diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,9 +1,9 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 3.483309143686502,
+  "epoch": 4.644412191582003,
   "eval_steps": 300,
-  "global_step": 3600,
+  "global_step": 4800,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
@@ -25303,6 +25303,8438 @@
       "eval_samples_per_second": 4.802,
       "eval_steps_per_second": 0.601,
       "step": 3600
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 1.398797631263733,
+      "learning_rate": 1.000417014178482e-05,
+      "loss": 0.7477,
+      "step": 3601
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 1.5675469636917114,
+      "learning_rate": 1e-05,
+      "loss": 0.6492,
+      "step": 3602
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 1.36086905002594,
+      "learning_rate": 9.995829858215181e-06,
+      "loss": 0.8345,
+      "step": 3603
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 1.302552342414856,
+      "learning_rate": 9.99165971643036e-06,
+      "loss": 0.6947,
+      "step": 3604
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 1.0472469329833984,
+      "learning_rate": 9.987489574645538e-06,
+      "loss": 0.7053,
+      "step": 3605
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 1.2574175596237183,
+      "learning_rate": 9.983319432860718e-06,
+      "loss": 0.7126,
+      "step": 3606
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 1.4842021465301514,
+      "learning_rate": 9.979149291075898e-06,
+      "loss": 0.7707,
+      "step": 3607
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 1.20749831199646,
+      "learning_rate": 9.974979149291076e-06,
+      "loss": 0.6888,
+      "step": 3608
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 1.147926926612854,
+      "learning_rate": 9.970809007506256e-06,
+      "loss": 0.5688,
+      "step": 3609
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 1.2449201345443726,
+      "learning_rate": 9.966638865721436e-06,
+      "loss": 0.7079,
+      "step": 3610
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 1.2286193370819092,
+      "learning_rate": 9.962468723936614e-06,
+      "loss": 0.7146,
+      "step": 3611
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 1.2378441095352173,
+      "learning_rate": 9.958298582151794e-06,
+      "loss": 0.6854,
+      "step": 3612
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 1.1023712158203125,
+      "learning_rate": 9.954128440366973e-06,
+      "loss": 0.6053,
+      "step": 3613
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 1.265261173248291,
+      "learning_rate": 9.949958298582153e-06,
+      "loss": 0.9018,
+      "step": 3614
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 1.1046671867370605,
+      "learning_rate": 9.945788156797331e-06,
+      "loss": 0.6952,
+      "step": 3615
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 1.394481897354126,
+      "learning_rate": 9.941618015012511e-06,
+      "loss": 0.7404,
+      "step": 3616
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 1.2353272438049316,
+      "learning_rate": 9.93744787322769e-06,
+      "loss": 0.7205,
+      "step": 3617
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 1.3304924964904785,
+      "learning_rate": 9.93327773144287e-06,
+      "loss": 0.7631,
+      "step": 3618
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 1.1558955907821655,
+      "learning_rate": 9.92910758965805e-06,
+      "loss": 0.7562,
+      "step": 3619
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 1.4406038522720337,
+      "learning_rate": 9.924937447873228e-06,
+      "loss": 0.8024,
+      "step": 3620
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 1.1018840074539185,
+      "learning_rate": 9.920767306088408e-06,
+      "loss": 0.6209,
+      "step": 3621
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 1.299937129020691,
+      "learning_rate": 9.916597164303586e-06,
+      "loss": 0.775,
+      "step": 3622
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 1.8240668773651123,
+      "learning_rate": 9.912427022518766e-06,
+      "loss": 0.7077,
+      "step": 3623
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 1.3343193531036377,
+      "learning_rate": 9.908256880733946e-06,
+      "loss": 0.7171,
+      "step": 3624
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 1.530676007270813,
+      "learning_rate": 9.904086738949125e-06,
+      "loss": 0.6222,
+      "step": 3625
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 1.612051010131836,
+      "learning_rate": 9.899916597164303e-06,
+      "loss": 0.6966,
+      "step": 3626
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 1.2645072937011719,
+      "learning_rate": 9.895746455379483e-06,
+      "loss": 0.6714,
+      "step": 3627
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 0.970465898513794,
+      "learning_rate": 9.891576313594663e-06,
+      "loss": 0.7006,
+      "step": 3628
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 1.1989374160766602,
+      "learning_rate": 9.887406171809841e-06,
+      "loss": 0.6727,
+      "step": 3629
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 1.3782232999801636,
+      "learning_rate": 9.883236030025021e-06,
+      "loss": 0.922,
+      "step": 3630
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 1.1525388956069946,
+      "learning_rate": 9.879065888240202e-06,
+      "loss": 0.8822,
+      "step": 3631
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 1.0232452154159546,
+      "learning_rate": 9.87489574645538e-06,
+      "loss": 0.6676,
+      "step": 3632
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 1.1694217920303345,
+      "learning_rate": 9.87072560467056e-06,
+      "loss": 0.8334,
+      "step": 3633
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 1.4995096921920776,
+      "learning_rate": 9.866555462885738e-06,
+      "loss": 0.6476,
+      "step": 3634
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 1.2373857498168945,
+      "learning_rate": 9.862385321100918e-06,
+      "loss": 0.727,
+      "step": 3635
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 1.32621169090271,
+      "learning_rate": 9.858215179316098e-06,
+      "loss": 0.6985,
+      "step": 3636
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 1.0579652786254883,
+      "learning_rate": 9.854045037531277e-06,
+      "loss": 0.6741,
+      "step": 3637
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 1.0375930070877075,
+      "learning_rate": 9.849874895746455e-06,
+      "loss": 0.5241,
+      "step": 3638
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 1.08954918384552,
+      "learning_rate": 9.845704753961635e-06,
+      "loss": 0.6364,
+      "step": 3639
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 1.3299015760421753,
+      "learning_rate": 9.841534612176815e-06,
+      "loss": 0.7207,
+      "step": 3640
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 1.180511236190796,
+      "learning_rate": 9.837364470391993e-06,
+      "loss": 0.8255,
+      "step": 3641
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 1.2076964378356934,
+      "learning_rate": 9.833194328607173e-06,
+      "loss": 0.6078,
+      "step": 3642
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 1.210663080215454,
+      "learning_rate": 9.829024186822352e-06,
+      "loss": 0.7695,
+      "step": 3643
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 1.2857632637023926,
+      "learning_rate": 9.824854045037532e-06,
+      "loss": 0.6456,
+      "step": 3644
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 1.383831262588501,
+      "learning_rate": 9.820683903252712e-06,
+      "loss": 0.6638,
+      "step": 3645
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 1.1777606010437012,
+      "learning_rate": 9.81651376146789e-06,
+      "loss": 0.7401,
+      "step": 3646
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 1.6303627490997314,
+      "learning_rate": 9.81234361968307e-06,
+      "loss": 0.6611,
+      "step": 3647
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 1.566899061203003,
+      "learning_rate": 9.808173477898249e-06,
+      "loss": 0.6503,
+      "step": 3648
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 1.554189920425415,
+      "learning_rate": 9.804003336113429e-06,
+      "loss": 0.7261,
+      "step": 3649
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 1.2478817701339722,
+      "learning_rate": 9.799833194328607e-06,
+      "loss": 0.786,
+      "step": 3650
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 1.0713328123092651,
+      "learning_rate": 9.795663052543787e-06,
+      "loss": 0.6749,
+      "step": 3651
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 1.0255318880081177,
+      "learning_rate": 9.791492910758967e-06,
+      "loss": 0.6317,
+      "step": 3652
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 1.5766685009002686,
+      "learning_rate": 9.787322768974145e-06,
+      "loss": 0.7716,
+      "step": 3653
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 1.2392951250076294,
+      "learning_rate": 9.783152627189325e-06,
+      "loss": 0.7043,
+      "step": 3654
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 1.043566107749939,
+      "learning_rate": 9.778982485404504e-06,
+      "loss": 0.5966,
+      "step": 3655
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 1.1397144794464111,
+      "learning_rate": 9.774812343619684e-06,
+      "loss": 0.7369,
+      "step": 3656
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 1.215667486190796,
+      "learning_rate": 9.770642201834864e-06,
+      "loss": 0.7371,
+      "step": 3657
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 1.9744904041290283,
+      "learning_rate": 9.766472060050042e-06,
+      "loss": 0.7856,
+      "step": 3658
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 1.0514297485351562,
+      "learning_rate": 9.76230191826522e-06,
+      "loss": 0.7465,
+      "step": 3659
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 1.0875815153121948,
+      "learning_rate": 9.7581317764804e-06,
+      "loss": 0.6965,
+      "step": 3660
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 1.1832997798919678,
+      "learning_rate": 9.75396163469558e-06,
+      "loss": 0.6994,
+      "step": 3661
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 1.108242392539978,
+      "learning_rate": 9.749791492910759e-06,
+      "loss": 0.685,
+      "step": 3662
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 1.3324147462844849,
+      "learning_rate": 9.745621351125939e-06,
+      "loss": 0.6214,
+      "step": 3663
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 1.1734163761138916,
+      "learning_rate": 9.741451209341117e-06,
+      "loss": 0.7048,
+      "step": 3664
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 1.1073940992355347,
+      "learning_rate": 9.737281067556297e-06,
+      "loss": 0.5457,
+      "step": 3665
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 1.361920714378357,
+      "learning_rate": 9.733110925771477e-06,
+      "loss": 0.8382,
+      "step": 3666
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 1.220529317855835,
+      "learning_rate": 9.728940783986656e-06,
+      "loss": 0.5753,
+      "step": 3667
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 1.112630009651184,
+      "learning_rate": 9.724770642201836e-06,
+      "loss": 0.6776,
+      "step": 3668
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 1.415000557899475,
+      "learning_rate": 9.720600500417016e-06,
+      "loss": 0.7272,
+      "step": 3669
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 1.3768407106399536,
+      "learning_rate": 9.716430358632194e-06,
+      "loss": 0.9133,
+      "step": 3670
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 1.0739002227783203,
+      "learning_rate": 9.712260216847372e-06,
+      "loss": 0.6825,
+      "step": 3671
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 1.6859723329544067,
+      "learning_rate": 9.708090075062552e-06,
+      "loss": 0.6812,
+      "step": 3672
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 1.2905598878860474,
+      "learning_rate": 9.703919933277733e-06,
+      "loss": 0.6885,
+      "step": 3673
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 1.417656660079956,
+      "learning_rate": 9.699749791492911e-06,
+      "loss": 0.6986,
+      "step": 3674
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 1.1009142398834229,
+      "learning_rate": 9.695579649708091e-06,
+      "loss": 0.6903,
+      "step": 3675
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 1.9833390712738037,
+      "learning_rate": 9.69140950792327e-06,
+      "loss": 0.7584,
+      "step": 3676
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 1.8520958423614502,
+      "learning_rate": 9.68723936613845e-06,
+      "loss": 0.6074,
+      "step": 3677
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 1.2613887786865234,
+      "learning_rate": 9.68306922435363e-06,
+      "loss": 0.7776,
+      "step": 3678
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 1.8603888750076294,
+      "learning_rate": 9.678899082568808e-06,
+      "loss": 0.8068,
+      "step": 3679
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 1.218383550643921,
+      "learning_rate": 9.674728940783988e-06,
+      "loss": 0.8195,
+      "step": 3680
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 1.3218454122543335,
+      "learning_rate": 9.670558798999166e-06,
+      "loss": 0.7639,
+      "step": 3681
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 1.1455937623977661,
+      "learning_rate": 9.666388657214346e-06,
+      "loss": 0.5993,
+      "step": 3682
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 1.2614988088607788,
+      "learning_rate": 9.662218515429524e-06,
+      "loss": 0.5664,
+      "step": 3683
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 1.880332589149475,
+      "learning_rate": 9.658048373644704e-06,
+      "loss": 0.7618,
+      "step": 3684
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 1.4358258247375488,
+      "learning_rate": 9.653878231859883e-06,
+      "loss": 0.9282,
+      "step": 3685
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 1.2800836563110352,
+      "learning_rate": 9.649708090075063e-06,
+      "loss": 0.8026,
+      "step": 3686
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 1.3461788892745972,
+      "learning_rate": 9.645537948290243e-06,
+      "loss": 0.7689,
+      "step": 3687
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 1.6501184701919556,
+      "learning_rate": 9.641367806505421e-06,
+      "loss": 0.6133,
+      "step": 3688
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 1.2736846208572388,
+      "learning_rate": 9.637197664720601e-06,
+      "loss": 0.8256,
+      "step": 3689
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 1.0804342031478882,
+      "learning_rate": 9.633027522935781e-06,
+      "loss": 0.645,
+      "step": 3690
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 1.1882388591766357,
+      "learning_rate": 9.62885738115096e-06,
+      "loss": 0.738,
+      "step": 3691
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 1.4110417366027832,
+      "learning_rate": 9.624687239366138e-06,
+      "loss": 0.6724,
+      "step": 3692
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 1.164905071258545,
+      "learning_rate": 9.620517097581318e-06,
+      "loss": 0.6826,
+      "step": 3693
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 1.250449776649475,
+      "learning_rate": 9.616346955796498e-06,
+      "loss": 0.7408,
+      "step": 3694
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 1.728674054145813,
+      "learning_rate": 9.612176814011676e-06,
+      "loss": 0.9098,
+      "step": 3695
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 1.9090059995651245,
+      "learning_rate": 9.608006672226856e-06,
+      "loss": 0.7101,
+      "step": 3696
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 1.120461106300354,
+      "learning_rate": 9.603836530442035e-06,
+      "loss": 0.6108,
+      "step": 3697
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 1.108109951019287,
+      "learning_rate": 9.599666388657215e-06,
+      "loss": 0.7364,
+      "step": 3698
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 1.2246441841125488,
+      "learning_rate": 9.595496246872395e-06,
+      "loss": 0.8161,
+      "step": 3699
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 1.3570644855499268,
+      "learning_rate": 9.591326105087573e-06,
+      "loss": 0.6529,
+      "step": 3700
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 1.089701533317566,
+      "learning_rate": 9.587155963302753e-06,
+      "loss": 0.7329,
+      "step": 3701
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 1.3717594146728516,
+      "learning_rate": 9.582985821517933e-06,
+      "loss": 0.7759,
+      "step": 3702
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 1.5683867931365967,
+      "learning_rate": 9.578815679733112e-06,
+      "loss": 0.7888,
+      "step": 3703
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 1.0901628732681274,
+      "learning_rate": 9.57464553794829e-06,
+      "loss": 0.6703,
+      "step": 3704
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 1.1070640087127686,
+      "learning_rate": 9.57047539616347e-06,
+      "loss": 0.8831,
+      "step": 3705
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 1.2490179538726807,
+      "learning_rate": 9.566305254378648e-06,
+      "loss": 0.6944,
+      "step": 3706
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 1.61417818069458,
+      "learning_rate": 9.562135112593828e-06,
+      "loss": 0.8053,
+      "step": 3707
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 1.1403989791870117,
+      "learning_rate": 9.557964970809008e-06,
+      "loss": 0.5259,
+      "step": 3708
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 1.4693373441696167,
+      "learning_rate": 9.553794829024187e-06,
+      "loss": 0.7022,
+      "step": 3709
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 1.109744668006897,
+      "learning_rate": 9.549624687239367e-06,
+      "loss": 0.8134,
+      "step": 3710
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 1.321455955505371,
+      "learning_rate": 9.545454545454547e-06,
+      "loss": 0.7871,
+      "step": 3711
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 1.0245014429092407,
+      "learning_rate": 9.541284403669725e-06,
+      "loss": 0.5754,
+      "step": 3712
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 1.4864879846572876,
+      "learning_rate": 9.537114261884905e-06,
+      "loss": 0.7603,
+      "step": 3713
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 1.1182165145874023,
+      "learning_rate": 9.532944120100083e-06,
+      "loss": 0.7357,
+      "step": 3714
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 1.3495372533798218,
+      "learning_rate": 9.528773978315264e-06,
+      "loss": 0.7678,
+      "step": 3715
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 1.2772372961044312,
+      "learning_rate": 9.524603836530442e-06,
+      "loss": 0.655,
+      "step": 3716
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 1.4288204908370972,
+      "learning_rate": 9.520433694745622e-06,
+      "loss": 0.8159,
+      "step": 3717
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 1.1076468229293823,
+      "learning_rate": 9.5162635529608e-06,
+      "loss": 0.7313,
+      "step": 3718
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 1.0478272438049316,
+      "learning_rate": 9.51209341117598e-06,
+      "loss": 0.6276,
+      "step": 3719
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 1.0887244939804077,
+      "learning_rate": 9.50792326939116e-06,
+      "loss": 0.4719,
+      "step": 3720
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 1.290826678276062,
+      "learning_rate": 9.503753127606339e-06,
+      "loss": 0.7153,
+      "step": 3721
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 1.2441591024398804,
+      "learning_rate": 9.499582985821519e-06,
+      "loss": 0.5785,
+      "step": 3722
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 1.3765219449996948,
+      "learning_rate": 9.495412844036699e-06,
+      "loss": 0.7279,
+      "step": 3723
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 1.2894562482833862,
+      "learning_rate": 9.491242702251877e-06,
+      "loss": 0.6829,
+      "step": 3724
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 1.8220261335372925,
+      "learning_rate": 9.487072560467055e-06,
+      "loss": 0.6256,
+      "step": 3725
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 1.4330283403396606,
+      "learning_rate": 9.482902418682235e-06,
+      "loss": 0.6358,
+      "step": 3726
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 1.0621377229690552,
+      "learning_rate": 9.478732276897414e-06,
+      "loss": 0.6017,
+      "step": 3727
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 1.6693440675735474,
+      "learning_rate": 9.474562135112594e-06,
+      "loss": 0.6218,
+      "step": 3728
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 1.2444632053375244,
+      "learning_rate": 9.470391993327774e-06,
+      "loss": 0.8163,
+      "step": 3729
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 1.4920762777328491,
+      "learning_rate": 9.466221851542952e-06,
+      "loss": 0.8798,
+      "step": 3730
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 1.4989265203475952,
+      "learning_rate": 9.462051709758132e-06,
+      "loss": 0.6288,
+      "step": 3731
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 1.5583096742630005,
+      "learning_rate": 9.457881567973312e-06,
+      "loss": 0.7625,
+      "step": 3732
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 1.2095603942871094,
+      "learning_rate": 9.45371142618849e-06,
+      "loss": 0.7579,
+      "step": 3733
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 1.6411110162734985,
+      "learning_rate": 9.44954128440367e-06,
+      "loss": 0.7544,
+      "step": 3734
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 1.5525989532470703,
+      "learning_rate": 9.44537114261885e-06,
+      "loss": 0.6798,
+      "step": 3735
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 1.4102756977081299,
+      "learning_rate": 9.441201000834029e-06,
+      "loss": 0.6924,
+      "step": 3736
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 1.1991088390350342,
+      "learning_rate": 9.437030859049207e-06,
+      "loss": 0.8546,
+      "step": 3737
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 1.2599495649337769,
+      "learning_rate": 9.432860717264387e-06,
+      "loss": 0.8464,
+      "step": 3738
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 1.1184576749801636,
+      "learning_rate": 9.428690575479566e-06,
+      "loss": 0.6982,
+      "step": 3739
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 1.1469248533248901,
+      "learning_rate": 9.424520433694746e-06,
+      "loss": 0.579,
+      "step": 3740
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 1.3901900053024292,
+      "learning_rate": 9.420350291909926e-06,
+      "loss": 0.583,
+      "step": 3741
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 1.032654881477356,
+      "learning_rate": 9.416180150125104e-06,
+      "loss": 0.7967,
+      "step": 3742
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 1.558066487312317,
+      "learning_rate": 9.412010008340284e-06,
+      "loss": 0.8004,
+      "step": 3743
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 1.0906705856323242,
+      "learning_rate": 9.407839866555464e-06,
+      "loss": 0.6494,
+      "step": 3744
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 1.29024076461792,
+      "learning_rate": 9.403669724770643e-06,
+      "loss": 0.8209,
+      "step": 3745
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 1.4281020164489746,
+      "learning_rate": 9.399499582985823e-06,
+      "loss": 0.7458,
+      "step": 3746
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 1.579883337020874,
+      "learning_rate": 9.395329441201001e-06,
+      "loss": 0.7569,
+      "step": 3747
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 1.3795017004013062,
+      "learning_rate": 9.39115929941618e-06,
+      "loss": 0.7465,
+      "step": 3748
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 1.039727807044983,
+      "learning_rate": 9.38698915763136e-06,
+      "loss": 0.6091,
+      "step": 3749
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 1.275769591331482,
+      "learning_rate": 9.38281901584654e-06,
+      "loss": 0.8901,
+      "step": 3750
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 0.8903956413269043,
+      "learning_rate": 9.378648874061718e-06,
+      "loss": 0.6723,
+      "step": 3751
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 1.2578043937683105,
+      "learning_rate": 9.374478732276898e-06,
+      "loss": 0.7633,
+      "step": 3752
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 1.2086877822875977,
+      "learning_rate": 9.370308590492078e-06,
+      "loss": 0.7965,
+      "step": 3753
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 1.6043813228607178,
+      "learning_rate": 9.366138448707256e-06,
+      "loss": 0.6879,
+      "step": 3754
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 1.2092528343200684,
+      "learning_rate": 9.361968306922436e-06,
+      "loss": 0.7433,
+      "step": 3755
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 1.9154289960861206,
+      "learning_rate": 9.357798165137616e-06,
+      "loss": 0.7641,
+      "step": 3756
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 1.324974775314331,
+      "learning_rate": 9.353628023352795e-06,
+      "loss": 0.7824,
+      "step": 3757
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 1.2821885347366333,
+      "learning_rate": 9.349457881567973e-06,
+      "loss": 0.769,
+      "step": 3758
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 1.1568022966384888,
+      "learning_rate": 9.345287739783153e-06,
+      "loss": 0.6623,
+      "step": 3759
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 1.1679307222366333,
+      "learning_rate": 9.341117597998331e-06,
+      "loss": 0.7793,
+      "step": 3760
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 1.2440667152404785,
+      "learning_rate": 9.336947456213511e-06,
+      "loss": 0.8573,
+      "step": 3761
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 1.2248564958572388,
+      "learning_rate": 9.332777314428691e-06,
+      "loss": 0.7017,
+      "step": 3762
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 1.140850305557251,
+      "learning_rate": 9.32860717264387e-06,
+      "loss": 0.741,
+      "step": 3763
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 1.4085427522659302,
+      "learning_rate": 9.32443703085905e-06,
+      "loss": 0.7385,
+      "step": 3764
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 1.1393663883209229,
+      "learning_rate": 9.32026688907423e-06,
+      "loss": 0.6256,
+      "step": 3765
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 1.5232007503509521,
+      "learning_rate": 9.316096747289408e-06,
+      "loss": 0.7036,
+      "step": 3766
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 1.3290067911148071,
+      "learning_rate": 9.311926605504588e-06,
+      "loss": 0.6962,
+      "step": 3767
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 1.3733572959899902,
+      "learning_rate": 9.307756463719768e-06,
+      "loss": 0.7121,
+      "step": 3768
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 1.4415980577468872,
+      "learning_rate": 9.303586321934945e-06,
+      "loss": 0.7647,
+      "step": 3769
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 1.3391342163085938,
+      "learning_rate": 9.299416180150125e-06,
+      "loss": 0.768,
+      "step": 3770
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 1.2893701791763306,
+      "learning_rate": 9.295246038365305e-06,
+      "loss": 0.667,
+      "step": 3771
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 1.296198844909668,
+      "learning_rate": 9.291075896580483e-06,
+      "loss": 0.7745,
+      "step": 3772
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 1.2993522882461548,
+      "learning_rate": 9.286905754795663e-06,
+      "loss": 0.8202,
+      "step": 3773
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 1.1732268333435059,
+      "learning_rate": 9.282735613010843e-06,
+      "loss": 0.9365,
+      "step": 3774
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 1.590893268585205,
+      "learning_rate": 9.278565471226022e-06,
+      "loss": 0.7263,
+      "step": 3775
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 1.3125766515731812,
+      "learning_rate": 9.274395329441202e-06,
+      "loss": 0.8581,
+      "step": 3776
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 1.445427417755127,
+      "learning_rate": 9.270225187656382e-06,
+      "loss": 0.687,
+      "step": 3777
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 1.2969833612442017,
+      "learning_rate": 9.26605504587156e-06,
+      "loss": 0.7846,
+      "step": 3778
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 1.3622808456420898,
+      "learning_rate": 9.26188490408674e-06,
+      "loss": 0.6802,
+      "step": 3779
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 1.2291429042816162,
+      "learning_rate": 9.257714762301918e-06,
+      "loss": 0.7613,
+      "step": 3780
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 1.2298214435577393,
+      "learning_rate": 9.253544620517097e-06,
+      "loss": 0.8673,
+      "step": 3781
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 1.1689229011535645,
+      "learning_rate": 9.249374478732277e-06,
+      "loss": 0.8588,
+      "step": 3782
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 1.2919254302978516,
+      "learning_rate": 9.245204336947457e-06,
+      "loss": 0.9037,
+      "step": 3783
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 0.9366986155509949,
+      "learning_rate": 9.241034195162635e-06,
+      "loss": 0.6829,
+      "step": 3784
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 1.0638045072555542,
+      "learning_rate": 9.236864053377815e-06,
+      "loss": 0.6312,
+      "step": 3785
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 1.3222174644470215,
+      "learning_rate": 9.232693911592995e-06,
+      "loss": 0.7944,
+      "step": 3786
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 1.158786654472351,
+      "learning_rate": 9.228523769808174e-06,
+      "loss": 0.7255,
+      "step": 3787
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 1.1780650615692139,
+      "learning_rate": 9.224353628023354e-06,
+      "loss": 0.7522,
+      "step": 3788
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 1.5175025463104248,
+      "learning_rate": 9.220183486238534e-06,
+      "loss": 0.8647,
+      "step": 3789
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 1.585942268371582,
+      "learning_rate": 9.216013344453712e-06,
+      "loss": 0.6549,
+      "step": 3790
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 1.1129939556121826,
+      "learning_rate": 9.21184320266889e-06,
+      "loss": 0.7661,
+      "step": 3791
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 1.2599798440933228,
+      "learning_rate": 9.20767306088407e-06,
+      "loss": 0.8126,
+      "step": 3792
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 1.6986596584320068,
+      "learning_rate": 9.203502919099249e-06,
+      "loss": 0.711,
+      "step": 3793
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 1.230600357055664,
+      "learning_rate": 9.199332777314429e-06,
+      "loss": 0.8057,
+      "step": 3794
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 1.1111856698989868,
+      "learning_rate": 9.195162635529609e-06,
+      "loss": 0.6925,
+      "step": 3795
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 1.2809500694274902,
+      "learning_rate": 9.190992493744787e-06,
+      "loss": 0.701,
+      "step": 3796
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 1.0989046096801758,
+      "learning_rate": 9.186822351959967e-06,
+      "loss": 0.62,
+      "step": 3797
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 1.066048264503479,
+      "learning_rate": 9.182652210175147e-06,
+      "loss": 0.7419,
+      "step": 3798
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 1.0855743885040283,
+      "learning_rate": 9.178482068390326e-06,
+      "loss": 0.7767,
+      "step": 3799
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 1.243959903717041,
+      "learning_rate": 9.174311926605506e-06,
+      "loss": 0.8384,
+      "step": 3800
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 1.0355478525161743,
+      "learning_rate": 9.170141784820686e-06,
+      "loss": 0.7668,
+      "step": 3801
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 1.559926152229309,
+      "learning_rate": 9.165971643035862e-06,
+      "loss": 0.7143,
+      "step": 3802
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 1.1760704517364502,
+      "learning_rate": 9.161801501251042e-06,
+      "loss": 0.8026,
+      "step": 3803
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 1.2300641536712646,
+      "learning_rate": 9.157631359466222e-06,
+      "loss": 0.7846,
+      "step": 3804
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 1.4027724266052246,
+      "learning_rate": 9.1534612176814e-06,
+      "loss": 0.7137,
+      "step": 3805
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 1.0805466175079346,
+      "learning_rate": 9.14929107589658e-06,
+      "loss": 0.6632,
+      "step": 3806
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 1.2802335023880005,
+      "learning_rate": 9.14512093411176e-06,
+      "loss": 0.8155,
+      "step": 3807
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 1.4101649522781372,
+      "learning_rate": 9.140950792326939e-06,
+      "loss": 0.7731,
+      "step": 3808
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 1.4881744384765625,
+      "learning_rate": 9.136780650542119e-06,
+      "loss": 0.7518,
+      "step": 3809
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 1.190462589263916,
+      "learning_rate": 9.132610508757299e-06,
+      "loss": 0.6099,
+      "step": 3810
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 1.277464509010315,
+      "learning_rate": 9.128440366972477e-06,
+      "loss": 0.7867,
+      "step": 3811
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 1.3371671438217163,
+      "learning_rate": 9.124270225187658e-06,
+      "loss": 0.7767,
+      "step": 3812
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 1.0814464092254639,
+      "learning_rate": 9.120100083402836e-06,
+      "loss": 0.6586,
+      "step": 3813
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 1.6695940494537354,
+      "learning_rate": 9.115929941618014e-06,
+      "loss": 0.659,
+      "step": 3814
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 1.0612856149673462,
+      "learning_rate": 9.111759799833194e-06,
+      "loss": 0.5992,
+      "step": 3815
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 1.6444050073623657,
+      "learning_rate": 9.107589658048374e-06,
+      "loss": 0.6657,
+      "step": 3816
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 1.5341229438781738,
+      "learning_rate": 9.103419516263553e-06,
+      "loss": 0.6647,
+      "step": 3817
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 1.2004408836364746,
+      "learning_rate": 9.099249374478733e-06,
+      "loss": 0.7232,
+      "step": 3818
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 1.0692825317382812,
+      "learning_rate": 9.095079232693913e-06,
+      "loss": 0.7102,
+      "step": 3819
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 1.3460806608200073,
+      "learning_rate": 9.090909090909091e-06,
+      "loss": 0.707,
+      "step": 3820
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 1.273742914199829,
+      "learning_rate": 9.086738949124271e-06,
+      "loss": 0.7478,
+      "step": 3821
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 2.021270275115967,
+      "learning_rate": 9.082568807339451e-06,
+      "loss": 0.812,
+      "step": 3822
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 1.2000283002853394,
+      "learning_rate": 9.07839866555463e-06,
+      "loss": 0.6991,
+      "step": 3823
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 1.7958741188049316,
+      "learning_rate": 9.074228523769808e-06,
+      "loss": 0.6708,
+      "step": 3824
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 0.9178566336631775,
+      "learning_rate": 9.070058381984988e-06,
+      "loss": 0.7075,
+      "step": 3825
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 1.2584969997406006,
+      "learning_rate": 9.065888240200166e-06,
+      "loss": 0.629,
+      "step": 3826
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 1.2469345331192017,
+      "learning_rate": 9.061718098415346e-06,
+      "loss": 0.7066,
+      "step": 3827
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 1.5744034051895142,
+      "learning_rate": 9.057547956630526e-06,
+      "loss": 0.7953,
+      "step": 3828
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 1.2157225608825684,
+      "learning_rate": 9.053377814845705e-06,
+      "loss": 0.7811,
+      "step": 3829
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 1.3481709957122803,
+      "learning_rate": 9.049207673060885e-06,
+      "loss": 0.7233,
+      "step": 3830
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 1.3199782371520996,
+      "learning_rate": 9.045037531276065e-06,
+      "loss": 0.8076,
+      "step": 3831
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 1.0716701745986938,
+      "learning_rate": 9.040867389491243e-06,
+      "loss": 0.6526,
+      "step": 3832
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 1.422723650932312,
+      "learning_rate": 9.036697247706423e-06,
+      "loss": 0.616,
+      "step": 3833
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 1.0690348148345947,
+      "learning_rate": 9.032527105921603e-06,
+      "loss": 0.6838,
+      "step": 3834
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 1.8414225578308105,
+      "learning_rate": 9.02835696413678e-06,
+      "loss": 0.7595,
+      "step": 3835
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 1.3578447103500366,
+      "learning_rate": 9.02418682235196e-06,
+      "loss": 0.6945,
+      "step": 3836
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 1.7697968482971191,
+      "learning_rate": 9.02001668056714e-06,
+      "loss": 0.7137,
+      "step": 3837
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 1.4124974012374878,
+      "learning_rate": 9.015846538782318e-06,
+      "loss": 0.7721,
+      "step": 3838
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 1.0588980913162231,
+      "learning_rate": 9.011676396997498e-06,
+      "loss": 0.6732,
+      "step": 3839
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 1.1191587448120117,
+      "learning_rate": 9.007506255212678e-06,
+      "loss": 0.6778,
+      "step": 3840
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 1.2233281135559082,
+      "learning_rate": 9.003336113427857e-06,
+      "loss": 0.8189,
+      "step": 3841
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 1.1825919151306152,
+      "learning_rate": 8.999165971643037e-06,
+      "loss": 0.7156,
+      "step": 3842
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 1.2899632453918457,
+      "learning_rate": 8.994995829858217e-06,
+      "loss": 0.6679,
+      "step": 3843
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 1.245190978050232,
+      "learning_rate": 8.990825688073395e-06,
+      "loss": 0.6828,
+      "step": 3844
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 1.7574058771133423,
+      "learning_rate": 8.986655546288575e-06,
+      "loss": 0.9988,
+      "step": 3845
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 1.1397372484207153,
+      "learning_rate": 8.982485404503753e-06,
+      "loss": 0.6545,
+      "step": 3846
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 1.2123494148254395,
+      "learning_rate": 8.978315262718932e-06,
+      "loss": 0.6628,
+      "step": 3847
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 1.0750784873962402,
+      "learning_rate": 8.974145120934112e-06,
+      "loss": 0.7138,
+      "step": 3848
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 1.2346811294555664,
+      "learning_rate": 8.969974979149292e-06,
+      "loss": 0.6846,
+      "step": 3849
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 1.402255892753601,
+      "learning_rate": 8.96580483736447e-06,
+      "loss": 0.787,
+      "step": 3850
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 1.5618852376937866,
+      "learning_rate": 8.96163469557965e-06,
+      "loss": 0.6784,
+      "step": 3851
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 1.0520151853561401,
+      "learning_rate": 8.95746455379483e-06,
+      "loss": 0.6024,
+      "step": 3852
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 1.403063178062439,
+      "learning_rate": 8.953294412010008e-06,
+      "loss": 0.7569,
+      "step": 3853
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 1.5110641717910767,
+      "learning_rate": 8.949124270225189e-06,
+      "loss": 0.7434,
+      "step": 3854
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 1.1577140092849731,
+      "learning_rate": 8.944954128440369e-06,
+      "loss": 0.6905,
+      "step": 3855
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 1.5566719770431519,
+      "learning_rate": 8.940783986655547e-06,
+      "loss": 0.8057,
+      "step": 3856
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 1.5115667581558228,
+      "learning_rate": 8.936613844870725e-06,
+      "loss": 0.5612,
+      "step": 3857
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 1.4218279123306274,
+      "learning_rate": 8.932443703085905e-06,
+      "loss": 0.7115,
+      "step": 3858
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 1.1728657484054565,
+      "learning_rate": 8.928273561301084e-06,
+      "loss": 0.6655,
+      "step": 3859
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 1.3531938791275024,
+      "learning_rate": 8.924103419516264e-06,
+      "loss": 0.8297,
+      "step": 3860
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 1.8712323904037476,
+      "learning_rate": 8.919933277731444e-06,
+      "loss": 0.792,
+      "step": 3861
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 1.3992277383804321,
+      "learning_rate": 8.915763135946622e-06,
+      "loss": 0.7692,
+      "step": 3862
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 1.2772537469863892,
+      "learning_rate": 8.911592994161802e-06,
+      "loss": 0.7146,
+      "step": 3863
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 1.083439588546753,
+      "learning_rate": 8.907422852376982e-06,
+      "loss": 0.6436,
+      "step": 3864
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 1.0760661363601685,
+      "learning_rate": 8.90325271059216e-06,
+      "loss": 0.6675,
+      "step": 3865
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 1.0748087167739868,
+      "learning_rate": 8.89908256880734e-06,
+      "loss": 0.8525,
+      "step": 3866
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 1.5770659446716309,
+      "learning_rate": 8.89491242702252e-06,
+      "loss": 0.7182,
+      "step": 3867
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 1.2480491399765015,
+      "learning_rate": 8.890742285237697e-06,
+      "loss": 0.6618,
+      "step": 3868
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 0.9434835314750671,
+      "learning_rate": 8.886572143452877e-06,
+      "loss": 0.6799,
+      "step": 3869
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 1.2592473030090332,
+      "learning_rate": 8.882402001668057e-06,
+      "loss": 0.5887,
+      "step": 3870
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 0.9747094511985779,
+      "learning_rate": 8.878231859883236e-06,
+      "loss": 0.659,
+      "step": 3871
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 1.5506322383880615,
+      "learning_rate": 8.874061718098416e-06,
+      "loss": 0.6788,
+      "step": 3872
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 1.4279855489730835,
+      "learning_rate": 8.869891576313596e-06,
+      "loss": 0.7228,
+      "step": 3873
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 1.0204380750656128,
+      "learning_rate": 8.865721434528774e-06,
+      "loss": 0.6567,
+      "step": 3874
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 2.258882999420166,
+      "learning_rate": 8.861551292743954e-06,
+      "loss": 0.7533,
+      "step": 3875
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 1.316502332687378,
+      "learning_rate": 8.857381150959134e-06,
+      "loss": 0.6603,
+      "step": 3876
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 1.306138515472412,
+      "learning_rate": 8.853211009174312e-06,
+      "loss": 0.7205,
+      "step": 3877
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 1.134323000907898,
+      "learning_rate": 8.849040867389492e-06,
+      "loss": 0.7431,
+      "step": 3878
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 1.4522703886032104,
+      "learning_rate": 8.84487072560467e-06,
+      "loss": 0.7373,
+      "step": 3879
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 1.2387404441833496,
+      "learning_rate": 8.840700583819849e-06,
+      "loss": 0.7731,
+      "step": 3880
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 1.2427064180374146,
+      "learning_rate": 8.83653044203503e-06,
+      "loss": 0.8667,
+      "step": 3881
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 1.855582356452942,
+      "learning_rate": 8.83236030025021e-06,
+      "loss": 0.8088,
+      "step": 3882
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 1.252475380897522,
+      "learning_rate": 8.828190158465388e-06,
+      "loss": 0.7236,
+      "step": 3883
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 1.447736144065857,
+      "learning_rate": 8.824020016680568e-06,
+      "loss": 0.8166,
+      "step": 3884
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 1.1489137411117554,
+      "learning_rate": 8.819849874895748e-06,
+      "loss": 0.709,
+      "step": 3885
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 1.1987931728363037,
+      "learning_rate": 8.815679733110926e-06,
+      "loss": 0.7412,
+      "step": 3886
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 1.2565351724624634,
+      "learning_rate": 8.811509591326106e-06,
+      "loss": 0.6736,
+      "step": 3887
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 1.304922103881836,
+      "learning_rate": 8.807339449541286e-06,
+      "loss": 0.6911,
+      "step": 3888
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 1.3182557821273804,
+      "learning_rate": 8.803169307756464e-06,
+      "loss": 0.9893,
+      "step": 3889
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 1.0852586030960083,
+      "learning_rate": 8.798999165971643e-06,
+      "loss": 0.8067,
+      "step": 3890
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 1.0428166389465332,
+      "learning_rate": 8.794829024186823e-06,
+      "loss": 0.5376,
+      "step": 3891
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 1.490378975868225,
+      "learning_rate": 8.790658882402001e-06,
+      "loss": 0.788,
+      "step": 3892
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 1.7355320453643799,
+      "learning_rate": 8.786488740617181e-06,
+      "loss": 0.6874,
+      "step": 3893
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 1.0040160417556763,
+      "learning_rate": 8.782318598832361e-06,
+      "loss": 0.6798,
+      "step": 3894
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 1.1138432025909424,
+      "learning_rate": 8.77814845704754e-06,
+      "loss": 0.7166,
+      "step": 3895
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 1.5286980867385864,
+      "learning_rate": 8.77397831526272e-06,
+      "loss": 0.7808,
+      "step": 3896
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 1.0869489908218384,
+      "learning_rate": 8.7698081734779e-06,
+      "loss": 0.6303,
+      "step": 3897
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 1.4481172561645508,
+      "learning_rate": 8.765638031693078e-06,
+      "loss": 0.8856,
+      "step": 3898
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 1.4681679010391235,
+      "learning_rate": 8.761467889908258e-06,
+      "loss": 0.6628,
+      "step": 3899
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 1.4028898477554321,
+      "learning_rate": 8.757297748123438e-06,
+      "loss": 0.8009,
+      "step": 3900
+    },
+    {
+      "epoch": 3.77,
+      "eval_loss": 0.8551230430603027,
+      "eval_runtime": 858.1586,
+      "eval_samples_per_second": 4.817,
+      "eval_steps_per_second": 0.602,
+      "step": 3900
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 1.1668781042099,
+      "learning_rate": 8.753127606338615e-06,
+      "loss": 0.7094,
+      "step": 3901
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 1.279202938079834,
+      "learning_rate": 8.748957464553795e-06,
+      "loss": 0.8539,
+      "step": 3902
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 1.3793323040008545,
+      "learning_rate": 8.744787322768975e-06,
+      "loss": 0.5602,
+      "step": 3903
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 1.2167240381240845,
+      "learning_rate": 8.740617180984153e-06,
+      "loss": 0.7458,
+      "step": 3904
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 1.5817376375198364,
+      "learning_rate": 8.736447039199333e-06,
+      "loss": 0.6797,
+      "step": 3905
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 1.3980896472930908,
+      "learning_rate": 8.732276897414513e-06,
+      "loss": 0.6803,
+      "step": 3906
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 1.4637523889541626,
+      "learning_rate": 8.728106755629691e-06,
+      "loss": 0.9646,
+      "step": 3907
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 1.1062219142913818,
+      "learning_rate": 8.723936613844871e-06,
+      "loss": 0.6811,
+      "step": 3908
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 1.5182491540908813,
+      "learning_rate": 8.719766472060052e-06,
+      "loss": 0.8207,
+      "step": 3909
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 1.0973973274230957,
+      "learning_rate": 8.71559633027523e-06,
+      "loss": 0.7436,
+      "step": 3910
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 1.0992844104766846,
+      "learning_rate": 8.71142618849041e-06,
+      "loss": 0.8696,
+      "step": 3911
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 1.2279597520828247,
+      "learning_rate": 8.707256046705588e-06,
+      "loss": 0.7107,
+      "step": 3912
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 1.3053665161132812,
+      "learning_rate": 8.703085904920767e-06,
+      "loss": 0.7919,
+      "step": 3913
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 1.193116545677185,
+      "learning_rate": 8.698915763135947e-06,
+      "loss": 0.7822,
+      "step": 3914
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 1.24557363986969,
+      "learning_rate": 8.694745621351127e-06,
+      "loss": 0.6917,
+      "step": 3915
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 1.1103416681289673,
+      "learning_rate": 8.690575479566305e-06,
+      "loss": 0.6902,
+      "step": 3916
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 1.5970569849014282,
+      "learning_rate": 8.686405337781485e-06,
+      "loss": 0.715,
+      "step": 3917
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 1.3219919204711914,
+      "learning_rate": 8.682235195996665e-06,
+      "loss": 0.668,
+      "step": 3918
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 1.3247804641723633,
+      "learning_rate": 8.678065054211843e-06,
+      "loss": 0.7101,
+      "step": 3919
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 1.8229241371154785,
+      "learning_rate": 8.673894912427023e-06,
+      "loss": 0.7737,
+      "step": 3920
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 1.1297950744628906,
+      "learning_rate": 8.669724770642203e-06,
+      "loss": 0.7004,
+      "step": 3921
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 1.413045048713684,
+      "learning_rate": 8.665554628857382e-06,
+      "loss": 0.7086,
+      "step": 3922
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 1.4547609090805054,
+      "learning_rate": 8.66138448707256e-06,
+      "loss": 0.8475,
+      "step": 3923
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 1.1815779209136963,
+      "learning_rate": 8.65721434528774e-06,
+      "loss": 0.5953,
+      "step": 3924
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 1.0510928630828857,
+      "learning_rate": 8.653044203502919e-06,
+      "loss": 0.6837,
+      "step": 3925
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 1.6641714572906494,
+      "learning_rate": 8.648874061718099e-06,
+      "loss": 0.7985,
+      "step": 3926
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 1.7786951065063477,
+      "learning_rate": 8.644703919933279e-06,
+      "loss": 0.68,
+      "step": 3927
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 1.3749067783355713,
+      "learning_rate": 8.640533778148457e-06,
+      "loss": 0.7351,
+      "step": 3928
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 1.4634582996368408,
+      "learning_rate": 8.636363636363637e-06,
+      "loss": 0.7863,
+      "step": 3929
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 1.6887547969818115,
+      "learning_rate": 8.632193494578817e-06,
+      "loss": 0.8711,
+      "step": 3930
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 1.2339435815811157,
+      "learning_rate": 8.628023352793995e-06,
+      "loss": 0.6633,
+      "step": 3931
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 1.174820899963379,
+      "learning_rate": 8.623853211009175e-06,
+      "loss": 0.633,
+      "step": 3932
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 1.3371484279632568,
+      "learning_rate": 8.619683069224355e-06,
+      "loss": 0.6297,
+      "step": 3933
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 1.1373400688171387,
+      "learning_rate": 8.615512927439532e-06,
+      "loss": 0.758,
+      "step": 3934
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 1.9303170442581177,
+      "learning_rate": 8.611342785654712e-06,
+      "loss": 0.7414,
+      "step": 3935
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 1.1726877689361572,
+      "learning_rate": 8.607172643869892e-06,
+      "loss": 0.671,
+      "step": 3936
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 1.918506383895874,
+      "learning_rate": 8.60300250208507e-06,
+      "loss": 0.6406,
+      "step": 3937
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 1.0149345397949219,
+      "learning_rate": 8.59883236030025e-06,
+      "loss": 0.6896,
+      "step": 3938
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 1.8189246654510498,
+      "learning_rate": 8.59466221851543e-06,
+      "loss": 0.6587,
+      "step": 3939
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 1.7515846490859985,
+      "learning_rate": 8.590492076730609e-06,
+      "loss": 0.9204,
+      "step": 3940
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 1.0661392211914062,
+      "learning_rate": 8.586321934945789e-06,
+      "loss": 0.6879,
+      "step": 3941
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 1.293162226676941,
+      "learning_rate": 8.582151793160969e-06,
+      "loss": 0.6811,
+      "step": 3942
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 1.080329179763794,
+      "learning_rate": 8.577981651376147e-06,
+      "loss": 0.8401,
+      "step": 3943
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 1.2835712432861328,
+      "learning_rate": 8.573811509591327e-06,
+      "loss": 0.6676,
+      "step": 3944
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 1.0308789014816284,
+      "learning_rate": 8.569641367806506e-06,
+      "loss": 0.6094,
+      "step": 3945
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 1.296432375907898,
+      "learning_rate": 8.565471226021684e-06,
+      "loss": 0.8018,
+      "step": 3946
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 1.3173983097076416,
+      "learning_rate": 8.561301084236864e-06,
+      "loss": 0.751,
+      "step": 3947
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 1.3164054155349731,
+      "learning_rate": 8.557130942452044e-06,
+      "loss": 0.7981,
+      "step": 3948
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 1.1949673891067505,
+      "learning_rate": 8.552960800667222e-06,
+      "loss": 0.7503,
+      "step": 3949
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 1.2497044801712036,
+      "learning_rate": 8.548790658882402e-06,
+      "loss": 0.8542,
+      "step": 3950
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 1.3445051908493042,
+      "learning_rate": 8.544620517097583e-06,
+      "loss": 0.843,
+      "step": 3951
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 1.0741132497787476,
+      "learning_rate": 8.540450375312761e-06,
+      "loss": 0.6759,
+      "step": 3952
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 1.1905039548873901,
+      "learning_rate": 8.536280233527941e-06,
+      "loss": 0.6363,
+      "step": 3953
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 1.234527826309204,
+      "learning_rate": 8.532110091743121e-06,
+      "loss": 0.7328,
+      "step": 3954
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 1.1649458408355713,
+      "learning_rate": 8.5279399499583e-06,
+      "loss": 0.556,
+      "step": 3955
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 1.1444815397262573,
+      "learning_rate": 8.523769808173478e-06,
+      "loss": 0.6184,
+      "step": 3956
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 1.1928359270095825,
+      "learning_rate": 8.519599666388658e-06,
+      "loss": 0.598,
+      "step": 3957
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 1.1508607864379883,
+      "learning_rate": 8.515429524603836e-06,
+      "loss": 0.5703,
+      "step": 3958
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 1.59660005569458,
+      "learning_rate": 8.511259382819016e-06,
+      "loss": 0.6351,
+      "step": 3959
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 1.2995659112930298,
+      "learning_rate": 8.507089241034196e-06,
+      "loss": 0.8417,
+      "step": 3960
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 1.7801694869995117,
+      "learning_rate": 8.502919099249374e-06,
+      "loss": 0.9495,
+      "step": 3961
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 1.3981291055679321,
+      "learning_rate": 8.498748957464554e-06,
+      "loss": 0.7076,
+      "step": 3962
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 1.372262716293335,
+      "learning_rate": 8.494578815679734e-06,
+      "loss": 0.8684,
+      "step": 3963
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 1.3083829879760742,
+      "learning_rate": 8.490408673894913e-06,
+      "loss": 0.816,
+      "step": 3964
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 1.0977733135223389,
+      "learning_rate": 8.486238532110093e-06,
+      "loss": 0.6485,
+      "step": 3965
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 1.4571865797042847,
+      "learning_rate": 8.482068390325273e-06,
+      "loss": 0.8095,
+      "step": 3966
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 1.2152897119522095,
+      "learning_rate": 8.47789824854045e-06,
+      "loss": 0.8019,
+      "step": 3967
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 1.812170147895813,
+      "learning_rate": 8.47372810675563e-06,
+      "loss": 0.7086,
+      "step": 3968
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 1.2257304191589355,
+      "learning_rate": 8.46955796497081e-06,
+      "loss": 0.7329,
+      "step": 3969
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 1.5328458547592163,
+      "learning_rate": 8.465387823185988e-06,
+      "loss": 0.8077,
+      "step": 3970
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 1.4491785764694214,
+      "learning_rate": 8.461217681401168e-06,
+      "loss": 0.8759,
+      "step": 3971
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 1.3397568464279175,
+      "learning_rate": 8.457047539616348e-06,
+      "loss": 0.6961,
+      "step": 3972
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 1.3516210317611694,
+      "learning_rate": 8.452877397831526e-06,
+      "loss": 0.7509,
+      "step": 3973
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 1.0660499334335327,
+      "learning_rate": 8.448707256046706e-06,
+      "loss": 0.5578,
+      "step": 3974
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 1.2426209449768066,
+      "learning_rate": 8.444537114261886e-06,
+      "loss": 0.7431,
+      "step": 3975
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 1.277725100517273,
+      "learning_rate": 8.440366972477065e-06,
+      "loss": 0.7016,
+      "step": 3976
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 1.1149075031280518,
+      "learning_rate": 8.436196830692245e-06,
+      "loss": 0.6526,
+      "step": 3977
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 1.565482258796692,
+      "learning_rate": 8.432026688907423e-06,
+      "loss": 0.8724,
+      "step": 3978
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 1.3156458139419556,
+      "learning_rate": 8.427856547122602e-06,
+      "loss": 0.6484,
+      "step": 3979
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 1.5746185779571533,
+      "learning_rate": 8.423686405337782e-06,
+      "loss": 0.8209,
+      "step": 3980
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 1.2292309999465942,
+      "learning_rate": 8.419516263552962e-06,
+      "loss": 0.6328,
+      "step": 3981
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 1.4444047212600708,
+      "learning_rate": 8.41534612176814e-06,
+      "loss": 0.7564,
+      "step": 3982
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 1.1935038566589355,
+      "learning_rate": 8.41117597998332e-06,
+      "loss": 0.7848,
+      "step": 3983
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 1.3232645988464355,
+      "learning_rate": 8.4070058381985e-06,
+      "loss": 0.6391,
+      "step": 3984
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 1.1306418180465698,
+      "learning_rate": 8.402835696413678e-06,
+      "loss": 0.7451,
+      "step": 3985
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 1.3414762020111084,
+      "learning_rate": 8.398665554628858e-06,
+      "loss": 0.5525,
+      "step": 3986
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 1.1248849630355835,
+      "learning_rate": 8.394495412844038e-06,
+      "loss": 0.774,
+      "step": 3987
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 1.3943297863006592,
+      "learning_rate": 8.390325271059217e-06,
+      "loss": 0.7663,
+      "step": 3988
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 1.0571784973144531,
+      "learning_rate": 8.386155129274395e-06,
+      "loss": 0.6862,
+      "step": 3989
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 1.528711199760437,
+      "learning_rate": 8.381984987489575e-06,
+      "loss": 0.7129,
+      "step": 3990
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 1.2378162145614624,
+      "learning_rate": 8.377814845704753e-06,
+      "loss": 0.6722,
+      "step": 3991
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 1.1124138832092285,
+      "learning_rate": 8.373644703919933e-06,
+      "loss": 0.7745,
+      "step": 3992
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 1.20100998878479,
+      "learning_rate": 8.369474562135114e-06,
+      "loss": 0.7916,
+      "step": 3993
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 1.234928011894226,
+      "learning_rate": 8.365304420350292e-06,
+      "loss": 0.8269,
+      "step": 3994
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 1.1601148843765259,
+      "learning_rate": 8.361134278565472e-06,
+      "loss": 0.6957,
+      "step": 3995
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 1.1746742725372314,
+      "learning_rate": 8.356964136780652e-06,
+      "loss": 0.6987,
+      "step": 3996
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 1.366929292678833,
+      "learning_rate": 8.35279399499583e-06,
+      "loss": 0.8521,
+      "step": 3997
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 1.2353994846343994,
+      "learning_rate": 8.34862385321101e-06,
+      "loss": 0.6475,
+      "step": 3998
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 1.0730762481689453,
+      "learning_rate": 8.344453711426189e-06,
+      "loss": 0.6401,
+      "step": 3999
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 1.1085494756698608,
+      "learning_rate": 8.340283569641367e-06,
+      "loss": 0.6696,
+      "step": 4000
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 1.027856707572937,
+      "learning_rate": 8.336113427856547e-06,
+      "loss": 0.7014,
+      "step": 4001
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 1.6614447832107544,
+      "learning_rate": 8.331943286071727e-06,
+      "loss": 0.7025,
+      "step": 4002
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 1.4471124410629272,
+      "learning_rate": 8.327773144286905e-06,
+      "loss": 0.6179,
+      "step": 4003
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 1.5548070669174194,
+      "learning_rate": 8.323603002502085e-06,
+      "loss": 0.831,
+      "step": 4004
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 1.3854544162750244,
+      "learning_rate": 8.319432860717265e-06,
+      "loss": 0.7213,
+      "step": 4005
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 1.2448898553848267,
+      "learning_rate": 8.315262718932444e-06,
+      "loss": 0.9163,
+      "step": 4006
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 1.0090829133987427,
+      "learning_rate": 8.311092577147624e-06,
+      "loss": 0.8564,
+      "step": 4007
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 1.4596858024597168,
+      "learning_rate": 8.306922435362804e-06,
+      "loss": 0.728,
+      "step": 4008
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 1.7371070384979248,
+      "learning_rate": 8.302752293577982e-06,
+      "loss": 0.7635,
+      "step": 4009
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 1.4701354503631592,
+      "learning_rate": 8.298582151793162e-06,
+      "loss": 0.8342,
+      "step": 4010
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 1.4418973922729492,
+      "learning_rate": 8.29441201000834e-06,
+      "loss": 0.6969,
+      "step": 4011
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 1.3696964979171753,
+      "learning_rate": 8.290241868223519e-06,
+      "loss": 0.8933,
+      "step": 4012
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 1.378226637840271,
+      "learning_rate": 8.286071726438699e-06,
+      "loss": 0.7117,
+      "step": 4013
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 1.2607053518295288,
+      "learning_rate": 8.281901584653879e-06,
+      "loss": 0.7669,
+      "step": 4014
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 1.0556697845458984,
+      "learning_rate": 8.277731442869057e-06,
+      "loss": 0.6564,
+      "step": 4015
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 1.4814832210540771,
+      "learning_rate": 8.273561301084237e-06,
+      "loss": 0.6241,
+      "step": 4016
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 1.2367101907730103,
+      "learning_rate": 8.269391159299417e-06,
+      "loss": 0.7911,
+      "step": 4017
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 1.853064775466919,
+      "learning_rate": 8.265221017514596e-06,
+      "loss": 0.79,
+      "step": 4018
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 1.3849866390228271,
+      "learning_rate": 8.261050875729776e-06,
+      "loss": 0.6966,
+      "step": 4019
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 1.6763454675674438,
+      "learning_rate": 8.256880733944954e-06,
+      "loss": 0.6965,
+      "step": 4020
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 1.173415184020996,
+      "learning_rate": 8.252710592160134e-06,
+      "loss": 0.6555,
+      "step": 4021
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 1.315529704093933,
+      "learning_rate": 8.248540450375313e-06,
+      "loss": 0.7649,
+      "step": 4022
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 1.5528346300125122,
+      "learning_rate": 8.244370308590493e-06,
+      "loss": 0.834,
+      "step": 4023
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 1.5623164176940918,
+      "learning_rate": 8.240200166805671e-06,
+      "loss": 0.6323,
+      "step": 4024
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 1.2017834186553955,
+      "learning_rate": 8.236030025020851e-06,
+      "loss": 0.5434,
+      "step": 4025
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 1.257459044456482,
+      "learning_rate": 8.231859883236031e-06,
+      "loss": 1.0339,
+      "step": 4026
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 1.421531081199646,
+      "learning_rate": 8.22768974145121e-06,
+      "loss": 0.7466,
+      "step": 4027
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 1.3751212358474731,
+      "learning_rate": 8.22351959966639e-06,
+      "loss": 0.593,
+      "step": 4028
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 1.1514464616775513,
+      "learning_rate": 8.21934945788157e-06,
+      "loss": 0.7281,
+      "step": 4029
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 1.050018548965454,
+      "learning_rate": 8.215179316096748e-06,
+      "loss": 0.6659,
+      "step": 4030
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 1.6007755994796753,
+      "learning_rate": 8.211009174311928e-06,
+      "loss": 0.6255,
+      "step": 4031
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 1.3381264209747314,
+      "learning_rate": 8.206839032527106e-06,
+      "loss": 0.7391,
+      "step": 4032
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 1.1450636386871338,
+      "learning_rate": 8.202668890742284e-06,
+      "loss": 0.6375,
+      "step": 4033
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 1.4435229301452637,
+      "learning_rate": 8.198498748957465e-06,
+      "loss": 0.7139,
+      "step": 4034
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 1.1717262268066406,
+      "learning_rate": 8.194328607172645e-06,
+      "loss": 0.6299,
+      "step": 4035
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 1.3287850618362427,
+      "learning_rate": 8.190158465387823e-06,
+      "loss": 0.8284,
+      "step": 4036
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 1.2789373397827148,
+      "learning_rate": 8.185988323603003e-06,
+      "loss": 0.6759,
+      "step": 4037
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 1.2459635734558105,
+      "learning_rate": 8.181818181818183e-06,
+      "loss": 0.7471,
+      "step": 4038
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 1.3551286458969116,
+      "learning_rate": 8.177648040033361e-06,
+      "loss": 0.7156,
+      "step": 4039
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 1.843151569366455,
+      "learning_rate": 8.173477898248541e-06,
+      "loss": 0.6114,
+      "step": 4040
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 1.3559409379959106,
+      "learning_rate": 8.16930775646372e-06,
+      "loss": 0.8308,
+      "step": 4041
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 1.3529644012451172,
+      "learning_rate": 8.1651376146789e-06,
+      "loss": 0.7349,
+      "step": 4042
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 1.9761481285095215,
+      "learning_rate": 8.16096747289408e-06,
+      "loss": 0.6383,
+      "step": 4043
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 1.4816498756408691,
+      "learning_rate": 8.156797331109258e-06,
+      "loss": 0.6025,
+      "step": 4044
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 1.2485196590423584,
+      "learning_rate": 8.152627189324436e-06,
+      "loss": 0.598,
+      "step": 4045
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 1.2901933193206787,
+      "learning_rate": 8.148457047539616e-06,
+      "loss": 0.6969,
+      "step": 4046
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 1.6511998176574707,
+      "learning_rate": 8.144286905754796e-06,
+      "loss": 0.7126,
+      "step": 4047
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 1.236632227897644,
+      "learning_rate": 8.140116763969975e-06,
+      "loss": 0.7123,
+      "step": 4048
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 1.117444634437561,
+      "learning_rate": 8.135946622185155e-06,
+      "loss": 0.7651,
+      "step": 4049
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 1.4278165102005005,
+      "learning_rate": 8.131776480400335e-06,
+      "loss": 0.746,
+      "step": 4050
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 1.3722543716430664,
+      "learning_rate": 8.127606338615513e-06,
+      "loss": 0.6405,
+      "step": 4051
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 1.6733298301696777,
+      "learning_rate": 8.123436196830693e-06,
+      "loss": 0.7232,
+      "step": 4052
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 1.2124762535095215,
+      "learning_rate": 8.119266055045872e-06,
+      "loss": 0.649,
+      "step": 4053
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 1.4258729219436646,
+      "learning_rate": 8.115095913261052e-06,
+      "loss": 0.7848,
+      "step": 4054
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 1.8051178455352783,
+      "learning_rate": 8.11092577147623e-06,
+      "loss": 0.8977,
+      "step": 4055
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 1.4249836206436157,
+      "learning_rate": 8.10675562969141e-06,
+      "loss": 0.664,
+      "step": 4056
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 1.1285873651504517,
+      "learning_rate": 8.102585487906588e-06,
+      "loss": 0.7577,
+      "step": 4057
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 1.891027808189392,
+      "learning_rate": 8.098415346121768e-06,
+      "loss": 0.8048,
+      "step": 4058
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 1.2006330490112305,
+      "learning_rate": 8.094245204336948e-06,
+      "loss": 0.8765,
+      "step": 4059
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 1.6823763847351074,
+      "learning_rate": 8.090075062552127e-06,
+      "loss": 0.8365,
+      "step": 4060
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 1.236812710762024,
+      "learning_rate": 8.085904920767307e-06,
+      "loss": 0.6771,
+      "step": 4061
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 1.1466010808944702,
+      "learning_rate": 8.081734778982485e-06,
+      "loss": 0.6461,
+      "step": 4062
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 1.853227972984314,
+      "learning_rate": 8.077564637197665e-06,
+      "loss": 0.5851,
+      "step": 4063
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 1.1324950456619263,
+      "learning_rate": 8.073394495412845e-06,
+      "loss": 0.5844,
+      "step": 4064
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 1.4123047590255737,
+      "learning_rate": 8.069224353628024e-06,
+      "loss": 0.6301,
+      "step": 4065
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 1.2132854461669922,
+      "learning_rate": 8.065054211843202e-06,
+      "loss": 0.7212,
+      "step": 4066
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 1.379394292831421,
+      "learning_rate": 8.060884070058382e-06,
+      "loss": 0.7104,
+      "step": 4067
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 1.1606208086013794,
+      "learning_rate": 8.056713928273562e-06,
+      "loss": 0.6059,
+      "step": 4068
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 1.2338708639144897,
+      "learning_rate": 8.05254378648874e-06,
+      "loss": 0.7739,
+      "step": 4069
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 1.2605102062225342,
+      "learning_rate": 8.04837364470392e-06,
+      "loss": 0.7205,
+      "step": 4070
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 1.2045714855194092,
+      "learning_rate": 8.0442035029191e-06,
+      "loss": 0.7695,
+      "step": 4071
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 1.4576596021652222,
+      "learning_rate": 8.040033361134279e-06,
+      "loss": 0.5976,
+      "step": 4072
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 1.3542989492416382,
+      "learning_rate": 8.035863219349459e-06,
+      "loss": 0.6175,
+      "step": 4073
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 1.817543625831604,
+      "learning_rate": 8.031693077564637e-06,
+      "loss": 0.6522,
+      "step": 4074
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 1.2661129236221313,
+      "learning_rate": 8.027522935779817e-06,
+      "loss": 0.6654,
+      "step": 4075
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 1.2577834129333496,
+      "learning_rate": 8.023352793994997e-06,
+      "loss": 0.7401,
+      "step": 4076
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 1.3352525234222412,
+      "learning_rate": 8.019182652210176e-06,
+      "loss": 0.8261,
+      "step": 4077
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 1.4374886751174927,
+      "learning_rate": 8.015012510425354e-06,
+      "loss": 0.8182,
+      "step": 4078
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 1.1310259103775024,
+      "learning_rate": 8.010842368640534e-06,
+      "loss": 0.7272,
+      "step": 4079
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 1.3984516859054565,
+      "learning_rate": 8.006672226855714e-06,
+      "loss": 0.8062,
+      "step": 4080
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 1.0844422578811646,
+      "learning_rate": 8.002502085070892e-06,
+      "loss": 0.6869,
+      "step": 4081
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 1.5735132694244385,
+      "learning_rate": 7.998331943286072e-06,
+      "loss": 0.7919,
+      "step": 4082
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 1.1832239627838135,
+      "learning_rate": 7.99416180150125e-06,
+      "loss": 0.6256,
+      "step": 4083
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 1.2862153053283691,
+      "learning_rate": 7.98999165971643e-06,
+      "loss": 0.6671,
+      "step": 4084
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 1.399275779724121,
+      "learning_rate": 7.98582151793161e-06,
+      "loss": 0.6625,
+      "step": 4085
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 1.5161666870117188,
+      "learning_rate": 7.981651376146789e-06,
+      "loss": 0.5437,
+      "step": 4086
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 1.5804264545440674,
+      "learning_rate": 7.977481234361969e-06,
+      "loss": 0.8329,
+      "step": 4087
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 1.2821407318115234,
+      "learning_rate": 7.973311092577147e-06,
+      "loss": 0.8995,
+      "step": 4088
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 1.3997056484222412,
+      "learning_rate": 7.969140950792327e-06,
+      "loss": 0.6828,
+      "step": 4089
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 1.1985435485839844,
+      "learning_rate": 7.964970809007506e-06,
+      "loss": 0.6269,
+      "step": 4090
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 1.3204777240753174,
+      "learning_rate": 7.960800667222686e-06,
+      "loss": 0.8216,
+      "step": 4091
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 1.1255851984024048,
+      "learning_rate": 7.956630525437866e-06,
+      "loss": 0.6771,
+      "step": 4092
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 1.0128809213638306,
+      "learning_rate": 7.952460383653044e-06,
+      "loss": 0.5821,
+      "step": 4093
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 1.311255693435669,
+      "learning_rate": 7.948290241868224e-06,
+      "loss": 0.7216,
+      "step": 4094
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 1.481868863105774,
+      "learning_rate": 7.944120100083403e-06,
+      "loss": 0.8287,
+      "step": 4095
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 1.1370958089828491,
+      "learning_rate": 7.939949958298583e-06,
+      "loss": 0.7117,
+      "step": 4096
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 1.6860215663909912,
+      "learning_rate": 7.935779816513763e-06,
+      "loss": 0.6729,
+      "step": 4097
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 1.8059720993041992,
+      "learning_rate": 7.931609674728941e-06,
+      "loss": 0.6416,
+      "step": 4098
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 1.1263718605041504,
+      "learning_rate": 7.927439532944121e-06,
+      "loss": 0.7189,
+      "step": 4099
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 1.3962024450302124,
+      "learning_rate": 7.9232693911593e-06,
+      "loss": 0.6085,
+      "step": 4100
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 1.22867751121521,
+      "learning_rate": 7.91909924937448e-06,
+      "loss": 0.8057,
+      "step": 4101
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 1.3705651760101318,
+      "learning_rate": 7.914929107589658e-06,
+      "loss": 0.7628,
+      "step": 4102
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 1.2708830833435059,
+      "learning_rate": 7.910758965804838e-06,
+      "loss": 0.8526,
+      "step": 4103
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 1.4503384828567505,
+      "learning_rate": 7.906588824020016e-06,
+      "loss": 0.7616,
+      "step": 4104
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 1.2485220432281494,
+      "learning_rate": 7.902418682235196e-06,
+      "loss": 0.6749,
+      "step": 4105
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 1.2287968397140503,
+      "learning_rate": 7.898248540450376e-06,
+      "loss": 0.8152,
+      "step": 4106
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 1.4547765254974365,
+      "learning_rate": 7.894078398665555e-06,
+      "loss": 0.8658,
+      "step": 4107
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 1.3261808156967163,
+      "learning_rate": 7.889908256880735e-06,
+      "loss": 0.6446,
+      "step": 4108
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 1.2493927478790283,
+      "learning_rate": 7.885738115095915e-06,
+      "loss": 0.6547,
+      "step": 4109
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 1.3406943082809448,
+      "learning_rate": 7.881567973311093e-06,
+      "loss": 0.7077,
+      "step": 4110
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 1.1841052770614624,
+      "learning_rate": 7.877397831526271e-06,
+      "loss": 0.763,
+      "step": 4111
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 1.1067297458648682,
+      "learning_rate": 7.873227689741451e-06,
+      "loss": 0.7358,
+      "step": 4112
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 1.1573944091796875,
+      "learning_rate": 7.869057547956631e-06,
+      "loss": 0.5685,
+      "step": 4113
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 1.2256602048873901,
+      "learning_rate": 7.86488740617181e-06,
+      "loss": 0.5798,
+      "step": 4114
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 1.2226641178131104,
+      "learning_rate": 7.86071726438699e-06,
+      "loss": 0.7897,
+      "step": 4115
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 1.1671439409255981,
+      "learning_rate": 7.856547122602168e-06,
+      "loss": 0.7308,
+      "step": 4116
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 0.9899999499320984,
+      "learning_rate": 7.852376980817348e-06,
+      "loss": 0.6807,
+      "step": 4117
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 1.2406564950942993,
+      "learning_rate": 7.848206839032528e-06,
+      "loss": 0.7886,
+      "step": 4118
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 1.2072851657867432,
+      "learning_rate": 7.844036697247707e-06,
+      "loss": 0.7571,
+      "step": 4119
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 1.3472397327423096,
+      "learning_rate": 7.839866555462887e-06,
+      "loss": 0.6397,
+      "step": 4120
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 1.2625677585601807,
+      "learning_rate": 7.835696413678067e-06,
+      "loss": 0.6698,
+      "step": 4121
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 1.5513256788253784,
+      "learning_rate": 7.831526271893245e-06,
+      "loss": 0.8605,
+      "step": 4122
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 1.3021574020385742,
+      "learning_rate": 7.827356130108423e-06,
+      "loss": 0.59,
+      "step": 4123
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 1.3085097074508667,
+      "learning_rate": 7.823185988323603e-06,
+      "loss": 0.7135,
+      "step": 4124
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 1.7146204710006714,
+      "learning_rate": 7.819015846538782e-06,
+      "loss": 0.7301,
+      "step": 4125
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 1.4580557346343994,
+      "learning_rate": 7.814845704753962e-06,
+      "loss": 0.5922,
+      "step": 4126
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 1.3994919061660767,
+      "learning_rate": 7.810675562969142e-06,
+      "loss": 0.7901,
+      "step": 4127
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 1.262799859046936,
+      "learning_rate": 7.80650542118432e-06,
+      "loss": 0.7913,
+      "step": 4128
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 1.7014191150665283,
+      "learning_rate": 7.8023352793995e-06,
+      "loss": 0.6946,
+      "step": 4129
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 1.3678264617919922,
+      "learning_rate": 7.79816513761468e-06,
+      "loss": 0.6873,
+      "step": 4130
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 1.4684674739837646,
+      "learning_rate": 7.793994995829859e-06,
+      "loss": 0.7323,
+      "step": 4131
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 1.1394392251968384,
+      "learning_rate": 7.789824854045039e-06,
+      "loss": 0.8703,
+      "step": 4132
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 0.9769808650016785,
+      "learning_rate": 7.785654712260217e-06,
+      "loss": 0.7576,
+      "step": 4133
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 1.3571994304656982,
+      "learning_rate": 7.781484570475397e-06,
+      "loss": 0.6476,
+      "step": 4134
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 1.3787708282470703,
+      "learning_rate": 7.777314428690575e-06,
+      "loss": 0.5787,
+      "step": 4135
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 1.407253384590149,
+      "learning_rate": 7.773144286905755e-06,
+      "loss": 0.5968,
+      "step": 4136
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 1.2448159456253052,
+      "learning_rate": 7.768974145120934e-06,
+      "loss": 0.6409,
+      "step": 4137
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 1.0467275381088257,
+      "learning_rate": 7.764804003336114e-06,
+      "loss": 0.6802,
+      "step": 4138
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 1.128260850906372,
+      "learning_rate": 7.760633861551294e-06,
+      "loss": 0.7138,
+      "step": 4139
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 1.056241750717163,
+      "learning_rate": 7.756463719766472e-06,
+      "loss": 0.6125,
+      "step": 4140
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 1.4507592916488647,
+      "learning_rate": 7.752293577981652e-06,
+      "loss": 0.6628,
+      "step": 4141
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 1.4795253276824951,
+      "learning_rate": 7.748123436196832e-06,
+      "loss": 0.6709,
+      "step": 4142
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 1.2754091024398804,
+      "learning_rate": 7.74395329441201e-06,
+      "loss": 0.6158,
+      "step": 4143
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 1.1661182641983032,
+      "learning_rate": 7.739783152627189e-06,
+      "loss": 0.6366,
+      "step": 4144
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 1.0788447856903076,
+      "learning_rate": 7.735613010842369e-06,
+      "loss": 0.6567,
+      "step": 4145
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 1.242462158203125,
+      "learning_rate": 7.731442869057547e-06,
+      "loss": 0.6916,
+      "step": 4146
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 1.7863349914550781,
+      "learning_rate": 7.727272727272727e-06,
+      "loss": 0.8489,
+      "step": 4147
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 1.997514247894287,
+      "learning_rate": 7.723102585487907e-06,
+      "loss": 0.6475,
+      "step": 4148
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 1.5353741645812988,
+      "learning_rate": 7.718932443703086e-06,
+      "loss": 0.4718,
+      "step": 4149
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 1.6328120231628418,
+      "learning_rate": 7.714762301918266e-06,
+      "loss": 0.5622,
+      "step": 4150
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 1.7414747476577759,
+      "learning_rate": 7.710592160133446e-06,
+      "loss": 0.7286,
+      "step": 4151
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 1.3826340436935425,
+      "learning_rate": 7.706422018348624e-06,
+      "loss": 0.6723,
+      "step": 4152
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 1.1786415576934814,
+      "learning_rate": 7.702251876563804e-06,
+      "loss": 0.7049,
+      "step": 4153
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 1.5943514108657837,
+      "learning_rate": 7.698081734778984e-06,
+      "loss": 0.7895,
+      "step": 4154
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 1.3432366847991943,
+      "learning_rate": 7.693911592994162e-06,
+      "loss": 0.8594,
+      "step": 4155
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 1.395645022392273,
+      "learning_rate": 7.68974145120934e-06,
+      "loss": 0.6955,
+      "step": 4156
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 1.2574201822280884,
+      "learning_rate": 7.68557130942452e-06,
+      "loss": 0.6216,
+      "step": 4157
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 1.43873929977417,
+      "learning_rate": 7.681401167639699e-06,
+      "loss": 0.7228,
+      "step": 4158
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 1.8955302238464355,
+      "learning_rate": 7.67723102585488e-06,
+      "loss": 0.6689,
+      "step": 4159
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 1.3884892463684082,
+      "learning_rate": 7.67306088407006e-06,
+      "loss": 0.6353,
+      "step": 4160
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 1.4339274168014526,
+      "learning_rate": 7.668890742285238e-06,
+      "loss": 0.7154,
+      "step": 4161
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 1.2962485551834106,
+      "learning_rate": 7.664720600500418e-06,
+      "loss": 0.6518,
+      "step": 4162
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 1.14412522315979,
+      "learning_rate": 7.660550458715598e-06,
+      "loss": 0.6968,
+      "step": 4163
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 1.1335933208465576,
+      "learning_rate": 7.656380316930776e-06,
+      "loss": 0.6509,
+      "step": 4164
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 1.2859175205230713,
+      "learning_rate": 7.652210175145956e-06,
+      "loss": 0.7919,
+      "step": 4165
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 1.3848085403442383,
+      "learning_rate": 7.648040033361134e-06,
+      "loss": 0.6549,
+      "step": 4166
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 1.1246529817581177,
+      "learning_rate": 7.643869891576313e-06,
+      "loss": 0.605,
+      "step": 4167
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 1.286983609199524,
+      "learning_rate": 7.639699749791493e-06,
+      "loss": 0.6526,
+      "step": 4168
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 1.1350047588348389,
+      "learning_rate": 7.635529608006673e-06,
+      "loss": 0.5825,
+      "step": 4169
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 1.4251058101654053,
+      "learning_rate": 7.631359466221851e-06,
+      "loss": 0.7617,
+      "step": 4170
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 1.2032486200332642,
+      "learning_rate": 7.627189324437031e-06,
+      "loss": 0.7039,
+      "step": 4171
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 1.334125280380249,
+      "learning_rate": 7.62301918265221e-06,
+      "loss": 0.7031,
+      "step": 4172
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 1.3071873188018799,
+      "learning_rate": 7.61884904086739e-06,
+      "loss": 0.7319,
+      "step": 4173
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 1.1474069356918335,
+      "learning_rate": 7.6146788990825695e-06,
+      "loss": 0.719,
+      "step": 4174
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 1.1719615459442139,
+      "learning_rate": 7.610508757297749e-06,
+      "loss": 0.7048,
+      "step": 4175
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 1.4841634035110474,
+      "learning_rate": 7.606338615512929e-06,
+      "loss": 0.7132,
+      "step": 4176
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 1.193687915802002,
+      "learning_rate": 7.602168473728106e-06,
+      "loss": 0.6186,
+      "step": 4177
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 1.4975613355636597,
+      "learning_rate": 7.597998331943286e-06,
+      "loss": 0.6453,
+      "step": 4178
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 1.264892339706421,
+      "learning_rate": 7.5938281901584655e-06,
+      "loss": 0.6619,
+      "step": 4179
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 1.7290844917297363,
+      "learning_rate": 7.589658048373645e-06,
+      "loss": 0.7351,
+      "step": 4180
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 1.3005568981170654,
+      "learning_rate": 7.585487906588825e-06,
+      "loss": 0.7248,
+      "step": 4181
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 1.3518257141113281,
+      "learning_rate": 7.581317764804004e-06,
+      "loss": 0.7722,
+      "step": 4182
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 1.853224277496338,
+      "learning_rate": 7.577147623019183e-06,
+      "loss": 0.6742,
+      "step": 4183
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 1.3090929985046387,
+      "learning_rate": 7.572977481234362e-06,
+      "loss": 0.7192,
+      "step": 4184
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 1.3071833848953247,
+      "learning_rate": 7.568807339449542e-06,
+      "loss": 0.6447,
+      "step": 4185
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 1.6736398935317993,
+      "learning_rate": 7.5646371976647215e-06,
+      "loss": 0.5652,
+      "step": 4186
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 1.0377569198608398,
+      "learning_rate": 7.560467055879901e-06,
+      "loss": 0.5316,
+      "step": 4187
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 1.1365035772323608,
+      "learning_rate": 7.556296914095079e-06,
+      "loss": 0.777,
+      "step": 4188
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 1.4174660444259644,
+      "learning_rate": 7.552126772310258e-06,
+      "loss": 0.6789,
+      "step": 4189
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 1.4120662212371826,
+      "learning_rate": 7.547956630525438e-06,
+      "loss": 0.7055,
+      "step": 4190
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 1.2537157535552979,
+      "learning_rate": 7.5437864887406174e-06,
+      "loss": 0.6143,
+      "step": 4191
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 1.7365707159042358,
+      "learning_rate": 7.539616346955797e-06,
+      "loss": 0.8578,
+      "step": 4192
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 1.385790467262268,
+      "learning_rate": 7.535446205170976e-06,
+      "loss": 0.6827,
+      "step": 4193
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 1.1795783042907715,
+      "learning_rate": 7.531276063386156e-06,
+      "loss": 0.5481,
+      "step": 4194
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 1.4059340953826904,
+      "learning_rate": 7.527105921601335e-06,
+      "loss": 0.657,
+      "step": 4195
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 1.4970194101333618,
+      "learning_rate": 7.522935779816514e-06,
+      "loss": 0.7697,
+      "step": 4196
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 1.3211995363235474,
+      "learning_rate": 7.518765638031694e-06,
+      "loss": 0.7945,
+      "step": 4197
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 1.497291922569275,
+      "learning_rate": 7.5145954962468735e-06,
+      "loss": 0.5842,
+      "step": 4198
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 1.3536211252212524,
+      "learning_rate": 7.510425354462052e-06,
+      "loss": 0.8181,
+      "step": 4199
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 1.2283011674880981,
+      "learning_rate": 7.506255212677231e-06,
+      "loss": 0.8476,
+      "step": 4200
+    },
+    {
+      "epoch": 4.06,
+      "eval_loss": 0.8597999811172485,
+      "eval_runtime": 861.3569,
+      "eval_samples_per_second": 4.799,
+      "eval_steps_per_second": 0.6,
+      "step": 4200
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 1.6468404531478882,
+      "learning_rate": 7.50208507089241e-06,
+      "loss": 0.8944,
+      "step": 4201
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 1.2757965326309204,
+      "learning_rate": 7.49791492910759e-06,
+      "loss": 0.5931,
+      "step": 4202
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 1.816794991493225,
+      "learning_rate": 7.493744787322769e-06,
+      "loss": 0.7,
+      "step": 4203
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 1.759100317955017,
+      "learning_rate": 7.489574645537949e-06,
+      "loss": 0.7633,
+      "step": 4204
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 1.4398976564407349,
+      "learning_rate": 7.485404503753128e-06,
+      "loss": 0.813,
+      "step": 4205
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 1.2360817193984985,
+      "learning_rate": 7.481234361968308e-06,
+      "loss": 0.6071,
+      "step": 4206
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 1.3215487003326416,
+      "learning_rate": 7.477064220183487e-06,
+      "loss": 0.7152,
+      "step": 4207
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 1.0810505151748657,
+      "learning_rate": 7.472894078398666e-06,
+      "loss": 0.6263,
+      "step": 4208
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 1.23736572265625,
+      "learning_rate": 7.468723936613846e-06,
+      "loss": 0.8244,
+      "step": 4209
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 1.5156235694885254,
+      "learning_rate": 7.464553794829024e-06,
+      "loss": 0.6425,
+      "step": 4210
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 1.2087029218673706,
+      "learning_rate": 7.460383653044204e-06,
+      "loss": 0.7957,
+      "step": 4211
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 1.286115288734436,
+      "learning_rate": 7.456213511259383e-06,
+      "loss": 0.5744,
+      "step": 4212
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 1.2504583597183228,
+      "learning_rate": 7.452043369474562e-06,
+      "loss": 0.6125,
+      "step": 4213
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 1.2574108839035034,
+      "learning_rate": 7.447873227689741e-06,
+      "loss": 0.6948,
+      "step": 4214
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 1.2414987087249756,
+      "learning_rate": 7.443703085904921e-06,
+      "loss": 0.5922,
+      "step": 4215
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 1.456743836402893,
+      "learning_rate": 7.4395329441201005e-06,
+      "loss": 0.6379,
+      "step": 4216
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 1.852477788925171,
+      "learning_rate": 7.43536280233528e-06,
+      "loss": 0.6489,
+      "step": 4217
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 2.1483681201934814,
+      "learning_rate": 7.43119266055046e-06,
+      "loss": 0.6942,
+      "step": 4218
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 1.3294167518615723,
+      "learning_rate": 7.427022518765639e-06,
+      "loss": 0.6328,
+      "step": 4219
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 1.067713975906372,
+      "learning_rate": 7.422852376980818e-06,
+      "loss": 0.7014,
+      "step": 4220
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 1.8215084075927734,
+      "learning_rate": 7.4186822351959965e-06,
+      "loss": 0.7014,
+      "step": 4221
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 1.0847300291061401,
+      "learning_rate": 7.414512093411176e-06,
+      "loss": 0.6824,
+      "step": 4222
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 1.9889841079711914,
+      "learning_rate": 7.410341951626356e-06,
+      "loss": 0.5876,
+      "step": 4223
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 1.2415003776550293,
+      "learning_rate": 7.406171809841535e-06,
+      "loss": 0.6037,
+      "step": 4224
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 1.4081982374191284,
+      "learning_rate": 7.402001668056714e-06,
+      "loss": 0.6475,
+      "step": 4225
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 2.0091793537139893,
+      "learning_rate": 7.397831526271893e-06,
+      "loss": 0.8408,
+      "step": 4226
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 1.2206426858901978,
+      "learning_rate": 7.393661384487073e-06,
+      "loss": 0.7029,
+      "step": 4227
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 1.1421000957489014,
+      "learning_rate": 7.3894912427022525e-06,
+      "loss": 0.7084,
+      "step": 4228
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 1.6849780082702637,
+      "learning_rate": 7.385321100917432e-06,
+      "loss": 0.7911,
+      "step": 4229
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 1.2689474821090698,
+      "learning_rate": 7.381150959132612e-06,
+      "loss": 0.5672,
+      "step": 4230
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 1.1003674268722534,
+      "learning_rate": 7.376980817347791e-06,
+      "loss": 0.7208,
+      "step": 4231
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 1.3360180854797363,
+      "learning_rate": 7.372810675562969e-06,
+      "loss": 0.6315,
+      "step": 4232
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 2.1548523902893066,
+      "learning_rate": 7.3686405337781484e-06,
+      "loss": 0.8421,
+      "step": 4233
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 1.4975672960281372,
+      "learning_rate": 7.364470391993328e-06,
+      "loss": 0.7763,
+      "step": 4234
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 1.7296528816223145,
+      "learning_rate": 7.360300250208507e-06,
+      "loss": 0.5566,
+      "step": 4235
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 1.404763102531433,
+      "learning_rate": 7.356130108423687e-06,
+      "loss": 0.6582,
+      "step": 4236
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 1.8087592124938965,
+      "learning_rate": 7.351959966638866e-06,
+      "loss": 0.7118,
+      "step": 4237
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 1.3908724784851074,
+      "learning_rate": 7.347789824854045e-06,
+      "loss": 0.7868,
+      "step": 4238
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 1.4540789127349854,
+      "learning_rate": 7.343619683069225e-06,
+      "loss": 0.611,
+      "step": 4239
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 1.3166730403900146,
+      "learning_rate": 7.3394495412844045e-06,
+      "loss": 0.7433,
+      "step": 4240
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 1.3730759620666504,
+      "learning_rate": 7.335279399499584e-06,
+      "loss": 0.741,
+      "step": 4241
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 1.1578608751296997,
+      "learning_rate": 7.331109257714764e-06,
+      "loss": 0.6909,
+      "step": 4242
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 1.2952829599380493,
+      "learning_rate": 7.326939115929941e-06,
+      "loss": 0.7476,
+      "step": 4243
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 1.559354305267334,
+      "learning_rate": 7.322768974145121e-06,
+      "loss": 0.7231,
+      "step": 4244
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 1.7259479761123657,
+      "learning_rate": 7.3185988323603e-06,
+      "loss": 0.7192,
+      "step": 4245
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 1.373360276222229,
+      "learning_rate": 7.31442869057548e-06,
+      "loss": 0.7247,
+      "step": 4246
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 1.2092353105545044,
+      "learning_rate": 7.310258548790659e-06,
+      "loss": 0.6721,
+      "step": 4247
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 1.6677278280258179,
+      "learning_rate": 7.306088407005839e-06,
+      "loss": 0.6647,
+      "step": 4248
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 1.4248433113098145,
+      "learning_rate": 7.301918265221018e-06,
+      "loss": 0.6858,
+      "step": 4249
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 1.6589043140411377,
+      "learning_rate": 7.297748123436197e-06,
+      "loss": 0.7623,
+      "step": 4250
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 1.4277273416519165,
+      "learning_rate": 7.293577981651377e-06,
+      "loss": 0.8315,
+      "step": 4251
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 1.2882044315338135,
+      "learning_rate": 7.289407839866556e-06,
+      "loss": 0.68,
+      "step": 4252
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 1.6518248319625854,
+      "learning_rate": 7.285237698081736e-06,
+      "loss": 0.6924,
+      "step": 4253
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 1.9300150871276855,
+      "learning_rate": 7.281067556296914e-06,
+      "loss": 0.729,
+      "step": 4254
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 1.2841802835464478,
+      "learning_rate": 7.276897414512093e-06,
+      "loss": 0.5875,
+      "step": 4255
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 1.8010226488113403,
+      "learning_rate": 7.272727272727272e-06,
+      "loss": 0.6811,
+      "step": 4256
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 1.115170955657959,
+      "learning_rate": 7.268557130942452e-06,
+      "loss": 0.6961,
+      "step": 4257
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 1.9197747707366943,
+      "learning_rate": 7.2643869891576315e-06,
+      "loss": 0.6216,
+      "step": 4258
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 1.5886878967285156,
+      "learning_rate": 7.260216847372811e-06,
+      "loss": 0.6624,
+      "step": 4259
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 1.5277332067489624,
+      "learning_rate": 7.256046705587991e-06,
+      "loss": 0.7778,
+      "step": 4260
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 1.5861066579818726,
+      "learning_rate": 7.25187656380317e-06,
+      "loss": 0.7441,
+      "step": 4261
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 1.7800103425979614,
+      "learning_rate": 7.247706422018349e-06,
+      "loss": 0.7026,
+      "step": 4262
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 1.6363444328308105,
+      "learning_rate": 7.243536280233529e-06,
+      "loss": 0.7031,
+      "step": 4263
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 1.7111765146255493,
+      "learning_rate": 7.239366138448708e-06,
+      "loss": 0.6233,
+      "step": 4264
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 1.126967191696167,
+      "learning_rate": 7.235195996663887e-06,
+      "loss": 0.5954,
+      "step": 4265
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 1.6149272918701172,
+      "learning_rate": 7.231025854879066e-06,
+      "loss": 0.7709,
+      "step": 4266
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 1.5248738527297974,
+      "learning_rate": 7.226855713094245e-06,
+      "loss": 0.6216,
+      "step": 4267
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 1.8891106843948364,
+      "learning_rate": 7.222685571309424e-06,
+      "loss": 0.8195,
+      "step": 4268
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 1.798587441444397,
+      "learning_rate": 7.218515429524604e-06,
+      "loss": 0.6679,
+      "step": 4269
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 1.6670140027999878,
+      "learning_rate": 7.2143452877397835e-06,
+      "loss": 0.6251,
+      "step": 4270
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 1.3214702606201172,
+      "learning_rate": 7.210175145954963e-06,
+      "loss": 0.5787,
+      "step": 4271
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 2.0496490001678467,
+      "learning_rate": 7.206005004170143e-06,
+      "loss": 0.6629,
+      "step": 4272
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 1.1761988401412964,
+      "learning_rate": 7.201834862385322e-06,
+      "loss": 0.7753,
+      "step": 4273
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 1.1243436336517334,
+      "learning_rate": 7.197664720600501e-06,
+      "loss": 0.602,
+      "step": 4274
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 1.3555951118469238,
+      "learning_rate": 7.193494578815681e-06,
+      "loss": 0.5541,
+      "step": 4275
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 1.6746872663497925,
+      "learning_rate": 7.189324437030859e-06,
+      "loss": 0.7901,
+      "step": 4276
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 1.5441886186599731,
+      "learning_rate": 7.185154295246038e-06,
+      "loss": 0.7559,
+      "step": 4277
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 1.4386554956436157,
+      "learning_rate": 7.180984153461218e-06,
+      "loss": 0.6865,
+      "step": 4278
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 1.2164535522460938,
+      "learning_rate": 7.176814011676397e-06,
+      "loss": 0.7552,
+      "step": 4279
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 1.222366213798523,
+      "learning_rate": 7.172643869891576e-06,
+      "loss": 0.506,
+      "step": 4280
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 1.8190783262252808,
+      "learning_rate": 7.168473728106756e-06,
+      "loss": 0.8186,
+      "step": 4281
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 1.2464604377746582,
+      "learning_rate": 7.1643035863219355e-06,
+      "loss": 0.8343,
+      "step": 4282
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 1.315718650817871,
+      "learning_rate": 7.160133444537115e-06,
+      "loss": 0.6059,
+      "step": 4283
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 1.3518524169921875,
+      "learning_rate": 7.155963302752295e-06,
+      "loss": 0.634,
+      "step": 4284
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 1.4598798751831055,
+      "learning_rate": 7.151793160967474e-06,
+      "loss": 0.7169,
+      "step": 4285
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 1.5651257038116455,
+      "learning_rate": 7.147623019182653e-06,
+      "loss": 0.7569,
+      "step": 4286
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 1.135716199874878,
+      "learning_rate": 7.143452877397831e-06,
+      "loss": 0.5594,
+      "step": 4287
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 1.3022222518920898,
+      "learning_rate": 7.139282735613011e-06,
+      "loss": 0.6758,
+      "step": 4288
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 1.3710997104644775,
+      "learning_rate": 7.13511259382819e-06,
+      "loss": 0.6633,
+      "step": 4289
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 1.335264801979065,
+      "learning_rate": 7.13094245204337e-06,
+      "loss": 0.8073,
+      "step": 4290
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 1.6815357208251953,
+      "learning_rate": 7.126772310258549e-06,
+      "loss": 0.6771,
+      "step": 4291
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 1.7518526315689087,
+      "learning_rate": 7.122602168473728e-06,
+      "loss": 0.6287,
+      "step": 4292
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 1.338187336921692,
+      "learning_rate": 7.118432026688908e-06,
+      "loss": 0.5854,
+      "step": 4293
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 1.319826364517212,
+      "learning_rate": 7.114261884904087e-06,
+      "loss": 0.5895,
+      "step": 4294
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 1.6188857555389404,
+      "learning_rate": 7.110091743119267e-06,
+      "loss": 0.909,
+      "step": 4295
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 1.8238894939422607,
+      "learning_rate": 7.105921601334447e-06,
+      "loss": 0.6531,
+      "step": 4296
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 1.8153358697891235,
+      "learning_rate": 7.101751459549626e-06,
+      "loss": 0.7412,
+      "step": 4297
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 1.539629340171814,
+      "learning_rate": 7.097581317764803e-06,
+      "loss": 0.6625,
+      "step": 4298
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 1.680781364440918,
+      "learning_rate": 7.093411175979983e-06,
+      "loss": 0.5307,
+      "step": 4299
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 1.341411828994751,
+      "learning_rate": 7.0892410341951625e-06,
+      "loss": 0.7209,
+      "step": 4300
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 1.168107032775879,
+      "learning_rate": 7.085070892410342e-06,
+      "loss": 0.593,
+      "step": 4301
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 1.2601464986801147,
+      "learning_rate": 7.080900750625522e-06,
+      "loss": 0.6123,
+      "step": 4302
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 1.542240023612976,
+      "learning_rate": 7.076730608840701e-06,
+      "loss": 0.6779,
+      "step": 4303
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 1.5732543468475342,
+      "learning_rate": 7.07256046705588e-06,
+      "loss": 0.713,
+      "step": 4304
+    },
+    {
+      "epoch": 4.17,
+      "grad_norm": 1.2263343334197998,
+      "learning_rate": 7.06839032527106e-06,
+      "loss": 0.6758,
+      "step": 4305
+    },
+    {
+      "epoch": 4.17,
+      "grad_norm": 1.4326815605163574,
+      "learning_rate": 7.064220183486239e-06,
+      "loss": 0.5931,
+      "step": 4306
+    },
+    {
+      "epoch": 4.17,
+      "grad_norm": 1.255967617034912,
+      "learning_rate": 7.0600500417014186e-06,
+      "loss": 0.6113,
+      "step": 4307
+    },
+    {
+      "epoch": 4.17,
+      "grad_norm": 1.4633044004440308,
+      "learning_rate": 7.055879899916598e-06,
+      "loss": 0.7899,
+      "step": 4308
+    },
+    {
+      "epoch": 4.17,
+      "grad_norm": 1.2892683744430542,
+      "learning_rate": 7.051709758131776e-06,
+      "loss": 0.7507,
+      "step": 4309
+    },
+    {
+      "epoch": 4.17,
+      "grad_norm": 2.049952268600464,
+      "learning_rate": 7.047539616346955e-06,
+      "loss": 0.6878,
+      "step": 4310
+    },
+    {
+      "epoch": 4.17,
+      "grad_norm": 2.016397714614868,
+      "learning_rate": 7.043369474562135e-06,
+      "loss": 0.832,
+      "step": 4311
+    },
+    {
+      "epoch": 4.17,
+      "grad_norm": 1.3233511447906494,
+      "learning_rate": 7.0391993327773145e-06,
+      "loss": 0.6509,
+      "step": 4312
+    },
+    {
+      "epoch": 4.17,
+      "grad_norm": 1.4653106927871704,
+      "learning_rate": 7.035029190992494e-06,
+      "loss": 0.6382,
+      "step": 4313
+    },
+    {
+      "epoch": 4.17,
+      "grad_norm": 1.0743006467819214,
+      "learning_rate": 7.030859049207674e-06,
+      "loss": 0.5874,
+      "step": 4314
+    },
+    {
+      "epoch": 4.18,
+      "grad_norm": 1.4268132448196411,
+      "learning_rate": 7.026688907422853e-06,
+      "loss": 0.7449,
+      "step": 4315
+    },
+    {
+      "epoch": 4.18,
+      "grad_norm": 1.418232798576355,
+      "learning_rate": 7.022518765638032e-06,
+      "loss": 0.7174,
+      "step": 4316
+    },
+    {
+      "epoch": 4.18,
+      "grad_norm": 1.6750450134277344,
+      "learning_rate": 7.018348623853212e-06,
+      "loss": 0.7305,
+      "step": 4317
+    },
+    {
+      "epoch": 4.18,
+      "grad_norm": 1.2780274152755737,
+      "learning_rate": 7.014178482068391e-06,
+      "loss": 0.6194,
+      "step": 4318
+    },
+    {
+      "epoch": 4.18,
+      "grad_norm": 1.248227596282959,
+      "learning_rate": 7.0100083402835705e-06,
+      "loss": 0.7412,
+      "step": 4319
+    },
+    {
+      "epoch": 4.18,
+      "grad_norm": 1.4439641237258911,
+      "learning_rate": 7.005838198498749e-06,
+      "loss": 0.6121,
+      "step": 4320
+    },
+    {
+      "epoch": 4.18,
+      "grad_norm": 1.3450074195861816,
+      "learning_rate": 7.001668056713928e-06,
+      "loss": 0.7011,
+      "step": 4321
+    },
+    {
+      "epoch": 4.18,
+      "grad_norm": 1.401066780090332,
+      "learning_rate": 6.997497914929107e-06,
+      "loss": 0.7435,
+      "step": 4322
+    },
+    {
+      "epoch": 4.18,
+      "grad_norm": 1.3789832592010498,
+      "learning_rate": 6.993327773144287e-06,
+      "loss": 0.7078,
+      "step": 4323
+    },
+    {
+      "epoch": 4.18,
+      "grad_norm": 1.165249228477478,
+      "learning_rate": 6.9891576313594665e-06,
+      "loss": 0.6393,
+      "step": 4324
+    },
+    {
+      "epoch": 4.18,
+      "grad_norm": 1.4255436658859253,
+      "learning_rate": 6.984987489574646e-06,
+      "loss": 0.7127,
+      "step": 4325
+    },
+    {
+      "epoch": 4.19,
+      "grad_norm": 1.549740195274353,
+      "learning_rate": 6.980817347789826e-06,
+      "loss": 0.5712,
+      "step": 4326
+    },
+    {
+      "epoch": 4.19,
+      "grad_norm": 1.22221839427948,
+      "learning_rate": 6.976647206005005e-06,
+      "loss": 0.5935,
+      "step": 4327
+    },
+    {
+      "epoch": 4.19,
+      "grad_norm": 1.5324398279190063,
+      "learning_rate": 6.972477064220184e-06,
+      "loss": 0.6328,
+      "step": 4328
+    },
+    {
+      "epoch": 4.19,
+      "grad_norm": 1.7590183019638062,
+      "learning_rate": 6.968306922435363e-06,
+      "loss": 0.609,
+      "step": 4329
+    },
+    {
+      "epoch": 4.19,
+      "grad_norm": 1.6595736742019653,
+      "learning_rate": 6.964136780650543e-06,
+      "loss": 0.7005,
+      "step": 4330
+    },
+    {
+      "epoch": 4.19,
+      "grad_norm": 1.2330102920532227,
+      "learning_rate": 6.959966638865721e-06,
+      "loss": 0.5565,
+      "step": 4331
+    },
+    {
+      "epoch": 4.19,
+      "grad_norm": 1.4129624366760254,
+      "learning_rate": 6.955796497080901e-06,
+      "loss": 0.7066,
+      "step": 4332
+    },
+    {
+      "epoch": 4.19,
+      "grad_norm": 1.362594723701477,
+      "learning_rate": 6.95162635529608e-06,
+      "loss": 0.7155,
+      "step": 4333
+    },
+    {
+      "epoch": 4.19,
+      "grad_norm": 1.6006284952163696,
+      "learning_rate": 6.947456213511259e-06,
+      "loss": 0.8129,
+      "step": 4334
+    },
+    {
+      "epoch": 4.19,
+      "grad_norm": 1.730517029762268,
+      "learning_rate": 6.943286071726439e-06,
+      "loss": 0.6247,
+      "step": 4335
+    },
+    {
+      "epoch": 4.2,
+      "grad_norm": 1.2271586656570435,
+      "learning_rate": 6.939115929941618e-06,
+      "loss": 0.5687,
+      "step": 4336
+    },
+    {
+      "epoch": 4.2,
+      "grad_norm": 1.6972774267196655,
+      "learning_rate": 6.934945788156798e-06,
+      "loss": 0.7686,
+      "step": 4337
+    },
+    {
+      "epoch": 4.2,
+      "grad_norm": 1.1939290761947632,
+      "learning_rate": 6.930775646371978e-06,
+      "loss": 0.6312,
+      "step": 4338
+    },
+    {
+      "epoch": 4.2,
+      "grad_norm": 1.2736120223999023,
+      "learning_rate": 6.926605504587157e-06,
+      "loss": 0.6084,
+      "step": 4339
+    },
+    {
+      "epoch": 4.2,
+      "grad_norm": 1.7349547147750854,
+      "learning_rate": 6.922435362802336e-06,
+      "loss": 0.5477,
+      "step": 4340
+    },
+    {
+      "epoch": 4.2,
+      "grad_norm": 1.8467882871627808,
+      "learning_rate": 6.918265221017515e-06,
+      "loss": 0.6469,
+      "step": 4341
+    },
+    {
+      "epoch": 4.2,
+      "grad_norm": 1.3428188562393188,
+      "learning_rate": 6.9140950792326936e-06,
+      "loss": 0.6622,
+      "step": 4342
+    },
+    {
+      "epoch": 4.2,
+      "grad_norm": 1.5804131031036377,
+      "learning_rate": 6.909924937447873e-06,
+      "loss": 0.8024,
+      "step": 4343
+    },
+    {
+      "epoch": 4.2,
+      "grad_norm": 1.5460153818130493,
+      "learning_rate": 6.905754795663053e-06,
+      "loss": 0.7976,
+      "step": 4344
+    },
+    {
+      "epoch": 4.2,
+      "grad_norm": 1.3329187631607056,
+      "learning_rate": 6.901584653878232e-06,
+      "loss": 0.6699,
+      "step": 4345
+    },
+    {
+      "epoch": 4.21,
+      "grad_norm": 1.2377336025238037,
+      "learning_rate": 6.897414512093411e-06,
+      "loss": 0.6386,
+      "step": 4346
+    },
+    {
+      "epoch": 4.21,
+      "grad_norm": 1.1283926963806152,
+      "learning_rate": 6.893244370308591e-06,
+      "loss": 0.6808,
+      "step": 4347
+    },
+    {
+      "epoch": 4.21,
+      "grad_norm": 1.1400607824325562,
+      "learning_rate": 6.88907422852377e-06,
+      "loss": 0.6433,
+      "step": 4348
+    },
+    {
+      "epoch": 4.21,
+      "grad_norm": 1.4880428314208984,
+      "learning_rate": 6.8849040867389496e-06,
+      "loss": 0.7377,
+      "step": 4349
+    },
+    {
+      "epoch": 4.21,
+      "grad_norm": 1.3610410690307617,
+      "learning_rate": 6.880733944954129e-06,
+      "loss": 0.6365,
+      "step": 4350
+    },
+    {
+      "epoch": 4.21,
+      "grad_norm": 1.3585680723190308,
+      "learning_rate": 6.876563803169309e-06,
+      "loss": 0.7175,
+      "step": 4351
+    },
+    {
+      "epoch": 4.21,
+      "grad_norm": 1.3938407897949219,
+      "learning_rate": 6.872393661384488e-06,
+      "loss": 0.7103,
+      "step": 4352
+    },
+    {
+      "epoch": 4.21,
+      "grad_norm": 1.8672634363174438,
+      "learning_rate": 6.868223519599666e-06,
+      "loss": 0.785,
+      "step": 4353
+    },
+    {
+      "epoch": 4.21,
+      "grad_norm": 1.3515386581420898,
+      "learning_rate": 6.8640533778148455e-06,
+      "loss": 0.6009,
+      "step": 4354
+    },
+    {
+      "epoch": 4.21,
+      "grad_norm": 1.4289917945861816,
+      "learning_rate": 6.859883236030025e-06,
+      "loss": 0.6041,
+      "step": 4355
+    },
+    {
+      "epoch": 4.21,
+      "grad_norm": 1.355471134185791,
+      "learning_rate": 6.855713094245205e-06,
+      "loss": 0.8069,
+      "step": 4356
+    },
+    {
+      "epoch": 4.22,
+      "grad_norm": 1.3696022033691406,
+      "learning_rate": 6.851542952460384e-06,
+      "loss": 0.8306,
+      "step": 4357
+    },
+    {
+      "epoch": 4.22,
+      "grad_norm": 1.4903416633605957,
+      "learning_rate": 6.847372810675563e-06,
+      "loss": 0.6864,
+      "step": 4358
+    },
+    {
+      "epoch": 4.22,
+      "grad_norm": 1.4738951921463013,
+      "learning_rate": 6.843202668890743e-06,
+      "loss": 0.5804,
+      "step": 4359
+    },
+    {
+      "epoch": 4.22,
+      "grad_norm": 1.2567520141601562,
+      "learning_rate": 6.839032527105922e-06,
+      "loss": 0.6282,
+      "step": 4360
+    },
+    {
+      "epoch": 4.22,
+      "grad_norm": 1.3491474390029907,
+      "learning_rate": 6.8348623853211015e-06,
+      "loss": 0.6883,
+      "step": 4361
+    },
+    {
+      "epoch": 4.22,
+      "grad_norm": 1.539475440979004,
+      "learning_rate": 6.830692243536281e-06,
+      "loss": 0.6455,
+      "step": 4362
+    },
+    {
+      "epoch": 4.22,
+      "grad_norm": 1.4578179121017456,
+      "learning_rate": 6.826522101751461e-06,
+      "loss": 0.7266,
+      "step": 4363
+    },
+    {
+      "epoch": 4.22,
+      "grad_norm": 1.6439467668533325,
+      "learning_rate": 6.822351959966638e-06,
+      "loss": 0.7019,
+      "step": 4364
+    },
+    {
+      "epoch": 4.22,
+      "grad_norm": 1.4401473999023438,
+      "learning_rate": 6.818181818181818e-06,
+      "loss": 0.7081,
+      "step": 4365
+    },
+    {
+      "epoch": 4.22,
+      "grad_norm": 1.6413389444351196,
+      "learning_rate": 6.8140116763969975e-06,
+      "loss": 0.7683,
+      "step": 4366
+    },
+    {
+      "epoch": 4.23,
+      "grad_norm": 1.5531343221664429,
+      "learning_rate": 6.809841534612177e-06,
+      "loss": 0.7655,
+      "step": 4367
+    },
+    {
+      "epoch": 4.23,
+      "grad_norm": 1.2082351446151733,
+      "learning_rate": 6.805671392827357e-06,
+      "loss": 0.6311,
+      "step": 4368
+    },
+    {
+      "epoch": 4.23,
+      "grad_norm": 1.2384322881698608,
+      "learning_rate": 6.801501251042536e-06,
+      "loss": 0.7649,
+      "step": 4369
+    },
+    {
+      "epoch": 4.23,
+      "grad_norm": 1.3738021850585938,
+      "learning_rate": 6.797331109257715e-06,
+      "loss": 0.8453,
+      "step": 4370
+    },
+    {
+      "epoch": 4.23,
+      "grad_norm": 1.4350394010543823,
+      "learning_rate": 6.793160967472894e-06,
+      "loss": 0.9358,
+      "step": 4371
+    },
+    {
+      "epoch": 4.23,
+      "grad_norm": 1.2021479606628418,
+      "learning_rate": 6.788990825688074e-06,
+      "loss": 0.5609,
+      "step": 4372
+    },
+    {
+      "epoch": 4.23,
+      "grad_norm": 1.453177571296692,
+      "learning_rate": 6.7848206839032535e-06,
+      "loss": 0.6782,
+      "step": 4373
+    },
+    {
+      "epoch": 4.23,
+      "grad_norm": 1.809272050857544,
+      "learning_rate": 6.780650542118433e-06,
+      "loss": 0.6943,
+      "step": 4374
+    },
+    {
+      "epoch": 4.23,
+      "grad_norm": 1.274755597114563,
+      "learning_rate": 6.776480400333611e-06,
+      "loss": 0.8202,
+      "step": 4375
+    },
+    {
+      "epoch": 4.23,
+      "grad_norm": 1.4974430799484253,
+      "learning_rate": 6.77231025854879e-06,
+      "loss": 0.6515,
+      "step": 4376
+    },
+    {
+      "epoch": 4.24,
+      "grad_norm": 1.6731135845184326,
+      "learning_rate": 6.76814011676397e-06,
+      "loss": 0.6141,
+      "step": 4377
+    },
+    {
+      "epoch": 4.24,
+      "grad_norm": 1.4311158657073975,
+      "learning_rate": 6.763969974979149e-06,
+      "loss": 0.5637,
+      "step": 4378
+    },
+    {
+      "epoch": 4.24,
+      "grad_norm": 1.4362891912460327,
+      "learning_rate": 6.759799833194329e-06,
+      "loss": 0.6189,
+      "step": 4379
+    },
+    {
+      "epoch": 4.24,
+      "grad_norm": 1.370519757270813,
+      "learning_rate": 6.755629691409509e-06,
+      "loss": 0.6415,
+      "step": 4380
+    },
+    {
+      "epoch": 4.24,
+      "grad_norm": 1.4324407577514648,
+      "learning_rate": 6.751459549624688e-06,
+      "loss": 0.6882,
+      "step": 4381
+    },
+    {
+      "epoch": 4.24,
+      "grad_norm": 1.5416909456253052,
+      "learning_rate": 6.747289407839867e-06,
+      "loss": 0.7743,
+      "step": 4382
+    },
+    {
+      "epoch": 4.24,
+      "grad_norm": 1.38045334815979,
+      "learning_rate": 6.743119266055046e-06,
+      "loss": 0.838,
+      "step": 4383
+    },
+    {
+      "epoch": 4.24,
+      "grad_norm": 1.2203946113586426,
+      "learning_rate": 6.738949124270226e-06,
+      "loss": 0.6709,
+      "step": 4384
+    },
+    {
+      "epoch": 4.24,
+      "grad_norm": 1.570999026298523,
+      "learning_rate": 6.7347789824854054e-06,
+      "loss": 0.8299,
+      "step": 4385
+    },
+    {
+      "epoch": 4.24,
+      "grad_norm": 1.1222865581512451,
+      "learning_rate": 6.730608840700584e-06,
+      "loss": 0.4957,
+      "step": 4386
+    },
+    {
+      "epoch": 4.24,
+      "grad_norm": 1.0991058349609375,
+      "learning_rate": 6.726438698915763e-06,
+      "loss": 0.6796,
+      "step": 4387
+    },
+    {
+      "epoch": 4.25,
+      "grad_norm": 1.5178134441375732,
+      "learning_rate": 6.722268557130942e-06,
+      "loss": 0.6211,
+      "step": 4388
+    },
+    {
+      "epoch": 4.25,
+      "grad_norm": 1.367569923400879,
+      "learning_rate": 6.718098415346122e-06,
+      "loss": 0.7229,
+      "step": 4389
+    },
+    {
+      "epoch": 4.25,
+      "grad_norm": 1.4267053604125977,
+      "learning_rate": 6.713928273561301e-06,
+      "loss": 0.7702,
+      "step": 4390
+    },
+    {
+      "epoch": 4.25,
+      "grad_norm": 1.4790523052215576,
+      "learning_rate": 6.7097581317764806e-06,
+      "loss": 0.7489,
+      "step": 4391
+    },
+    {
+      "epoch": 4.25,
+      "grad_norm": 1.9144536256790161,
+      "learning_rate": 6.70558798999166e-06,
+      "loss": 0.6801,
+      "step": 4392
+    },
+    {
+      "epoch": 4.25,
+      "grad_norm": 1.2850652933120728,
+      "learning_rate": 6.70141784820684e-06,
+      "loss": 0.7403,
+      "step": 4393
+    },
+    {
+      "epoch": 4.25,
+      "grad_norm": 1.5940477848052979,
+      "learning_rate": 6.697247706422019e-06,
+      "loss": 0.6171,
+      "step": 4394
+    },
+    {
+      "epoch": 4.25,
+      "grad_norm": 1.6070002317428589,
+      "learning_rate": 6.693077564637198e-06,
+      "loss": 0.6765,
+      "step": 4395
+    },
+    {
+      "epoch": 4.25,
+      "grad_norm": 1.9825026988983154,
+      "learning_rate": 6.688907422852378e-06,
+      "loss": 0.6512,
+      "step": 4396
+    },
+    {
+      "epoch": 4.25,
+      "grad_norm": 1.339361548423767,
+      "learning_rate": 6.684737281067556e-06,
+      "loss": 0.7988,
+      "step": 4397
+    },
+    {
+      "epoch": 4.26,
+      "grad_norm": 1.8375968933105469,
+      "learning_rate": 6.680567139282736e-06,
+      "loss": 0.7436,
+      "step": 4398
+    },
+    {
+      "epoch": 4.26,
+      "grad_norm": 1.5017917156219482,
+      "learning_rate": 6.676396997497915e-06,
+      "loss": 0.7136,
+      "step": 4399
+    },
+    {
+      "epoch": 4.26,
+      "grad_norm": 1.284619927406311,
+      "learning_rate": 6.672226855713094e-06,
+      "loss": 0.6345,
+      "step": 4400
+    },
+    {
+      "epoch": 4.26,
+      "grad_norm": 1.629049301147461,
+      "learning_rate": 6.668056713928274e-06,
+      "loss": 0.6055,
+      "step": 4401
+    },
+    {
+      "epoch": 4.26,
+      "grad_norm": 2.0935773849487305,
+      "learning_rate": 6.663886572143453e-06,
+      "loss": 0.6613,
+      "step": 4402
+    },
+    {
+      "epoch": 4.26,
+      "grad_norm": 1.2761112451553345,
+      "learning_rate": 6.6597164303586325e-06,
+      "loss": 0.9318,
+      "step": 4403
+    },
+    {
+      "epoch": 4.26,
+      "grad_norm": 1.524101972579956,
+      "learning_rate": 6.655546288573812e-06,
+      "loss": 0.6462,
+      "step": 4404
+    },
+    {
+      "epoch": 4.26,
+      "grad_norm": 1.6486200094223022,
+      "learning_rate": 6.651376146788992e-06,
+      "loss": 0.7126,
+      "step": 4405
+    },
+    {
+      "epoch": 4.26,
+      "grad_norm": 1.5298010110855103,
+      "learning_rate": 6.647206005004171e-06,
+      "loss": 0.8588,
+      "step": 4406
+    },
+    {
+      "epoch": 4.26,
+      "grad_norm": 1.5817596912384033,
+      "learning_rate": 6.64303586321935e-06,
+      "loss": 0.7953,
+      "step": 4407
+    },
+    {
+      "epoch": 4.27,
+      "grad_norm": 1.8760782480239868,
+      "learning_rate": 6.6388657214345285e-06,
+      "loss": 0.5902,
+      "step": 4408
+    },
+    {
+      "epoch": 4.27,
+      "grad_norm": 1.343523383140564,
+      "learning_rate": 6.634695579649708e-06,
+      "loss": 0.5988,
+      "step": 4409
+    },
+    {
+      "epoch": 4.27,
+      "grad_norm": 1.6053376197814941,
+      "learning_rate": 6.630525437864888e-06,
+      "loss": 0.6569,
+      "step": 4410
+    },
+    {
+      "epoch": 4.27,
+      "grad_norm": 1.8453056812286377,
+      "learning_rate": 6.626355296080067e-06,
+      "loss": 0.7616,
+      "step": 4411
+    },
+    {
+      "epoch": 4.27,
+      "grad_norm": 1.554426908493042,
+      "learning_rate": 6.622185154295246e-06,
+      "loss": 0.7056,
+      "step": 4412
+    },
+    {
+      "epoch": 4.27,
+      "grad_norm": 1.5776360034942627,
+      "learning_rate": 6.618015012510425e-06,
+      "loss": 0.7216,
+      "step": 4413
+    },
+    {
+      "epoch": 4.27,
+      "grad_norm": 1.8021432161331177,
+      "learning_rate": 6.613844870725605e-06,
+      "loss": 0.5638,
+      "step": 4414
+    },
+    {
+      "epoch": 4.27,
+      "grad_norm": 1.8947653770446777,
+      "learning_rate": 6.6096747289407845e-06,
+      "loss": 0.631,
+      "step": 4415
+    },
+    {
+      "epoch": 4.27,
+      "grad_norm": 1.4564845561981201,
+      "learning_rate": 6.605504587155964e-06,
+      "loss": 0.7454,
+      "step": 4416
+    },
+    {
+      "epoch": 4.27,
+      "grad_norm": 1.5228424072265625,
+      "learning_rate": 6.601334445371144e-06,
+      "loss": 0.5376,
+      "step": 4417
+    },
+    {
+      "epoch": 4.27,
+      "grad_norm": 1.2071088552474976,
+      "learning_rate": 6.597164303586323e-06,
+      "loss": 0.6799,
+      "step": 4418
+    },
+    {
+      "epoch": 4.28,
+      "grad_norm": 1.364287257194519,
+      "learning_rate": 6.592994161801501e-06,
+      "loss": 0.6515,
+      "step": 4419
+    },
+    {
+      "epoch": 4.28,
+      "grad_norm": 1.24534010887146,
+      "learning_rate": 6.5888240200166804e-06,
+      "loss": 0.6459,
+      "step": 4420
+    },
+    {
+      "epoch": 4.28,
+      "grad_norm": 1.8855162858963013,
+      "learning_rate": 6.58465387823186e-06,
+      "loss": 0.6589,
+      "step": 4421
+    },
+    {
+      "epoch": 4.28,
+      "grad_norm": 1.6058472394943237,
+      "learning_rate": 6.58048373644704e-06,
+      "loss": 0.7044,
+      "step": 4422
+    },
+    {
+      "epoch": 4.28,
+      "grad_norm": 1.393979787826538,
+      "learning_rate": 6.576313594662219e-06,
+      "loss": 0.7708,
+      "step": 4423
+    },
+    {
+      "epoch": 4.28,
+      "grad_norm": 1.3308145999908447,
+      "learning_rate": 6.572143452877398e-06,
+      "loss": 0.7672,
+      "step": 4424
+    },
+    {
+      "epoch": 4.28,
+      "grad_norm": 1.3571292161941528,
+      "learning_rate": 6.567973311092577e-06,
+      "loss": 0.7612,
+      "step": 4425
+    },
+    {
+      "epoch": 4.28,
+      "grad_norm": 1.1570378541946411,
+      "learning_rate": 6.563803169307757e-06,
+      "loss": 0.7035,
+      "step": 4426
+    },
+    {
+      "epoch": 4.28,
+      "grad_norm": 1.7388896942138672,
+      "learning_rate": 6.5596330275229364e-06,
+      "loss": 0.9574,
+      "step": 4427
+    },
+    {
+      "epoch": 4.28,
+      "grad_norm": 1.3100439310073853,
+      "learning_rate": 6.555462885738116e-06,
+      "loss": 0.7397,
+      "step": 4428
+    },
+    {
+      "epoch": 4.29,
+      "grad_norm": 1.8462955951690674,
+      "learning_rate": 6.551292743953296e-06,
+      "loss": 0.6368,
+      "step": 4429
+    },
+    {
+      "epoch": 4.29,
+      "grad_norm": 1.5478241443634033,
+      "learning_rate": 6.547122602168473e-06,
+      "loss": 0.4956,
+      "step": 4430
+    },
+    {
+      "epoch": 4.29,
+      "grad_norm": 1.5611926317214966,
+      "learning_rate": 6.542952460383653e-06,
+      "loss": 0.5935,
+      "step": 4431
+    },
+    {
+      "epoch": 4.29,
+      "grad_norm": 1.588344931602478,
+      "learning_rate": 6.538782318598832e-06,
+      "loss": 0.7343,
+      "step": 4432
+    },
+    {
+      "epoch": 4.29,
+      "grad_norm": 1.6632417440414429,
+      "learning_rate": 6.5346121768140116e-06,
+      "loss": 0.888,
+      "step": 4433
+    },
+    {
+      "epoch": 4.29,
+      "grad_norm": 1.861594796180725,
+      "learning_rate": 6.530442035029191e-06,
+      "loss": 0.6662,
+      "step": 4434
+    },
+    {
+      "epoch": 4.29,
+      "grad_norm": 1.4945757389068604,
+      "learning_rate": 6.526271893244371e-06,
+      "loss": 0.7169,
+      "step": 4435
+    },
+    {
+      "epoch": 4.29,
+      "grad_norm": 1.4567769765853882,
+      "learning_rate": 6.52210175145955e-06,
+      "loss": 0.856,
+      "step": 4436
+    },
+    {
+      "epoch": 4.29,
+      "grad_norm": 1.3844677209854126,
+      "learning_rate": 6.517931609674729e-06,
+      "loss": 0.5879,
+      "step": 4437
+    },
+    {
+      "epoch": 4.29,
+      "grad_norm": 1.4528354406356812,
+      "learning_rate": 6.513761467889909e-06,
+      "loss": 0.6289,
+      "step": 4438
+    },
+    {
+      "epoch": 4.3,
+      "grad_norm": 1.5714360475540161,
+      "learning_rate": 6.509591326105088e-06,
+      "loss": 0.753,
+      "step": 4439
+    },
+    {
+      "epoch": 4.3,
+      "grad_norm": 1.2027802467346191,
+      "learning_rate": 6.505421184320268e-06,
+      "loss": 0.6496,
+      "step": 4440
+    },
+    {
+      "epoch": 4.3,
+      "grad_norm": 1.3950735330581665,
+      "learning_rate": 6.501251042535446e-06,
+      "loss": 0.6583,
+      "step": 4441
+    },
+    {
+      "epoch": 4.3,
+      "grad_norm": 1.5818397998809814,
+      "learning_rate": 6.497080900750625e-06,
+      "loss": 0.6201,
+      "step": 4442
+    },
+    {
+      "epoch": 4.3,
+      "grad_norm": 1.5390669107437134,
+      "learning_rate": 6.492910758965805e-06,
+      "loss": 0.6856,
+      "step": 4443
+    },
+    {
+      "epoch": 4.3,
+      "grad_norm": 1.4448585510253906,
+      "learning_rate": 6.488740617180984e-06,
+      "loss": 0.6811,
+      "step": 4444
+    },
+    {
+      "epoch": 4.3,
+      "grad_norm": 1.9134057760238647,
+      "learning_rate": 6.4845704753961635e-06,
+      "loss": 0.6728,
+      "step": 4445
+    },
+    {
+      "epoch": 4.3,
+      "grad_norm": 1.314792513847351,
+      "learning_rate": 6.480400333611343e-06,
+      "loss": 0.6442,
+      "step": 4446
+    },
+    {
+      "epoch": 4.3,
+      "grad_norm": 1.6837114095687866,
+      "learning_rate": 6.476230191826523e-06,
+      "loss": 0.6764,
+      "step": 4447
+    },
+    {
+      "epoch": 4.3,
+      "grad_norm": 1.4650620222091675,
+      "learning_rate": 6.472060050041702e-06,
+      "loss": 0.7084,
+      "step": 4448
+    },
+    {
+      "epoch": 4.3,
+      "grad_norm": 1.4765480756759644,
+      "learning_rate": 6.467889908256881e-06,
+      "loss": 0.7154,
+      "step": 4449
+    },
+    {
+      "epoch": 4.31,
+      "grad_norm": 1.4001424312591553,
+      "learning_rate": 6.463719766472061e-06,
+      "loss": 0.7793,
+      "step": 4450
+    },
+    {
+      "epoch": 4.31,
+      "grad_norm": 1.1573011875152588,
+      "learning_rate": 6.45954962468724e-06,
+      "loss": 0.7458,
+      "step": 4451
+    },
+    {
+      "epoch": 4.31,
+      "grad_norm": 1.374753475189209,
+      "learning_rate": 6.455379482902419e-06,
+      "loss": 0.6044,
+      "step": 4452
+    },
+    {
+      "epoch": 4.31,
+      "grad_norm": 1.8408690690994263,
+      "learning_rate": 6.451209341117598e-06,
+      "loss": 0.7081,
+      "step": 4453
+    },
+    {
+      "epoch": 4.31,
+      "grad_norm": 1.4238532781600952,
+      "learning_rate": 6.447039199332777e-06,
+      "loss": 0.6827,
+      "step": 4454
+    },
+    {
+      "epoch": 4.31,
+      "grad_norm": 1.4446330070495605,
+      "learning_rate": 6.442869057547956e-06,
+      "loss": 0.5796,
+      "step": 4455
+    },
+    {
+      "epoch": 4.31,
+      "grad_norm": 1.597171664237976,
+      "learning_rate": 6.438698915763136e-06,
+      "loss": 0.7198,
+      "step": 4456
+    },
+    {
+      "epoch": 4.31,
+      "grad_norm": 1.1514348983764648,
+      "learning_rate": 6.4345287739783155e-06,
+      "loss": 0.6486,
+      "step": 4457
+    },
+    {
+      "epoch": 4.31,
+      "grad_norm": 1.5550312995910645,
+      "learning_rate": 6.430358632193495e-06,
+      "loss": 0.6427,
+      "step": 4458
+    },
+    {
+      "epoch": 4.31,
+      "grad_norm": 1.2647738456726074,
+      "learning_rate": 6.426188490408675e-06,
+      "loss": 0.9576,
+      "step": 4459
+    },
+    {
+      "epoch": 4.32,
+      "grad_norm": 1.4543766975402832,
+      "learning_rate": 6.422018348623854e-06,
+      "loss": 0.755,
+      "step": 4460
+    },
+    {
+      "epoch": 4.32,
+      "grad_norm": 1.6130422353744507,
+      "learning_rate": 6.417848206839033e-06,
+      "loss": 0.911,
+      "step": 4461
+    },
+    {
+      "epoch": 4.32,
+      "grad_norm": 1.3784844875335693,
+      "learning_rate": 6.413678065054213e-06,
+      "loss": 0.8042,
+      "step": 4462
+    },
+    {
+      "epoch": 4.32,
+      "grad_norm": 1.572872281074524,
+      "learning_rate": 6.409507923269391e-06,
+      "loss": 0.6032,
+      "step": 4463
+    },
+    {
+      "epoch": 4.32,
+      "grad_norm": 1.4436098337173462,
+      "learning_rate": 6.405337781484571e-06,
+      "loss": 0.7356,
+      "step": 4464
+    },
+    {
+      "epoch": 4.32,
+      "grad_norm": 1.6711442470550537,
+      "learning_rate": 6.40116763969975e-06,
+      "loss": 0.7637,
+      "step": 4465
+    },
+    {
+      "epoch": 4.32,
+      "grad_norm": 1.4850529432296753,
+      "learning_rate": 6.396997497914929e-06,
+      "loss": 0.8872,
+      "step": 4466
+    },
+    {
+      "epoch": 4.32,
+      "grad_norm": 1.4410306215286255,
+      "learning_rate": 6.392827356130108e-06,
+      "loss": 0.7155,
+      "step": 4467
+    },
+    {
+      "epoch": 4.32,
+      "grad_norm": 1.6716467142105103,
+      "learning_rate": 6.388657214345288e-06,
+      "loss": 0.7096,
+      "step": 4468
+    },
+    {
+      "epoch": 4.32,
+      "grad_norm": 1.4089220762252808,
+      "learning_rate": 6.3844870725604674e-06,
+      "loss": 0.7065,
+      "step": 4469
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 1.375309705734253,
+      "learning_rate": 6.380316930775647e-06,
+      "loss": 0.6433,
+      "step": 4470
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 1.5627754926681519,
+      "learning_rate": 6.376146788990827e-06,
+      "loss": 0.9394,
+      "step": 4471
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 1.535332202911377,
+      "learning_rate": 6.371976647206006e-06,
+      "loss": 0.9066,
+      "step": 4472
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 1.3067853450775146,
+      "learning_rate": 6.367806505421185e-06,
+      "loss": 0.6955,
+      "step": 4473
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 1.353558897972107,
+      "learning_rate": 6.363636363636363e-06,
+      "loss": 0.6122,
+      "step": 4474
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 1.277823805809021,
+      "learning_rate": 6.3594662218515426e-06,
+      "loss": 0.708,
+      "step": 4475
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 1.5617762804031372,
+      "learning_rate": 6.355296080066722e-06,
+      "loss": 0.8191,
+      "step": 4476
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 1.3509758710861206,
+      "learning_rate": 6.351125938281902e-06,
+      "loss": 0.7159,
+      "step": 4477
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 1.5985627174377441,
+      "learning_rate": 6.346955796497081e-06,
+      "loss": 0.6212,
+      "step": 4478
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 1.2670369148254395,
+      "learning_rate": 6.34278565471226e-06,
+      "loss": 0.5706,
+      "step": 4479
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 1.6988168954849243,
+      "learning_rate": 6.33861551292744e-06,
+      "loss": 0.7364,
+      "step": 4480
+    },
+    {
+      "epoch": 4.34,
+      "grad_norm": 1.737405776977539,
+      "learning_rate": 6.334445371142619e-06,
+      "loss": 0.7654,
+      "step": 4481
+    },
+    {
+      "epoch": 4.34,
+      "grad_norm": 1.70065438747406,
+      "learning_rate": 6.330275229357799e-06,
+      "loss": 0.8737,
+      "step": 4482
+    },
+    {
+      "epoch": 4.34,
+      "grad_norm": 1.3618711233139038,
+      "learning_rate": 6.326105087572979e-06,
+      "loss": 0.609,
+      "step": 4483
+    },
+    {
+      "epoch": 4.34,
+      "grad_norm": 1.207494854927063,
+      "learning_rate": 6.321934945788158e-06,
+      "loss": 0.6795,
+      "step": 4484
+    },
+    {
+      "epoch": 4.34,
+      "grad_norm": 1.6335123777389526,
+      "learning_rate": 6.317764804003336e-06,
+      "loss": 0.5714,
+      "step": 4485
+    },
+    {
+      "epoch": 4.34,
+      "grad_norm": 1.358859658241272,
+      "learning_rate": 6.313594662218515e-06,
+      "loss": 0.6437,
+      "step": 4486
+    },
+    {
+      "epoch": 4.34,
+      "grad_norm": 1.5137168169021606,
+      "learning_rate": 6.3094245204336945e-06,
+      "loss": 0.6404,
+      "step": 4487
+    },
+    {
+      "epoch": 4.34,
+      "grad_norm": 1.8898518085479736,
+      "learning_rate": 6.305254378648874e-06,
+      "loss": 0.6857,
+      "step": 4488
+    },
+    {
+      "epoch": 4.34,
+      "grad_norm": 1.2113492488861084,
+      "learning_rate": 6.301084236864054e-06,
+      "loss": 0.6162,
+      "step": 4489
+    },
+    {
+      "epoch": 4.34,
+      "grad_norm": 1.317917823791504,
+      "learning_rate": 6.296914095079233e-06,
+      "loss": 0.6325,
+      "step": 4490
+    },
+    {
+      "epoch": 4.35,
+      "grad_norm": 1.5444402694702148,
+      "learning_rate": 6.292743953294412e-06,
+      "loss": 0.6928,
+      "step": 4491
+    },
+    {
+      "epoch": 4.35,
+      "grad_norm": 1.3579732179641724,
+      "learning_rate": 6.288573811509592e-06,
+      "loss": 0.6092,
+      "step": 4492
+    },
+    {
+      "epoch": 4.35,
+      "grad_norm": 1.6278555393218994,
+      "learning_rate": 6.284403669724771e-06,
+      "loss": 0.79,
+      "step": 4493
+    },
+    {
+      "epoch": 4.35,
+      "grad_norm": 1.4639973640441895,
+      "learning_rate": 6.2802335279399505e-06,
+      "loss": 0.8683,
+      "step": 4494
+    },
+    {
+      "epoch": 4.35,
+      "grad_norm": 1.2048999071121216,
+      "learning_rate": 6.276063386155131e-06,
+      "loss": 0.5256,
+      "step": 4495
+    },
+    {
+      "epoch": 4.35,
+      "grad_norm": 1.336359977722168,
+      "learning_rate": 6.271893244370308e-06,
+      "loss": 0.5147,
+      "step": 4496
+    },
+    {
+      "epoch": 4.35,
+      "grad_norm": 1.4179655313491821,
+      "learning_rate": 6.267723102585487e-06,
+      "loss": 0.7034,
+      "step": 4497
+    },
+    {
+      "epoch": 4.35,
+      "grad_norm": 1.2926969528198242,
+      "learning_rate": 6.263552960800667e-06,
+      "loss": 0.6835,
+      "step": 4498
+    },
+    {
+      "epoch": 4.35,
+      "grad_norm": 1.225582480430603,
+      "learning_rate": 6.2593828190158465e-06,
+      "loss": 0.7043,
+      "step": 4499
+    },
+    {
+      "epoch": 4.35,
+      "grad_norm": 1.4773812294006348,
+      "learning_rate": 6.255212677231026e-06,
+      "loss": 0.9061,
+      "step": 4500
+    },
+    {
+      "epoch": 4.35,
+      "eval_loss": 0.8581743240356445,
+      "eval_runtime": 856.5442,
+      "eval_samples_per_second": 4.826,
+      "eval_steps_per_second": 0.604,
+      "step": 4500
+    },
+    {
+      "epoch": 4.36,
+      "grad_norm": 1.2046254873275757,
+      "learning_rate": 6.251042535446206e-06,
+      "loss": 0.6437,
+      "step": 4501
+    },
+    {
+      "epoch": 4.36,
+      "grad_norm": 1.9686743021011353,
+      "learning_rate": 6.246872393661385e-06,
+      "loss": 0.7816,
+      "step": 4502
+    },
+    {
+      "epoch": 4.36,
+      "grad_norm": 1.4191972017288208,
+      "learning_rate": 6.242702251876564e-06,
+      "loss": 0.6423,
+      "step": 4503
+    },
+    {
+      "epoch": 4.36,
+      "grad_norm": 1.3162946701049805,
+      "learning_rate": 6.238532110091744e-06,
+      "loss": 0.7523,
+      "step": 4504
+    },
+    {
+      "epoch": 4.36,
+      "grad_norm": 1.2248839139938354,
+      "learning_rate": 6.2343619683069225e-06,
+      "loss": 0.6698,
+      "step": 4505
+    },
+    {
+      "epoch": 4.36,
+      "grad_norm": 1.5480225086212158,
+      "learning_rate": 6.230191826522102e-06,
+      "loss": 0.7132,
+      "step": 4506
+    },
+    {
+      "epoch": 4.36,
+      "grad_norm": 1.264034390449524,
+      "learning_rate": 6.226021684737282e-06,
+      "loss": 0.5244,
+      "step": 4507
+    },
+    {
+      "epoch": 4.36,
+      "grad_norm": 1.3124220371246338,
+      "learning_rate": 6.221851542952461e-06,
+      "loss": 0.846,
+      "step": 4508
+    },
+    {
+      "epoch": 4.36,
+      "grad_norm": 1.2050153017044067,
+      "learning_rate": 6.21768140116764e-06,
+      "loss": 0.7297,
+      "step": 4509
+    },
+    {
+      "epoch": 4.36,
+      "grad_norm": 1.1116827726364136,
+      "learning_rate": 6.213511259382819e-06,
+      "loss": 0.5178,
+      "step": 4510
+    },
+    {
+      "epoch": 4.36,
+      "grad_norm": 1.7788770198822021,
+      "learning_rate": 6.2093411175979984e-06,
+      "loss": 0.6892,
+      "step": 4511
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 1.5678691864013672,
+      "learning_rate": 6.205170975813178e-06,
+      "loss": 0.7617,
+      "step": 4512
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 1.6809147596359253,
+      "learning_rate": 6.201000834028358e-06,
+      "loss": 0.6395,
+      "step": 4513
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 1.4206522703170776,
+      "learning_rate": 6.196830692243537e-06,
+      "loss": 0.7164,
+      "step": 4514
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 1.2952232360839844,
+      "learning_rate": 6.192660550458716e-06,
+      "loss": 0.6791,
+      "step": 4515
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 1.8799912929534912,
+      "learning_rate": 6.188490408673895e-06,
+      "loss": 0.6621,
+      "step": 4516
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 1.5345474481582642,
+      "learning_rate": 6.1843202668890744e-06,
+      "loss": 0.7264,
+      "step": 4517
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 1.6303728818893433,
+      "learning_rate": 6.180150125104254e-06,
+      "loss": 0.5539,
+      "step": 4518
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 1.459893822669983,
+      "learning_rate": 6.175979983319434e-06,
+      "loss": 0.7534,
+      "step": 4519
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 1.3992021083831787,
+      "learning_rate": 6.171809841534613e-06,
+      "loss": 0.5496,
+      "step": 4520
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 1.4364956617355347,
+      "learning_rate": 6.167639699749791e-06,
+      "loss": 0.7307,
+      "step": 4521
+    },
+    {
+      "epoch": 4.38,
+      "grad_norm": 1.596460223197937,
+      "learning_rate": 6.163469557964971e-06,
+      "loss": 0.6842,
+      "step": 4522
+    },
+    {
+      "epoch": 4.38,
+      "grad_norm": 1.453177809715271,
+      "learning_rate": 6.15929941618015e-06,
+      "loss": 0.6237,
+      "step": 4523
+    },
+    {
+      "epoch": 4.38,
+      "grad_norm": 1.6033985614776611,
+      "learning_rate": 6.15512927439533e-06,
+      "loss": 0.7595,
+      "step": 4524
+    },
+    {
+      "epoch": 4.38,
+      "grad_norm": 1.551035761833191,
+      "learning_rate": 6.15095913261051e-06,
+      "loss": 0.6417,
+      "step": 4525
+    },
+    {
+      "epoch": 4.38,
+      "grad_norm": 1.6841317415237427,
+      "learning_rate": 6.146788990825689e-06,
+      "loss": 0.7573,
+      "step": 4526
+    },
+    {
+      "epoch": 4.38,
+      "grad_norm": 1.8649938106536865,
+      "learning_rate": 6.142618849040867e-06,
+      "loss": 0.5851,
+      "step": 4527
+    },
+    {
+      "epoch": 4.38,
+      "grad_norm": 1.2951748371124268,
+      "learning_rate": 6.138448707256047e-06,
+      "loss": 0.6367,
+      "step": 4528
+    },
+    {
+      "epoch": 4.38,
+      "grad_norm": 1.4893923997879028,
+      "learning_rate": 6.134278565471226e-06,
+      "loss": 0.7389,
+      "step": 4529
+    },
+    {
+      "epoch": 4.38,
+      "grad_norm": 1.517099380493164,
+      "learning_rate": 6.1301084236864056e-06,
+      "loss": 0.6981,
+      "step": 4530
+    },
+    {
+      "epoch": 4.38,
+      "grad_norm": 0.9570285677909851,
+      "learning_rate": 6.125938281901586e-06,
+      "loss": 0.5932,
+      "step": 4531
+    },
+    {
+      "epoch": 4.39,
+      "grad_norm": 1.7340022325515747,
+      "learning_rate": 6.121768140116764e-06,
+      "loss": 0.6693,
+      "step": 4532
+    },
+    {
+      "epoch": 4.39,
+      "grad_norm": 1.3229211568832397,
+      "learning_rate": 6.117597998331943e-06,
+      "loss": 0.6984,
+      "step": 4533
+    },
+    {
+      "epoch": 4.39,
+      "grad_norm": 1.7112501859664917,
+      "learning_rate": 6.113427856547123e-06,
+      "loss": 0.7863,
+      "step": 4534
+    },
+    {
+      "epoch": 4.39,
+      "grad_norm": 1.1070160865783691,
+      "learning_rate": 6.109257714762302e-06,
+      "loss": 0.6884,
+      "step": 4535
+    },
+    {
+      "epoch": 4.39,
+      "grad_norm": 1.218097448348999,
+      "learning_rate": 6.1050875729774816e-06,
+      "loss": 0.8119,
+      "step": 4536
+    },
+    {
+      "epoch": 4.39,
+      "grad_norm": 1.3918700218200684,
+      "learning_rate": 6.100917431192662e-06,
+      "loss": 0.7109,
+      "step": 4537
+    },
+    {
+      "epoch": 4.39,
+      "grad_norm": 1.3033515214920044,
+      "learning_rate": 6.09674728940784e-06,
+      "loss": 0.7304,
+      "step": 4538
+    },
+    {
+      "epoch": 4.39,
+      "grad_norm": 1.5123662948608398,
+      "learning_rate": 6.092577147623019e-06,
+      "loss": 0.8365,
+      "step": 4539
+    },
+    {
+      "epoch": 4.39,
+      "grad_norm": 1.6895707845687866,
+      "learning_rate": 6.088407005838199e-06,
+      "loss": 0.5216,
+      "step": 4540
+    },
+    {
+      "epoch": 4.39,
+      "grad_norm": 1.752480149269104,
+      "learning_rate": 6.084236864053378e-06,
+      "loss": 0.7785,
+      "step": 4541
+    },
+    {
+      "epoch": 4.39,
+      "grad_norm": 0.9798446893692017,
+      "learning_rate": 6.0800667222685575e-06,
+      "loss": 0.71,
+      "step": 4542
+    },
+    {
+      "epoch": 4.4,
+      "grad_norm": 1.2920374870300293,
+      "learning_rate": 6.075896580483737e-06,
+      "loss": 0.9092,
+      "step": 4543
+    },
+    {
+      "epoch": 4.4,
+      "grad_norm": 1.741518259048462,
+      "learning_rate": 6.071726438698916e-06,
+      "loss": 0.6685,
+      "step": 4544
+    },
+    {
+      "epoch": 4.4,
+      "grad_norm": 1.4512648582458496,
+      "learning_rate": 6.067556296914095e-06,
+      "loss": 0.6378,
+      "step": 4545
+    },
+    {
+      "epoch": 4.4,
+      "grad_norm": 1.3239641189575195,
+      "learning_rate": 6.063386155129275e-06,
+      "loss": 0.7009,
+      "step": 4546
+    },
+    {
+      "epoch": 4.4,
+      "grad_norm": 1.6167960166931152,
+      "learning_rate": 6.059216013344454e-06,
+      "loss": 0.6332,
+      "step": 4547
+    },
+    {
+      "epoch": 4.4,
+      "grad_norm": 1.438257098197937,
+      "learning_rate": 6.0550458715596335e-06,
+      "loss": 0.6658,
+      "step": 4548
+    },
+    {
+      "epoch": 4.4,
+      "grad_norm": 1.4437922239303589,
+      "learning_rate": 6.050875729774813e-06,
+      "loss": 0.6923,
+      "step": 4549
+    },
+    {
+      "epoch": 4.4,
+      "grad_norm": 1.660217523574829,
+      "learning_rate": 6.046705587989992e-06,
+      "loss": 0.641,
+      "step": 4550
+    },
+    {
+      "epoch": 4.4,
+      "grad_norm": 1.4629638195037842,
+      "learning_rate": 6.042535446205171e-06,
+      "loss": 0.7885,
+      "step": 4551
+    },
+    {
+      "epoch": 4.4,
+      "grad_norm": 1.5988290309906006,
+      "learning_rate": 6.038365304420351e-06,
+      "loss": 0.7277,
+      "step": 4552
+    },
+    {
+      "epoch": 4.41,
+      "grad_norm": 1.421746850013733,
+      "learning_rate": 6.03419516263553e-06,
+      "loss": 0.5789,
+      "step": 4553
+    },
+    {
+      "epoch": 4.41,
+      "grad_norm": 1.5449638366699219,
+      "learning_rate": 6.030025020850709e-06,
+      "loss": 0.5776,
+      "step": 4554
+    },
+    {
+      "epoch": 4.41,
+      "grad_norm": 1.4405105113983154,
+      "learning_rate": 6.025854879065889e-06,
+      "loss": 0.6907,
+      "step": 4555
+    },
+    {
+      "epoch": 4.41,
+      "grad_norm": 1.2792209386825562,
+      "learning_rate": 6.021684737281068e-06,
+      "loss": 0.5526,
+      "step": 4556
+    },
+    {
+      "epoch": 4.41,
+      "grad_norm": 1.272682547569275,
+      "learning_rate": 6.017514595496247e-06,
+      "loss": 0.6382,
+      "step": 4557
+    },
+    {
+      "epoch": 4.41,
+      "grad_norm": 1.6121176481246948,
+      "learning_rate": 6.013344453711427e-06,
+      "loss": 0.608,
+      "step": 4558
+    },
+    {
+      "epoch": 4.41,
+      "grad_norm": 1.5273442268371582,
+      "learning_rate": 6.009174311926606e-06,
+      "loss": 0.7358,
+      "step": 4559
+    },
+    {
+      "epoch": 4.41,
+      "grad_norm": 1.7222824096679688,
+      "learning_rate": 6.005004170141785e-06,
+      "loss": 0.7476,
+      "step": 4560
+    },
+    {
+      "epoch": 4.41,
+      "grad_norm": 1.130391001701355,
+      "learning_rate": 6.000834028356965e-06,
+      "loss": 0.5936,
+      "step": 4561
+    },
+    {
+      "epoch": 4.41,
+      "grad_norm": 1.3474061489105225,
+      "learning_rate": 5.996663886572144e-06,
+      "loss": 0.5635,
+      "step": 4562
+    },
+    {
+      "epoch": 4.42,
+      "grad_norm": 1.3879218101501465,
+      "learning_rate": 5.992493744787323e-06,
+      "loss": 0.7199,
+      "step": 4563
+    },
+    {
+      "epoch": 4.42,
+      "grad_norm": 1.6851611137390137,
+      "learning_rate": 5.988323603002502e-06,
+      "loss": 0.6714,
+      "step": 4564
+    },
+    {
+      "epoch": 4.42,
+      "grad_norm": 1.5315393209457397,
+      "learning_rate": 5.984153461217681e-06,
+      "loss": 0.5827,
+      "step": 4565
+    },
+    {
+      "epoch": 4.42,
+      "grad_norm": 1.298017978668213,
+      "learning_rate": 5.979983319432861e-06,
+      "loss": 0.8551,
+      "step": 4566
+    },
+    {
+      "epoch": 4.42,
+      "grad_norm": 1.3128635883331299,
+      "learning_rate": 5.975813177648041e-06,
+      "loss": 0.682,
+      "step": 4567
+    },
+    {
+      "epoch": 4.42,
+      "grad_norm": 1.4132344722747803,
+      "learning_rate": 5.97164303586322e-06,
+      "loss": 0.6471,
+      "step": 4568
+    },
+    {
+      "epoch": 4.42,
+      "grad_norm": 1.35657799243927,
+      "learning_rate": 5.967472894078399e-06,
+      "loss": 0.4753,
+      "step": 4569
+    },
+    {
+      "epoch": 4.42,
+      "grad_norm": 1.211188554763794,
+      "learning_rate": 5.963302752293578e-06,
+      "loss": 0.7416,
+      "step": 4570
+    },
+    {
+      "epoch": 4.42,
+      "grad_norm": 1.1546083688735962,
+      "learning_rate": 5.959132610508757e-06,
+      "loss": 0.7779,
+      "step": 4571
+    },
+    {
+      "epoch": 4.42,
+      "grad_norm": 1.3571324348449707,
+      "learning_rate": 5.9549624687239366e-06,
+      "loss": 0.6944,
+      "step": 4572
+    },
+    {
+      "epoch": 4.42,
+      "grad_norm": 1.4888793230056763,
+      "learning_rate": 5.950792326939117e-06,
+      "loss": 0.6427,
+      "step": 4573
+    },
+    {
+      "epoch": 4.43,
+      "grad_norm": 1.563929796218872,
+      "learning_rate": 5.946622185154296e-06,
+      "loss": 0.7502,
+      "step": 4574
+    },
+    {
+      "epoch": 4.43,
+      "grad_norm": 1.6636128425598145,
+      "learning_rate": 5.942452043369475e-06,
+      "loss": 0.6738,
+      "step": 4575
+    },
+    {
+      "epoch": 4.43,
+      "grad_norm": 1.431536078453064,
+      "learning_rate": 5.938281901584654e-06,
+      "loss": 0.751,
+      "step": 4576
+    },
+    {
+      "epoch": 4.43,
+      "grad_norm": 1.067686676979065,
+      "learning_rate": 5.934111759799833e-06,
+      "loss": 0.5663,
+      "step": 4577
+    },
+    {
+      "epoch": 4.43,
+      "grad_norm": 1.6124565601348877,
+      "learning_rate": 5.9299416180150126e-06,
+      "loss": 0.6439,
+      "step": 4578
+    },
+    {
+      "epoch": 4.43,
+      "grad_norm": 1.589220404624939,
+      "learning_rate": 5.925771476230193e-06,
+      "loss": 0.7915,
+      "step": 4579
+    },
+    {
+      "epoch": 4.43,
+      "grad_norm": 1.9252246618270874,
+      "learning_rate": 5.921601334445372e-06,
+      "loss": 0.8538,
+      "step": 4580
+    },
+    {
+      "epoch": 4.43,
+      "grad_norm": 1.3919744491577148,
+      "learning_rate": 5.917431192660551e-06,
+      "loss": 0.5826,
+      "step": 4581
+    },
+    {
+      "epoch": 4.43,
+      "grad_norm": 1.2920315265655518,
+      "learning_rate": 5.91326105087573e-06,
+      "loss": 0.6896,
+      "step": 4582
+    },
+    {
+      "epoch": 4.43,
+      "grad_norm": 1.8945282697677612,
+      "learning_rate": 5.909090909090909e-06,
+      "loss": 0.6634,
+      "step": 4583
+    },
+    {
+      "epoch": 4.44,
+      "grad_norm": 1.4104385375976562,
+      "learning_rate": 5.9049207673060885e-06,
+      "loss": 0.6774,
+      "step": 4584
+    },
+    {
+      "epoch": 4.44,
+      "grad_norm": 0.9925658106803894,
+      "learning_rate": 5.9007506255212686e-06,
+      "loss": 0.7068,
+      "step": 4585
+    },
+    {
+      "epoch": 4.44,
+      "grad_norm": 1.6425079107284546,
+      "learning_rate": 5.896580483736448e-06,
+      "loss": 0.6906,
+      "step": 4586
+    },
+    {
+      "epoch": 4.44,
+      "grad_norm": 1.734883427619934,
+      "learning_rate": 5.892410341951626e-06,
+      "loss": 0.6661,
+      "step": 4587
+    },
+    {
+      "epoch": 4.44,
+      "grad_norm": 1.314497709274292,
+      "learning_rate": 5.888240200166806e-06,
+      "loss": 0.6166,
+      "step": 4588
+    },
+    {
+      "epoch": 4.44,
+      "grad_norm": 1.7467992305755615,
+      "learning_rate": 5.884070058381985e-06,
+      "loss": 0.6317,
+      "step": 4589
+    },
+    {
+      "epoch": 4.44,
+      "grad_norm": 1.6375136375427246,
+      "learning_rate": 5.8798999165971645e-06,
+      "loss": 0.8575,
+      "step": 4590
+    },
+    {
+      "epoch": 4.44,
+      "grad_norm": 1.6072413921356201,
+      "learning_rate": 5.875729774812344e-06,
+      "loss": 0.6524,
+      "step": 4591
+    },
+    {
+      "epoch": 4.44,
+      "grad_norm": 1.4089336395263672,
+      "learning_rate": 5.871559633027524e-06,
+      "loss": 0.7012,
+      "step": 4592
+    },
+    {
+      "epoch": 4.44,
+      "grad_norm": 1.2882603406906128,
+      "learning_rate": 5.867389491242702e-06,
+      "loss": 0.6848,
+      "step": 4593
+    },
+    {
+      "epoch": 4.45,
+      "grad_norm": 1.2387679815292358,
+      "learning_rate": 5.863219349457882e-06,
+      "loss": 0.8055,
+      "step": 4594
+    },
+    {
+      "epoch": 4.45,
+      "grad_norm": 1.8764599561691284,
+      "learning_rate": 5.859049207673061e-06,
+      "loss": 0.578,
+      "step": 4595
+    },
+    {
+      "epoch": 4.45,
+      "grad_norm": 1.4414983987808228,
+      "learning_rate": 5.8548790658882405e-06,
+      "loss": 0.6162,
+      "step": 4596
+    },
+    {
+      "epoch": 4.45,
+      "grad_norm": 1.491419792175293,
+      "learning_rate": 5.85070892410342e-06,
+      "loss": 0.5989,
+      "step": 4597
+    },
+    {
+      "epoch": 4.45,
+      "grad_norm": 2.5183537006378174,
+      "learning_rate": 5.846538782318599e-06,
+      "loss": 0.7089,
+      "step": 4598
+    },
+    {
+      "epoch": 4.45,
+      "grad_norm": 1.0582003593444824,
+      "learning_rate": 5.842368640533778e-06,
+      "loss": 0.6531,
+      "step": 4599
+    },
+    {
+      "epoch": 4.45,
+      "grad_norm": 1.6829962730407715,
+      "learning_rate": 5.838198498748958e-06,
+      "loss": 0.7932,
+      "step": 4600
+    },
+    {
+      "epoch": 4.45,
+      "grad_norm": 1.9470819234848022,
+      "learning_rate": 5.834028356964137e-06,
+      "loss": 0.6358,
+      "step": 4601
+    },
+    {
+      "epoch": 4.45,
+      "grad_norm": 1.2453104257583618,
+      "learning_rate": 5.8298582151793165e-06,
+      "loss": 0.6782,
+      "step": 4602
+    },
+    {
+      "epoch": 4.45,
+      "grad_norm": 2.341174840927124,
+      "learning_rate": 5.825688073394496e-06,
+      "loss": 0.7806,
+      "step": 4603
+    },
+    {
+      "epoch": 4.45,
+      "grad_norm": 1.486384391784668,
+      "learning_rate": 5.821517931609675e-06,
+      "loss": 0.741,
+      "step": 4604
+    },
+    {
+      "epoch": 4.46,
+      "grad_norm": 1.264007329940796,
+      "learning_rate": 5.817347789824854e-06,
+      "loss": 0.6896,
+      "step": 4605
+    },
+    {
+      "epoch": 4.46,
+      "grad_norm": 1.740382432937622,
+      "learning_rate": 5.813177648040034e-06,
+      "loss": 0.6706,
+      "step": 4606
+    },
+    {
+      "epoch": 4.46,
+      "grad_norm": 1.2103074789047241,
+      "learning_rate": 5.809007506255213e-06,
+      "loss": 0.6736,
+      "step": 4607
+    },
+    {
+      "epoch": 4.46,
+      "grad_norm": 1.4383554458618164,
+      "learning_rate": 5.8048373644703924e-06,
+      "loss": 0.7079,
+      "step": 4608
+    },
+    {
+      "epoch": 4.46,
+      "grad_norm": 1.181028962135315,
+      "learning_rate": 5.800667222685572e-06,
+      "loss": 0.6232,
+      "step": 4609
+    },
+    {
+      "epoch": 4.46,
+      "grad_norm": 1.425830602645874,
+      "learning_rate": 5.796497080900751e-06,
+      "loss": 0.7142,
+      "step": 4610
+    },
+    {
+      "epoch": 4.46,
+      "grad_norm": 1.0202964544296265,
+      "learning_rate": 5.79232693911593e-06,
+      "loss": 0.6243,
+      "step": 4611
+    },
+    {
+      "epoch": 4.46,
+      "grad_norm": 1.389682650566101,
+      "learning_rate": 5.788156797331109e-06,
+      "loss": 0.7684,
+      "step": 4612
+    },
+    {
+      "epoch": 4.46,
+      "grad_norm": 2.0149354934692383,
+      "learning_rate": 5.783986655546289e-06,
+      "loss": 0.647,
+      "step": 4613
+    },
+    {
+      "epoch": 4.46,
+      "grad_norm": 1.9424692392349243,
+      "learning_rate": 5.7798165137614684e-06,
+      "loss": 0.6478,
+      "step": 4614
+    },
+    {
+      "epoch": 4.47,
+      "grad_norm": 1.4337615966796875,
+      "learning_rate": 5.775646371976648e-06,
+      "loss": 0.6507,
+      "step": 4615
+    },
+    {
+      "epoch": 4.47,
+      "grad_norm": 1.2378499507904053,
+      "learning_rate": 5.771476230191827e-06,
+      "loss": 0.6993,
+      "step": 4616
+    },
+    {
+      "epoch": 4.47,
+      "grad_norm": 1.8119548559188843,
+      "learning_rate": 5.767306088407006e-06,
+      "loss": 0.6861,
+      "step": 4617
+    },
+    {
+      "epoch": 4.47,
+      "grad_norm": 1.6021716594696045,
+      "learning_rate": 5.763135946622185e-06,
+      "loss": 0.6991,
+      "step": 4618
+    },
+    {
+      "epoch": 4.47,
+      "grad_norm": 1.6414045095443726,
+      "learning_rate": 5.758965804837365e-06,
+      "loss": 0.6302,
+      "step": 4619
+    },
+    {
+      "epoch": 4.47,
+      "grad_norm": 1.7944785356521606,
+      "learning_rate": 5.7547956630525436e-06,
+      "loss": 0.7703,
+      "step": 4620
+    },
+    {
+      "epoch": 4.47,
+      "grad_norm": 1.0491467714309692,
+      "learning_rate": 5.750625521267724e-06,
+      "loss": 0.6449,
+      "step": 4621
+    },
+    {
+      "epoch": 4.47,
+      "grad_norm": 1.6654112339019775,
+      "learning_rate": 5.746455379482903e-06,
+      "loss": 0.6433,
+      "step": 4622
+    },
+    {
+      "epoch": 4.47,
+      "grad_norm": 1.2329283952713013,
+      "learning_rate": 5.742285237698082e-06,
+      "loss": 0.6477,
+      "step": 4623
+    },
+    {
+      "epoch": 4.47,
+      "grad_norm": 1.6013000011444092,
+      "learning_rate": 5.738115095913261e-06,
+      "loss": 0.6302,
+      "step": 4624
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 1.8500837087631226,
+      "learning_rate": 5.733944954128441e-06,
+      "loss": 0.7395,
+      "step": 4625
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 1.3515931367874146,
+      "learning_rate": 5.7297748123436195e-06,
+      "loss": 0.6717,
+      "step": 4626
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 1.6684017181396484,
+      "learning_rate": 5.7256046705587996e-06,
+      "loss": 0.6337,
+      "step": 4627
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 1.1044994592666626,
+      "learning_rate": 5.721434528773979e-06,
+      "loss": 0.6495,
+      "step": 4628
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 1.3723560571670532,
+      "learning_rate": 5.717264386989158e-06,
+      "loss": 0.7101,
+      "step": 4629
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 1.2639974355697632,
+      "learning_rate": 5.713094245204337e-06,
+      "loss": 0.7862,
+      "step": 4630
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 1.0320762395858765,
+      "learning_rate": 5.708924103419516e-06,
+      "loss": 0.594,
+      "step": 4631
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 1.7549233436584473,
+      "learning_rate": 5.7047539616346955e-06,
+      "loss": 0.6192,
+      "step": 4632
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 1.311571717262268,
+      "learning_rate": 5.700583819849875e-06,
+      "loss": 0.6548,
+      "step": 4633
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 1.782296895980835,
+      "learning_rate": 5.696413678065055e-06,
+      "loss": 0.6356,
+      "step": 4634
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 1.3225741386413574,
+      "learning_rate": 5.692243536280234e-06,
+      "loss": 0.8134,
+      "step": 4635
+    },
+    {
+      "epoch": 4.49,
+      "grad_norm": 1.4524954557418823,
+      "learning_rate": 5.688073394495413e-06,
+      "loss": 0.7109,
+      "step": 4636
+    },
+    {
+      "epoch": 4.49,
+      "grad_norm": 1.2796392440795898,
+      "learning_rate": 5.683903252710592e-06,
+      "loss": 0.7489,
+      "step": 4637
+    },
+    {
+      "epoch": 4.49,
+      "grad_norm": 1.9327563047409058,
+      "learning_rate": 5.6797331109257715e-06,
+      "loss": 0.6422,
+      "step": 4638
+    },
+    {
+      "epoch": 4.49,
+      "grad_norm": 2.186600923538208,
+      "learning_rate": 5.675562969140951e-06,
+      "loss": 0.5712,
+      "step": 4639
+    },
+    {
+      "epoch": 4.49,
+      "grad_norm": 1.4998987913131714,
+      "learning_rate": 5.671392827356131e-06,
+      "loss": 0.6776,
+      "step": 4640
+    },
+    {
+      "epoch": 4.49,
+      "grad_norm": 1.1831672191619873,
+      "learning_rate": 5.66722268557131e-06,
+      "loss": 0.6532,
+      "step": 4641
+    },
+    {
+      "epoch": 4.49,
+      "grad_norm": 1.3108035326004028,
+      "learning_rate": 5.663052543786489e-06,
+      "loss": 0.6943,
+      "step": 4642
+    },
+    {
+      "epoch": 4.49,
+      "grad_norm": 1.7191115617752075,
+      "learning_rate": 5.658882402001668e-06,
+      "loss": 0.7349,
+      "step": 4643
+    },
+    {
+      "epoch": 4.49,
+      "grad_norm": 1.2873550653457642,
+      "learning_rate": 5.6547122602168475e-06,
+      "loss": 0.8601,
+      "step": 4644
+    },
+    {
+      "epoch": 4.49,
+      "grad_norm": 1.3496503829956055,
+      "learning_rate": 5.650542118432027e-06,
+      "loss": 0.6711,
+      "step": 4645
+    },
+    {
+      "epoch": 4.5,
+      "grad_norm": 1.552755355834961,
+      "learning_rate": 5.646371976647207e-06,
+      "loss": 0.6595,
+      "step": 4646
+    },
+    {
+      "epoch": 4.5,
+      "grad_norm": 1.2598237991333008,
+      "learning_rate": 5.642201834862386e-06,
+      "loss": 0.8592,
+      "step": 4647
+    },
+    {
+      "epoch": 4.5,
+      "grad_norm": 1.4796096086502075,
+      "learning_rate": 5.638031693077565e-06,
+      "loss": 0.8437,
+      "step": 4648
+    },
+    {
+      "epoch": 4.5,
+      "grad_norm": 1.364597201347351,
+      "learning_rate": 5.633861551292744e-06,
+      "loss": 0.5901,
+      "step": 4649
+    },
+    {
+      "epoch": 4.5,
+      "grad_norm": 1.2805659770965576,
+      "learning_rate": 5.6296914095079234e-06,
+      "loss": 0.7921,
+      "step": 4650
+    },
+    {
+      "epoch": 4.5,
+      "grad_norm": 1.4497544765472412,
+      "learning_rate": 5.625521267723103e-06,
+      "loss": 0.6249,
+      "step": 4651
+    },
+    {
+      "epoch": 4.5,
+      "grad_norm": 1.7044733762741089,
+      "learning_rate": 5.621351125938283e-06,
+      "loss": 0.7502,
+      "step": 4652
+    },
+    {
+      "epoch": 4.5,
+      "grad_norm": 1.466091275215149,
+      "learning_rate": 5.617180984153461e-06,
+      "loss": 0.6731,
+      "step": 4653
+    },
+    {
+      "epoch": 4.5,
+      "grad_norm": 1.43474280834198,
+      "learning_rate": 5.61301084236864e-06,
+      "loss": 0.6992,
+      "step": 4654
+    },
+    {
+      "epoch": 4.5,
+      "grad_norm": 1.197882056236267,
+      "learning_rate": 5.60884070058382e-06,
+      "loss": 0.6655,
+      "step": 4655
+    },
+    {
+      "epoch": 4.51,
+      "grad_norm": 1.6035302877426147,
+      "learning_rate": 5.6046705587989994e-06,
+      "loss": 0.7569,
+      "step": 4656
+    },
+    {
+      "epoch": 4.51,
+      "grad_norm": 1.7309684753417969,
+      "learning_rate": 5.600500417014179e-06,
+      "loss": 0.7687,
+      "step": 4657
+    },
+    {
+      "epoch": 4.51,
+      "grad_norm": 1.495622158050537,
+      "learning_rate": 5.596330275229359e-06,
+      "loss": 0.6196,
+      "step": 4658
+    },
+    {
+      "epoch": 4.51,
+      "grad_norm": 1.9390276670455933,
+      "learning_rate": 5.592160133444537e-06,
+      "loss": 0.7622,
+      "step": 4659
+    },
+    {
+      "epoch": 4.51,
+      "grad_norm": 1.416953682899475,
+      "learning_rate": 5.587989991659716e-06,
+      "loss": 0.5089,
+      "step": 4660
+    },
+    {
+      "epoch": 4.51,
+      "grad_norm": 1.546241283416748,
+      "learning_rate": 5.583819849874896e-06,
+      "loss": 0.6598,
+      "step": 4661
+    },
+    {
+      "epoch": 4.51,
+      "grad_norm": 1.6533225774765015,
+      "learning_rate": 5.579649708090075e-06,
+      "loss": 0.9055,
+      "step": 4662
+    },
+    {
+      "epoch": 4.51,
+      "grad_norm": 1.3580639362335205,
+      "learning_rate": 5.575479566305255e-06,
+      "loss": 0.8042,
+      "step": 4663
+    },
+    {
+      "epoch": 4.51,
+      "grad_norm": 1.56376314163208,
+      "learning_rate": 5.571309424520434e-06,
+      "loss": 0.8632,
+      "step": 4664
+    },
+    {
+      "epoch": 4.51,
+      "grad_norm": 1.5229129791259766,
+      "learning_rate": 5.567139282735613e-06,
+      "loss": 0.6711,
+      "step": 4665
+    },
+    {
+      "epoch": 4.51,
+      "grad_norm": 1.6064221858978271,
+      "learning_rate": 5.562969140950792e-06,
+      "loss": 0.8422,
+      "step": 4666
+    },
+    {
+      "epoch": 4.52,
+      "grad_norm": 1.3323804140090942,
+      "learning_rate": 5.558798999165972e-06,
+      "loss": 0.6602,
+      "step": 4667
+    },
+    {
+      "epoch": 4.52,
+      "grad_norm": 1.4172663688659668,
+      "learning_rate": 5.554628857381151e-06,
+      "loss": 0.6033,
+      "step": 4668
+    },
+    {
+      "epoch": 4.52,
+      "grad_norm": 1.5368646383285522,
+      "learning_rate": 5.5504587155963306e-06,
+      "loss": 0.6856,
+      "step": 4669
+    },
+    {
+      "epoch": 4.52,
+      "grad_norm": 1.359999656677246,
+      "learning_rate": 5.54628857381151e-06,
+      "loss": 0.6784,
+      "step": 4670
+    },
+    {
+      "epoch": 4.52,
+      "grad_norm": 1.7071532011032104,
+      "learning_rate": 5.542118432026689e-06,
+      "loss": 0.748,
+      "step": 4671
+    },
+    {
+      "epoch": 4.52,
+      "grad_norm": 1.4840952157974243,
+      "learning_rate": 5.537948290241868e-06,
+      "loss": 0.767,
+      "step": 4672
+    },
+    {
+      "epoch": 4.52,
+      "grad_norm": 1.385696291923523,
+      "learning_rate": 5.533778148457048e-06,
+      "loss": 0.6435,
+      "step": 4673
+    },
+    {
+      "epoch": 4.52,
+      "grad_norm": 1.4366190433502197,
+      "learning_rate": 5.529608006672227e-06,
+      "loss": 0.5813,
+      "step": 4674
+    },
+    {
+      "epoch": 4.52,
+      "grad_norm": 1.3593374490737915,
+      "learning_rate": 5.525437864887406e-06,
+      "loss": 0.6451,
+      "step": 4675
+    },
+    {
+      "epoch": 4.52,
+      "grad_norm": 1.4528684616088867,
+      "learning_rate": 5.521267723102586e-06,
+      "loss": 0.7298,
+      "step": 4676
+    },
+    {
+      "epoch": 4.53,
+      "grad_norm": 1.4020411968231201,
+      "learning_rate": 5.517097581317765e-06,
+      "loss": 0.6495,
+      "step": 4677
+    },
+    {
+      "epoch": 4.53,
+      "grad_norm": 1.2791095972061157,
+      "learning_rate": 5.512927439532944e-06,
+      "loss": 0.596,
+      "step": 4678
+    },
+    {
+      "epoch": 4.53,
+      "grad_norm": 1.73211669921875,
+      "learning_rate": 5.508757297748124e-06,
+      "loss": 0.7312,
+      "step": 4679
+    },
+    {
+      "epoch": 4.53,
+      "grad_norm": 1.699086308479309,
+      "learning_rate": 5.504587155963303e-06,
+      "loss": 0.5438,
+      "step": 4680
+    },
+    {
+      "epoch": 4.53,
+      "grad_norm": 1.799531102180481,
+      "learning_rate": 5.500417014178482e-06,
+      "loss": 0.7692,
+      "step": 4681
+    },
+    {
+      "epoch": 4.53,
+      "grad_norm": 1.633374810218811,
+      "learning_rate": 5.496246872393662e-06,
+      "loss": 0.7507,
+      "step": 4682
+    },
+    {
+      "epoch": 4.53,
+      "grad_norm": 1.4437192678451538,
+      "learning_rate": 5.492076730608841e-06,
+      "loss": 0.7854,
+      "step": 4683
+    },
+    {
+      "epoch": 4.53,
+      "grad_norm": 1.8305168151855469,
+      "learning_rate": 5.48790658882402e-06,
+      "loss": 0.604,
+      "step": 4684
+    },
+    {
+      "epoch": 4.53,
+      "grad_norm": 1.6433649063110352,
+      "learning_rate": 5.4837364470392e-06,
+      "loss": 0.7629,
+      "step": 4685
+    },
+    {
+      "epoch": 4.53,
+      "grad_norm": 1.6655476093292236,
+      "learning_rate": 5.4795663052543785e-06,
+      "loss": 0.7122,
+      "step": 4686
+    },
+    {
+      "epoch": 4.54,
+      "grad_norm": 1.635684609413147,
+      "learning_rate": 5.475396163469558e-06,
+      "loss": 0.6979,
+      "step": 4687
+    },
+    {
+      "epoch": 4.54,
+      "grad_norm": 1.8280525207519531,
+      "learning_rate": 5.471226021684738e-06,
+      "loss": 0.6252,
+      "step": 4688
+    },
+    {
+      "epoch": 4.54,
+      "grad_norm": 1.7299901247024536,
+      "learning_rate": 5.467055879899917e-06,
+      "loss": 0.5505,
+      "step": 4689
+    },
+    {
+      "epoch": 4.54,
+      "grad_norm": 1.6322293281555176,
+      "learning_rate": 5.462885738115096e-06,
+      "loss": 0.7157,
+      "step": 4690
+    },
+    {
+      "epoch": 4.54,
+      "grad_norm": 2.1160528659820557,
+      "learning_rate": 5.458715596330276e-06,
+      "loss": 0.7681,
+      "step": 4691
+    },
+    {
+      "epoch": 4.54,
+      "grad_norm": 1.3451793193817139,
+      "learning_rate": 5.4545454545454545e-06,
+      "loss": 0.6274,
+      "step": 4692
+    },
+    {
+      "epoch": 4.54,
+      "grad_norm": 1.2206450700759888,
+      "learning_rate": 5.450375312760634e-06,
+      "loss": 0.6895,
+      "step": 4693
+    },
+    {
+      "epoch": 4.54,
+      "grad_norm": 1.297421932220459,
+      "learning_rate": 5.446205170975814e-06,
+      "loss": 0.7076,
+      "step": 4694
+    },
+    {
+      "epoch": 4.54,
+      "grad_norm": 1.439626693725586,
+      "learning_rate": 5.442035029190993e-06,
+      "loss": 0.7507,
+      "step": 4695
+    },
+    {
+      "epoch": 4.54,
+      "grad_norm": 1.2901383638381958,
+      "learning_rate": 5.437864887406172e-06,
+      "loss": 0.5414,
+      "step": 4696
+    },
+    {
+      "epoch": 4.54,
+      "grad_norm": 1.759224772453308,
+      "learning_rate": 5.433694745621351e-06,
+      "loss": 0.7401,
+      "step": 4697
+    },
+    {
+      "epoch": 4.55,
+      "grad_norm": 1.2545334100723267,
+      "learning_rate": 5.4295246038365304e-06,
+      "loss": 0.689,
+      "step": 4698
+    },
+    {
+      "epoch": 4.55,
+      "grad_norm": 1.627297282218933,
+      "learning_rate": 5.42535446205171e-06,
+      "loss": 0.8067,
+      "step": 4699
+    },
+    {
+      "epoch": 4.55,
+      "grad_norm": 1.56171452999115,
+      "learning_rate": 5.42118432026689e-06,
+      "loss": 0.7103,
+      "step": 4700
+    },
+    {
+      "epoch": 4.55,
+      "grad_norm": 1.5313225984573364,
+      "learning_rate": 5.417014178482069e-06,
+      "loss": 0.5791,
+      "step": 4701
+    },
+    {
+      "epoch": 4.55,
+      "grad_norm": 1.573706030845642,
+      "learning_rate": 5.412844036697248e-06,
+      "loss": 0.7821,
+      "step": 4702
+    },
+    {
+      "epoch": 4.55,
+      "grad_norm": 1.542401671409607,
+      "learning_rate": 5.408673894912427e-06,
+      "loss": 0.5661,
+      "step": 4703
+    },
+    {
+      "epoch": 4.55,
+      "grad_norm": 1.9369721412658691,
+      "learning_rate": 5.404503753127606e-06,
+      "loss": 0.7782,
+      "step": 4704
+    },
+    {
+      "epoch": 4.55,
+      "grad_norm": 1.4803828001022339,
+      "learning_rate": 5.400333611342786e-06,
+      "loss": 0.7395,
+      "step": 4705
+    },
+    {
+      "epoch": 4.55,
+      "grad_norm": 1.692859411239624,
+      "learning_rate": 5.396163469557966e-06,
+      "loss": 0.6787,
+      "step": 4706
+    },
+    {
+      "epoch": 4.55,
+      "grad_norm": 1.4623578786849976,
+      "learning_rate": 5.391993327773145e-06,
+      "loss": 0.6155,
+      "step": 4707
+    },
+    {
+      "epoch": 4.56,
+      "grad_norm": 1.4121465682983398,
+      "learning_rate": 5.387823185988323e-06,
+      "loss": 0.6422,
+      "step": 4708
+    },
+    {
+      "epoch": 4.56,
+      "grad_norm": 1.2926206588745117,
+      "learning_rate": 5.383653044203503e-06,
+      "loss": 0.7233,
+      "step": 4709
+    },
+    {
+      "epoch": 4.56,
+      "grad_norm": 1.1981099843978882,
+      "learning_rate": 5.379482902418682e-06,
+      "loss": 0.7078,
+      "step": 4710
+    },
+    {
+      "epoch": 4.56,
+      "grad_norm": 1.4239003658294678,
+      "learning_rate": 5.375312760633862e-06,
+      "loss": 0.6135,
+      "step": 4711
+    },
+    {
+      "epoch": 4.56,
+      "grad_norm": 1.8219012022018433,
+      "learning_rate": 5.371142618849042e-06,
+      "loss": 0.6569,
+      "step": 4712
+    },
+    {
+      "epoch": 4.56,
+      "grad_norm": 1.68372642993927,
+      "learning_rate": 5.366972477064221e-06,
+      "loss": 0.4832,
+      "step": 4713
+    },
+    {
+      "epoch": 4.56,
+      "grad_norm": 1.3844177722930908,
+      "learning_rate": 5.362802335279399e-06,
+      "loss": 0.7542,
+      "step": 4714
+    },
+    {
+      "epoch": 4.56,
+      "grad_norm": 1.6422113180160522,
+      "learning_rate": 5.358632193494579e-06,
+      "loss": 0.6599,
+      "step": 4715
+    },
+    {
+      "epoch": 4.56,
+      "grad_norm": 1.2676887512207031,
+      "learning_rate": 5.354462051709758e-06,
+      "loss": 0.6356,
+      "step": 4716
+    },
+    {
+      "epoch": 4.56,
+      "grad_norm": 1.4765608310699463,
+      "learning_rate": 5.3502919099249376e-06,
+      "loss": 0.5737,
+      "step": 4717
+    },
+    {
+      "epoch": 4.57,
+      "grad_norm": 1.6787502765655518,
+      "learning_rate": 5.346121768140118e-06,
+      "loss": 0.5737,
+      "step": 4718
+    },
+    {
+      "epoch": 4.57,
+      "grad_norm": 1.5113779306411743,
+      "learning_rate": 5.341951626355296e-06,
+      "loss": 0.7146,
+      "step": 4719
+    },
+    {
+      "epoch": 4.57,
+      "grad_norm": 1.2630488872528076,
+      "learning_rate": 5.337781484570475e-06,
+      "loss": 0.5984,
+      "step": 4720
+    },
+    {
+      "epoch": 4.57,
+      "grad_norm": 1.5241637229919434,
+      "learning_rate": 5.333611342785655e-06,
+      "loss": 1.0374,
+      "step": 4721
+    },
+    {
+      "epoch": 4.57,
+      "grad_norm": 1.288524866104126,
+      "learning_rate": 5.329441201000834e-06,
+      "loss": 0.6198,
+      "step": 4722
+    },
+    {
+      "epoch": 4.57,
+      "grad_norm": 1.4171669483184814,
+      "learning_rate": 5.3252710592160135e-06,
+      "loss": 0.6126,
+      "step": 4723
+    },
+    {
+      "epoch": 4.57,
+      "grad_norm": 1.2440285682678223,
+      "learning_rate": 5.3211009174311936e-06,
+      "loss": 0.7572,
+      "step": 4724
+    },
+    {
+      "epoch": 4.57,
+      "grad_norm": 1.20148503780365,
+      "learning_rate": 5.316930775646372e-06,
+      "loss": 0.6161,
+      "step": 4725
+    },
+    {
+      "epoch": 4.57,
+      "grad_norm": 1.493597388267517,
+      "learning_rate": 5.312760633861551e-06,
+      "loss": 0.6352,
+      "step": 4726
+    },
+    {
+      "epoch": 4.57,
+      "grad_norm": 1.736620306968689,
+      "learning_rate": 5.308590492076731e-06,
+      "loss": 0.7501,
+      "step": 4727
+    },
+    {
+      "epoch": 4.57,
+      "grad_norm": 1.3761045932769775,
+      "learning_rate": 5.30442035029191e-06,
+      "loss": 0.7785,
+      "step": 4728
+    },
+    {
+      "epoch": 4.58,
+      "grad_norm": 1.5715200901031494,
+      "learning_rate": 5.3002502085070895e-06,
+      "loss": 0.658,
+      "step": 4729
+    },
+    {
+      "epoch": 4.58,
+      "grad_norm": 2.1754579544067383,
+      "learning_rate": 5.296080066722269e-06,
+      "loss": 0.7418,
+      "step": 4730
+    },
+    {
+      "epoch": 4.58,
+      "grad_norm": 1.5906949043273926,
+      "learning_rate": 5.291909924937448e-06,
+      "loss": 0.641,
+      "step": 4731
+    },
+    {
+      "epoch": 4.58,
+      "grad_norm": 1.359338402748108,
+      "learning_rate": 5.287739783152627e-06,
+      "loss": 0.5666,
+      "step": 4732
+    },
+    {
+      "epoch": 4.58,
+      "grad_norm": 1.6645625829696655,
+      "learning_rate": 5.283569641367807e-06,
+      "loss": 0.8261,
+      "step": 4733
+    },
+    {
+      "epoch": 4.58,
+      "grad_norm": 1.646085262298584,
+      "learning_rate": 5.279399499582986e-06,
+      "loss": 0.7408,
+      "step": 4734
+    },
+    {
+      "epoch": 4.58,
+      "grad_norm": 1.2797610759735107,
+      "learning_rate": 5.2752293577981655e-06,
+      "loss": 0.7439,
+      "step": 4735
+    },
+    {
+      "epoch": 4.58,
+      "grad_norm": 2.1849143505096436,
+      "learning_rate": 5.271059216013345e-06,
+      "loss": 0.76,
+      "step": 4736
+    },
+    {
+      "epoch": 4.58,
+      "grad_norm": 1.5100131034851074,
+      "learning_rate": 5.266889074228524e-06,
+      "loss": 0.7065,
+      "step": 4737
+    },
+    {
+      "epoch": 4.58,
+      "grad_norm": 1.7423347234725952,
+      "learning_rate": 5.262718932443703e-06,
+      "loss": 0.6817,
+      "step": 4738
+    },
+    {
+      "epoch": 4.59,
+      "grad_norm": 1.381550669670105,
+      "learning_rate": 5.258548790658883e-06,
+      "loss": 0.6844,
+      "step": 4739
+    },
+    {
+      "epoch": 4.59,
+      "grad_norm": 1.2365972995758057,
+      "learning_rate": 5.254378648874062e-06,
+      "loss": 0.7174,
+      "step": 4740
+    },
+    {
+      "epoch": 4.59,
+      "grad_norm": 1.9256683588027954,
+      "learning_rate": 5.250208507089241e-06,
+      "loss": 0.8264,
+      "step": 4741
+    },
+    {
+      "epoch": 4.59,
+      "grad_norm": 1.5130013227462769,
+      "learning_rate": 5.246038365304421e-06,
+      "loss": 0.6881,
+      "step": 4742
+    },
+    {
+      "epoch": 4.59,
+      "grad_norm": 1.4686535596847534,
+      "learning_rate": 5.2418682235196e-06,
+      "loss": 0.7101,
+      "step": 4743
+    },
+    {
+      "epoch": 4.59,
+      "grad_norm": 1.4216227531433105,
+      "learning_rate": 5.237698081734779e-06,
+      "loss": 0.7207,
+      "step": 4744
+    },
+    {
+      "epoch": 4.59,
+      "grad_norm": 1.66473388671875,
+      "learning_rate": 5.233527939949959e-06,
+      "loss": 0.7675,
+      "step": 4745
+    },
+    {
+      "epoch": 4.59,
+      "grad_norm": 1.5849100351333618,
+      "learning_rate": 5.229357798165138e-06,
+      "loss": 0.6856,
+      "step": 4746
+    },
+    {
+      "epoch": 4.59,
+      "grad_norm": 1.4400815963745117,
+      "learning_rate": 5.225187656380317e-06,
+      "loss": 0.6889,
+      "step": 4747
+    },
+    {
+      "epoch": 4.59,
+      "grad_norm": 1.6706531047821045,
+      "learning_rate": 5.221017514595497e-06,
+      "loss": 0.7243,
+      "step": 4748
+    },
+    {
+      "epoch": 4.6,
+      "grad_norm": 1.865055799484253,
+      "learning_rate": 5.216847372810676e-06,
+      "loss": 0.7578,
+      "step": 4749
+    },
+    {
+      "epoch": 4.6,
+      "grad_norm": 1.6698907613754272,
+      "learning_rate": 5.212677231025855e-06,
+      "loss": 0.6211,
+      "step": 4750
+    },
+    {
+      "epoch": 4.6,
+      "grad_norm": 1.5708190202713013,
+      "learning_rate": 5.208507089241035e-06,
+      "loss": 0.7566,
+      "step": 4751
+    },
+    {
+      "epoch": 4.6,
+      "grad_norm": 2.0286459922790527,
+      "learning_rate": 5.204336947456213e-06,
+      "loss": 0.6628,
+      "step": 4752
+    },
+    {
+      "epoch": 4.6,
+      "grad_norm": 1.241782307624817,
+      "learning_rate": 5.200166805671393e-06,
+      "loss": 0.6194,
+      "step": 4753
+    },
+    {
+      "epoch": 4.6,
+      "grad_norm": 1.5183227062225342,
+      "learning_rate": 5.195996663886573e-06,
+      "loss": 0.7809,
+      "step": 4754
+    },
+    {
+      "epoch": 4.6,
+      "grad_norm": 1.5377471446990967,
+      "learning_rate": 5.191826522101752e-06,
+      "loss": 0.8283,
+      "step": 4755
+    },
+    {
+      "epoch": 4.6,
+      "grad_norm": 1.3237930536270142,
+      "learning_rate": 5.187656380316931e-06,
+      "loss": 0.7827,
+      "step": 4756
+    },
+    {
+      "epoch": 4.6,
+      "grad_norm": 1.519793152809143,
+      "learning_rate": 5.183486238532111e-06,
+      "loss": 0.6665,
+      "step": 4757
+    },
+    {
+      "epoch": 4.6,
+      "grad_norm": 1.4032889604568481,
+      "learning_rate": 5.179316096747289e-06,
+      "loss": 0.6367,
+      "step": 4758
+    },
+    {
+      "epoch": 4.6,
+      "grad_norm": 1.611026406288147,
+      "learning_rate": 5.1751459549624686e-06,
+      "loss": 0.6187,
+      "step": 4759
+    },
+    {
+      "epoch": 4.61,
+      "grad_norm": 1.4174565076828003,
+      "learning_rate": 5.170975813177649e-06,
+      "loss": 1.0453,
+      "step": 4760
+    },
+    {
+      "epoch": 4.61,
+      "grad_norm": 1.5549442768096924,
+      "learning_rate": 5.166805671392828e-06,
+      "loss": 0.6455,
+      "step": 4761
+    },
+    {
+      "epoch": 4.61,
+      "grad_norm": 1.5228101015090942,
+      "learning_rate": 5.162635529608007e-06,
+      "loss": 0.6295,
+      "step": 4762
+    },
+    {
+      "epoch": 4.61,
+      "grad_norm": 1.405816674232483,
+      "learning_rate": 5.158465387823186e-06,
+      "loss": 0.6324,
+      "step": 4763
+    },
+    {
+      "epoch": 4.61,
+      "grad_norm": 1.5264065265655518,
+      "learning_rate": 5.154295246038365e-06,
+      "loss": 0.6643,
+      "step": 4764
+    },
+    {
+      "epoch": 4.61,
+      "grad_norm": 1.7057868242263794,
+      "learning_rate": 5.1501251042535445e-06,
+      "loss": 0.638,
+      "step": 4765
+    },
+    {
+      "epoch": 4.61,
+      "grad_norm": 1.5288913249969482,
+      "learning_rate": 5.1459549624687246e-06,
+      "loss": 0.7412,
+      "step": 4766
+    },
+    {
+      "epoch": 4.61,
+      "grad_norm": 1.7368863821029663,
+      "learning_rate": 5.141784820683904e-06,
+      "loss": 0.6938,
+      "step": 4767
+    },
+    {
+      "epoch": 4.61,
+      "grad_norm": 2.2275304794311523,
+      "learning_rate": 5.137614678899083e-06,
+      "loss": 0.7366,
+      "step": 4768
+    },
+    {
+      "epoch": 4.61,
+      "grad_norm": 1.8191800117492676,
+      "learning_rate": 5.133444537114262e-06,
+      "loss": 0.7383,
+      "step": 4769
+    },
+    {
+      "epoch": 4.62,
+      "grad_norm": 1.4102978706359863,
+      "learning_rate": 5.129274395329441e-06,
+      "loss": 0.74,
+      "step": 4770
+    },
+    {
+      "epoch": 4.62,
+      "grad_norm": 1.8225984573364258,
+      "learning_rate": 5.1251042535446205e-06,
+      "loss": 0.721,
+      "step": 4771
+    },
+    {
+      "epoch": 4.62,
+      "grad_norm": 1.3556174039840698,
+      "learning_rate": 5.1209341117598006e-06,
+      "loss": 0.7782,
+      "step": 4772
+    },
+    {
+      "epoch": 4.62,
+      "grad_norm": 1.751905918121338,
+      "learning_rate": 5.11676396997498e-06,
+      "loss": 0.8165,
+      "step": 4773
+    },
+    {
+      "epoch": 4.62,
+      "grad_norm": 1.4879447221755981,
+      "learning_rate": 5.112593828190159e-06,
+      "loss": 0.6494,
+      "step": 4774
+    },
+    {
+      "epoch": 4.62,
+      "grad_norm": 1.6665265560150146,
+      "learning_rate": 5.108423686405338e-06,
+      "loss": 0.5807,
+      "step": 4775
+    },
+    {
+      "epoch": 4.62,
+      "grad_norm": 1.3465780019760132,
+      "learning_rate": 5.104253544620517e-06,
+      "loss": 0.7655,
+      "step": 4776
+    },
+    {
+      "epoch": 4.62,
+      "grad_norm": 1.2939397096633911,
+      "learning_rate": 5.1000834028356965e-06,
+      "loss": 0.6662,
+      "step": 4777
+    },
+    {
+      "epoch": 4.62,
+      "grad_norm": 1.625292420387268,
+      "learning_rate": 5.0959132610508765e-06,
+      "loss": 0.7895,
+      "step": 4778
+    },
+    {
+      "epoch": 4.62,
+      "grad_norm": 1.1519323587417603,
+      "learning_rate": 5.091743119266056e-06,
+      "loss": 0.748,
+      "step": 4779
+    },
+    {
+      "epoch": 4.63,
+      "grad_norm": 1.2725067138671875,
+      "learning_rate": 5.087572977481234e-06,
+      "loss": 0.8611,
+      "step": 4780
+    },
+    {
+      "epoch": 4.63,
+      "grad_norm": 1.5201002359390259,
+      "learning_rate": 5.083402835696414e-06,
+      "loss": 0.6974,
+      "step": 4781
+    },
+    {
+      "epoch": 4.63,
+      "grad_norm": 1.7186866998672485,
+      "learning_rate": 5.079232693911593e-06,
+      "loss": 0.6424,
+      "step": 4782
+    },
+    {
+      "epoch": 4.63,
+      "grad_norm": 1.6808973550796509,
+      "learning_rate": 5.0750625521267725e-06,
+      "loss": 0.6133,
+      "step": 4783
+    },
+    {
+      "epoch": 4.63,
+      "grad_norm": 1.4706043004989624,
+      "learning_rate": 5.0708924103419525e-06,
+      "loss": 0.7042,
+      "step": 4784
+    },
+    {
+      "epoch": 4.63,
+      "grad_norm": 1.3771601915359497,
+      "learning_rate": 5.066722268557132e-06,
+      "loss": 0.6956,
+      "step": 4785
+    },
+    {
+      "epoch": 4.63,
+      "grad_norm": 1.5026473999023438,
+      "learning_rate": 5.06255212677231e-06,
+      "loss": 0.6319,
+      "step": 4786
+    },
+    {
+      "epoch": 4.63,
+      "grad_norm": 1.444579839706421,
+      "learning_rate": 5.05838198498749e-06,
+      "loss": 0.7188,
+      "step": 4787
+    },
+    {
+      "epoch": 4.63,
+      "grad_norm": 1.5965029001235962,
+      "learning_rate": 5.054211843202669e-06,
+      "loss": 0.6373,
+      "step": 4788
+    },
+    {
+      "epoch": 4.63,
+      "grad_norm": 1.3108526468276978,
+      "learning_rate": 5.0500417014178485e-06,
+      "loss": 0.6486,
+      "step": 4789
+    },
+    {
+      "epoch": 4.63,
+      "grad_norm": 1.3351519107818604,
+      "learning_rate": 5.045871559633028e-06,
+      "loss": 0.61,
+      "step": 4790
+    },
+    {
+      "epoch": 4.64,
+      "grad_norm": 1.2217323780059814,
+      "learning_rate": 5.041701417848207e-06,
+      "loss": 0.6368,
+      "step": 4791
+    },
+    {
+      "epoch": 4.64,
+      "grad_norm": 1.9024354219436646,
+      "learning_rate": 5.037531276063386e-06,
+      "loss": 0.7607,
+      "step": 4792
+    },
+    {
+      "epoch": 4.64,
+      "grad_norm": 1.520367980003357,
+      "learning_rate": 5.033361134278566e-06,
+      "loss": 0.6358,
+      "step": 4793
+    },
+    {
+      "epoch": 4.64,
+      "grad_norm": 1.536771297454834,
+      "learning_rate": 5.029190992493745e-06,
+      "loss": 0.7668,
+      "step": 4794
+    },
+    {
+      "epoch": 4.64,
+      "grad_norm": 1.5564029216766357,
+      "learning_rate": 5.0250208507089244e-06,
+      "loss": 0.7745,
+      "step": 4795
+    },
+    {
+      "epoch": 4.64,
+      "grad_norm": 1.4659467935562134,
+      "learning_rate": 5.020850708924104e-06,
+      "loss": 0.6083,
+      "step": 4796
+    },
+    {
+      "epoch": 4.64,
+      "grad_norm": 1.8061070442199707,
+      "learning_rate": 5.016680567139283e-06,
+      "loss": 0.8638,
+      "step": 4797
+    },
+    {
+      "epoch": 4.64,
+      "grad_norm": 1.3351647853851318,
+      "learning_rate": 5.012510425354462e-06,
+      "loss": 0.8551,
+      "step": 4798
+    },
+    {
+      "epoch": 4.64,
+      "grad_norm": 1.2969155311584473,
+      "learning_rate": 5.008340283569642e-06,
+      "loss": 0.612,
+      "step": 4799
+    },
+    {
+      "epoch": 4.64,
+      "grad_norm": 1.4245238304138184,
+      "learning_rate": 5.004170141784821e-06,
+      "loss": 0.6597,
+      "step": 4800
+    },
+    {
+      "epoch": 4.64,
+      "eval_loss": 0.8592209815979004,
+      "eval_runtime": 856.5817,
+      "eval_samples_per_second": 4.826,
+      "eval_steps_per_second": 0.604,
+      "step": 4800
     }
   ],
   "logging_steps": 1,
@@ -25310,7 +33742,7 @@
   "num_input_tokens_seen": 0,
   "num_train_epochs": 6,
   "save_steps": 300,
-  "total_flos": 1.566461114115588e+18,
+  "total_flos": 2.092122684281979e+18,
   "train_batch_size": 4,
   "trial_name": null,
   "trial_params": null