913580246913,
+      "grad_norm": 3.328125,
+      "learning_rate": 8.440546011905426e-05,
+      "loss": 7.1956,
+      "step": 338
+    },
+    {
+      "epoch": 8.271604938271604,
+      "grad_norm": 3.203125,
+      "learning_rate": 8.217393011227537e-05,
+      "loss": 9.5874,
+      "step": 339
+    },
+    {
+      "epoch": 8.296296296296296,
+      "grad_norm": 2.890625,
+      "learning_rate": 7.996965181282417e-05,
+      "loss": 8.7425,
+      "step": 340
+    },
+    {
+      "epoch": 8.320987654320987,
+      "grad_norm": 3.515625,
+      "learning_rate": 7.779276898948734e-05,
+      "loss": 5.3657,
+      "step": 341
+    },
+    {
+      "epoch": 8.345679012345679,
+      "grad_norm": 3.109375,
+      "learning_rate": 7.564342362424714e-05,
+      "loss": 6.8744,
+      "step": 342
+    },
+    {
+      "epoch": 8.37037037037037,
+      "grad_norm": 3.421875,
+      "learning_rate": 7.352175590302063e-05,
+      "loss": 6.373,
+      "step": 343
+    },
+    {
+      "epoch": 8.395061728395062,
+      "grad_norm": 3.21875,
+      "learning_rate": 7.14279042065169e-05,
+      "loss": 6.7817,
+      "step": 344
+    },
+    {
+      "epoch": 8.419753086419753,
+      "grad_norm": 3.203125,
+      "learning_rate": 6.93620051012111e-05,
+      "loss": 6.8337,
+      "step": 345
+    },
+    {
+      "epoch": 8.444444444444445,
+      "grad_norm": 2.953125,
+      "learning_rate": 6.732419333043755e-05,
+      "loss": 8.7099,
+      "step": 346
+    },
+    {
+      "epoch": 8.469135802469136,
+      "grad_norm": 2.984375,
+      "learning_rate": 6.531460180560111e-05,
+      "loss": 6.9555,
+      "step": 347
+    },
+    {
+      "epoch": 8.493827160493828,
+      "grad_norm": 3.8125,
+      "learning_rate": 6.333336159750869e-05,
+      "loss": 7.0301,
+      "step": 348
+    },
+    {
+      "epoch": 8.518518518518519,
+      "grad_norm": 3.46875,
+      "learning_rate": 6.138060192782002e-05,
+      "loss": 6.0988,
+      "step": 349
+    },
+    {
+      "epoch": 8.54320987654321,
+      "grad_norm": 3.0625,
+      "learning_rate": 5.945645016062001e-05,
+      "loss": 7.8126,
       "step": 350
     },
     {
-      "epoch": 0.2370417193426043,
-      "grad_norm": 1.5,
-      "learning_rate": 0.0002725218800571637,
-      "loss": 8.8709,
+      "epoch": 8.567901234567902,
+      "grad_norm": 3.046875,
+      "learning_rate": 5.7561031794111316e-05,
+      "loss": 6.7753,
+      "step": 351
+    },
+    {
+      "epoch": 8.592592592592592,
+      "grad_norm": 3.359375,
+      "learning_rate": 5.569447045242931e-05,
+      "loss": 6.4113,
+      "step": 352
+    },
+    {
+      "epoch": 8.617283950617283,
+      "grad_norm": 3.125,
+      "learning_rate": 5.385688787757864e-05,
+      "loss": 6.0227,
+      "step": 353
+    },
+    {
+      "epoch": 8.641975308641975,
+      "grad_norm": 3.171875,
+      "learning_rate": 5.204840392149329e-05,
+      "loss": 6.8429,
+      "step": 354
+    },
+    {
+      "epoch": 8.666666666666666,
+      "grad_norm": 3.109375,
+      "learning_rate": 5.02691365382193e-05,
+      "loss": 7.7754,
+      "step": 355
+    },
+    {
+      "epoch": 8.691358024691358,
+      "grad_norm": 3.140625,
+      "learning_rate": 4.851920177622143e-05,
+      "loss": 7.8351,
+      "step": 356
+    },
+    {
+      "epoch": 8.716049382716049,
+      "grad_norm": 2.96875,
+      "learning_rate": 4.6798713770814624e-05,
+      "loss": 8.5431,
+      "step": 357
+    },
+    {
+      "epoch": 8.74074074074074,
+      "grad_norm": 3.203125,
+      "learning_rate": 4.5107784736718946e-05,
+      "loss": 6.7501,
+      "step": 358
+    },
+    {
+      "epoch": 8.765432098765432,
+      "grad_norm": 3.265625,
+      "learning_rate": 4.344652496074142e-05,
+      "loss": 8.5514,
+      "step": 359
+    },
+    {
+      "epoch": 8.790123456790123,
+      "grad_norm": 3.1875,
+      "learning_rate": 4.181504279458287e-05,
+      "loss": 8.5261,
+      "step": 360
+    },
+    {
+      "epoch": 8.814814814814815,
+      "grad_norm": 3.484375,
+      "learning_rate": 4.021344464777038e-05,
+      "loss": 5.1709,
+      "step": 361
+    },
+    {
+      "epoch": 8.839506172839506,
+      "grad_norm": 3.234375,
+      "learning_rate": 3.864183498071699e-05,
+      "loss": 6.7079,
+      "step": 362
+    },
+    {
+      "epoch": 8.864197530864198,
+      "grad_norm": 3.140625,
+      "learning_rate": 3.710031629790928e-05,
+      "loss": 6.9687,
+      "step": 363
+    },
+    {
+      "epoch": 8.88888888888889,
+      "grad_norm": 3.28125,
+      "learning_rate": 3.558898914122088e-05,
+      "loss": 6.912,
+      "step": 364
+    },
+    {
+      "epoch": 8.91358024691358,
+      "grad_norm": 3.0,
+      "learning_rate": 3.4107952083355545e-05,
+      "loss": 9.5947,
+      "step": 365
+    },
+    {
+      "epoch": 8.938271604938272,
+      "grad_norm": 2.984375,
+      "learning_rate": 3.265730172141757e-05,
+      "loss": 9.588,
+      "step": 366
+    },
+    {
+      "epoch": 8.962962962962964,
+      "grad_norm": 4.28125,
+      "learning_rate": 3.1237132670611455e-05,
+      "loss": 5.5477,
+      "step": 367
+    },
+    {
+      "epoch": 8.987654320987655,
+      "grad_norm": 3.015625,
+      "learning_rate": 2.9847537558071126e-05,
+      "loss": 6.8754,
+      "step": 368
+    },
+    {
+      "epoch": 9.0,
+      "grad_norm": 4.46875,
+      "learning_rate": 2.848860701681849e-05,
+      "loss": 5.978,
+      "step": 369
+    },
+    {
+      "epoch": 9.024691358024691,
+      "grad_norm": 3.0625,
+      "learning_rate": 2.7160429679851938e-05,
+      "loss": 7.7472,
+      "step": 370
+    },
+    {
+      "epoch": 9.049382716049383,
+      "grad_norm": 3.1875,
+      "learning_rate": 2.5863092174365555e-05,
+      "loss": 7.6777,
+      "step": 371
+    },
+    {
+      "epoch": 9.074074074074074,
+      "grad_norm": 3.84375,
+      "learning_rate": 2.4596679116099086e-05,
+      "loss": 6.3852,
+      "step": 372
+    },
+    {
+      "epoch": 9.098765432098766,
+      "grad_norm": 3.0,
+      "learning_rate": 2.336127310381919e-05,
+      "loss": 7.4578,
+      "step": 373
+    },
+    {
+      "epoch": 9.123456790123457,
+      "grad_norm": 3.328125,
+      "learning_rate": 2.2156954713931974e-05,
+      "loss": 4.2617,
+      "step": 374
+    },
+    {
+      "epoch": 9.148148148148149,
+      "grad_norm": 3.0,
+      "learning_rate": 2.0983802495227522e-05,
+      "loss": 8.7434,
       "step": 375
     },
     {
-      "epoch": 0.2528445006321112,
-      "grad_norm": 1.5546875,
-      "learning_rate": 0.0002678315228861933,
-      "loss": 8.8332,
-      "step": 400
+      "epoch": 9.17283950617284,
+      "grad_norm": 3.28125,
+      "learning_rate": 1.984189296375699e-05,
+      "loss": 7.671,
+      "step": 376
     },
     {
-      "epoch": 0.2686472819216182,
-      "grad_norm": 1.640625,
-      "learning_rate": 0.00026281905703410164,
-      "loss": 8.7912,
-      "step": 425
+      "epoch": 9.197530864197532,
+      "grad_norm": 2.9375,
+      "learning_rate": 1.873130059784184e-05,
+      "loss": 7.8471,
+      "step": 377
     },
     {
-      "epoch": 0.28445006321112515,
-      "grad_norm": 1.75,
-      "learning_rate": 0.0002574981847659977,
-      "loss": 8.7849,
-      "step": 450
+      "epoch": 9.222222222222221,
+      "grad_norm": 3.03125,
+      "learning_rate": 1.7652097833216195e-05,
+      "loss": 6.0614,
+      "step": 378
     },
     {
-      "epoch": 0.3002528445006321,
-      "grad_norm": 1.75,
-      "learning_rate": 0.0002518834514183646,
-      "loss": 8.7603,
-      "step": 475
+      "epoch": 9.246913580246913,
+      "grad_norm": 3.21875,
+      "learning_rate": 1.660435505830249e-05,
+      "loss": 6.3376,
+      "step": 379
     },
     {
-      "epoch": 0.31605562579013907,
-      "grad_norm": 1.828125,
-      "learning_rate": 0.00024599020563738084,
-      "loss": 8.7401,
-      "step": 500
+      "epoch": 9.271604938271604,
+      "grad_norm": 3.15625,
+      "learning_rate": 1.5588140609620326e-05,
+      "loss": 7.9326,
+      "step": 380
     },
     {
-      "epoch": 0.33185840707964603,
-      "grad_norm": 1.875,
-      "learning_rate": 0.00023983455742128404,
-      "loss": 8.739,
-      "step": 525
+      "epoch": 9.296296296296296,
+      "grad_norm": 3.265625,
+      "learning_rate": 1.4603520767329681e-05,
+      "loss": 5.339,
+      "step": 381
     },
     {
-      "epoch": 0.347661188369153,
-      "grad_norm": 1.984375,
-      "learning_rate": 0.00023343333408147417,
-      "loss": 8.746,
-      "step": 550
+      "epoch": 9.320987654320987,
+      "grad_norm": 3.078125,
+      "learning_rate": 1.3650559750907732e-05,
+      "loss": 5.9893,
+      "step": 382
     },
     {
-      "epoch": 0.3634639696586599,
-      "grad_norm": 2.046875,
-      "learning_rate": 0.00022680403424274226,
-      "loss": 8.7452,
-      "step": 575
+      "epoch": 9.345679012345679,
+      "grad_norm": 3.265625,
+      "learning_rate": 1.2729319714960552e-05,
+      "loss": 7.7263,
+      "step": 383
     },
     {
-      "epoch": 0.37926675094816686,
-      "grad_norm": 2.078125,
-      "learning_rate": 0.00021996478000837203,
-      "loss": 8.7349,
-      "step": 600
+      "epoch": 9.37037037037037,
+      "grad_norm": 3.171875,
+      "learning_rate": 1.1839860745168628e-05,
+      "loss": 7.8087,
+      "step": 384
     },
     {
-      "epoch": 0.3950695322376738,
-      "grad_norm": 2.109375,
-      "learning_rate": 0.00021293426742087664,
-      "loss": 8.7307,
-      "step": 625
+      "epoch": 9.395061728395062,
+      "grad_norm": 3.4375,
+      "learning_rate": 1.0982240854368775e-05,
+      "loss": 5.376,
+      "step": 385
     },
     {
-      "epoch": 0.4108723135271808,
-      "grad_norm": 2.125,
-      "learning_rate": 0.00020573171535379444,
-      "loss": 8.7263,
-      "step": 650
+      "epoch": 9.419753086419753,
+      "grad_norm": 3.09375,
+      "learning_rate": 1.0156515978769597e-05,
+      "loss": 6.8875,
+      "step": 386
     },
     {
-      "epoch": 0.42667509481668775,
-      "grad_norm": 2.25,
-      "learning_rate": 0.0001983768129742532,
-      "loss": 8.7183,
-      "step": 675
+      "epoch": 9.444444444444445,
+      "grad_norm": 3.359375,
+      "learning_rate": 9.362739974303758e-06,
+      "loss": 6.0189,
+      "step": 387
     },
     {
-      "epoch": 0.4424778761061947,
-      "grad_norm": 2.234375,
-      "learning_rate": 0.00019088966591992288,
-      "loss": 8.7314,
-      "step": 700
+      "epoch": 9.469135802469136,
+      "grad_norm": 3.09375,
+      "learning_rate": 8.600964613114937e-06,
+      "loss": 6.8963,
+      "step": 388
     },
     {
-      "epoch": 0.4582806573957016,
-      "grad_norm": 2.078125,
-      "learning_rate": 0.0001832907413374885,
-      "loss": 8.7217,
-      "step": 725
+      "epoch": 9.493827160493828,
+      "grad_norm": 3.203125,
+      "learning_rate": 7.871239580181489e-06,
+      "loss": 8.6681,
+      "step": 389
     },
     {
-      "epoch": 0.4740834386852086,
-      "grad_norm": 2.15625,
-      "learning_rate": 0.00017560081193288848,
-      "loss": 8.7156,
-      "step": 750
+      "epoch": 9.518518518518519,
+      "grad_norm": 2.875,
+      "learning_rate": 7.173612470075419e-06,
+      "loss": 7.8249,
+      "step": 390
     },
     {
-      "epoch": 0.48988621997471554,
-      "grad_norm": 2.140625,
-      "learning_rate": 0.0001678408991862646,
-      "loss": 8.7147,
-      "step": 775
+      "epoch": 9.54320987654321,
+      "grad_norm": 3.265625,
+      "learning_rate": 6.508128783858647e-06,
+      "loss": 6.0478,
+      "step": 391
     },
     {
-      "epoch": 0.5056890012642224,
-      "grad_norm": 2.265625,
-      "learning_rate": 0.00016003221588685362,
-      "loss": 8.713,
-      "step": 800
+      "epoch": 9.567901234567902,
+      "grad_norm": 3.109375,
+      "learning_rate": 5.874831926114932e-06,
+      "loss": 8.4802,
+      "step": 392
+    },
+    {
+      "epoch": 9.592592592592592,
+      "grad_norm": 3.515625,
+      "learning_rate": 5.2737632021189196e-06,
+      "loss": 6.0807,
+      "step": 393
+    },
+    {
+      "epoch": 9.617283950617283,
+      "grad_norm": 2.96875,
+      "learning_rate": 4.704961815142406e-06,
+      "loss": 9.5407,
+      "step": 394
+    },
+    {
+      "epoch": 9.641975308641975,
+      "grad_norm": 3.546875,
+      "learning_rate": 4.168464863897159e-06,
+      "loss": 6.9195,
+      "step": 395
+    },
+    {
+      "epoch": 9.666666666666666,
+      "grad_norm": 3.328125,
+      "learning_rate": 3.664307340115247e-06,
+      "loss": 6.8413,
+      "step": 396
+    },
+    {
+      "epoch": 9.691358024691358,
+      "grad_norm": 3.234375,
+      "learning_rate": 3.1925221262668615e-06,
+      "loss": 6.9524,
+      "step": 397
+    },
+    {
+      "epoch": 9.716049382716049,
+      "grad_norm": 3.078125,
+      "learning_rate": 2.753139993415532e-06,
+      "loss": 9.5263,
+      "step": 398
+    },
+    {
+      "epoch": 9.74074074074074,
+      "grad_norm": 3.15625,
+      "learning_rate": 2.3461895992114014e-06,
+      "loss": 6.9148,
+      "step": 399
+    },
+    {
+      "epoch": 9.765432098765432,
+      "grad_norm": 3.1875,
+      "learning_rate": 1.971697486021773e-06,
+      "loss": 7.6293,
+      "step": 400
+    },
+    {
+      "epoch": 9.790123456790123,
+      "grad_norm": 3.171875,
+      "learning_rate": 1.6296880792001646e-06,
+      "loss": 7.6517,
+      "step": 401
     },
     {
-      "epoch": 0.5214917825537294,
-      "grad_norm": 2.171875,
-      "learning_rate": 0.00015219610814490908,
-      "loss": 8.7145,
-      "step": 825
+      "epoch": 9.814814814814815,
+      "grad_norm": 3.828125,
+      "learning_rate": 1.3201836854931926e-06,
+      "loss": 7.0111,
+      "step": 402
     },
     {
-      "epoch": 0.5372945638432364,
-      "grad_norm": 2.203125,
-      "learning_rate": 0.00014435399703917158,
-      "loss": 8.7091,
-      "step": 850
+      "epoch": 9.839506172839506,
+      "grad_norm": 2.984375,
+      "learning_rate": 1.0432044915855699e-06,
+      "loss": 8.7232,
+      "step": 403
     },
     {
-      "epoch": 0.5530973451327433,
-      "grad_norm": 2.171875,
-      "learning_rate": 0.00013652732005940295,
-      "loss": 8.711,
-      "step": 875
+      "epoch": 9.864197530864198,
+      "grad_norm": 2.9375,
+      "learning_rate": 7.987685627836027e-07,
+      "loss": 8.6614,
+      "step": 404
     },
     {
-      "epoch": 0.5689001264222503,
-      "grad_norm": 2.203125,
-      "learning_rate": 0.0001287374725040577,
-      "loss": 8.7163,
-      "step": 900
+      "epoch": 9.88888888888889,
+      "grad_norm": 3.0625,
+      "learning_rate": 5.868918418368563e-07,
+      "loss": 7.741,
+      "step": 405
     },
     {
-      "epoch": 0.5847029077117573,
-      "grad_norm": 2.265625,
-      "learning_rate": 0.00012100574899329124,
-      "loss": 8.7016,
-      "step": 925
+      "epoch": 9.91358024691358,
+      "grad_norm": 3.265625,
+      "learning_rate": 4.07588147898319e-07,
+      "loss": 4.983,
+      "step": 406
     },
     {
-      "epoch": 0.6005056890012642,
-      "grad_norm": 2.171875,
-      "learning_rate": 0.00011335328525718533,
-      "loss": 8.7065,
-      "step": 950
+      "epoch": 9.938271604938272,
+      "grad_norm": 2.875,
+      "learning_rate": 2.6086917562317956e-07,
+      "loss": 7.8299,
+      "step": 407
     },
     {
-      "epoch": 0.6163084702907712,
-      "grad_norm": 2.1875,
-      "learning_rate": 0.00010580100035832173,
-      "loss": 8.7081,
-      "step": 975
+      "epoch": 9.962962962962964,
+      "grad_norm": 3.21875,
+      "learning_rate": 1.4674449440582606e-07,
+      "loss": 7.1008,
+      "step": 408
     },
     {
-      "epoch": 0.6321112515802781,
-      "grad_norm": 2.28125,
-      "learning_rate": 9.836953950664638e-05,
-      "loss": 8.6993,
-      "step": 1000
+      "epoch": 9.987654320987655,
+      "grad_norm": 3.28125,
+      "learning_rate": 6.522154775606693e-08,
+      "loss": 6.8156,
+      "step": 409
     },
     {
-      "epoch": 0.6321112515802781,
-      "eval_loss": 8.754563331604004,
-      "eval_runtime": 6.2727,
-      "eval_samples_per_second": 32.362,
-      "eval_steps_per_second": 1.116,
-      "step": 1000
+      "epoch": 10.0,
+      "grad_norm": 3.96875,
+      "learning_rate": 1.6305652813353166e-08,
+      "loss": 5.9423,
+      "step": 410
     }
   ],
-  "logging_steps": 25,
-  "max_steps": 1582,
+  "logging_steps": 1,
+  "max_steps": 410,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 1,
+  "num_train_epochs": 10,
   "save_steps": 1000,
   "stateful_callbacks": {
     "TrainerControl": {
@@ -310,13 +2892,13 @@
         "should_evaluate": false,
         "should_log": false,
         "should_save": true,
-        "should_training_stop": false
+        "should_training_stop": true
       },
       "attributes": {}
     }
   },
-  "total_flos": 1.0188945762340045e+18,
-  "train_batch_size": 16,
+  "total_flos": 2.579076841734144e+16,
+  "train_batch_size": 4,
   "trial_name": null,
   "trial_params": null
 }

last-checkpoint/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4e910fd348e698394defcd79f420663259360f2de743e0565ce5dc8f50a65876
+oid sha256:21dfcc140b20b5859602309c4b2a7d5ca7cc6bb8c090092b61e00047f79e7a4f
 size 5777