diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5810 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9999697418983933,
+  "eval_steps": 500,
+  "global_step": 8262,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 44.79621505737305,
+      "learning_rate": 2.5000000000000004e-07,
+      "loss": 6.2908,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 43.94892501831055,
+      "learning_rate": 2.5e-06,
+      "loss": 5.9803,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 23.46196174621582,
+      "learning_rate": 5e-06,
+      "loss": 4.5474,
+      "step": 20
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 13.829703330993652,
+      "learning_rate": 7.5e-06,
+      "loss": 2.7988,
+      "step": 30
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 6.53416633605957,
+      "learning_rate": 1e-05,
+      "loss": 1.6724,
+      "step": 40
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.733919620513916,
+      "learning_rate": 1.25e-05,
+      "loss": 1.0302,
+      "step": 50
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.498934268951416,
+      "learning_rate": 1.5e-05,
+      "loss": 0.8931,
+      "step": 60
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.673191547393799,
+      "learning_rate": 1.75e-05,
+      "loss": 0.8022,
+      "step": 70
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.530500888824463,
+      "learning_rate": 2e-05,
+      "loss": 0.7569,
+      "step": 80
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.044854164123535,
+      "learning_rate": 2.25e-05,
+      "loss": 0.7018,
+      "step": 90
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.3081376552581787,
+      "learning_rate": 2.5e-05,
+      "loss": 0.7441,
+      "step": 100
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.033691167831421,
+      "learning_rate": 2.7500000000000004e-05,
+      "loss": 0.7261,
+      "step": 110
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.066368818283081,
+      "learning_rate": 3e-05,
+      "loss": 0.7129,
+      "step": 120
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.072371482849121,
+      "learning_rate": 3.2500000000000004e-05,
+      "loss": 0.6918,
+      "step": 130
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.3042588233947754,
+      "learning_rate": 3.5e-05,
+      "loss": 0.6664,
+      "step": 140
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.430894374847412,
+      "learning_rate": 3.7500000000000003e-05,
+      "loss": 0.6787,
+      "step": 150
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.5288453102111816,
+      "learning_rate": 4e-05,
+      "loss": 0.6568,
+      "step": 160
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.180598020553589,
+      "learning_rate": 4.25e-05,
+      "loss": 0.6806,
+      "step": 170
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.3031888008117676,
+      "learning_rate": 4.5e-05,
+      "loss": 0.6649,
+      "step": 180
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.2974207401275635,
+      "learning_rate": 4.75e-05,
+      "loss": 0.6582,
+      "step": 190
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.340533971786499,
+      "learning_rate": 5e-05,
+      "loss": 0.6166,
+      "step": 200
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.238347291946411,
+      "learning_rate": 4.996937025238913e-05,
+      "loss": 0.6613,
+      "step": 210
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.9168024063110352,
+      "learning_rate": 4.993874050477824e-05,
+      "loss": 0.6526,
+      "step": 220
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 7.175029754638672,
+      "learning_rate": 4.9908110757167364e-05,
+      "loss": 0.612,
+      "step": 230
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.5029752254486084,
+      "learning_rate": 4.987748100955648e-05,
+      "loss": 0.6545,
+      "step": 240
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.5689034461975098,
+      "learning_rate": 4.98468512619456e-05,
+      "loss": 0.5908,
+      "step": 250
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.646583080291748,
+      "learning_rate": 4.9816221514334726e-05,
+      "loss": 0.619,
+      "step": 260
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.582393765449524,
+      "learning_rate": 4.9785591766723845e-05,
+      "loss": 0.5844,
+      "step": 270
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.5414413213729858,
+      "learning_rate": 4.975496201911296e-05,
+      "loss": 0.6054,
+      "step": 280
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.632761836051941,
+      "learning_rate": 4.972433227150208e-05,
+      "loss": 0.584,
+      "step": 290
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.6687579154968262,
+      "learning_rate": 4.9693702523891206e-05,
+      "loss": 0.6424,
+      "step": 300
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.4187216758728027,
+      "learning_rate": 4.966307277628033e-05,
+      "loss": 0.5808,
+      "step": 310
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.4725416898727417,
+      "learning_rate": 4.963244302866944e-05,
+      "loss": 0.5662,
+      "step": 320
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.9247803688049316,
+      "learning_rate": 4.960181328105857e-05,
+      "loss": 0.5577,
+      "step": 330
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.6531754732131958,
+      "learning_rate": 4.957118353344769e-05,
+      "loss": 0.554,
+      "step": 340
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.3355740308761597,
+      "learning_rate": 4.9540553785836805e-05,
+      "loss": 0.5791,
+      "step": 350
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.8228713274002075,
+      "learning_rate": 4.950992403822593e-05,
+      "loss": 0.5796,
+      "step": 360
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.9731523990631104,
+      "learning_rate": 4.947929429061505e-05,
+      "loss": 0.5759,
+      "step": 370
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.4433354139328003,
+      "learning_rate": 4.944866454300417e-05,
+      "loss": 0.5389,
+      "step": 380
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.272991418838501,
+      "learning_rate": 4.9418034795393285e-05,
+      "loss": 0.5303,
+      "step": 390
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.1282144784927368,
+      "learning_rate": 4.938740504778241e-05,
+      "loss": 0.5306,
+      "step": 400
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.1406899690628052,
+      "learning_rate": 4.935677530017153e-05,
+      "loss": 0.5412,
+      "step": 410
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.461844801902771,
+      "learning_rate": 4.932614555256065e-05,
+      "loss": 0.556,
+      "step": 420
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.4814263582229614,
+      "learning_rate": 4.929551580494977e-05,
+      "loss": 0.5381,
+      "step": 430
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.1951137781143188,
+      "learning_rate": 4.926488605733889e-05,
+      "loss": 0.5582,
+      "step": 440
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.3326703310012817,
+      "learning_rate": 4.923425630972801e-05,
+      "loss": 0.5075,
+      "step": 450
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.4733253717422485,
+      "learning_rate": 4.9203626562117134e-05,
+      "loss": 0.5135,
+      "step": 460
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.4258859157562256,
+      "learning_rate": 4.917299681450625e-05,
+      "loss": 0.5497,
+      "step": 470
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.315460205078125,
+      "learning_rate": 4.914236706689537e-05,
+      "loss": 0.4966,
+      "step": 480
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.0685410499572754,
+      "learning_rate": 4.911173731928449e-05,
+      "loss": 0.4967,
+      "step": 490
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.591590166091919,
+      "learning_rate": 4.9081107571673614e-05,
+      "loss": 0.4755,
+      "step": 500
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.3233535289764404,
+      "learning_rate": 4.9050477824062726e-05,
+      "loss": 0.4999,
+      "step": 510
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.4207834005355835,
+      "learning_rate": 4.901984807645185e-05,
+      "loss": 0.4936,
+      "step": 520
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.4395300149917603,
+      "learning_rate": 4.8989218328840976e-05,
+      "loss": 0.4954,
+      "step": 530
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.7463901042938232,
+      "learning_rate": 4.8958588581230095e-05,
+      "loss": 0.4964,
+      "step": 540
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.2900428771972656,
+      "learning_rate": 4.892795883361921e-05,
+      "loss": 0.5025,
+      "step": 550
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.444162130355835,
+      "learning_rate": 4.889732908600833e-05,
+      "loss": 0.5223,
+      "step": 560
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.340908408164978,
+      "learning_rate": 4.886669933839746e-05,
+      "loss": 0.4691,
+      "step": 570
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.3857574462890625,
+      "learning_rate": 4.8836069590786575e-05,
+      "loss": 0.4834,
+      "step": 580
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.3462368249893188,
+      "learning_rate": 4.880543984317569e-05,
+      "loss": 0.4798,
+      "step": 590
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.5114245414733887,
+      "learning_rate": 4.877481009556482e-05,
+      "loss": 0.4424,
+      "step": 600
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.418565273284912,
+      "learning_rate": 4.874418034795393e-05,
+      "loss": 0.4685,
+      "step": 610
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.2730127573013306,
+      "learning_rate": 4.8713550600343055e-05,
+      "loss": 0.4696,
+      "step": 620
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.4079108238220215,
+      "learning_rate": 4.868292085273218e-05,
+      "loss": 0.4681,
+      "step": 630
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.4005239009857178,
+      "learning_rate": 4.865229110512129e-05,
+      "loss": 0.4473,
+      "step": 640
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.169668197631836,
+      "learning_rate": 4.862166135751042e-05,
+      "loss": 0.4486,
+      "step": 650
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.5414419174194336,
+      "learning_rate": 4.8591031609899536e-05,
+      "loss": 0.4665,
+      "step": 660
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.3343652486801147,
+      "learning_rate": 4.8560401862288654e-05,
+      "loss": 0.4432,
+      "step": 670
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.4561718702316284,
+      "learning_rate": 4.852977211467778e-05,
+      "loss": 0.4793,
+      "step": 680
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.444230556488037,
+      "learning_rate": 4.84991423670669e-05,
+      "loss": 0.4663,
+      "step": 690
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.0794904232025146,
+      "learning_rate": 4.8468512619456016e-05,
+      "loss": 0.4611,
+      "step": 700
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3252354860305786,
+      "learning_rate": 4.8437882871845134e-05,
+      "loss": 0.4532,
+      "step": 710
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3652528524398804,
+      "learning_rate": 4.840725312423426e-05,
+      "loss": 0.4186,
+      "step": 720
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.8579872846603394,
+      "learning_rate": 4.8376623376623384e-05,
+      "loss": 0.4091,
+      "step": 730
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.7196540832519531,
+      "learning_rate": 4.8345993629012496e-05,
+      "loss": 0.4478,
+      "step": 740
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3259663581848145,
+      "learning_rate": 4.831536388140162e-05,
+      "loss": 0.4333,
+      "step": 750
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.2580039501190186,
+      "learning_rate": 4.828473413379074e-05,
+      "loss": 0.4224,
+      "step": 760
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.5129112005233765,
+      "learning_rate": 4.825410438617986e-05,
+      "loss": 0.4456,
+      "step": 770
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3986139297485352,
+      "learning_rate": 4.822347463856898e-05,
+      "loss": 0.4053,
+      "step": 780
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.2099454402923584,
+      "learning_rate": 4.81928448909581e-05,
+      "loss": 0.422,
+      "step": 790
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.071122646331787,
+      "learning_rate": 4.816221514334722e-05,
+      "loss": 0.4387,
+      "step": 800
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.465188980102539,
+      "learning_rate": 4.813158539573634e-05,
+      "loss": 0.3949,
+      "step": 810
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.3556270599365234,
+      "learning_rate": 4.810095564812546e-05,
+      "loss": 0.394,
+      "step": 820
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.4712861776351929,
+      "learning_rate": 4.807032590051458e-05,
+      "loss": 0.3831,
+      "step": 830
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.6468979120254517,
+      "learning_rate": 4.80396961529037e-05,
+      "loss": 0.4113,
+      "step": 840
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.6901087760925293,
+      "learning_rate": 4.8009066405292825e-05,
+      "loss": 0.4219,
+      "step": 850
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.4851107597351074,
+      "learning_rate": 4.7978436657681944e-05,
+      "loss": 0.4135,
+      "step": 860
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.6299171447753906,
+      "learning_rate": 4.794780691007106e-05,
+      "loss": 0.3947,
+      "step": 870
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.5189801454544067,
+      "learning_rate": 4.791717716246019e-05,
+      "loss": 0.4008,
+      "step": 880
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.386616826057434,
+      "learning_rate": 4.7886547414849305e-05,
+      "loss": 0.4107,
+      "step": 890
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.5139007568359375,
+      "learning_rate": 4.7855917667238424e-05,
+      "loss": 0.4225,
+      "step": 900
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.7688578367233276,
+      "learning_rate": 4.782528791962754e-05,
+      "loss": 0.4063,
+      "step": 910
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.5936031341552734,
+      "learning_rate": 4.779465817201667e-05,
+      "loss": 0.3858,
+      "step": 920
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.3068761825561523,
+      "learning_rate": 4.7764028424405786e-05,
+      "loss": 0.3875,
+      "step": 930
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.721594214439392,
+      "learning_rate": 4.7733398676794904e-05,
+      "loss": 0.4035,
+      "step": 940
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.5830979347229004,
+      "learning_rate": 4.770276892918403e-05,
+      "loss": 0.4263,
+      "step": 950
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.4326703548431396,
+      "learning_rate": 4.767213918157315e-05,
+      "loss": 0.3945,
+      "step": 960
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.4239739179611206,
+      "learning_rate": 4.7641509433962266e-05,
+      "loss": 0.3893,
+      "step": 970
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.5029996633529663,
+      "learning_rate": 4.761087968635139e-05,
+      "loss": 0.3832,
+      "step": 980
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.542860507965088,
+      "learning_rate": 4.758024993874051e-05,
+      "loss": 0.394,
+      "step": 990
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.5541925430297852,
+      "learning_rate": 4.754962019112963e-05,
+      "loss": 0.3766,
+      "step": 1000
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.515035629272461,
+      "learning_rate": 4.7518990443518746e-05,
+      "loss": 0.3931,
+      "step": 1010
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.4406639337539673,
+      "learning_rate": 4.748836069590787e-05,
+      "loss": 0.3944,
+      "step": 1020
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.340774416923523,
+      "learning_rate": 4.745773094829698e-05,
+      "loss": 0.3839,
+      "step": 1030
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.888028621673584,
+      "learning_rate": 4.742710120068611e-05,
+      "loss": 0.379,
+      "step": 1040
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.4663670063018799,
+      "learning_rate": 4.739647145307523e-05,
+      "loss": 0.3973,
+      "step": 1050
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.2871754169464111,
+      "learning_rate": 4.7365841705464345e-05,
+      "loss": 0.3574,
+      "step": 1060
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.1411924362182617,
+      "learning_rate": 4.733521195785347e-05,
+      "loss": 0.3599,
+      "step": 1070
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.6594351530075073,
+      "learning_rate": 4.730458221024259e-05,
+      "loss": 0.4149,
+      "step": 1080
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.224915862083435,
+      "learning_rate": 4.727395246263171e-05,
+      "loss": 0.3603,
+      "step": 1090
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.6931504011154175,
+      "learning_rate": 4.724332271502083e-05,
+      "loss": 0.3834,
+      "step": 1100
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.5090054273605347,
+      "learning_rate": 4.721269296740995e-05,
+      "loss": 0.3779,
+      "step": 1110
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.4564669132232666,
+      "learning_rate": 4.718206321979907e-05,
+      "loss": 0.3817,
+      "step": 1120
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.5073637962341309,
+      "learning_rate": 4.715143347218819e-05,
+      "loss": 0.3453,
+      "step": 1130
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.3254026174545288,
+      "learning_rate": 4.712080372457731e-05,
+      "loss": 0.3627,
+      "step": 1140
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.5201656818389893,
+      "learning_rate": 4.709017397696644e-05,
+      "loss": 0.3711,
+      "step": 1150
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.2688968181610107,
+      "learning_rate": 4.705954422935555e-05,
+      "loss": 0.3629,
+      "step": 1160
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.046254873275757,
+      "learning_rate": 4.7028914481744674e-05,
+      "loss": 0.3638,
+      "step": 1170
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.4509512186050415,
+      "learning_rate": 4.699828473413379e-05,
+      "loss": 0.3526,
+      "step": 1180
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.6802263259887695,
+      "learning_rate": 4.696765498652291e-05,
+      "loss": 0.3794,
+      "step": 1190
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.5504150390625,
+      "learning_rate": 4.6937025238912036e-05,
+      "loss": 0.3547,
+      "step": 1200
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.4444814920425415,
+      "learning_rate": 4.6906395491301154e-05,
+      "loss": 0.3637,
+      "step": 1210
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.6725267171859741,
+      "learning_rate": 4.687576574369027e-05,
+      "loss": 0.3445,
+      "step": 1220
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.4031798839569092,
+      "learning_rate": 4.684513599607939e-05,
+      "loss": 0.3883,
+      "step": 1230
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.853564977645874,
+      "learning_rate": 4.6814506248468516e-05,
+      "loss": 0.3967,
+      "step": 1240
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.5319392681121826,
+      "learning_rate": 4.6783876500857635e-05,
+      "loss": 0.3548,
+      "step": 1250
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.3312479257583618,
+      "learning_rate": 4.675324675324675e-05,
+      "loss": 0.3658,
+      "step": 1260
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.9926903247833252,
+      "learning_rate": 4.672261700563588e-05,
+      "loss": 0.334,
+      "step": 1270
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.3891220092773438,
+      "learning_rate": 4.6691987258024996e-05,
+      "loss": 0.3309,
+      "step": 1280
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.2705334424972534,
+      "learning_rate": 4.6661357510414115e-05,
+      "loss": 0.3312,
+      "step": 1290
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.541327714920044,
+      "learning_rate": 4.663072776280324e-05,
+      "loss": 0.3491,
+      "step": 1300
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.216559648513794,
+      "learning_rate": 4.660009801519236e-05,
+      "loss": 0.3229,
+      "step": 1310
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.2766779661178589,
+      "learning_rate": 4.656946826758148e-05,
+      "loss": 0.331,
+      "step": 1320
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.5856209993362427,
+      "learning_rate": 4.6538838519970595e-05,
+      "loss": 0.3475,
+      "step": 1330
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.344161033630371,
+      "learning_rate": 4.650820877235972e-05,
+      "loss": 0.3407,
+      "step": 1340
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.7017796039581299,
+      "learning_rate": 4.647757902474884e-05,
+      "loss": 0.3415,
+      "step": 1350
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.6224266290664673,
+      "learning_rate": 4.644694927713796e-05,
+      "loss": 0.3653,
+      "step": 1360
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.68430495262146,
+      "learning_rate": 4.641631952952708e-05,
+      "loss": 0.3636,
+      "step": 1370
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.495883822441101,
+      "learning_rate": 4.63856897819162e-05,
+      "loss": 0.3435,
+      "step": 1380
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.3294087648391724,
+      "learning_rate": 4.635506003430532e-05,
+      "loss": 0.3008,
+      "step": 1390
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.7772376537322998,
+      "learning_rate": 4.6324430286694444e-05,
+      "loss": 0.3289,
+      "step": 1400
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.5760084390640259,
+      "learning_rate": 4.629380053908356e-05,
+      "loss": 0.3368,
+      "step": 1410
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.2715150117874146,
+      "learning_rate": 4.626317079147268e-05,
+      "loss": 0.3259,
+      "step": 1420
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.6388717889785767,
+      "learning_rate": 4.62325410438618e-05,
+      "loss": 0.344,
+      "step": 1430
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.3928598165512085,
+      "learning_rate": 4.6201911296250924e-05,
+      "loss": 0.3397,
+      "step": 1440
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.310430884361267,
+      "learning_rate": 4.617128154864004e-05,
+      "loss": 0.3235,
+      "step": 1450
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.0350983142852783,
+      "learning_rate": 4.614065180102916e-05,
+      "loss": 0.3306,
+      "step": 1460
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.626246452331543,
+      "learning_rate": 4.6110022053418286e-05,
+      "loss": 0.3156,
+      "step": 1470
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.236250638961792,
+      "learning_rate": 4.60793923058074e-05,
+      "loss": 0.3145,
+      "step": 1480
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.6814018487930298,
+      "learning_rate": 4.604876255819652e-05,
+      "loss": 0.3517,
+      "step": 1490
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.8516896963119507,
+      "learning_rate": 4.601813281058565e-05,
+      "loss": 0.3308,
+      "step": 1500
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.634451985359192,
+      "learning_rate": 4.598750306297476e-05,
+      "loss": 0.3284,
+      "step": 1510
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.8714609146118164,
+      "learning_rate": 4.5956873315363885e-05,
+      "loss": 0.3264,
+      "step": 1520
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.3867897987365723,
+      "learning_rate": 4.5926243567753e-05,
+      "loss": 0.3227,
+      "step": 1530
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.2322877645492554,
+      "learning_rate": 4.589561382014212e-05,
+      "loss": 0.3227,
+      "step": 1540
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.8601164817810059,
+      "learning_rate": 4.586498407253124e-05,
+      "loss": 0.3529,
+      "step": 1550
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.336089849472046,
+      "learning_rate": 4.5834354324920365e-05,
+      "loss": 0.3306,
+      "step": 1560
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.3384654521942139,
+      "learning_rate": 4.580372457730949e-05,
+      "loss": 0.3179,
+      "step": 1570
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.7255178689956665,
+      "learning_rate": 4.57730948296986e-05,
+      "loss": 0.2986,
+      "step": 1580
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.4233754873275757,
+      "learning_rate": 4.574246508208773e-05,
+      "loss": 0.3327,
+      "step": 1590
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 2.0262444019317627,
+      "learning_rate": 4.5711835334476845e-05,
+      "loss": 0.3184,
+      "step": 1600
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.5325473546981812,
+      "learning_rate": 4.5681205586865964e-05,
+      "loss": 0.3393,
+      "step": 1610
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.3027546405792236,
+      "learning_rate": 4.565057583925509e-05,
+      "loss": 0.3117,
+      "step": 1620
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.3333057165145874,
+      "learning_rate": 4.561994609164421e-05,
+      "loss": 0.2971,
+      "step": 1630
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.5342081785202026,
+      "learning_rate": 4.5589316344033326e-05,
+      "loss": 0.3332,
+      "step": 1640
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.941325306892395,
+      "learning_rate": 4.5558686596422444e-05,
+      "loss": 0.3111,
+      "step": 1650
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.0228817462921143,
+      "learning_rate": 4.552805684881157e-05,
+      "loss": 0.3522,
+      "step": 1660
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.4719650745391846,
+      "learning_rate": 4.549742710120069e-05,
+      "loss": 0.3004,
+      "step": 1670
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.722540020942688,
+      "learning_rate": 4.5466797353589806e-05,
+      "loss": 0.3295,
+      "step": 1680
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.4042344093322754,
+      "learning_rate": 4.543616760597893e-05,
+      "loss": 0.3099,
+      "step": 1690
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.457822322845459,
+      "learning_rate": 4.540553785836805e-05,
+      "loss": 0.3071,
+      "step": 1700
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.4983888864517212,
+      "learning_rate": 4.537490811075717e-05,
+      "loss": 0.2697,
+      "step": 1710
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.7834343910217285,
+      "learning_rate": 4.534427836314629e-05,
+      "loss": 0.3017,
+      "step": 1720
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.6020764112472534,
+      "learning_rate": 4.531364861553541e-05,
+      "loss": 0.3131,
+      "step": 1730
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.1770570278167725,
+      "learning_rate": 4.528301886792453e-05,
+      "loss": 0.3224,
+      "step": 1740
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.6841928958892822,
+      "learning_rate": 4.525238912031365e-05,
+      "loss": 0.3358,
+      "step": 1750
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.511427879333496,
+      "learning_rate": 4.522175937270277e-05,
+      "loss": 0.2918,
+      "step": 1760
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.1187732219696045,
+      "learning_rate": 4.519112962509189e-05,
+      "loss": 0.2959,
+      "step": 1770
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.5806143283843994,
+      "learning_rate": 4.516049987748101e-05,
+      "loss": 0.2746,
+      "step": 1780
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.5360352993011475,
+      "learning_rate": 4.5129870129870135e-05,
+      "loss": 0.3212,
+      "step": 1790
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.458064079284668,
+      "learning_rate": 4.509924038225925e-05,
+      "loss": 0.2629,
+      "step": 1800
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.633918046951294,
+      "learning_rate": 4.506861063464837e-05,
+      "loss": 0.3177,
+      "step": 1810
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.3995980024337769,
+      "learning_rate": 4.50379808870375e-05,
+      "loss": 0.2958,
+      "step": 1820
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.7348047494888306,
+      "learning_rate": 4.5007351139426615e-05,
+      "loss": 0.289,
+      "step": 1830
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.2848684787750244,
+      "learning_rate": 4.4976721391815734e-05,
+      "loss": 0.3134,
+      "step": 1840
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.4572398662567139,
+      "learning_rate": 4.494609164420485e-05,
+      "loss": 0.308,
+      "step": 1850
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.4925013780593872,
+      "learning_rate": 4.491546189659398e-05,
+      "loss": 0.2779,
+      "step": 1860
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.2891759872436523,
+      "learning_rate": 4.4884832148983095e-05,
+      "loss": 0.2685,
+      "step": 1870
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.4620780944824219,
+      "learning_rate": 4.4854202401372214e-05,
+      "loss": 0.2843,
+      "step": 1880
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.8115986585617065,
+      "learning_rate": 4.482357265376134e-05,
+      "loss": 0.312,
+      "step": 1890
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.4076948165893555,
+      "learning_rate": 4.479294290615045e-05,
+      "loss": 0.2962,
+      "step": 1900
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.4485734701156616,
+      "learning_rate": 4.4762313158539576e-05,
+      "loss": 0.288,
+      "step": 1910
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.4280623197555542,
+      "learning_rate": 4.47316834109287e-05,
+      "loss": 0.2585,
+      "step": 1920
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.8238214254379272,
+      "learning_rate": 4.470105366331781e-05,
+      "loss": 0.2935,
+      "step": 1930
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.5512504577636719,
+      "learning_rate": 4.467042391570694e-05,
+      "loss": 0.2634,
+      "step": 1940
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.3456778526306152,
+      "learning_rate": 4.4639794168096056e-05,
+      "loss": 0.3108,
+      "step": 1950
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.5383868217468262,
+      "learning_rate": 4.4609164420485174e-05,
+      "loss": 0.2889,
+      "step": 1960
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.5723869800567627,
+      "learning_rate": 4.45785346728743e-05,
+      "loss": 0.2935,
+      "step": 1970
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.4342284202575684,
+      "learning_rate": 4.454790492526342e-05,
+      "loss": 0.2893,
+      "step": 1980
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.1003159284591675,
+      "learning_rate": 4.451727517765254e-05,
+      "loss": 0.3122,
+      "step": 1990
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.4138984680175781,
+      "learning_rate": 4.4486645430041655e-05,
+      "loss": 0.2935,
+      "step": 2000
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.1630557775497437,
+      "learning_rate": 4.445601568243078e-05,
+      "loss": 0.2906,
+      "step": 2010
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.767928957939148,
+      "learning_rate": 4.4425385934819905e-05,
+      "loss": 0.2973,
+      "step": 2020
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.570387601852417,
+      "learning_rate": 4.4394756187209017e-05,
+      "loss": 0.2844,
+      "step": 2030
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 2.0096559524536133,
+      "learning_rate": 4.436412643959814e-05,
+      "loss": 0.3146,
+      "step": 2040
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.328818917274475,
+      "learning_rate": 4.433349669198726e-05,
+      "loss": 0.288,
+      "step": 2050
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.590100646018982,
+      "learning_rate": 4.430286694437638e-05,
+      "loss": 0.2675,
+      "step": 2060
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.5741573572158813,
+      "learning_rate": 4.42722371967655e-05,
+      "loss": 0.2993,
+      "step": 2070
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.4999186992645264,
+      "learning_rate": 4.424160744915462e-05,
+      "loss": 0.303,
+      "step": 2080
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.2271442413330078,
+      "learning_rate": 4.421097770154374e-05,
+      "loss": 0.2727,
+      "step": 2090
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.9721845388412476,
+      "learning_rate": 4.418034795393286e-05,
+      "loss": 0.2967,
+      "step": 2100
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.5282645225524902,
+      "learning_rate": 4.4149718206321984e-05,
+      "loss": 0.291,
+      "step": 2110
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.165602445602417,
+      "learning_rate": 4.41190884587111e-05,
+      "loss": 0.2951,
+      "step": 2120
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.3674421310424805,
+      "learning_rate": 4.408845871110022e-05,
+      "loss": 0.2638,
+      "step": 2130
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.3786592483520508,
+      "learning_rate": 4.4057828963489346e-05,
+      "loss": 0.2933,
+      "step": 2140
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.3042171001434326,
+      "learning_rate": 4.4027199215878464e-05,
+      "loss": 0.304,
+      "step": 2150
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.3430255651474,
+      "learning_rate": 4.399656946826758e-05,
+      "loss": 0.2883,
+      "step": 2160
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 2.224194049835205,
+      "learning_rate": 4.39659397206567e-05,
+      "loss": 0.2804,
+      "step": 2170
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.5343328714370728,
+      "learning_rate": 4.3935309973045826e-05,
+      "loss": 0.2764,
+      "step": 2180
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.113574743270874,
+      "learning_rate": 4.3904680225434944e-05,
+      "loss": 0.2765,
+      "step": 2190
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.4001346826553345,
+      "learning_rate": 4.387405047782406e-05,
+      "loss": 0.2748,
+      "step": 2200
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.3673542737960815,
+      "learning_rate": 4.384342073021319e-05,
+      "loss": 0.2891,
+      "step": 2210
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.684692621231079,
+      "learning_rate": 4.3812790982602306e-05,
+      "loss": 0.2772,
+      "step": 2220
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.785766839981079,
+      "learning_rate": 4.3782161234991425e-05,
+      "loss": 0.3077,
+      "step": 2230
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.5469920635223389,
+      "learning_rate": 4.375153148738055e-05,
+      "loss": 0.298,
+      "step": 2240
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.457453727722168,
+      "learning_rate": 4.372090173976967e-05,
+      "loss": 0.2857,
+      "step": 2250
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.80753755569458,
+      "learning_rate": 4.3690271992158786e-05,
+      "loss": 0.2673,
+      "step": 2260
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.3318232297897339,
+      "learning_rate": 4.3659642244547905e-05,
+      "loss": 0.2386,
+      "step": 2270
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 2.2377285957336426,
+      "learning_rate": 4.362901249693703e-05,
+      "loss": 0.2817,
+      "step": 2280
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.4423060417175293,
+      "learning_rate": 4.359838274932615e-05,
+      "loss": 0.2801,
+      "step": 2290
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.263735055923462,
+      "learning_rate": 4.356775300171527e-05,
+      "loss": 0.2742,
+      "step": 2300
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.1926136016845703,
+      "learning_rate": 4.353712325410439e-05,
+      "loss": 0.2746,
+      "step": 2310
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.570346474647522,
+      "learning_rate": 4.3506493506493503e-05,
+      "loss": 0.2858,
+      "step": 2320
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.5538769960403442,
+      "learning_rate": 4.347586375888263e-05,
+      "loss": 0.2469,
+      "step": 2330
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.6215842962265015,
+      "learning_rate": 4.3445234011271754e-05,
+      "loss": 0.252,
+      "step": 2340
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.5585484504699707,
+      "learning_rate": 4.3414604263660865e-05,
+      "loss": 0.2606,
+      "step": 2350
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.0863536596298218,
+      "learning_rate": 4.338397451604999e-05,
+      "loss": 0.2731,
+      "step": 2360
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.6124608516693115,
+      "learning_rate": 4.335334476843911e-05,
+      "loss": 0.2747,
+      "step": 2370
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.7818570137023926,
+      "learning_rate": 4.332271502082823e-05,
+      "loss": 0.2888,
+      "step": 2380
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.1063569784164429,
+      "learning_rate": 4.329208527321735e-05,
+      "loss": 0.2526,
+      "step": 2390
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.675656795501709,
+      "learning_rate": 4.326145552560647e-05,
+      "loss": 0.2629,
+      "step": 2400
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.4156547784805298,
+      "learning_rate": 4.3230825777995596e-05,
+      "loss": 0.2701,
+      "step": 2410
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.373252272605896,
+      "learning_rate": 4.320019603038471e-05,
+      "loss": 0.2665,
+      "step": 2420
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.4938608407974243,
+      "learning_rate": 4.316956628277383e-05,
+      "loss": 0.2568,
+      "step": 2430
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.7306394577026367,
+      "learning_rate": 4.313893653516296e-05,
+      "loss": 0.2606,
+      "step": 2440
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.5790083408355713,
+      "learning_rate": 4.310830678755207e-05,
+      "loss": 0.2731,
+      "step": 2450
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.1762480735778809,
+      "learning_rate": 4.3077677039941195e-05,
+      "loss": 0.2857,
+      "step": 2460
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.5262078046798706,
+      "learning_rate": 4.304704729233031e-05,
+      "loss": 0.2697,
+      "step": 2470
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.5879408121109009,
+      "learning_rate": 4.301641754471943e-05,
+      "loss": 0.2982,
+      "step": 2480
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.4351246356964111,
+      "learning_rate": 4.2985787797108556e-05,
+      "loss": 0.2735,
+      "step": 2490
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.6204626560211182,
+      "learning_rate": 4.2955158049497675e-05,
+      "loss": 0.2425,
+      "step": 2500
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.3296780586242676,
+      "learning_rate": 4.292452830188679e-05,
+      "loss": 0.2579,
+      "step": 2510
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.7536988258361816,
+      "learning_rate": 4.289389855427591e-05,
+      "loss": 0.2637,
+      "step": 2520
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.5759055614471436,
+      "learning_rate": 4.286326880666504e-05,
+      "loss": 0.279,
+      "step": 2530
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.7171709537506104,
+      "learning_rate": 4.2832639059054155e-05,
+      "loss": 0.2557,
+      "step": 2540
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.6710295677185059,
+      "learning_rate": 4.2802009311443273e-05,
+      "loss": 0.2784,
+      "step": 2550
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.6412403583526611,
+      "learning_rate": 4.27713795638324e-05,
+      "loss": 0.2875,
+      "step": 2560
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.5678155422210693,
+      "learning_rate": 4.274074981622152e-05,
+      "loss": 0.2361,
+      "step": 2570
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.6160064935684204,
+      "learning_rate": 4.2710120068610635e-05,
+      "loss": 0.264,
+      "step": 2580
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.323428988456726,
+      "learning_rate": 4.2679490320999754e-05,
+      "loss": 0.2584,
+      "step": 2590
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.2934988737106323,
+      "learning_rate": 4.264886057338888e-05,
+      "loss": 0.2505,
+      "step": 2600
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.6210576295852661,
+      "learning_rate": 4.2618230825778e-05,
+      "loss": 0.2513,
+      "step": 2610
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.3180238008499146,
+      "learning_rate": 4.2587601078167116e-05,
+      "loss": 0.2378,
+      "step": 2620
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.365300178527832,
+      "learning_rate": 4.255697133055624e-05,
+      "loss": 0.2541,
+      "step": 2630
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.389766812324524,
+      "learning_rate": 4.252634158294536e-05,
+      "loss": 0.2601,
+      "step": 2640
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.6183340549468994,
+      "learning_rate": 4.249571183533448e-05,
+      "loss": 0.2355,
+      "step": 2650
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.4444153308868408,
+      "learning_rate": 4.24650820877236e-05,
+      "loss": 0.2797,
+      "step": 2660
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.5106576681137085,
+      "learning_rate": 4.243445234011272e-05,
+      "loss": 0.2507,
+      "step": 2670
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 2.0144007205963135,
+      "learning_rate": 4.240382259250184e-05,
+      "loss": 0.266,
+      "step": 2680
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.5243396759033203,
+      "learning_rate": 4.237319284489096e-05,
+      "loss": 0.2249,
+      "step": 2690
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.9910212755203247,
+      "learning_rate": 4.234256309728008e-05,
+      "loss": 0.2555,
+      "step": 2700
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.6437292098999023,
+      "learning_rate": 4.23119333496692e-05,
+      "loss": 0.2634,
+      "step": 2710
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.5106643438339233,
+      "learning_rate": 4.228130360205832e-05,
+      "loss": 0.2428,
+      "step": 2720
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 57.060325622558594,
+      "learning_rate": 4.2250673854447445e-05,
+      "loss": 0.26,
+      "step": 2730
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.6187288761138916,
+      "learning_rate": 4.2220044106836556e-05,
+      "loss": 0.2471,
+      "step": 2740
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.3576431274414062,
+      "learning_rate": 4.218941435922568e-05,
+      "loss": 0.2628,
+      "step": 2750
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.2909529209136963,
+      "learning_rate": 4.215878461161481e-05,
+      "loss": 0.2557,
+      "step": 2760
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.2785391807556152,
+      "learning_rate": 4.212815486400392e-05,
+      "loss": 0.2584,
+      "step": 2770
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.528359293937683,
+      "learning_rate": 4.209752511639304e-05,
+      "loss": 0.2389,
+      "step": 2780
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.818306565284729,
+      "learning_rate": 4.206689536878216e-05,
+      "loss": 0.2906,
+      "step": 2790
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.5244324207305908,
+      "learning_rate": 4.203626562117129e-05,
+      "loss": 0.2546,
+      "step": 2800
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.5336534976959229,
+      "learning_rate": 4.2005635873560405e-05,
+      "loss": 0.2386,
+      "step": 2810
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.6018409729003906,
+      "learning_rate": 4.1975006125949524e-05,
+      "loss": 0.2452,
+      "step": 2820
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.3080527782440186,
+      "learning_rate": 4.194437637833865e-05,
+      "loss": 0.2641,
+      "step": 2830
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.4105219841003418,
+      "learning_rate": 4.191374663072776e-05,
+      "loss": 0.2401,
+      "step": 2840
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.7197917699813843,
+      "learning_rate": 4.1883116883116886e-05,
+      "loss": 0.2464,
+      "step": 2850
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.3024113178253174,
+      "learning_rate": 4.185248713550601e-05,
+      "loss": 0.2784,
+      "step": 2860
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.3709226846694946,
+      "learning_rate": 4.182185738789512e-05,
+      "loss": 0.2339,
+      "step": 2870
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.4183152914047241,
+      "learning_rate": 4.179122764028425e-05,
+      "loss": 0.2892,
+      "step": 2880
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.1996904611587524,
+      "learning_rate": 4.1760597892673366e-05,
+      "loss": 0.2206,
+      "step": 2890
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.350128173828125,
+      "learning_rate": 4.1729968145062484e-05,
+      "loss": 0.2414,
+      "step": 2900
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.3338422775268555,
+      "learning_rate": 4.169933839745161e-05,
+      "loss": 0.2443,
+      "step": 2910
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.930519700050354,
+      "learning_rate": 4.166870864984073e-05,
+      "loss": 0.2541,
+      "step": 2920
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.3033095598220825,
+      "learning_rate": 4.1638078902229846e-05,
+      "loss": 0.2412,
+      "step": 2930
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.4129502773284912,
+      "learning_rate": 4.1607449154618964e-05,
+      "loss": 0.2403,
+      "step": 2940
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 2.1562488079071045,
+      "learning_rate": 4.157681940700809e-05,
+      "loss": 0.2608,
+      "step": 2950
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.5833796262741089,
+      "learning_rate": 4.154618965939721e-05,
+      "loss": 0.2434,
+      "step": 2960
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.8136632442474365,
+      "learning_rate": 4.1515559911786326e-05,
+      "loss": 0.2263,
+      "step": 2970
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.982427954673767,
+      "learning_rate": 4.148493016417545e-05,
+      "loss": 0.2528,
+      "step": 2980
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.3271403312683105,
+      "learning_rate": 4.145430041656457e-05,
+      "loss": 0.2392,
+      "step": 2990
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.8585102558135986,
+      "learning_rate": 4.142367066895369e-05,
+      "loss": 0.2651,
+      "step": 3000
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.6813907623291016,
+      "learning_rate": 4.139304092134281e-05,
+      "loss": 0.2564,
+      "step": 3010
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.7984387874603271,
+      "learning_rate": 4.136241117373193e-05,
+      "loss": 0.2467,
+      "step": 3020
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.5527530908584595,
+      "learning_rate": 4.133178142612105e-05,
+      "loss": 0.2256,
+      "step": 3030
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.5511258840560913,
+      "learning_rate": 4.130115167851017e-05,
+      "loss": 0.2488,
+      "step": 3040
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.8069953918457031,
+      "learning_rate": 4.1270521930899294e-05,
+      "loss": 0.2552,
+      "step": 3050
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.1661847829818726,
+      "learning_rate": 4.123989218328841e-05,
+      "loss": 0.2337,
+      "step": 3060
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.448970913887024,
+      "learning_rate": 4.120926243567753e-05,
+      "loss": 0.236,
+      "step": 3070
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.424075961112976,
+      "learning_rate": 4.1178632688066655e-05,
+      "loss": 0.2383,
+      "step": 3080
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 2.000866651535034,
+      "learning_rate": 4.1148002940455774e-05,
+      "loss": 0.2294,
+      "step": 3090
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.829994559288025,
+      "learning_rate": 4.111737319284489e-05,
+      "loss": 0.2536,
+      "step": 3100
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.3467332124710083,
+      "learning_rate": 4.108674344523401e-05,
+      "loss": 0.2422,
+      "step": 3110
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.3595892190933228,
+      "learning_rate": 4.1056113697623136e-05,
+      "loss": 0.2564,
+      "step": 3120
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.7775511741638184,
+      "learning_rate": 4.1025483950012254e-05,
+      "loss": 0.2244,
+      "step": 3130
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.349105715751648,
+      "learning_rate": 4.099485420240137e-05,
+      "loss": 0.2351,
+      "step": 3140
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 2.176692008972168,
+      "learning_rate": 4.09642244547905e-05,
+      "loss": 0.2477,
+      "step": 3150
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.1634658575057983,
+      "learning_rate": 4.093359470717961e-05,
+      "loss": 0.2538,
+      "step": 3160
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.511855959892273,
+      "learning_rate": 4.0902964959568734e-05,
+      "loss": 0.2209,
+      "step": 3170
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.200825810432434,
+      "learning_rate": 4.087233521195786e-05,
+      "loss": 0.2271,
+      "step": 3180
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.0485877990722656,
+      "learning_rate": 4.084170546434697e-05,
+      "loss": 0.2347,
+      "step": 3190
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.416651964187622,
+      "learning_rate": 4.0811075716736096e-05,
+      "loss": 0.2628,
+      "step": 3200
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.6312257051467896,
+      "learning_rate": 4.0780445969125215e-05,
+      "loss": 0.2335,
+      "step": 3210
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.495829463005066,
+      "learning_rate": 4.074981622151434e-05,
+      "loss": 0.2271,
+      "step": 3220
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.4085733890533447,
+      "learning_rate": 4.071918647390346e-05,
+      "loss": 0.2535,
+      "step": 3230
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.937908411026001,
+      "learning_rate": 4.0688556726292577e-05,
+      "loss": 0.2336,
+      "step": 3240
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.8848986625671387,
+      "learning_rate": 4.06579269786817e-05,
+      "loss": 0.2348,
+      "step": 3250
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.6204856634140015,
+      "learning_rate": 4.062729723107081e-05,
+      "loss": 0.2624,
+      "step": 3260
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.7245253324508667,
+      "learning_rate": 4.059666748345994e-05,
+      "loss": 0.2439,
+      "step": 3270
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.2971001863479614,
+      "learning_rate": 4.0566037735849064e-05,
+      "loss": 0.2456,
+      "step": 3280
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.1006497144699097,
+      "learning_rate": 4.0535407988238175e-05,
+      "loss": 0.2112,
+      "step": 3290
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.4925537109375,
+      "learning_rate": 4.05047782406273e-05,
+      "loss": 0.2256,
+      "step": 3300
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.4216618537902832,
+      "learning_rate": 4.047414849301642e-05,
+      "loss": 0.2385,
+      "step": 3310
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.2764017581939697,
+      "learning_rate": 4.044351874540554e-05,
+      "loss": 0.2201,
+      "step": 3320
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.4631826877593994,
+      "learning_rate": 4.041288899779466e-05,
+      "loss": 0.2408,
+      "step": 3330
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.6391186714172363,
+      "learning_rate": 4.038225925018378e-05,
+      "loss": 0.2445,
+      "step": 3340
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.7994047403335571,
+      "learning_rate": 4.03516295025729e-05,
+      "loss": 0.2405,
+      "step": 3350
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.5009214878082275,
+      "learning_rate": 4.032099975496202e-05,
+      "loss": 0.2128,
+      "step": 3360
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.2063971757888794,
+      "learning_rate": 4.029037000735114e-05,
+      "loss": 0.2456,
+      "step": 3370
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.7369698286056519,
+      "learning_rate": 4.025974025974026e-05,
+      "loss": 0.2347,
+      "step": 3380
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.1080302000045776,
+      "learning_rate": 4.022911051212938e-05,
+      "loss": 0.2154,
+      "step": 3390
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.581632375717163,
+      "learning_rate": 4.0198480764518504e-05,
+      "loss": 0.2375,
+      "step": 3400
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.2571433782577515,
+      "learning_rate": 4.016785101690762e-05,
+      "loss": 0.2273,
+      "step": 3410
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.6267669200897217,
+      "learning_rate": 4.013722126929674e-05,
+      "loss": 0.2327,
+      "step": 3420
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.3594502210617065,
+      "learning_rate": 4.0106591521685866e-05,
+      "loss": 0.2303,
+      "step": 3430
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.3065011501312256,
+      "learning_rate": 4.0075961774074985e-05,
+      "loss": 0.2456,
+      "step": 3440
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.3775655031204224,
+      "learning_rate": 4.00453320264641e-05,
+      "loss": 0.2215,
+      "step": 3450
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.0961012840270996,
+      "learning_rate": 4.001470227885322e-05,
+      "loss": 0.216,
+      "step": 3460
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.339247226715088,
+      "learning_rate": 3.9984072531242346e-05,
+      "loss": 0.2336,
+      "step": 3470
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.6671198606491089,
+      "learning_rate": 3.9953442783631465e-05,
+      "loss": 0.2285,
+      "step": 3480
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.443637490272522,
+      "learning_rate": 3.992281303602058e-05,
+      "loss": 0.2312,
+      "step": 3490
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.8523837327957153,
+      "learning_rate": 3.989218328840971e-05,
+      "loss": 0.2384,
+      "step": 3500
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.586118459701538,
+      "learning_rate": 3.986155354079883e-05,
+      "loss": 0.212,
+      "step": 3510
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.630063772201538,
+      "learning_rate": 3.9830923793187945e-05,
+      "loss": 0.2642,
+      "step": 3520
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 2.182255744934082,
+      "learning_rate": 3.980029404557707e-05,
+      "loss": 0.2064,
+      "step": 3530
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.0610166788101196,
+      "learning_rate": 3.976966429796619e-05,
+      "loss": 0.2168,
+      "step": 3540
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.6151827573776245,
+      "learning_rate": 3.973903455035531e-05,
+      "loss": 0.2456,
+      "step": 3550
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.9630222320556641,
+      "learning_rate": 3.9708404802744425e-05,
+      "loss": 0.2354,
+      "step": 3560
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.4157471656799316,
+      "learning_rate": 3.967777505513355e-05,
+      "loss": 0.2276,
+      "step": 3570
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.2138807773590088,
+      "learning_rate": 3.964714530752267e-05,
+      "loss": 0.2157,
+      "step": 3580
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.6903012990951538,
+      "learning_rate": 3.961651555991179e-05,
+      "loss": 0.2144,
+      "step": 3590
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.6704226732254028,
+      "learning_rate": 3.958588581230091e-05,
+      "loss": 0.2318,
+      "step": 3600
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.262870192527771,
+      "learning_rate": 3.9555256064690024e-05,
+      "loss": 0.2105,
+      "step": 3610
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 2.6524696350097656,
+      "learning_rate": 3.952462631707915e-05,
+      "loss": 0.2319,
+      "step": 3620
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.3405276536941528,
+      "learning_rate": 3.949399656946827e-05,
+      "loss": 0.2233,
+      "step": 3630
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.5943964719772339,
+      "learning_rate": 3.946336682185739e-05,
+      "loss": 0.2011,
+      "step": 3640
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.366666555404663,
+      "learning_rate": 3.943273707424651e-05,
+      "loss": 0.2458,
+      "step": 3650
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.6376194953918457,
+      "learning_rate": 3.940210732663563e-05,
+      "loss": 0.2181,
+      "step": 3660
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.3548191785812378,
+      "learning_rate": 3.9371477579024755e-05,
+      "loss": 0.2147,
+      "step": 3670
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.1803979873657227,
+      "learning_rate": 3.9340847831413866e-05,
+      "loss": 0.2266,
+      "step": 3680
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.8261361122131348,
+      "learning_rate": 3.931021808380299e-05,
+      "loss": 0.2085,
+      "step": 3690
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 4.506346702575684,
+      "learning_rate": 3.9279588336192116e-05,
+      "loss": 0.2282,
+      "step": 3700
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.4388922452926636,
+      "learning_rate": 3.924895858858123e-05,
+      "loss": 0.2241,
+      "step": 3710
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.1566166877746582,
+      "learning_rate": 3.921832884097035e-05,
+      "loss": 0.2266,
+      "step": 3720
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.451364278793335,
+      "learning_rate": 3.918769909335947e-05,
+      "loss": 0.2207,
+      "step": 3730
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.653739333152771,
+      "learning_rate": 3.915706934574859e-05,
+      "loss": 0.2165,
+      "step": 3740
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.654880404472351,
+      "learning_rate": 3.9126439598137715e-05,
+      "loss": 0.2293,
+      "step": 3750
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.49134361743927,
+      "learning_rate": 3.9095809850526833e-05,
+      "loss": 0.2166,
+      "step": 3760
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.3947138786315918,
+      "learning_rate": 3.906518010291595e-05,
+      "loss": 0.2239,
+      "step": 3770
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.2860134840011597,
+      "learning_rate": 3.903455035530507e-05,
+      "loss": 0.2021,
+      "step": 3780
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.611852765083313,
+      "learning_rate": 3.9003920607694195e-05,
+      "loss": 0.2164,
+      "step": 3790
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.9268925189971924,
+      "learning_rate": 3.8973290860083314e-05,
+      "loss": 0.2143,
+      "step": 3800
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 2.1335513591766357,
+      "learning_rate": 3.894266111247243e-05,
+      "loss": 0.2518,
+      "step": 3810
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.3358123302459717,
+      "learning_rate": 3.891203136486156e-05,
+      "loss": 0.2322,
+      "step": 3820
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.4572689533233643,
+      "learning_rate": 3.8881401617250676e-05,
+      "loss": 0.2023,
+      "step": 3830
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.6910518407821655,
+      "learning_rate": 3.8850771869639794e-05,
+      "loss": 0.2535,
+      "step": 3840
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.0582376718521118,
+      "learning_rate": 3.882014212202892e-05,
+      "loss": 0.2278,
+      "step": 3850
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.6501924991607666,
+      "learning_rate": 3.878951237441804e-05,
+      "loss": 0.1965,
+      "step": 3860
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.3161675930023193,
+      "learning_rate": 3.8758882626807156e-05,
+      "loss": 0.2104,
+      "step": 3870
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.3588526248931885,
+      "learning_rate": 3.8728252879196274e-05,
+      "loss": 0.2224,
+      "step": 3880
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.317922830581665,
+      "learning_rate": 3.86976231315854e-05,
+      "loss": 0.1909,
+      "step": 3890
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.4751834869384766,
+      "learning_rate": 3.866699338397452e-05,
+      "loss": 0.211,
+      "step": 3900
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.8289200067520142,
+      "learning_rate": 3.8636363636363636e-05,
+      "loss": 0.2339,
+      "step": 3910
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.8388066291809082,
+      "learning_rate": 3.860573388875276e-05,
+      "loss": 0.2095,
+      "step": 3920
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.1754611730575562,
+      "learning_rate": 3.857510414114188e-05,
+      "loss": 0.212,
+      "step": 3930
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.712350845336914,
+      "learning_rate": 3.8544474393531e-05,
+      "loss": 0.2238,
+      "step": 3940
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.5514004230499268,
+      "learning_rate": 3.851384464592012e-05,
+      "loss": 0.2219,
+      "step": 3950
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.831300973892212,
+      "learning_rate": 3.848321489830924e-05,
+      "loss": 0.2261,
+      "step": 3960
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.1789157390594482,
+      "learning_rate": 3.845258515069836e-05,
+      "loss": 0.1911,
+      "step": 3970
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.1377627849578857,
+      "learning_rate": 3.842195540308748e-05,
+      "loss": 0.2079,
+      "step": 3980
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.3747925758361816,
+      "learning_rate": 3.83913256554766e-05,
+      "loss": 0.2336,
+      "step": 3990
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.3656913042068481,
+      "learning_rate": 3.836069590786572e-05,
+      "loss": 0.2237,
+      "step": 4000
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.7472519874572754,
+      "learning_rate": 3.833006616025484e-05,
+      "loss": 0.2133,
+      "step": 4010
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.4794762134552002,
+      "learning_rate": 3.8299436412643965e-05,
+      "loss": 0.2194,
+      "step": 4020
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.1777324676513672,
+      "learning_rate": 3.826880666503308e-05,
+      "loss": 0.1951,
+      "step": 4030
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.6810731887817383,
+      "learning_rate": 3.82381769174222e-05,
+      "loss": 0.219,
+      "step": 4040
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 2.1886587142944336,
+      "learning_rate": 3.820754716981133e-05,
+      "loss": 0.1996,
+      "step": 4050
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.3431891202926636,
+      "learning_rate": 3.8176917422200445e-05,
+      "loss": 0.2188,
+      "step": 4060
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.4996401071548462,
+      "learning_rate": 3.8146287674589564e-05,
+      "loss": 0.219,
+      "step": 4070
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.1930394172668457,
+      "learning_rate": 3.811565792697868e-05,
+      "loss": 0.2148,
+      "step": 4080
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.9747510552406311,
+      "learning_rate": 3.808502817936781e-05,
+      "loss": 0.2087,
+      "step": 4090
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 2.048163652420044,
+      "learning_rate": 3.8054398431756926e-05,
+      "loss": 0.2419,
+      "step": 4100
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.1260502338409424,
+      "learning_rate": 3.8023768684146044e-05,
+      "loss": 0.2085,
+      "step": 4110
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.4994487762451172,
+      "learning_rate": 3.799313893653517e-05,
+      "loss": 0.1949,
+      "step": 4120
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.5391995906829834,
+      "learning_rate": 3.796250918892428e-05,
+      "loss": 0.2344,
+      "step": 4130
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.5020917654037476,
+      "learning_rate": 3.7931879441313406e-05,
+      "loss": 0.2194,
+      "step": 4140
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.186409831047058,
+      "learning_rate": 3.7901249693702524e-05,
+      "loss": 0.2105,
+      "step": 4150
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.4810385704040527,
+      "learning_rate": 3.787061994609164e-05,
+      "loss": 0.209,
+      "step": 4160
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.3342121839523315,
+      "learning_rate": 3.783999019848077e-05,
+      "loss": 0.2076,
+      "step": 4170
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.529411792755127,
+      "learning_rate": 3.7809360450869886e-05,
+      "loss": 0.2257,
+      "step": 4180
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.2292615175247192,
+      "learning_rate": 3.7778730703259005e-05,
+      "loss": 0.198,
+      "step": 4190
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.2979429960250854,
+      "learning_rate": 3.774810095564812e-05,
+      "loss": 0.1833,
+      "step": 4200
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.3138002157211304,
+      "learning_rate": 3.771747120803725e-05,
+      "loss": 0.2292,
+      "step": 4210
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.9133843183517456,
+      "learning_rate": 3.7686841460426367e-05,
+      "loss": 0.2154,
+      "step": 4220
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.0788285732269287,
+      "learning_rate": 3.7656211712815485e-05,
+      "loss": 0.1983,
+      "step": 4230
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.1172419786453247,
+      "learning_rate": 3.762558196520461e-05,
+      "loss": 0.211,
+      "step": 4240
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.2741632461547852,
+      "learning_rate": 3.759495221759373e-05,
+      "loss": 0.2192,
+      "step": 4250
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.5227195024490356,
+      "learning_rate": 3.756432246998285e-05,
+      "loss": 0.2028,
+      "step": 4260
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.2935453653335571,
+      "learning_rate": 3.753369272237197e-05,
+      "loss": 0.1964,
+      "step": 4270
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.7768583297729492,
+      "learning_rate": 3.750306297476109e-05,
+      "loss": 0.2205,
+      "step": 4280
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.0349539518356323,
+      "learning_rate": 3.747243322715021e-05,
+      "loss": 0.2167,
+      "step": 4290
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.185727834701538,
+      "learning_rate": 3.744180347953933e-05,
+      "loss": 0.1886,
+      "step": 4300
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.6226075887680054,
+      "learning_rate": 3.741117373192845e-05,
+      "loss": 0.211,
+      "step": 4310
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.487450122833252,
+      "learning_rate": 3.738054398431757e-05,
+      "loss": 0.1911,
+      "step": 4320
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.641066074371338,
+      "learning_rate": 3.734991423670669e-05,
+      "loss": 0.2106,
+      "step": 4330
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.1860120296478271,
+      "learning_rate": 3.7319284489095814e-05,
+      "loss": 0.2024,
+      "step": 4340
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.1305229663848877,
+      "learning_rate": 3.728865474148493e-05,
+      "loss": 0.1944,
+      "step": 4350
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.98978191614151,
+      "learning_rate": 3.725802499387405e-05,
+      "loss": 0.2004,
+      "step": 4360
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.2854243516921997,
+      "learning_rate": 3.7227395246263176e-05,
+      "loss": 0.1856,
+      "step": 4370
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.5476226806640625,
+      "learning_rate": 3.7196765498652294e-05,
+      "loss": 0.1911,
+      "step": 4380
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.7981789112091064,
+      "learning_rate": 3.716613575104141e-05,
+      "loss": 0.2123,
+      "step": 4390
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.5477904081344604,
+      "learning_rate": 3.713550600343053e-05,
+      "loss": 0.2167,
+      "step": 4400
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.7662404775619507,
+      "learning_rate": 3.7104876255819656e-05,
+      "loss": 0.2125,
+      "step": 4410
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.5097447633743286,
+      "learning_rate": 3.7074246508208775e-05,
+      "loss": 0.1964,
+      "step": 4420
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.2024070024490356,
+      "learning_rate": 3.704361676059789e-05,
+      "loss": 0.1943,
+      "step": 4430
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.331238865852356,
+      "learning_rate": 3.701298701298702e-05,
+      "loss": 0.2114,
+      "step": 4440
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.187941312789917,
+      "learning_rate": 3.698235726537613e-05,
+      "loss": 0.2077,
+      "step": 4450
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.1885707378387451,
+      "learning_rate": 3.6951727517765255e-05,
+      "loss": 0.1921,
+      "step": 4460
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.5863043069839478,
+      "learning_rate": 3.692109777015438e-05,
+      "loss": 0.181,
+      "step": 4470
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.060566782951355,
+      "learning_rate": 3.68904680225435e-05,
+      "loss": 0.1947,
+      "step": 4480
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.241894245147705,
+      "learning_rate": 3.685983827493262e-05,
+      "loss": 0.1939,
+      "step": 4490
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.567139983177185,
+      "learning_rate": 3.6829208527321735e-05,
+      "loss": 0.2042,
+      "step": 4500
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.5378930568695068,
+      "learning_rate": 3.679857877971086e-05,
+      "loss": 0.1863,
+      "step": 4510
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.2834867238998413,
+      "learning_rate": 3.676794903209998e-05,
+      "loss": 0.223,
+      "step": 4520
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.3938517570495605,
+      "learning_rate": 3.67373192844891e-05,
+      "loss": 0.2066,
+      "step": 4530
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.7854987382888794,
+      "learning_rate": 3.670668953687822e-05,
+      "loss": 0.2191,
+      "step": 4540
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 2.232520818710327,
+      "learning_rate": 3.6676059789267334e-05,
+      "loss": 0.196,
+      "step": 4550
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.4364125728607178,
+      "learning_rate": 3.664543004165646e-05,
+      "loss": 0.2014,
+      "step": 4560
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.329524278640747,
+      "learning_rate": 3.6614800294045584e-05,
+      "loss": 0.1979,
+      "step": 4570
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.4252960681915283,
+      "learning_rate": 3.6584170546434696e-05,
+      "loss": 0.224,
+      "step": 4580
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.5715065002441406,
+      "learning_rate": 3.655354079882382e-05,
+      "loss": 0.2015,
+      "step": 4590
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.33690345287323,
+      "learning_rate": 3.652291105121294e-05,
+      "loss": 0.2023,
+      "step": 4600
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.8494676351547241,
+      "learning_rate": 3.649228130360206e-05,
+      "loss": 0.2099,
+      "step": 4610
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.2794080972671509,
+      "learning_rate": 3.646165155599118e-05,
+      "loss": 0.2015,
+      "step": 4620
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.2285436391830444,
+      "learning_rate": 3.64310218083803e-05,
+      "loss": 0.2072,
+      "step": 4630
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.6561449766159058,
+      "learning_rate": 3.640039206076942e-05,
+      "loss": 0.1928,
+      "step": 4640
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.3827136754989624,
+      "learning_rate": 3.636976231315854e-05,
+      "loss": 0.2094,
+      "step": 4650
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.631355881690979,
+      "learning_rate": 3.633913256554766e-05,
+      "loss": 0.1855,
+      "step": 4660
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.7016180753707886,
+      "learning_rate": 3.630850281793678e-05,
+      "loss": 0.224,
+      "step": 4670
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.7572182416915894,
+      "learning_rate": 3.62778730703259e-05,
+      "loss": 0.1859,
+      "step": 4680
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.5234770774841309,
+      "learning_rate": 3.6247243322715025e-05,
+      "loss": 0.1931,
+      "step": 4690
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.4753092527389526,
+      "learning_rate": 3.621661357510414e-05,
+      "loss": 0.203,
+      "step": 4700
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.4132767915725708,
+      "learning_rate": 3.618598382749326e-05,
+      "loss": 0.2001,
+      "step": 4710
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.3893791437149048,
+      "learning_rate": 3.615535407988238e-05,
+      "loss": 0.1921,
+      "step": 4720
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.4078565835952759,
+      "learning_rate": 3.6124724332271505e-05,
+      "loss": 0.1893,
+      "step": 4730
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.5593945980072021,
+      "learning_rate": 3.6094094584660623e-05,
+      "loss": 0.2106,
+      "step": 4740
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.3134779930114746,
+      "learning_rate": 3.606346483704974e-05,
+      "loss": 0.2049,
+      "step": 4750
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.4126642942428589,
+      "learning_rate": 3.603283508943887e-05,
+      "loss": 0.2054,
+      "step": 4760
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.4551384449005127,
+      "learning_rate": 3.6002205341827985e-05,
+      "loss": 0.1943,
+      "step": 4770
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.3557885885238647,
+      "learning_rate": 3.5971575594217104e-05,
+      "loss": 0.1829,
+      "step": 4780
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.4521667957305908,
+      "learning_rate": 3.594094584660623e-05,
+      "loss": 0.1756,
+      "step": 4790
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.4649301767349243,
+      "learning_rate": 3.591031609899535e-05,
+      "loss": 0.1943,
+      "step": 4800
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.366057276725769,
+      "learning_rate": 3.5879686351384466e-05,
+      "loss": 0.2149,
+      "step": 4810
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.4880870580673218,
+      "learning_rate": 3.5849056603773584e-05,
+      "loss": 0.2083,
+      "step": 4820
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.6057343482971191,
+      "learning_rate": 3.581842685616271e-05,
+      "loss": 0.1877,
+      "step": 4830
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.4553625583648682,
+      "learning_rate": 3.578779710855183e-05,
+      "loss": 0.226,
+      "step": 4840
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.5336939096450806,
+      "learning_rate": 3.5757167360940946e-05,
+      "loss": 0.1839,
+      "step": 4850
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.4835028648376465,
+      "learning_rate": 3.572653761333007e-05,
+      "loss": 0.2107,
+      "step": 4860
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.4277873039245605,
+      "learning_rate": 3.569590786571918e-05,
+      "loss": 0.2147,
+      "step": 4870
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.6307196617126465,
+      "learning_rate": 3.566527811810831e-05,
+      "loss": 0.2024,
+      "step": 4880
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.3433023691177368,
+      "learning_rate": 3.563464837049743e-05,
+      "loss": 0.2017,
+      "step": 4890
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.2463970184326172,
+      "learning_rate": 3.560401862288655e-05,
+      "loss": 0.1742,
+      "step": 4900
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.650128722190857,
+      "learning_rate": 3.557338887527567e-05,
+      "loss": 0.2114,
+      "step": 4910
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.5994027853012085,
+      "learning_rate": 3.554275912766479e-05,
+      "loss": 0.1987,
+      "step": 4920
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.648940920829773,
+      "learning_rate": 3.551212938005391e-05,
+      "loss": 0.188,
+      "step": 4930
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.4793117046356201,
+      "learning_rate": 3.548149963244303e-05,
+      "loss": 0.1806,
+      "step": 4940
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.6560360193252563,
+      "learning_rate": 3.545086988483215e-05,
+      "loss": 0.2116,
+      "step": 4950
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.377936601638794,
+      "learning_rate": 3.5420240137221275e-05,
+      "loss": 0.1886,
+      "step": 4960
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.3976082801818848,
+      "learning_rate": 3.5389610389610387e-05,
+      "loss": 0.1899,
+      "step": 4970
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.7872759103775024,
+      "learning_rate": 3.535898064199951e-05,
+      "loss": 0.1968,
+      "step": 4980
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.084206461906433,
+      "learning_rate": 3.532835089438864e-05,
+      "loss": 0.2029,
+      "step": 4990
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.5632914304733276,
+      "learning_rate": 3.529772114677775e-05,
+      "loss": 0.1995,
+      "step": 5000
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.2333999872207642,
+      "learning_rate": 3.5267091399166874e-05,
+      "loss": 0.2267,
+      "step": 5010
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.4328374862670898,
+      "learning_rate": 3.523646165155599e-05,
+      "loss": 0.1935,
+      "step": 5020
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.2758526802062988,
+      "learning_rate": 3.520583190394511e-05,
+      "loss": 0.1917,
+      "step": 5030
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 3.819425106048584,
+      "learning_rate": 3.5175202156334236e-05,
+      "loss": 0.2066,
+      "step": 5040
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.3627198934555054,
+      "learning_rate": 3.5144572408723354e-05,
+      "loss": 0.1833,
+      "step": 5050
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.0906269550323486,
+      "learning_rate": 3.511394266111247e-05,
+      "loss": 0.1934,
+      "step": 5060
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.4054425954818726,
+      "learning_rate": 3.508331291350159e-05,
+      "loss": 0.171,
+      "step": 5070
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.4550117254257202,
+      "learning_rate": 3.5052683165890716e-05,
+      "loss": 0.1937,
+      "step": 5080
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.2433176040649414,
+      "learning_rate": 3.502205341827984e-05,
+      "loss": 0.1954,
+      "step": 5090
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.2884045839309692,
+      "learning_rate": 3.499142367066895e-05,
+      "loss": 0.212,
+      "step": 5100
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.0211153030395508,
+      "learning_rate": 3.496079392305808e-05,
+      "loss": 0.1919,
+      "step": 5110
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.4524012804031372,
+      "learning_rate": 3.4930164175447196e-05,
+      "loss": 0.1788,
+      "step": 5120
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.7646048069000244,
+      "learning_rate": 3.4899534427836314e-05,
+      "loss": 0.177,
+      "step": 5130
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.2974878549575806,
+      "learning_rate": 3.486890468022544e-05,
+      "loss": 0.2155,
+      "step": 5140
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.960159420967102,
+      "learning_rate": 3.483827493261456e-05,
+      "loss": 0.2001,
+      "step": 5150
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.0398203134536743,
+      "learning_rate": 3.4807645185003676e-05,
+      "loss": 0.181,
+      "step": 5160
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.6925101280212402,
+      "learning_rate": 3.4777015437392795e-05,
+      "loss": 0.1794,
+      "step": 5170
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.4617935419082642,
+      "learning_rate": 3.474638568978192e-05,
+      "loss": 0.2041,
+      "step": 5180
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.3038135766983032,
+      "learning_rate": 3.471575594217104e-05,
+      "loss": 0.1981,
+      "step": 5190
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.2198913097381592,
+      "learning_rate": 3.4685126194560157e-05,
+      "loss": 0.1945,
+      "step": 5200
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.8526829481124878,
+      "learning_rate": 3.465449644694928e-05,
+      "loss": 0.2128,
+      "step": 5210
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.4101462364196777,
+      "learning_rate": 3.46238666993384e-05,
+      "loss": 0.1854,
+      "step": 5220
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.36957848072052,
+      "learning_rate": 3.459323695172752e-05,
+      "loss": 0.1586,
+      "step": 5230
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.6012911796569824,
+      "learning_rate": 3.456260720411664e-05,
+      "loss": 0.2054,
+      "step": 5240
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.3765629529953003,
+      "learning_rate": 3.453197745650576e-05,
+      "loss": 0.1815,
+      "step": 5250
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.8125637769699097,
+      "learning_rate": 3.450134770889488e-05,
+      "loss": 0.1804,
+      "step": 5260
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.4987002611160278,
+      "learning_rate": 3.4470717961284e-05,
+      "loss": 0.1749,
+      "step": 5270
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.4361060857772827,
+      "learning_rate": 3.4440088213673124e-05,
+      "loss": 0.1935,
+      "step": 5280
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.114900827407837,
+      "learning_rate": 3.4409458466062235e-05,
+      "loss": 0.1801,
+      "step": 5290
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.6583603620529175,
+      "learning_rate": 3.437882871845136e-05,
+      "loss": 0.226,
+      "step": 5300
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.3105204105377197,
+      "learning_rate": 3.4348198970840486e-05,
+      "loss": 0.1973,
+      "step": 5310
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.6126765012741089,
+      "learning_rate": 3.4317569223229604e-05,
+      "loss": 0.2026,
+      "step": 5320
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.1288155317306519,
+      "learning_rate": 3.428693947561872e-05,
+      "loss": 0.1869,
+      "step": 5330
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.1172573566436768,
+      "learning_rate": 3.425630972800784e-05,
+      "loss": 0.1746,
+      "step": 5340
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.6315549612045288,
+      "learning_rate": 3.4225679980396966e-05,
+      "loss": 0.1775,
+      "step": 5350
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.1503742933273315,
+      "learning_rate": 3.4195050232786084e-05,
+      "loss": 0.2062,
+      "step": 5360
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.6012240648269653,
+      "learning_rate": 3.41644204851752e-05,
+      "loss": 0.1874,
+      "step": 5370
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.5097299814224243,
+      "learning_rate": 3.413379073756433e-05,
+      "loss": 0.1638,
+      "step": 5380
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.6788183450698853,
+      "learning_rate": 3.410316098995344e-05,
+      "loss": 0.1797,
+      "step": 5390
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.3478535413742065,
+      "learning_rate": 3.4072531242342565e-05,
+      "loss": 0.176,
+      "step": 5400
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.5208170413970947,
+      "learning_rate": 3.404190149473169e-05,
+      "loss": 0.1862,
+      "step": 5410
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.0249415636062622,
+      "learning_rate": 3.40112717471208e-05,
+      "loss": 0.1926,
+      "step": 5420
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.396653175354004,
+      "learning_rate": 3.3980641999509926e-05,
+      "loss": 0.192,
+      "step": 5430
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.1902254819869995,
+      "learning_rate": 3.3950012251899045e-05,
+      "loss": 0.2152,
+      "step": 5440
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.5217747688293457,
+      "learning_rate": 3.391938250428816e-05,
+      "loss": 0.2053,
+      "step": 5450
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.631514072418213,
+      "learning_rate": 3.388875275667729e-05,
+      "loss": 0.175,
+      "step": 5460
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.376104712486267,
+      "learning_rate": 3.385812300906641e-05,
+      "loss": 0.1945,
+      "step": 5470
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.5270049571990967,
+      "learning_rate": 3.3827493261455525e-05,
+      "loss": 0.1678,
+      "step": 5480
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.373671531677246,
+      "learning_rate": 3.3796863513844643e-05,
+      "loss": 0.1974,
+      "step": 5490
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.1565955877304077,
+      "learning_rate": 3.376623376623377e-05,
+      "loss": 0.2067,
+      "step": 5500
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.4378836154937744,
+      "learning_rate": 3.3735604018622894e-05,
+      "loss": 0.1968,
+      "step": 5510
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.3857978582382202,
+      "learning_rate": 3.3704974271012005e-05,
+      "loss": 0.1571,
+      "step": 5520
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.9464209079742432,
+      "learning_rate": 3.367434452340113e-05,
+      "loss": 0.1975,
+      "step": 5530
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.1438575983047485,
+      "learning_rate": 3.364371477579025e-05,
+      "loss": 0.1829,
+      "step": 5540
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 2.004664182662964,
+      "learning_rate": 3.361308502817937e-05,
+      "loss": 0.182,
+      "step": 5550
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.9720888137817383,
+      "learning_rate": 3.358245528056849e-05,
+      "loss": 0.1946,
+      "step": 5560
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.6719261407852173,
+      "learning_rate": 3.355182553295761e-05,
+      "loss": 0.1981,
+      "step": 5570
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.6081453561782837,
+      "learning_rate": 3.352119578534673e-05,
+      "loss": 0.1825,
+      "step": 5580
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.5715957880020142,
+      "learning_rate": 3.349056603773585e-05,
+      "loss": 0.1742,
+      "step": 5590
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.535361647605896,
+      "learning_rate": 3.345993629012497e-05,
+      "loss": 0.169,
+      "step": 5600
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.5899757146835327,
+      "learning_rate": 3.342930654251409e-05,
+      "loss": 0.1797,
+      "step": 5610
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.0383834838867188,
+      "learning_rate": 3.339867679490321e-05,
+      "loss": 0.1862,
+      "step": 5620
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.2471222877502441,
+      "learning_rate": 3.3368047047292335e-05,
+      "loss": 0.1804,
+      "step": 5630
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.9518430829048157,
+      "learning_rate": 3.333741729968145e-05,
+      "loss": 0.1718,
+      "step": 5640
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.4693788290023804,
+      "learning_rate": 3.330678755207057e-05,
+      "loss": 0.1564,
+      "step": 5650
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.2576932907104492,
+      "learning_rate": 3.3276157804459696e-05,
+      "loss": 0.1723,
+      "step": 5660
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 2.0755367279052734,
+      "learning_rate": 3.3245528056848815e-05,
+      "loss": 0.1797,
+      "step": 5670
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.4428520202636719,
+      "learning_rate": 3.321489830923793e-05,
+      "loss": 0.1802,
+      "step": 5680
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.2220004796981812,
+      "learning_rate": 3.318426856162705e-05,
+      "loss": 0.1814,
+      "step": 5690
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.5297048091888428,
+      "learning_rate": 3.315363881401618e-05,
+      "loss": 0.1677,
+      "step": 5700
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.548630714416504,
+      "learning_rate": 3.312300906640529e-05,
+      "loss": 0.188,
+      "step": 5710
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.1936557292938232,
+      "learning_rate": 3.3092379318794413e-05,
+      "loss": 0.191,
+      "step": 5720
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.197527289390564,
+      "learning_rate": 3.306174957118354e-05,
+      "loss": 0.1771,
+      "step": 5730
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.7312843799591064,
+      "learning_rate": 3.303111982357266e-05,
+      "loss": 0.2025,
+      "step": 5740
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.376200795173645,
+      "learning_rate": 3.3000490075961775e-05,
+      "loss": 0.1688,
+      "step": 5750
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.693571925163269,
+      "learning_rate": 3.2969860328350894e-05,
+      "loss": 0.1899,
+      "step": 5760
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.1624635457992554,
+      "learning_rate": 3.293923058074002e-05,
+      "loss": 0.1601,
+      "step": 5770
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.5829757452011108,
+      "learning_rate": 3.290860083312914e-05,
+      "loss": 0.1999,
+      "step": 5780
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.1708341836929321,
+      "learning_rate": 3.2877971085518256e-05,
+      "loss": 0.1984,
+      "step": 5790
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.3500685691833496,
+      "learning_rate": 3.284734133790738e-05,
+      "loss": 0.1639,
+      "step": 5800
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.4064408540725708,
+      "learning_rate": 3.281671159029649e-05,
+      "loss": 0.178,
+      "step": 5810
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.5399430990219116,
+      "learning_rate": 3.278608184268562e-05,
+      "loss": 0.1795,
+      "step": 5820
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.3644287586212158,
+      "learning_rate": 3.275545209507474e-05,
+      "loss": 0.161,
+      "step": 5830
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.8360896110534668,
+      "learning_rate": 3.2724822347463854e-05,
+      "loss": 0.1945,
+      "step": 5840
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.4975295066833496,
+      "learning_rate": 3.269419259985298e-05,
+      "loss": 0.1882,
+      "step": 5850
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.2881444692611694,
+      "learning_rate": 3.26635628522421e-05,
+      "loss": 0.1767,
+      "step": 5860
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.6968104839324951,
+      "learning_rate": 3.2632933104631216e-05,
+      "loss": 0.1886,
+      "step": 5870
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.5182228088378906,
+      "learning_rate": 3.260230335702034e-05,
+      "loss": 0.204,
+      "step": 5880
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.3683419227600098,
+      "learning_rate": 3.257167360940946e-05,
+      "loss": 0.1939,
+      "step": 5890
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.1811920404434204,
+      "learning_rate": 3.254104386179858e-05,
+      "loss": 0.1937,
+      "step": 5900
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.9215272665023804,
+      "learning_rate": 3.2510414114187696e-05,
+      "loss": 0.1594,
+      "step": 5910
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.2016764879226685,
+      "learning_rate": 3.247978436657682e-05,
+      "loss": 0.1799,
+      "step": 5920
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.8221676349639893,
+      "learning_rate": 3.244915461896595e-05,
+      "loss": 0.1848,
+      "step": 5930
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.8971909284591675,
+      "learning_rate": 3.241852487135506e-05,
+      "loss": 0.1953,
+      "step": 5940
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.4460947513580322,
+      "learning_rate": 3.2387895123744183e-05,
+      "loss": 0.1747,
+      "step": 5950
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.1138249635696411,
+      "learning_rate": 3.23572653761333e-05,
+      "loss": 0.1558,
+      "step": 5960
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.1532104015350342,
+      "learning_rate": 3.232663562852242e-05,
+      "loss": 0.1759,
+      "step": 5970
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.4475568532943726,
+      "learning_rate": 3.2296005880911545e-05,
+      "loss": 0.1704,
+      "step": 5980
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.2649788856506348,
+      "learning_rate": 3.2265376133300664e-05,
+      "loss": 0.1929,
+      "step": 5990
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.468369960784912,
+      "learning_rate": 3.223474638568978e-05,
+      "loss": 0.1938,
+      "step": 6000
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.6281121969223022,
+      "learning_rate": 3.22041166380789e-05,
+      "loss": 0.1786,
+      "step": 6010
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.4551911354064941,
+      "learning_rate": 3.2173486890468026e-05,
+      "loss": 0.1772,
+      "step": 6020
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.7859156131744385,
+      "learning_rate": 3.2142857142857144e-05,
+      "loss": 0.1917,
+      "step": 6030
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.6637579202651978,
+      "learning_rate": 3.211222739524626e-05,
+      "loss": 0.1564,
+      "step": 6040
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.7902681827545166,
+      "learning_rate": 3.208159764763539e-05,
+      "loss": 0.158,
+      "step": 6050
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.2490273714065552,
+      "learning_rate": 3.2050967900024506e-05,
+      "loss": 0.173,
+      "step": 6060
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.1261332035064697,
+      "learning_rate": 3.2020338152413624e-05,
+      "loss": 0.1798,
+      "step": 6070
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.464557409286499,
+      "learning_rate": 3.198970840480275e-05,
+      "loss": 0.2003,
+      "step": 6080
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.4368430376052856,
+      "learning_rate": 3.195907865719187e-05,
+      "loss": 0.1627,
+      "step": 6090
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.3397611379623413,
+      "learning_rate": 3.1928448909580986e-05,
+      "loss": 0.1895,
+      "step": 6100
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.1559630632400513,
+      "learning_rate": 3.1897819161970104e-05,
+      "loss": 0.1854,
+      "step": 6110
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.4304611682891846,
+      "learning_rate": 3.186718941435923e-05,
+      "loss": 0.1694,
+      "step": 6120
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.4819403886795044,
+      "learning_rate": 3.183655966674835e-05,
+      "loss": 0.1739,
+      "step": 6130
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.1274824142456055,
+      "learning_rate": 3.1805929919137466e-05,
+      "loss": 0.1627,
+      "step": 6140
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.2060612440109253,
+      "learning_rate": 3.177530017152659e-05,
+      "loss": 0.2006,
+      "step": 6150
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.165612816810608,
+      "learning_rate": 3.174467042391571e-05,
+      "loss": 0.179,
+      "step": 6160
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.30056631565094,
+      "learning_rate": 3.171404067630483e-05,
+      "loss": 0.1517,
+      "step": 6170
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.686110019683838,
+      "learning_rate": 3.168341092869395e-05,
+      "loss": 0.1803,
+      "step": 6180
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.522021770477295,
+      "learning_rate": 3.165278118108307e-05,
+      "loss": 0.1454,
+      "step": 6190
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.4369442462921143,
+      "learning_rate": 3.162215143347219e-05,
+      "loss": 0.1857,
+      "step": 6200
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.3451831340789795,
+      "learning_rate": 3.159152168586131e-05,
+      "loss": 0.19,
+      "step": 6210
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.656409740447998,
+      "learning_rate": 3.1560891938250434e-05,
+      "loss": 0.1636,
+      "step": 6220
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.8665788173675537,
+      "learning_rate": 3.1530262190639545e-05,
+      "loss": 0.1787,
+      "step": 6230
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.7516517639160156,
+      "learning_rate": 3.149963244302867e-05,
+      "loss": 0.1601,
+      "step": 6240
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.1921964883804321,
+      "learning_rate": 3.1469002695417795e-05,
+      "loss": 0.184,
+      "step": 6250
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.1232781410217285,
+      "learning_rate": 3.143837294780691e-05,
+      "loss": 0.1676,
+      "step": 6260
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.3251980543136597,
+      "learning_rate": 3.140774320019603e-05,
+      "loss": 0.1659,
+      "step": 6270
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.579655408859253,
+      "learning_rate": 3.137711345258515e-05,
+      "loss": 0.171,
+      "step": 6280
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.501421570777893,
+      "learning_rate": 3.134648370497427e-05,
+      "loss": 0.1801,
+      "step": 6290
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.5305452346801758,
+      "learning_rate": 3.1315853957363394e-05,
+      "loss": 0.1707,
+      "step": 6300
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.7370951175689697,
+      "learning_rate": 3.128522420975251e-05,
+      "loss": 0.1696,
+      "step": 6310
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.65492582321167,
+      "learning_rate": 3.125459446214163e-05,
+      "loss": 0.183,
+      "step": 6320
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.4975395202636719,
+      "learning_rate": 3.122396471453075e-05,
+      "loss": 0.1928,
+      "step": 6330
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.260558843612671,
+      "learning_rate": 3.1193334966919874e-05,
+      "loss": 0.1795,
+      "step": 6340
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.4968366622924805,
+      "learning_rate": 3.1162705219309e-05,
+      "loss": 0.1763,
+      "step": 6350
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.5058164596557617,
+      "learning_rate": 3.113207547169811e-05,
+      "loss": 0.2023,
+      "step": 6360
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.329582929611206,
+      "learning_rate": 3.1101445724087236e-05,
+      "loss": 0.1766,
+      "step": 6370
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.3644320964813232,
+      "learning_rate": 3.1070815976476355e-05,
+      "loss": 0.1722,
+      "step": 6380
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.2065993547439575,
+      "learning_rate": 3.104018622886547e-05,
+      "loss": 0.1846,
+      "step": 6390
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.4604241847991943,
+      "learning_rate": 3.10095564812546e-05,
+      "loss": 0.1652,
+      "step": 6400
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.6481974124908447,
+      "learning_rate": 3.0978926733643717e-05,
+      "loss": 0.1882,
+      "step": 6410
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.674408197402954,
+      "learning_rate": 3.0948296986032835e-05,
+      "loss": 0.1824,
+      "step": 6420
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.0522202253341675,
+      "learning_rate": 3.091766723842195e-05,
+      "loss": 0.1913,
+      "step": 6430
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.3425509929656982,
+      "learning_rate": 3.088703749081108e-05,
+      "loss": 0.1649,
+      "step": 6440
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.9756949543952942,
+      "learning_rate": 3.08564077432002e-05,
+      "loss": 0.1489,
+      "step": 6450
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.2803707122802734,
+      "learning_rate": 3.0825777995589315e-05,
+      "loss": 0.1553,
+      "step": 6460
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.1759666204452515,
+      "learning_rate": 3.079514824797844e-05,
+      "loss": 0.1671,
+      "step": 6470
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.2560944557189941,
+      "learning_rate": 3.076451850036756e-05,
+      "loss": 0.1706,
+      "step": 6480
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.4681024551391602,
+      "learning_rate": 3.073388875275668e-05,
+      "loss": 0.1607,
+      "step": 6490
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.558706283569336,
+      "learning_rate": 3.07032590051458e-05,
+      "loss": 0.164,
+      "step": 6500
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.6408095359802246,
+      "learning_rate": 3.067262925753492e-05,
+      "loss": 0.167,
+      "step": 6510
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.364626407623291,
+      "learning_rate": 3.064199950992404e-05,
+      "loss": 0.1957,
+      "step": 6520
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.2001851797103882,
+      "learning_rate": 3.061136976231316e-05,
+      "loss": 0.1563,
+      "step": 6530
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.8316638469696045,
+      "learning_rate": 3.058074001470228e-05,
+      "loss": 0.1471,
+      "step": 6540
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.423321008682251,
+      "learning_rate": 3.05501102670914e-05,
+      "loss": 0.1659,
+      "step": 6550
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.596010684967041,
+      "learning_rate": 3.051948051948052e-05,
+      "loss": 0.1975,
+      "step": 6560
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.2045807838439941,
+      "learning_rate": 3.048885077186964e-05,
+      "loss": 0.186,
+      "step": 6570
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.1752638816833496,
+      "learning_rate": 3.045822102425876e-05,
+      "loss": 0.1584,
+      "step": 6580
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.2570778131484985,
+      "learning_rate": 3.042759127664788e-05,
+      "loss": 0.1713,
+      "step": 6590
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.229406714439392,
+      "learning_rate": 3.0396961529037006e-05,
+      "loss": 0.1612,
+      "step": 6600
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.3653537034988403,
+      "learning_rate": 3.036633178142612e-05,
+      "loss": 0.1817,
+      "step": 6610
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.9214671850204468,
+      "learning_rate": 3.0335702033815243e-05,
+      "loss": 0.1611,
+      "step": 6620
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.596480369567871,
+      "learning_rate": 3.030507228620436e-05,
+      "loss": 0.1945,
+      "step": 6630
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.0726596117019653,
+      "learning_rate": 3.0274442538593483e-05,
+      "loss": 0.1815,
+      "step": 6640
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 2.0356228351593018,
+      "learning_rate": 3.0243812790982605e-05,
+      "loss": 0.1871,
+      "step": 6650
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.2258411645889282,
+      "learning_rate": 3.0213183043371723e-05,
+      "loss": 0.1862,
+      "step": 6660
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.4276363849639893,
+      "learning_rate": 3.0182553295760845e-05,
+      "loss": 0.1863,
+      "step": 6670
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.5817575454711914,
+      "learning_rate": 3.0151923548149963e-05,
+      "loss": 0.1684,
+      "step": 6680
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.6616454124450684,
+      "learning_rate": 3.0121293800539085e-05,
+      "loss": 0.1803,
+      "step": 6690
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.172263741493225,
+      "learning_rate": 3.0090664052928207e-05,
+      "loss": 0.1838,
+      "step": 6700
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.3011975288391113,
+      "learning_rate": 3.0060034305317325e-05,
+      "loss": 0.1621,
+      "step": 6710
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.00931715965271,
+      "learning_rate": 3.0029404557706447e-05,
+      "loss": 0.179,
+      "step": 6720
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 2.7549054622650146,
+      "learning_rate": 2.9998774810095565e-05,
+      "loss": 0.1821,
+      "step": 6730
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.609864592552185,
+      "learning_rate": 2.9968145062484687e-05,
+      "loss": 0.1734,
+      "step": 6740
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.521108865737915,
+      "learning_rate": 2.9937515314873805e-05,
+      "loss": 0.1724,
+      "step": 6750
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.6473437547683716,
+      "learning_rate": 2.9906885567262927e-05,
+      "loss": 0.1875,
+      "step": 6760
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.340497612953186,
+      "learning_rate": 2.987625581965205e-05,
+      "loss": 0.1693,
+      "step": 6770
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.6408023834228516,
+      "learning_rate": 2.9845626072041167e-05,
+      "loss": 0.1654,
+      "step": 6780
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.0924148559570312,
+      "learning_rate": 2.981499632443029e-05,
+      "loss": 0.1934,
+      "step": 6790
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.5360684394836426,
+      "learning_rate": 2.9784366576819404e-05,
+      "loss": 0.2011,
+      "step": 6800
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.1396403312683105,
+      "learning_rate": 2.975373682920853e-05,
+      "loss": 0.1765,
+      "step": 6810
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.193796157836914,
+      "learning_rate": 2.972310708159765e-05,
+      "loss": 0.1601,
+      "step": 6820
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.3440088033676147,
+      "learning_rate": 2.969247733398677e-05,
+      "loss": 0.1564,
+      "step": 6830
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.3894766569137573,
+      "learning_rate": 2.966184758637589e-05,
+      "loss": 0.1709,
+      "step": 6840
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.206499695777893,
+      "learning_rate": 2.9631217838765006e-05,
+      "loss": 0.1766,
+      "step": 6850
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.264289140701294,
+      "learning_rate": 2.960058809115413e-05,
+      "loss": 0.1965,
+      "step": 6860
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.3975270986557007,
+      "learning_rate": 2.9569958343543253e-05,
+      "loss": 0.163,
+      "step": 6870
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.3025867938995361,
+      "learning_rate": 2.9539328595932368e-05,
+      "loss": 0.1695,
+      "step": 6880
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.0210461616516113,
+      "learning_rate": 2.9508698848321493e-05,
+      "loss": 0.1809,
+      "step": 6890
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.4554662704467773,
+      "learning_rate": 2.9478069100710608e-05,
+      "loss": 0.1393,
+      "step": 6900
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.2963159084320068,
+      "learning_rate": 2.9447439353099733e-05,
+      "loss": 0.1704,
+      "step": 6910
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.4193776845932007,
+      "learning_rate": 2.9416809605488855e-05,
+      "loss": 0.134,
+      "step": 6920
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.3702312707901,
+      "learning_rate": 2.938617985787797e-05,
+      "loss": 0.1813,
+      "step": 6930
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.3123691082000732,
+      "learning_rate": 2.9355550110267095e-05,
+      "loss": 0.1569,
+      "step": 6940
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.4189176559448242,
+      "learning_rate": 2.932492036265621e-05,
+      "loss": 0.1922,
+      "step": 6950
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.4712148904800415,
+      "learning_rate": 2.9294290615045332e-05,
+      "loss": 0.1603,
+      "step": 6960
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.4863829612731934,
+      "learning_rate": 2.9263660867434457e-05,
+      "loss": 0.1693,
+      "step": 6970
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.9813027381896973,
+      "learning_rate": 2.9233031119823572e-05,
+      "loss": 0.1721,
+      "step": 6980
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.2544353008270264,
+      "learning_rate": 2.9202401372212694e-05,
+      "loss": 0.1609,
+      "step": 6990
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.2305036783218384,
+      "learning_rate": 2.9171771624601812e-05,
+      "loss": 0.1772,
+      "step": 7000
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.1962831020355225,
+      "learning_rate": 2.9141141876990934e-05,
+      "loss": 0.1648,
+      "step": 7010
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.3442049026489258,
+      "learning_rate": 2.911051212938006e-05,
+      "loss": 0.1614,
+      "step": 7020
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.282299518585205,
+      "learning_rate": 2.9079882381769174e-05,
+      "loss": 0.1605,
+      "step": 7030
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.1980081796646118,
+      "learning_rate": 2.9049252634158296e-05,
+      "loss": 0.1527,
+      "step": 7040
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.8890668153762817,
+      "learning_rate": 2.9018622886547414e-05,
+      "loss": 0.1788,
+      "step": 7050
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.5879610776901245,
+      "learning_rate": 2.8987993138936536e-05,
+      "loss": 0.1461,
+      "step": 7060
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.011800765991211,
+      "learning_rate": 2.8957363391325658e-05,
+      "loss": 0.1388,
+      "step": 7070
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.9925493001937866,
+      "learning_rate": 2.8926733643714776e-05,
+      "loss": 0.1758,
+      "step": 7080
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.9547811150550842,
+      "learning_rate": 2.8896103896103898e-05,
+      "loss": 0.1654,
+      "step": 7090
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.4539928436279297,
+      "learning_rate": 2.8865474148493016e-05,
+      "loss": 0.1504,
+      "step": 7100
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.6095476150512695,
+      "learning_rate": 2.8834844400882138e-05,
+      "loss": 0.1775,
+      "step": 7110
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.447981834411621,
+      "learning_rate": 2.880421465327126e-05,
+      "loss": 0.1543,
+      "step": 7120
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.9824799299240112,
+      "learning_rate": 2.8773584905660378e-05,
+      "loss": 0.1688,
+      "step": 7130
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.537764549255371,
+      "learning_rate": 2.87429551580495e-05,
+      "loss": 0.1823,
+      "step": 7140
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.4593915939331055,
+      "learning_rate": 2.8712325410438618e-05,
+      "loss": 0.1543,
+      "step": 7150
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.31696355342865,
+      "learning_rate": 2.868169566282774e-05,
+      "loss": 0.1511,
+      "step": 7160
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.6222825050354004,
+      "learning_rate": 2.8651065915216862e-05,
+      "loss": 0.177,
+      "step": 7170
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.5468302965164185,
+      "learning_rate": 2.862043616760598e-05,
+      "loss": 0.1587,
+      "step": 7180
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.7186715602874756,
+      "learning_rate": 2.8589806419995102e-05,
+      "loss": 0.1671,
+      "step": 7190
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.1558398008346558,
+      "learning_rate": 2.855917667238422e-05,
+      "loss": 0.153,
+      "step": 7200
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.890857458114624,
+      "learning_rate": 2.8528546924773342e-05,
+      "loss": 0.1698,
+      "step": 7210
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.917914867401123,
+      "learning_rate": 2.8497917177162464e-05,
+      "loss": 0.1691,
+      "step": 7220
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.1543052196502686,
+      "learning_rate": 2.8467287429551582e-05,
+      "loss": 0.1795,
+      "step": 7230
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.4072394371032715,
+      "learning_rate": 2.8436657681940704e-05,
+      "loss": 0.1693,
+      "step": 7240
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.8398141860961914,
+      "learning_rate": 2.8406027934329822e-05,
+      "loss": 0.1696,
+      "step": 7250
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.453688383102417,
+      "learning_rate": 2.8375398186718944e-05,
+      "loss": 0.1766,
+      "step": 7260
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.4082579612731934,
+      "learning_rate": 2.834476843910806e-05,
+      "loss": 0.1547,
+      "step": 7270
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.419756531715393,
+      "learning_rate": 2.8314138691497184e-05,
+      "loss": 0.1755,
+      "step": 7280
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.276735782623291,
+      "learning_rate": 2.8283508943886306e-05,
+      "loss": 0.1497,
+      "step": 7290
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.2450299263000488,
+      "learning_rate": 2.825287919627542e-05,
+      "loss": 0.1686,
+      "step": 7300
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.4796665906906128,
+      "learning_rate": 2.8222249448664546e-05,
+      "loss": 0.1652,
+      "step": 7310
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.361664056777954,
+      "learning_rate": 2.819161970105366e-05,
+      "loss": 0.1699,
+      "step": 7320
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.1027274131774902,
+      "learning_rate": 2.8160989953442786e-05,
+      "loss": 0.1636,
+      "step": 7330
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.5320936441421509,
+      "learning_rate": 2.8130360205831908e-05,
+      "loss": 0.1947,
+      "step": 7340
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.3628391027450562,
+      "learning_rate": 2.8099730458221023e-05,
+      "loss": 0.1622,
+      "step": 7350
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.9188462495803833,
+      "learning_rate": 2.8069100710610148e-05,
+      "loss": 0.1388,
+      "step": 7360
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.6645480394363403,
+      "learning_rate": 2.8038470962999263e-05,
+      "loss": 0.1695,
+      "step": 7370
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.2059977054595947,
+      "learning_rate": 2.8007841215388385e-05,
+      "loss": 0.1551,
+      "step": 7380
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.0181052684783936,
+      "learning_rate": 2.797721146777751e-05,
+      "loss": 0.14,
+      "step": 7390
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.2087146043777466,
+      "learning_rate": 2.7946581720166625e-05,
+      "loss": 0.154,
+      "step": 7400
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.3544085025787354,
+      "learning_rate": 2.7915951972555747e-05,
+      "loss": 0.1792,
+      "step": 7410
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.2395013570785522,
+      "learning_rate": 2.7885322224944865e-05,
+      "loss": 0.1452,
+      "step": 7420
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.326521635055542,
+      "learning_rate": 2.7854692477333987e-05,
+      "loss": 0.1762,
+      "step": 7430
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.1233044862747192,
+      "learning_rate": 2.7824062729723112e-05,
+      "loss": 0.1601,
+      "step": 7440
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.2976155281066895,
+      "learning_rate": 2.7793432982112227e-05,
+      "loss": 0.1559,
+      "step": 7450
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.2908642292022705,
+      "learning_rate": 2.776280323450135e-05,
+      "loss": 0.1571,
+      "step": 7460
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 2.082435369491577,
+      "learning_rate": 2.7732173486890467e-05,
+      "loss": 0.1686,
+      "step": 7470
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.3823455572128296,
+      "learning_rate": 2.770154373927959e-05,
+      "loss": 0.1834,
+      "step": 7480
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.6810978651046753,
+      "learning_rate": 2.767091399166871e-05,
+      "loss": 0.1686,
+      "step": 7490
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.3233197927474976,
+      "learning_rate": 2.764028424405783e-05,
+      "loss": 0.1404,
+      "step": 7500
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.1426876783370972,
+      "learning_rate": 2.760965449644695e-05,
+      "loss": 0.1557,
+      "step": 7510
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.426680088043213,
+      "learning_rate": 2.757902474883607e-05,
+      "loss": 0.1789,
+      "step": 7520
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 2.4810688495635986,
+      "learning_rate": 2.754839500122519e-05,
+      "loss": 0.1688,
+      "step": 7530
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.5051275491714478,
+      "learning_rate": 2.7517765253614313e-05,
+      "loss": 0.1629,
+      "step": 7540
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.5268399715423584,
+      "learning_rate": 2.748713550600343e-05,
+      "loss": 0.1619,
+      "step": 7550
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.3295812606811523,
+      "learning_rate": 2.7456505758392553e-05,
+      "loss": 0.1903,
+      "step": 7560
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.991754412651062,
+      "learning_rate": 2.742587601078167e-05,
+      "loss": 0.1784,
+      "step": 7570
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.1559555530548096,
+      "learning_rate": 2.7395246263170793e-05,
+      "loss": 0.1732,
+      "step": 7580
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.5855534076690674,
+      "learning_rate": 2.7364616515559915e-05,
+      "loss": 0.1681,
+      "step": 7590
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.5041630268096924,
+      "learning_rate": 2.7333986767949033e-05,
+      "loss": 0.1552,
+      "step": 7600
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.3512407541275024,
+      "learning_rate": 2.7303357020338155e-05,
+      "loss": 0.1594,
+      "step": 7610
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.3841335773468018,
+      "learning_rate": 2.7272727272727273e-05,
+      "loss": 0.152,
+      "step": 7620
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.3479012250900269,
+      "learning_rate": 2.7242097525116395e-05,
+      "loss": 0.1779,
+      "step": 7630
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.0527963638305664,
+      "learning_rate": 2.7211467777505517e-05,
+      "loss": 0.1667,
+      "step": 7640
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.7754255533218384,
+      "learning_rate": 2.7180838029894635e-05,
+      "loss": 0.1707,
+      "step": 7650
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.2524574995040894,
+      "learning_rate": 2.7150208282283757e-05,
+      "loss": 0.1579,
+      "step": 7660
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.2442678213119507,
+      "learning_rate": 2.7119578534672875e-05,
+      "loss": 0.1678,
+      "step": 7670
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.3260364532470703,
+      "learning_rate": 2.7088948787061997e-05,
+      "loss": 0.1693,
+      "step": 7680
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.2896537780761719,
+      "learning_rate": 2.705831903945112e-05,
+      "loss": 0.1474,
+      "step": 7690
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.0071998834609985,
+      "learning_rate": 2.7027689291840237e-05,
+      "loss": 0.1621,
+      "step": 7700
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.1180951595306396,
+      "learning_rate": 2.699705954422936e-05,
+      "loss": 0.1619,
+      "step": 7710
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.3478256464004517,
+      "learning_rate": 2.6966429796618474e-05,
+      "loss": 0.1372,
+      "step": 7720
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.1745206117630005,
+      "learning_rate": 2.69358000490076e-05,
+      "loss": 0.1462,
+      "step": 7730
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.4203293323516846,
+      "learning_rate": 2.690517030139672e-05,
+      "loss": 0.1427,
+      "step": 7740
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.16546630859375,
+      "learning_rate": 2.687454055378584e-05,
+      "loss": 0.1661,
+      "step": 7750
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.4038243293762207,
+      "learning_rate": 2.684391080617496e-05,
+      "loss": 0.1716,
+      "step": 7760
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.6690312623977661,
+      "learning_rate": 2.6813281058564076e-05,
+      "loss": 0.1836,
+      "step": 7770
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.225852370262146,
+      "learning_rate": 2.67826513109532e-05,
+      "loss": 0.1585,
+      "step": 7780
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.5048850774765015,
+      "learning_rate": 2.6752021563342316e-05,
+      "loss": 0.1483,
+      "step": 7790
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.8813077807426453,
+      "learning_rate": 2.6721391815731438e-05,
+      "loss": 0.1457,
+      "step": 7800
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.18279230594635,
+      "learning_rate": 2.6690762068120563e-05,
+      "loss": 0.1616,
+      "step": 7810
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.2559739351272583,
+      "learning_rate": 2.6660132320509678e-05,
+      "loss": 0.1671,
+      "step": 7820
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.8123114109039307,
+      "learning_rate": 2.66295025728988e-05,
+      "loss": 0.1687,
+      "step": 7830
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.171390414237976,
+      "learning_rate": 2.6598872825287918e-05,
+      "loss": 0.1669,
+      "step": 7840
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.0999939441680908,
+      "learning_rate": 2.656824307767704e-05,
+      "loss": 0.183,
+      "step": 7850
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.0986688137054443,
+      "learning_rate": 2.6537613330066165e-05,
+      "loss": 0.1517,
+      "step": 7860
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.43050217628479,
+      "learning_rate": 2.650698358245528e-05,
+      "loss": 0.1779,
+      "step": 7870
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.1102867126464844,
+      "learning_rate": 2.64763538348444e-05,
+      "loss": 0.1719,
+      "step": 7880
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.6433628797531128,
+      "learning_rate": 2.644572408723352e-05,
+      "loss": 0.1627,
+      "step": 7890
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.135524868965149,
+      "learning_rate": 2.641509433962264e-05,
+      "loss": 0.148,
+      "step": 7900
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.1294916868209839,
+      "learning_rate": 2.6384464592011763e-05,
+      "loss": 0.144,
+      "step": 7910
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.2410553693771362,
+      "learning_rate": 2.6353834844400882e-05,
+      "loss": 0.1513,
+      "step": 7920
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 2.2181475162506104,
+      "learning_rate": 2.6323205096790004e-05,
+      "loss": 0.1744,
+      "step": 7930
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.2367093563079834,
+      "learning_rate": 2.6292575349179122e-05,
+      "loss": 0.1733,
+      "step": 7940
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.2846170663833618,
+      "learning_rate": 2.6261945601568244e-05,
+      "loss": 0.1651,
+      "step": 7950
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.7288285493850708,
+      "learning_rate": 2.6231315853957365e-05,
+      "loss": 0.1713,
+      "step": 7960
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.2710671424865723,
+      "learning_rate": 2.6200686106346484e-05,
+      "loss": 0.1673,
+      "step": 7970
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.3041337728500366,
+      "learning_rate": 2.6170056358735606e-05,
+      "loss": 0.1743,
+      "step": 7980
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.030639410018921,
+      "learning_rate": 2.6139426611124724e-05,
+      "loss": 0.155,
+      "step": 7990
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 2.0335237979888916,
+      "learning_rate": 2.6108796863513846e-05,
+      "loss": 0.1449,
+      "step": 8000
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.7640258073806763,
+      "learning_rate": 2.6078167115902967e-05,
+      "loss": 0.1771,
+      "step": 8010
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.259002447128296,
+      "learning_rate": 2.6047537368292086e-05,
+      "loss": 0.1349,
+      "step": 8020
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.651031732559204,
+      "learning_rate": 2.6016907620681208e-05,
+      "loss": 0.1692,
+      "step": 8030
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.7378960847854614,
+      "learning_rate": 2.5986277873070326e-05,
+      "loss": 0.1594,
+      "step": 8040
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.9246784448623657,
+      "learning_rate": 2.5955648125459448e-05,
+      "loss": 0.1999,
+      "step": 8050
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.4016398191452026,
+      "learning_rate": 2.592501837784857e-05,
+      "loss": 0.1594,
+      "step": 8060
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.453421950340271,
+      "learning_rate": 2.5894388630237688e-05,
+      "loss": 0.1481,
+      "step": 8070
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.3118133544921875,
+      "learning_rate": 2.586375888262681e-05,
+      "loss": 0.1378,
+      "step": 8080
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.0037002563476562,
+      "learning_rate": 2.5833129135015928e-05,
+      "loss": 0.1409,
+      "step": 8090
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 2.507158041000366,
+      "learning_rate": 2.580249938740505e-05,
+      "loss": 0.1576,
+      "step": 8100
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.3732621669769287,
+      "learning_rate": 2.577186963979417e-05,
+      "loss": 0.1732,
+      "step": 8110
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.177677869796753,
+      "learning_rate": 2.574123989218329e-05,
+      "loss": 0.161,
+      "step": 8120
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.133461833000183,
+      "learning_rate": 2.571061014457241e-05,
+      "loss": 0.1522,
+      "step": 8130
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.411163091659546,
+      "learning_rate": 2.5679980396961527e-05,
+      "loss": 0.1362,
+      "step": 8140
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.4202933311462402,
+      "learning_rate": 2.5649350649350652e-05,
+      "loss": 0.1481,
+      "step": 8150
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.0804930925369263,
+      "learning_rate": 2.5618720901739774e-05,
+      "loss": 0.1514,
+      "step": 8160
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.3504397869110107,
+      "learning_rate": 2.5588091154128892e-05,
+      "loss": 0.151,
+      "step": 8170
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.161055088043213,
+      "learning_rate": 2.5557461406518014e-05,
+      "loss": 0.1499,
+      "step": 8180
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.7804343700408936,
+      "learning_rate": 2.552683165890713e-05,
+      "loss": 0.1447,
+      "step": 8190
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.501635193824768,
+      "learning_rate": 2.5496201911296254e-05,
+      "loss": 0.1801,
+      "step": 8200
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.0425118207931519,
+      "learning_rate": 2.5465572163685376e-05,
+      "loss": 0.1634,
+      "step": 8210
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.719896674156189,
+      "learning_rate": 2.543494241607449e-05,
+      "loss": 0.1583,
+      "step": 8220
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 1.4397616386413574,
+      "learning_rate": 2.5404312668463616e-05,
+      "loss": 0.1615,
+      "step": 8230
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 1.292136549949646,
+      "learning_rate": 2.537368292085273e-05,
+      "loss": 0.1594,
+      "step": 8240
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.8452214002609253,
+      "learning_rate": 2.5343053173241856e-05,
+      "loss": 0.1569,
+      "step": 8250
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 1.3470544815063477,
+      "learning_rate": 2.5312423425630978e-05,
+      "loss": 0.1585,
+      "step": 8260
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 16524,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 500,
+  "total_flos": 1.6100359460093952e+17,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}