diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,7826 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 20.0,
+  "eval_steps": 500,
+  "global_step": 11120,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.017985611510791366,
+      "grad_norm": 7.884528160095215,
+      "learning_rate": 1.7985611510791366e-06,
+      "loss": 1.3258,
+      "step": 10
+    },
+    {
+      "epoch": 0.03597122302158273,
+      "grad_norm": 7.888326644897461,
+      "learning_rate": 3.5971223021582732e-06,
+      "loss": 1.4075,
+      "step": 20
+    },
+    {
+      "epoch": 0.0539568345323741,
+      "grad_norm": 7.247873783111572,
+      "learning_rate": 5.3956834532374105e-06,
+      "loss": 1.3813,
+      "step": 30
+    },
+    {
+      "epoch": 0.07194244604316546,
+      "grad_norm": 6.167496204376221,
+      "learning_rate": 7.1942446043165465e-06,
+      "loss": 1.2313,
+      "step": 40
+    },
+    {
+      "epoch": 0.08992805755395683,
+      "grad_norm": 5.86017370223999,
+      "learning_rate": 8.992805755395683e-06,
+      "loss": 1.0879,
+      "step": 50
+    },
+    {
+      "epoch": 0.1079136690647482,
+      "grad_norm": 4.6050872802734375,
+      "learning_rate": 1.0791366906474821e-05,
+      "loss": 0.9481,
+      "step": 60
+    },
+    {
+      "epoch": 0.12589928057553956,
+      "grad_norm": 4.555079936981201,
+      "learning_rate": 1.2589928057553957e-05,
+      "loss": 0.7796,
+      "step": 70
+    },
+    {
+      "epoch": 0.14388489208633093,
+      "grad_norm": 5.556536674499512,
+      "learning_rate": 1.4388489208633093e-05,
+      "loss": 0.6536,
+      "step": 80
+    },
+    {
+      "epoch": 0.1618705035971223,
+      "grad_norm": 1.7559210062026978,
+      "learning_rate": 1.618705035971223e-05,
+      "loss": 0.5556,
+      "step": 90
+    },
+    {
+      "epoch": 0.17985611510791366,
+      "grad_norm": 3.5355215072631836,
+      "learning_rate": 1.7985611510791367e-05,
+      "loss": 0.4709,
+      "step": 100
+    },
+    {
+      "epoch": 0.19784172661870503,
+      "grad_norm": 1.762351632118225,
+      "learning_rate": 1.9784172661870504e-05,
+      "loss": 0.3749,
+      "step": 110
+    },
+    {
+      "epoch": 0.2158273381294964,
+      "grad_norm": 5.933515548706055,
+      "learning_rate": 2.1582733812949642e-05,
+      "loss": 0.3193,
+      "step": 120
+    },
+    {
+      "epoch": 0.23381294964028776,
+      "grad_norm": 56.398719787597656,
+      "learning_rate": 2.3381294964028776e-05,
+      "loss": 0.2682,
+      "step": 130
+    },
+    {
+      "epoch": 0.2517985611510791,
+      "grad_norm": 2.2381231784820557,
+      "learning_rate": 2.5179856115107914e-05,
+      "loss": 0.2628,
+      "step": 140
+    },
+    {
+      "epoch": 0.2697841726618705,
+      "grad_norm": 1.0526609420776367,
+      "learning_rate": 2.697841726618705e-05,
+      "loss": 0.2477,
+      "step": 150
+    },
+    {
+      "epoch": 0.28776978417266186,
+      "grad_norm": 1.2504981756210327,
+      "learning_rate": 2.8776978417266186e-05,
+      "loss": 0.219,
+      "step": 160
+    },
+    {
+      "epoch": 0.3057553956834532,
+      "grad_norm": 1.3793128728866577,
+      "learning_rate": 3.0575539568345324e-05,
+      "loss": 0.2022,
+      "step": 170
+    },
+    {
+      "epoch": 0.3237410071942446,
+      "grad_norm": 1.2026208639144897,
+      "learning_rate": 3.237410071942446e-05,
+      "loss": 0.1956,
+      "step": 180
+    },
+    {
+      "epoch": 0.34172661870503596,
+      "grad_norm": 1.0879606008529663,
+      "learning_rate": 3.41726618705036e-05,
+      "loss": 0.1757,
+      "step": 190
+    },
+    {
+      "epoch": 0.3597122302158273,
+      "grad_norm": 1.0885703563690186,
+      "learning_rate": 3.597122302158273e-05,
+      "loss": 0.1828,
+      "step": 200
+    },
+    {
+      "epoch": 0.3776978417266187,
+      "grad_norm": 1.349827527999878,
+      "learning_rate": 3.776978417266187e-05,
+      "loss": 0.1751,
+      "step": 210
+    },
+    {
+      "epoch": 0.39568345323741005,
+      "grad_norm": 1.4132412672042847,
+      "learning_rate": 3.956834532374101e-05,
+      "loss": 0.1674,
+      "step": 220
+    },
+    {
+      "epoch": 0.4136690647482014,
+      "grad_norm": 1.6034249067306519,
+      "learning_rate": 4.136690647482014e-05,
+      "loss": 0.1531,
+      "step": 230
+    },
+    {
+      "epoch": 0.4316546762589928,
+      "grad_norm": 2.5243616104125977,
+      "learning_rate": 4.3165467625899284e-05,
+      "loss": 0.1406,
+      "step": 240
+    },
+    {
+      "epoch": 0.44964028776978415,
+      "grad_norm": 1.1689270734786987,
+      "learning_rate": 4.496402877697842e-05,
+      "loss": 0.1358,
+      "step": 250
+    },
+    {
+      "epoch": 0.4676258992805755,
+      "grad_norm": 1.9412041902542114,
+      "learning_rate": 4.676258992805755e-05,
+      "loss": 0.1249,
+      "step": 260
+    },
+    {
+      "epoch": 0.4856115107913669,
+      "grad_norm": 0.9682784080505371,
+      "learning_rate": 4.8561151079136694e-05,
+      "loss": 0.1422,
+      "step": 270
+    },
+    {
+      "epoch": 0.5035971223021583,
+      "grad_norm": 2.2200927734375,
+      "learning_rate": 5.035971223021583e-05,
+      "loss": 0.1262,
+      "step": 280
+    },
+    {
+      "epoch": 0.5215827338129496,
+      "grad_norm": 0.920986533164978,
+      "learning_rate": 5.215827338129496e-05,
+      "loss": 0.1256,
+      "step": 290
+    },
+    {
+      "epoch": 0.539568345323741,
+      "grad_norm": 1.342592716217041,
+      "learning_rate": 5.39568345323741e-05,
+      "loss": 0.116,
+      "step": 300
+    },
+    {
+      "epoch": 0.5575539568345323,
+      "grad_norm": 1.378659725189209,
+      "learning_rate": 5.575539568345324e-05,
+      "loss": 0.1182,
+      "step": 310
+    },
+    {
+      "epoch": 0.5755395683453237,
+      "grad_norm": 0.9859443306922913,
+      "learning_rate": 5.755395683453237e-05,
+      "loss": 0.1134,
+      "step": 320
+    },
+    {
+      "epoch": 0.5935251798561151,
+      "grad_norm": 1.2858574390411377,
+      "learning_rate": 5.935251798561151e-05,
+      "loss": 0.1102,
+      "step": 330
+    },
+    {
+      "epoch": 0.6115107913669064,
+      "grad_norm": 1.122875452041626,
+      "learning_rate": 6.115107913669065e-05,
+      "loss": 0.1026,
+      "step": 340
+    },
+    {
+      "epoch": 0.6294964028776978,
+      "grad_norm": 1.0037808418273926,
+      "learning_rate": 6.294964028776978e-05,
+      "loss": 0.0951,
+      "step": 350
+    },
+    {
+      "epoch": 0.6474820143884892,
+      "grad_norm": 0.9888423085212708,
+      "learning_rate": 6.474820143884892e-05,
+      "loss": 0.098,
+      "step": 360
+    },
+    {
+      "epoch": 0.6654676258992805,
+      "grad_norm": 1.0353509187698364,
+      "learning_rate": 6.654676258992806e-05,
+      "loss": 0.0955,
+      "step": 370
+    },
+    {
+      "epoch": 0.6834532374100719,
+      "grad_norm": 0.9193432331085205,
+      "learning_rate": 6.83453237410072e-05,
+      "loss": 0.0922,
+      "step": 380
+    },
+    {
+      "epoch": 0.7014388489208633,
+      "grad_norm": 1.7486399412155151,
+      "learning_rate": 7.014388489208633e-05,
+      "loss": 0.0915,
+      "step": 390
+    },
+    {
+      "epoch": 0.7194244604316546,
+      "grad_norm": 1.4038766622543335,
+      "learning_rate": 7.194244604316547e-05,
+      "loss": 0.09,
+      "step": 400
+    },
+    {
+      "epoch": 0.737410071942446,
+      "grad_norm": 0.5766810178756714,
+      "learning_rate": 7.37410071942446e-05,
+      "loss": 0.0943,
+      "step": 410
+    },
+    {
+      "epoch": 0.7553956834532374,
+      "grad_norm": 1.0704598426818848,
+      "learning_rate": 7.553956834532374e-05,
+      "loss": 0.0869,
+      "step": 420
+    },
+    {
+      "epoch": 0.7733812949640287,
+      "grad_norm": 1.2633750438690186,
+      "learning_rate": 7.733812949640288e-05,
+      "loss": 0.0899,
+      "step": 430
+    },
+    {
+      "epoch": 0.7913669064748201,
+      "grad_norm": 0.8877132534980774,
+      "learning_rate": 7.913669064748202e-05,
+      "loss": 0.0828,
+      "step": 440
+    },
+    {
+      "epoch": 0.8093525179856115,
+      "grad_norm": 0.6728298664093018,
+      "learning_rate": 8.093525179856115e-05,
+      "loss": 0.0765,
+      "step": 450
+    },
+    {
+      "epoch": 0.8273381294964028,
+      "grad_norm": 0.8592141270637512,
+      "learning_rate": 8.273381294964029e-05,
+      "loss": 0.0788,
+      "step": 460
+    },
+    {
+      "epoch": 0.8453237410071942,
+      "grad_norm": 0.7244961261749268,
+      "learning_rate": 8.453237410071942e-05,
+      "loss": 0.0894,
+      "step": 470
+    },
+    {
+      "epoch": 0.8633093525179856,
+      "grad_norm": 1.1089451313018799,
+      "learning_rate": 8.633093525179857e-05,
+      "loss": 0.0886,
+      "step": 480
+    },
+    {
+      "epoch": 0.8812949640287769,
+      "grad_norm": 1.6618589162826538,
+      "learning_rate": 8.81294964028777e-05,
+      "loss": 0.0799,
+      "step": 490
+    },
+    {
+      "epoch": 0.8992805755395683,
+      "grad_norm": 0.9292100667953491,
+      "learning_rate": 8.992805755395684e-05,
+      "loss": 0.0748,
+      "step": 500
+    },
+    {
+      "epoch": 0.9172661870503597,
+      "grad_norm": 3.7133781909942627,
+      "learning_rate": 9.172661870503597e-05,
+      "loss": 0.0903,
+      "step": 510
+    },
+    {
+      "epoch": 0.935251798561151,
+      "grad_norm": 0.5967691540718079,
+      "learning_rate": 9.35251798561151e-05,
+      "loss": 0.0716,
+      "step": 520
+    },
+    {
+      "epoch": 0.9532374100719424,
+      "grad_norm": 1.0344717502593994,
+      "learning_rate": 9.532374100719424e-05,
+      "loss": 0.0791,
+      "step": 530
+    },
+    {
+      "epoch": 0.9712230215827338,
+      "grad_norm": 0.4440005123615265,
+      "learning_rate": 9.712230215827339e-05,
+      "loss": 0.0713,
+      "step": 540
+    },
+    {
+      "epoch": 0.9892086330935251,
+      "grad_norm": 0.6721171736717224,
+      "learning_rate": 9.892086330935252e-05,
+      "loss": 0.0703,
+      "step": 550
+    },
+    {
+      "epoch": 1.0071942446043165,
+      "grad_norm": 0.9711641669273376,
+      "learning_rate": 9.999996462447447e-05,
+      "loss": 0.0679,
+      "step": 560
+    },
+    {
+      "epoch": 1.025179856115108,
+      "grad_norm": 1.1041004657745361,
+      "learning_rate": 9.999956665038706e-05,
+      "loss": 0.0736,
+      "step": 570
+    },
+    {
+      "epoch": 1.0431654676258992,
+      "grad_norm": 1.3187923431396484,
+      "learning_rate": 9.999872648633667e-05,
+      "loss": 0.0775,
+      "step": 580
+    },
+    {
+      "epoch": 1.0611510791366907,
+      "grad_norm": 1.1964014768600464,
+      "learning_rate": 9.999744413975362e-05,
+      "loss": 0.0768,
+      "step": 590
+    },
+    {
+      "epoch": 1.079136690647482,
+      "grad_norm": 0.6774939894676208,
+      "learning_rate": 9.999571962197882e-05,
+      "loss": 0.0744,
+      "step": 600
+    },
+    {
+      "epoch": 1.0971223021582734,
+      "grad_norm": 1.4241557121276855,
+      "learning_rate": 9.999355294826369e-05,
+      "loss": 0.0809,
+      "step": 610
+    },
+    {
+      "epoch": 1.1151079136690647,
+      "grad_norm": 0.6569636464118958,
+      "learning_rate": 9.999094413777002e-05,
+      "loss": 0.0679,
+      "step": 620
+    },
+    {
+      "epoch": 1.1330935251798562,
+      "grad_norm": 0.7455745935440063,
+      "learning_rate": 9.998789321356983e-05,
+      "loss": 0.0696,
+      "step": 630
+    },
+    {
+      "epoch": 1.1510791366906474,
+      "grad_norm": 0.6449496150016785,
+      "learning_rate": 9.998440020264508e-05,
+      "loss": 0.0665,
+      "step": 640
+    },
+    {
+      "epoch": 1.169064748201439,
+      "grad_norm": 0.9546118378639221,
+      "learning_rate": 9.998046513588753e-05,
+      "loss": 0.0674,
+      "step": 650
+    },
+    {
+      "epoch": 1.1870503597122302,
+      "grad_norm": 0.6732275485992432,
+      "learning_rate": 9.997608804809845e-05,
+      "loss": 0.067,
+      "step": 660
+    },
+    {
+      "epoch": 1.2050359712230216,
+      "grad_norm": 0.8681954741477966,
+      "learning_rate": 9.997126897798825e-05,
+      "loss": 0.0686,
+      "step": 670
+    },
+    {
+      "epoch": 1.223021582733813,
+      "grad_norm": 0.9863311648368835,
+      "learning_rate": 9.996600796817617e-05,
+      "loss": 0.0724,
+      "step": 680
+    },
+    {
+      "epoch": 1.2410071942446044,
+      "grad_norm": 0.8919282555580139,
+      "learning_rate": 9.996030506518995e-05,
+      "loss": 0.0596,
+      "step": 690
+    },
+    {
+      "epoch": 1.2589928057553956,
+      "grad_norm": 0.5879567265510559,
+      "learning_rate": 9.995416031946534e-05,
+      "loss": 0.0683,
+      "step": 700
+    },
+    {
+      "epoch": 1.276978417266187,
+      "grad_norm": 0.8846353888511658,
+      "learning_rate": 9.994757378534572e-05,
+      "loss": 0.0773,
+      "step": 710
+    },
+    {
+      "epoch": 1.2949640287769784,
+      "grad_norm": 1.2975221872329712,
+      "learning_rate": 9.994054552108156e-05,
+      "loss": 0.0655,
+      "step": 720
+    },
+    {
+      "epoch": 1.3129496402877698,
+      "grad_norm": 1.3209266662597656,
+      "learning_rate": 9.993307558882999e-05,
+      "loss": 0.0722,
+      "step": 730
+    },
+    {
+      "epoch": 1.330935251798561,
+      "grad_norm": 0.6153837442398071,
+      "learning_rate": 9.992516405465414e-05,
+      "loss": 0.061,
+      "step": 740
+    },
+    {
+      "epoch": 1.3489208633093526,
+      "grad_norm": 0.4629378020763397,
+      "learning_rate": 9.991681098852264e-05,
+      "loss": 0.0675,
+      "step": 750
+    },
+    {
+      "epoch": 1.3669064748201438,
+      "grad_norm": 0.8385881781578064,
+      "learning_rate": 9.990801646430898e-05,
+      "loss": 0.0531,
+      "step": 760
+    },
+    {
+      "epoch": 1.3848920863309353,
+      "grad_norm": 0.5813649892807007,
+      "learning_rate": 9.989878055979085e-05,
+      "loss": 0.0596,
+      "step": 770
+    },
+    {
+      "epoch": 1.4028776978417266,
+      "grad_norm": 0.5986373424530029,
+      "learning_rate": 9.98891033566494e-05,
+      "loss": 0.0633,
+      "step": 780
+    },
+    {
+      "epoch": 1.420863309352518,
+      "grad_norm": 0.9480665922164917,
+      "learning_rate": 9.987898494046866e-05,
+      "loss": 0.0559,
+      "step": 790
+    },
+    {
+      "epoch": 1.4388489208633093,
+      "grad_norm": 0.600028932094574,
+      "learning_rate": 9.986842540073462e-05,
+      "loss": 0.0656,
+      "step": 800
+    },
+    {
+      "epoch": 1.4568345323741008,
+      "grad_norm": 0.6275537014007568,
+      "learning_rate": 9.985742483083455e-05,
+      "loss": 0.0582,
+      "step": 810
+    },
+    {
+      "epoch": 1.474820143884892,
+      "grad_norm": 0.8818138837814331,
+      "learning_rate": 9.984598332805612e-05,
+      "loss": 0.0622,
+      "step": 820
+    },
+    {
+      "epoch": 1.4928057553956835,
+      "grad_norm": 1.0676714181900024,
+      "learning_rate": 9.983410099358656e-05,
+      "loss": 0.0627,
+      "step": 830
+    },
+    {
+      "epoch": 1.5107913669064748,
+      "grad_norm": 0.6063862442970276,
+      "learning_rate": 9.982177793251178e-05,
+      "loss": 0.0579,
+      "step": 840
+    },
+    {
+      "epoch": 1.5287769784172662,
+      "grad_norm": 0.8405861258506775,
+      "learning_rate": 9.980901425381538e-05,
+      "loss": 0.0696,
+      "step": 850
+    },
+    {
+      "epoch": 1.5467625899280577,
+      "grad_norm": 1.1120002269744873,
+      "learning_rate": 9.979581007037776e-05,
+      "loss": 0.0629,
+      "step": 860
+    },
+    {
+      "epoch": 1.564748201438849,
+      "grad_norm": 0.622896671295166,
+      "learning_rate": 9.978216549897508e-05,
+      "loss": 0.0551,
+      "step": 870
+    },
+    {
+      "epoch": 1.5827338129496402,
+      "grad_norm": 0.7314297556877136,
+      "learning_rate": 9.976808066027823e-05,
+      "loss": 0.0535,
+      "step": 880
+    },
+    {
+      "epoch": 1.6007194244604317,
+      "grad_norm": 0.8343653082847595,
+      "learning_rate": 9.975355567885179e-05,
+      "loss": 0.0536,
+      "step": 890
+    },
+    {
+      "epoch": 1.6187050359712232,
+      "grad_norm": 0.8823739886283875,
+      "learning_rate": 9.973859068315288e-05,
+      "loss": 0.0571,
+      "step": 900
+    },
+    {
+      "epoch": 1.6366906474820144,
+      "grad_norm": 0.78243488073349,
+      "learning_rate": 9.972318580553007e-05,
+      "loss": 0.0571,
+      "step": 910
+    },
+    {
+      "epoch": 1.6546762589928057,
+      "grad_norm": 0.9111320972442627,
+      "learning_rate": 9.970734118222216e-05,
+      "loss": 0.0588,
+      "step": 920
+    },
+    {
+      "epoch": 1.6726618705035972,
+      "grad_norm": 0.6662073135375977,
+      "learning_rate": 9.969105695335707e-05,
+      "loss": 0.0565,
+      "step": 930
+    },
+    {
+      "epoch": 1.6906474820143886,
+      "grad_norm": 1.136330246925354,
+      "learning_rate": 9.96743332629505e-05,
+      "loss": 0.051,
+      "step": 940
+    },
+    {
+      "epoch": 1.70863309352518,
+      "grad_norm": 0.7974278926849365,
+      "learning_rate": 9.965717025890468e-05,
+      "loss": 0.0637,
+      "step": 950
+    },
+    {
+      "epoch": 1.7266187050359711,
+      "grad_norm": 0.8473669290542603,
+      "learning_rate": 9.963956809300708e-05,
+      "loss": 0.052,
+      "step": 960
+    },
+    {
+      "epoch": 1.7446043165467626,
+      "grad_norm": 0.6436973214149475,
+      "learning_rate": 9.962152692092908e-05,
+      "loss": 0.0491,
+      "step": 970
+    },
+    {
+      "epoch": 1.762589928057554,
+      "grad_norm": 0.960101306438446,
+      "learning_rate": 9.960304690222458e-05,
+      "loss": 0.0531,
+      "step": 980
+    },
+    {
+      "epoch": 1.7805755395683454,
+      "grad_norm": 0.6839088201522827,
+      "learning_rate": 9.958412820032853e-05,
+      "loss": 0.0528,
+      "step": 990
+    },
+    {
+      "epoch": 1.7985611510791366,
+      "grad_norm": 0.5344822406768799,
+      "learning_rate": 9.956477098255563e-05,
+      "loss": 0.0536,
+      "step": 1000
+    },
+    {
+      "epoch": 1.816546762589928,
+      "grad_norm": 0.8056336641311646,
+      "learning_rate": 9.954497542009868e-05,
+      "loss": 0.0526,
+      "step": 1010
+    },
+    {
+      "epoch": 1.8345323741007196,
+      "grad_norm": 0.6944713592529297,
+      "learning_rate": 9.952474168802719e-05,
+      "loss": 0.0533,
+      "step": 1020
+    },
+    {
+      "epoch": 1.8525179856115108,
+      "grad_norm": 1.0095305442810059,
+      "learning_rate": 9.950406996528579e-05,
+      "loss": 0.0504,
+      "step": 1030
+    },
+    {
+      "epoch": 1.870503597122302,
+      "grad_norm": 0.7429274320602417,
+      "learning_rate": 9.948296043469259e-05,
+      "loss": 0.0545,
+      "step": 1040
+    },
+    {
+      "epoch": 1.8884892086330936,
+      "grad_norm": 0.8363075256347656,
+      "learning_rate": 9.94614132829377e-05,
+      "loss": 0.0543,
+      "step": 1050
+    },
+    {
+      "epoch": 1.906474820143885,
+      "grad_norm": 0.9278216361999512,
+      "learning_rate": 9.943942870058144e-05,
+      "loss": 0.0538,
+      "step": 1060
+    },
+    {
+      "epoch": 1.9244604316546763,
+      "grad_norm": 0.6224967837333679,
+      "learning_rate": 9.941700688205274e-05,
+      "loss": 0.0541,
+      "step": 1070
+    },
+    {
+      "epoch": 1.9424460431654675,
+      "grad_norm": 0.6423580050468445,
+      "learning_rate": 9.939414802564735e-05,
+      "loss": 0.0568,
+      "step": 1080
+    },
+    {
+      "epoch": 1.960431654676259,
+      "grad_norm": 1.1750601530075073,
+      "learning_rate": 9.937085233352619e-05,
+      "loss": 0.0476,
+      "step": 1090
+    },
+    {
+      "epoch": 1.9784172661870505,
+      "grad_norm": 0.6462400555610657,
+      "learning_rate": 9.934712001171345e-05,
+      "loss": 0.0467,
+      "step": 1100
+    },
+    {
+      "epoch": 1.9964028776978417,
+      "grad_norm": 0.5434532761573792,
+      "learning_rate": 9.932295127009485e-05,
+      "loss": 0.0519,
+      "step": 1110
+    },
+    {
+      "epoch": 2.014388489208633,
+      "grad_norm": 0.925330638885498,
+      "learning_rate": 9.929834632241577e-05,
+      "loss": 0.06,
+      "step": 1120
+    },
+    {
+      "epoch": 2.0323741007194243,
+      "grad_norm": 0.6649676561355591,
+      "learning_rate": 9.927330538627926e-05,
+      "loss": 0.0565,
+      "step": 1130
+    },
+    {
+      "epoch": 2.050359712230216,
+      "grad_norm": 0.7511556148529053,
+      "learning_rate": 9.924782868314429e-05,
+      "loss": 0.0585,
+      "step": 1140
+    },
+    {
+      "epoch": 2.068345323741007,
+      "grad_norm": 1.022379755973816,
+      "learning_rate": 9.922191643832363e-05,
+      "loss": 0.0576,
+      "step": 1150
+    },
+    {
+      "epoch": 2.0863309352517985,
+      "grad_norm": 0.6581182479858398,
+      "learning_rate": 9.919556888098198e-05,
+      "loss": 0.0585,
+      "step": 1160
+    },
+    {
+      "epoch": 2.1043165467625897,
+      "grad_norm": 0.70713871717453,
+      "learning_rate": 9.916878624413386e-05,
+      "loss": 0.055,
+      "step": 1170
+    },
+    {
+      "epoch": 2.1223021582733814,
+      "grad_norm": 1.1394716501235962,
+      "learning_rate": 9.91415687646416e-05,
+      "loss": 0.055,
+      "step": 1180
+    },
+    {
+      "epoch": 2.1402877697841727,
+      "grad_norm": 0.8932343125343323,
+      "learning_rate": 9.911391668321321e-05,
+      "loss": 0.0574,
+      "step": 1190
+    },
+    {
+      "epoch": 2.158273381294964,
+      "grad_norm": 0.7466623187065125,
+      "learning_rate": 9.908583024440024e-05,
+      "loss": 0.0569,
+      "step": 1200
+    },
+    {
+      "epoch": 2.176258992805755,
+      "grad_norm": 0.7184358239173889,
+      "learning_rate": 9.905730969659569e-05,
+      "loss": 0.0486,
+      "step": 1210
+    },
+    {
+      "epoch": 2.194244604316547,
+      "grad_norm": 0.8060582280158997,
+      "learning_rate": 9.902835529203174e-05,
+      "loss": 0.0524,
+      "step": 1220
+    },
+    {
+      "epoch": 2.212230215827338,
+      "grad_norm": 0.6587594747543335,
+      "learning_rate": 9.899896728677757e-05,
+      "loss": 0.0569,
+      "step": 1230
+    },
+    {
+      "epoch": 2.2302158273381294,
+      "grad_norm": 0.6646767258644104,
+      "learning_rate": 9.896914594073703e-05,
+      "loss": 0.0447,
+      "step": 1240
+    },
+    {
+      "epoch": 2.2482014388489207,
+      "grad_norm": 0.45537829399108887,
+      "learning_rate": 9.89388915176464e-05,
+      "loss": 0.0506,
+      "step": 1250
+    },
+    {
+      "epoch": 2.2661870503597124,
+      "grad_norm": 0.6282981038093567,
+      "learning_rate": 9.890820428507206e-05,
+      "loss": 0.0482,
+      "step": 1260
+    },
+    {
+      "epoch": 2.2841726618705036,
+      "grad_norm": 0.7369111180305481,
+      "learning_rate": 9.88770845144081e-05,
+      "loss": 0.0506,
+      "step": 1270
+    },
+    {
+      "epoch": 2.302158273381295,
+      "grad_norm": 0.6427294611930847,
+      "learning_rate": 9.884553248087385e-05,
+      "loss": 0.0521,
+      "step": 1280
+    },
+    {
+      "epoch": 2.3201438848920866,
+      "grad_norm": 0.8718444108963013,
+      "learning_rate": 9.881354846351166e-05,
+      "loss": 0.0568,
+      "step": 1290
+    },
+    {
+      "epoch": 2.338129496402878,
+      "grad_norm": 0.4948193430900574,
+      "learning_rate": 9.878113274518415e-05,
+      "loss": 0.0534,
+      "step": 1300
+    },
+    {
+      "epoch": 2.356115107913669,
+      "grad_norm": 0.7721247673034668,
+      "learning_rate": 9.874828561257192e-05,
+      "loss": 0.0532,
+      "step": 1310
+    },
+    {
+      "epoch": 2.3741007194244603,
+      "grad_norm": 0.5891484618186951,
+      "learning_rate": 9.871500735617096e-05,
+      "loss": 0.0502,
+      "step": 1320
+    },
+    {
+      "epoch": 2.3920863309352516,
+      "grad_norm": 0.581562876701355,
+      "learning_rate": 9.868129827029002e-05,
+      "loss": 0.0409,
+      "step": 1330
+    },
+    {
+      "epoch": 2.4100719424460433,
+      "grad_norm": 0.5472557544708252,
+      "learning_rate": 9.864715865304808e-05,
+      "loss": 0.0493,
+      "step": 1340
+    },
+    {
+      "epoch": 2.4280575539568345,
+      "grad_norm": 0.5667267441749573,
+      "learning_rate": 9.861258880637167e-05,
+      "loss": 0.0438,
+      "step": 1350
+    },
+    {
+      "epoch": 2.446043165467626,
+      "grad_norm": 0.8043990731239319,
+      "learning_rate": 9.857758903599224e-05,
+      "loss": 0.0489,
+      "step": 1360
+    },
+    {
+      "epoch": 2.4640287769784175,
+      "grad_norm": 0.6041139364242554,
+      "learning_rate": 9.85421596514434e-05,
+      "loss": 0.0404,
+      "step": 1370
+    },
+    {
+      "epoch": 2.4820143884892087,
+      "grad_norm": 0.5199022889137268,
+      "learning_rate": 9.850630096605824e-05,
+      "loss": 0.0478,
+      "step": 1380
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.5527715086936951,
+      "learning_rate": 9.847001329696653e-05,
+      "loss": 0.043,
+      "step": 1390
+    },
+    {
+      "epoch": 2.5179856115107913,
+      "grad_norm": 0.6931111216545105,
+      "learning_rate": 9.843329696509189e-05,
+      "loss": 0.047,
+      "step": 1400
+    },
+    {
+      "epoch": 2.5359712230215825,
+      "grad_norm": 0.5406497716903687,
+      "learning_rate": 9.839615229514902e-05,
+      "loss": 0.0451,
+      "step": 1410
+    },
+    {
+      "epoch": 2.553956834532374,
+      "grad_norm": 0.9430763125419617,
+      "learning_rate": 9.835857961564077e-05,
+      "loss": 0.0542,
+      "step": 1420
+    },
+    {
+      "epoch": 2.5719424460431655,
+      "grad_norm": 0.8105902075767517,
+      "learning_rate": 9.832057925885525e-05,
+      "loss": 0.0503,
+      "step": 1430
+    },
+    {
+      "epoch": 2.5899280575539567,
+      "grad_norm": 1.1112093925476074,
+      "learning_rate": 9.828215156086292e-05,
+      "loss": 0.0537,
+      "step": 1440
+    },
+    {
+      "epoch": 2.6079136690647484,
+      "grad_norm": 0.8996851444244385,
+      "learning_rate": 9.824329686151354e-05,
+      "loss": 0.0512,
+      "step": 1450
+    },
+    {
+      "epoch": 2.6258992805755397,
+      "grad_norm": 0.629753589630127,
+      "learning_rate": 9.820401550443325e-05,
+      "loss": 0.0497,
+      "step": 1460
+    },
+    {
+      "epoch": 2.643884892086331,
+      "grad_norm": 0.6719884872436523,
+      "learning_rate": 9.816430783702154e-05,
+      "loss": 0.0533,
+      "step": 1470
+    },
+    {
+      "epoch": 2.661870503597122,
+      "grad_norm": 1.2444987297058105,
+      "learning_rate": 9.812417421044807e-05,
+      "loss": 0.0571,
+      "step": 1480
+    },
+    {
+      "epoch": 2.6798561151079134,
+      "grad_norm": 0.6962327361106873,
+      "learning_rate": 9.808361497964962e-05,
+      "loss": 0.0462,
+      "step": 1490
+    },
+    {
+      "epoch": 2.697841726618705,
+      "grad_norm": 1.1346492767333984,
+      "learning_rate": 9.804263050332703e-05,
+      "loss": 0.0554,
+      "step": 1500
+    },
+    {
+      "epoch": 2.7158273381294964,
+      "grad_norm": 0.7784806489944458,
+      "learning_rate": 9.800122114394193e-05,
+      "loss": 0.0539,
+      "step": 1510
+    },
+    {
+      "epoch": 2.7338129496402876,
+      "grad_norm": 0.7351565361022949,
+      "learning_rate": 9.795938726771355e-05,
+      "loss": 0.0475,
+      "step": 1520
+    },
+    {
+      "epoch": 2.7517985611510793,
+      "grad_norm": 0.6214766502380371,
+      "learning_rate": 9.791712924461546e-05,
+      "loss": 0.051,
+      "step": 1530
+    },
+    {
+      "epoch": 2.7697841726618706,
+      "grad_norm": 0.7485048174858093,
+      "learning_rate": 9.787444744837243e-05,
+      "loss": 0.0513,
+      "step": 1540
+    },
+    {
+      "epoch": 2.787769784172662,
+      "grad_norm": 0.6118731498718262,
+      "learning_rate": 9.783134225645693e-05,
+      "loss": 0.053,
+      "step": 1550
+    },
+    {
+      "epoch": 2.805755395683453,
+      "grad_norm": 0.8059982657432556,
+      "learning_rate": 9.778781405008596e-05,
+      "loss": 0.0485,
+      "step": 1560
+    },
+    {
+      "epoch": 2.8237410071942444,
+      "grad_norm": 0.4733278453350067,
+      "learning_rate": 9.774386321421754e-05,
+      "loss": 0.0456,
+      "step": 1570
+    },
+    {
+      "epoch": 2.841726618705036,
+      "grad_norm": 0.9504257440567017,
+      "learning_rate": 9.769949013754744e-05,
+      "loss": 0.0543,
+      "step": 1580
+    },
+    {
+      "epoch": 2.8597122302158273,
+      "grad_norm": 0.5240606069564819,
+      "learning_rate": 9.765469521250563e-05,
+      "loss": 0.0473,
+      "step": 1590
+    },
+    {
+      "epoch": 2.8776978417266186,
+      "grad_norm": 0.794711172580719,
+      "learning_rate": 9.760947883525287e-05,
+      "loss": 0.0459,
+      "step": 1600
+    },
+    {
+      "epoch": 2.8956834532374103,
+      "grad_norm": 0.8448195457458496,
+      "learning_rate": 9.756384140567718e-05,
+      "loss": 0.0444,
+      "step": 1610
+    },
+    {
+      "epoch": 2.9136690647482015,
+      "grad_norm": 0.8142967224121094,
+      "learning_rate": 9.751778332739033e-05,
+      "loss": 0.0437,
+      "step": 1620
+    },
+    {
+      "epoch": 2.931654676258993,
+      "grad_norm": 0.5138818621635437,
+      "learning_rate": 9.747130500772425e-05,
+      "loss": 0.0464,
+      "step": 1630
+    },
+    {
+      "epoch": 2.949640287769784,
+      "grad_norm": 0.8041589260101318,
+      "learning_rate": 9.742440685772745e-05,
+      "loss": 0.0451,
+      "step": 1640
+    },
+    {
+      "epoch": 2.9676258992805753,
+      "grad_norm": 0.9745432734489441,
+      "learning_rate": 9.737708929216131e-05,
+      "loss": 0.048,
+      "step": 1650
+    },
+    {
+      "epoch": 2.985611510791367,
+      "grad_norm": 0.41117537021636963,
+      "learning_rate": 9.732935272949655e-05,
+      "loss": 0.0442,
+      "step": 1660
+    },
+    {
+      "epoch": 3.0035971223021583,
+      "grad_norm": 0.41132697463035583,
+      "learning_rate": 9.728119759190939e-05,
+      "loss": 0.0412,
+      "step": 1670
+    },
+    {
+      "epoch": 3.0215827338129495,
+      "grad_norm": 0.5575608015060425,
+      "learning_rate": 9.723262430527786e-05,
+      "loss": 0.0496,
+      "step": 1680
+    },
+    {
+      "epoch": 3.039568345323741,
+      "grad_norm": 0.8312233686447144,
+      "learning_rate": 9.718363329917813e-05,
+      "loss": 0.0491,
+      "step": 1690
+    },
+    {
+      "epoch": 3.0575539568345325,
+      "grad_norm": 0.599704384803772,
+      "learning_rate": 9.713422500688057e-05,
+      "loss": 0.0415,
+      "step": 1700
+    },
+    {
+      "epoch": 3.0755395683453237,
+      "grad_norm": 0.6746256947517395,
+      "learning_rate": 9.708439986534595e-05,
+      "loss": 0.0499,
+      "step": 1710
+    },
+    {
+      "epoch": 3.093525179856115,
+      "grad_norm": 0.7200654745101929,
+      "learning_rate": 9.703415831522164e-05,
+      "loss": 0.0441,
+      "step": 1720
+    },
+    {
+      "epoch": 3.1115107913669067,
+      "grad_norm": 0.8032772541046143,
+      "learning_rate": 9.698350080083773e-05,
+      "loss": 0.0444,
+      "step": 1730
+    },
+    {
+      "epoch": 3.129496402877698,
+      "grad_norm": 0.39724618196487427,
+      "learning_rate": 9.693242777020295e-05,
+      "loss": 0.0398,
+      "step": 1740
+    },
+    {
+      "epoch": 3.147482014388489,
+      "grad_norm": 0.5301024317741394,
+      "learning_rate": 9.688093967500085e-05,
+      "loss": 0.0494,
+      "step": 1750
+    },
+    {
+      "epoch": 3.1654676258992804,
+      "grad_norm": 0.41485583782196045,
+      "learning_rate": 9.682903697058575e-05,
+      "loss": 0.0475,
+      "step": 1760
+    },
+    {
+      "epoch": 3.183453237410072,
+      "grad_norm": 0.5792023539543152,
+      "learning_rate": 9.677672011597874e-05,
+      "loss": 0.0432,
+      "step": 1770
+    },
+    {
+      "epoch": 3.2014388489208634,
+      "grad_norm": 0.5463465452194214,
+      "learning_rate": 9.672398957386358e-05,
+      "loss": 0.0425,
+      "step": 1780
+    },
+    {
+      "epoch": 3.2194244604316546,
+      "grad_norm": 0.8164083361625671,
+      "learning_rate": 9.667084581058266e-05,
+      "loss": 0.0479,
+      "step": 1790
+    },
+    {
+      "epoch": 3.237410071942446,
+      "grad_norm": 0.5091384053230286,
+      "learning_rate": 9.661728929613279e-05,
+      "loss": 0.0438,
+      "step": 1800
+    },
+    {
+      "epoch": 3.2553956834532376,
+      "grad_norm": 0.5957716703414917,
+      "learning_rate": 9.656332050416116e-05,
+      "loss": 0.0497,
+      "step": 1810
+    },
+    {
+      "epoch": 3.273381294964029,
+      "grad_norm": 0.4585723578929901,
+      "learning_rate": 9.650893991196109e-05,
+      "loss": 0.0395,
+      "step": 1820
+    },
+    {
+      "epoch": 3.29136690647482,
+      "grad_norm": 0.6154191493988037,
+      "learning_rate": 9.645414800046777e-05,
+      "loss": 0.0401,
+      "step": 1830
+    },
+    {
+      "epoch": 3.3093525179856114,
+      "grad_norm": 0.468321830034256,
+      "learning_rate": 9.639894525425406e-05,
+      "loss": 0.0411,
+      "step": 1840
+    },
+    {
+      "epoch": 3.327338129496403,
+      "grad_norm": 0.43552345037460327,
+      "learning_rate": 9.634333216152619e-05,
+      "loss": 0.038,
+      "step": 1850
+    },
+    {
+      "epoch": 3.3453237410071943,
+      "grad_norm": 0.665313184261322,
+      "learning_rate": 9.628730921411949e-05,
+      "loss": 0.0471,
+      "step": 1860
+    },
+    {
+      "epoch": 3.3633093525179856,
+      "grad_norm": 0.7994027733802795,
+      "learning_rate": 9.623087690749389e-05,
+      "loss": 0.0438,
+      "step": 1870
+    },
+    {
+      "epoch": 3.381294964028777,
+      "grad_norm": 0.6105477213859558,
+      "learning_rate": 9.617403574072977e-05,
+      "loss": 0.039,
+      "step": 1880
+    },
+    {
+      "epoch": 3.3992805755395685,
+      "grad_norm": 0.5954127311706543,
+      "learning_rate": 9.611678621652334e-05,
+      "loss": 0.0425,
+      "step": 1890
+    },
+    {
+      "epoch": 3.41726618705036,
+      "grad_norm": 0.4164612293243408,
+      "learning_rate": 9.605912884118227e-05,
+      "loss": 0.0524,
+      "step": 1900
+    },
+    {
+      "epoch": 3.435251798561151,
+      "grad_norm": 0.6321906447410583,
+      "learning_rate": 9.600106412462123e-05,
+      "loss": 0.0505,
+      "step": 1910
+    },
+    {
+      "epoch": 3.4532374100719423,
+      "grad_norm": 0.846747875213623,
+      "learning_rate": 9.594259258035738e-05,
+      "loss": 0.0442,
+      "step": 1920
+    },
+    {
+      "epoch": 3.471223021582734,
+      "grad_norm": 0.4675831198692322,
+      "learning_rate": 9.588371472550582e-05,
+      "loss": 0.0486,
+      "step": 1930
+    },
+    {
+      "epoch": 3.4892086330935252,
+      "grad_norm": 0.7421311736106873,
+      "learning_rate": 9.582443108077497e-05,
+      "loss": 0.0414,
+      "step": 1940
+    },
+    {
+      "epoch": 3.5071942446043165,
+      "grad_norm": 0.6084485054016113,
+      "learning_rate": 9.576474217046204e-05,
+      "loss": 0.0417,
+      "step": 1950
+    },
+    {
+      "epoch": 3.5251798561151078,
+      "grad_norm": 0.5336371660232544,
+      "learning_rate": 9.570464852244831e-05,
+      "loss": 0.0396,
+      "step": 1960
+    },
+    {
+      "epoch": 3.543165467625899,
+      "grad_norm": 0.43448108434677124,
+      "learning_rate": 9.564415066819459e-05,
+      "loss": 0.0464,
+      "step": 1970
+    },
+    {
+      "epoch": 3.5611510791366907,
+      "grad_norm": 0.7890388369560242,
+      "learning_rate": 9.558324914273637e-05,
+      "loss": 0.0496,
+      "step": 1980
+    },
+    {
+      "epoch": 3.579136690647482,
+      "grad_norm": 0.6465239524841309,
+      "learning_rate": 9.55219444846792e-05,
+      "loss": 0.0411,
+      "step": 1990
+    },
+    {
+      "epoch": 3.597122302158273,
+      "grad_norm": 1.1230480670928955,
+      "learning_rate": 9.546023723619386e-05,
+      "loss": 0.0394,
+      "step": 2000
+    },
+    {
+      "epoch": 3.615107913669065,
+      "grad_norm": 0.6308082938194275,
+      "learning_rate": 9.53981279430116e-05,
+      "loss": 0.042,
+      "step": 2010
+    },
+    {
+      "epoch": 3.633093525179856,
+      "grad_norm": 0.6348657011985779,
+      "learning_rate": 9.533561715441929e-05,
+      "loss": 0.0453,
+      "step": 2020
+    },
+    {
+      "epoch": 3.6510791366906474,
+      "grad_norm": 0.6183689832687378,
+      "learning_rate": 9.527270542325462e-05,
+      "loss": 0.0511,
+      "step": 2030
+    },
+    {
+      "epoch": 3.6690647482014387,
+      "grad_norm": 0.4541698694229126,
+      "learning_rate": 9.520939330590109e-05,
+      "loss": 0.0485,
+      "step": 2040
+    },
+    {
+      "epoch": 3.68705035971223,
+      "grad_norm": 0.5956753492355347,
+      "learning_rate": 9.514568136228324e-05,
+      "loss": 0.0443,
+      "step": 2050
+    },
+    {
+      "epoch": 3.7050359712230216,
+      "grad_norm": 0.4104117453098297,
+      "learning_rate": 9.508157015586159e-05,
+      "loss": 0.0423,
+      "step": 2060
+    },
+    {
+      "epoch": 3.723021582733813,
+      "grad_norm": 0.8681178092956543,
+      "learning_rate": 9.501706025362769e-05,
+      "loss": 0.0436,
+      "step": 2070
+    },
+    {
+      "epoch": 3.741007194244604,
+      "grad_norm": 0.37762942910194397,
+      "learning_rate": 9.49521522260991e-05,
+      "loss": 0.0407,
+      "step": 2080
+    },
+    {
+      "epoch": 3.758992805755396,
+      "grad_norm": 0.8141522407531738,
+      "learning_rate": 9.488684664731438e-05,
+      "loss": 0.0373,
+      "step": 2090
+    },
+    {
+      "epoch": 3.776978417266187,
+      "grad_norm": 0.46811094880104065,
+      "learning_rate": 9.482114409482794e-05,
+      "loss": 0.0361,
+      "step": 2100
+    },
+    {
+      "epoch": 3.7949640287769784,
+      "grad_norm": 0.45261967182159424,
+      "learning_rate": 9.475504514970503e-05,
+      "loss": 0.0399,
+      "step": 2110
+    },
+    {
+      "epoch": 3.81294964028777,
+      "grad_norm": 0.5435159802436829,
+      "learning_rate": 9.46885503965165e-05,
+      "loss": 0.0409,
+      "step": 2120
+    },
+    {
+      "epoch": 3.8309352517985613,
+      "grad_norm": 0.9027155041694641,
+      "learning_rate": 9.46216604233337e-05,
+      "loss": 0.046,
+      "step": 2130
+    },
+    {
+      "epoch": 3.8489208633093526,
+      "grad_norm": 0.8086517453193665,
+      "learning_rate": 9.455437582172326e-05,
+      "loss": 0.0457,
+      "step": 2140
+    },
+    {
+      "epoch": 3.866906474820144,
+      "grad_norm": 0.5380069613456726,
+      "learning_rate": 9.448669718674184e-05,
+      "loss": 0.0549,
+      "step": 2150
+    },
+    {
+      "epoch": 3.884892086330935,
+      "grad_norm": 0.7407916188240051,
+      "learning_rate": 9.441862511693086e-05,
+      "loss": 0.0406,
+      "step": 2160
+    },
+    {
+      "epoch": 3.902877697841727,
+      "grad_norm": 0.4355371296405792,
+      "learning_rate": 9.435016021431133e-05,
+      "loss": 0.0414,
+      "step": 2170
+    },
+    {
+      "epoch": 3.920863309352518,
+      "grad_norm": 0.48018592596054077,
+      "learning_rate": 9.428130308437829e-05,
+      "loss": 0.0435,
+      "step": 2180
+    },
+    {
+      "epoch": 3.9388489208633093,
+      "grad_norm": 0.6430679559707642,
+      "learning_rate": 9.421205433609568e-05,
+      "loss": 0.0435,
+      "step": 2190
+    },
+    {
+      "epoch": 3.956834532374101,
+      "grad_norm": 0.600344717502594,
+      "learning_rate": 9.414241458189081e-05,
+      "loss": 0.0397,
+      "step": 2200
+    },
+    {
+      "epoch": 3.9748201438848922,
+      "grad_norm": 0.3487362861633301,
+      "learning_rate": 9.407238443764908e-05,
+      "loss": 0.0352,
+      "step": 2210
+    },
+    {
+      "epoch": 3.9928057553956835,
+      "grad_norm": 0.4792194366455078,
+      "learning_rate": 9.400196452270834e-05,
+      "loss": 0.0413,
+      "step": 2220
+    },
+    {
+      "epoch": 4.010791366906475,
+      "grad_norm": 0.9044307470321655,
+      "learning_rate": 9.393115545985359e-05,
+      "loss": 0.0393,
+      "step": 2230
+    },
+    {
+      "epoch": 4.028776978417266,
+      "grad_norm": 0.6789924502372742,
+      "learning_rate": 9.385995787531141e-05,
+      "loss": 0.042,
+      "step": 2240
+    },
+    {
+      "epoch": 4.046762589928058,
+      "grad_norm": 0.5473588705062866,
+      "learning_rate": 9.37883723987444e-05,
+      "loss": 0.0413,
+      "step": 2250
+    },
+    {
+      "epoch": 4.0647482014388485,
+      "grad_norm": 0.5175599455833435,
+      "learning_rate": 9.371639966324564e-05,
+      "loss": 0.036,
+      "step": 2260
+    },
+    {
+      "epoch": 4.08273381294964,
+      "grad_norm": 0.7959902286529541,
+      "learning_rate": 9.364404030533305e-05,
+      "loss": 0.0399,
+      "step": 2270
+    },
+    {
+      "epoch": 4.100719424460432,
+      "grad_norm": 0.5999887585639954,
+      "learning_rate": 9.357129496494382e-05,
+      "loss": 0.0423,
+      "step": 2280
+    },
+    {
+      "epoch": 4.118705035971223,
+      "grad_norm": 0.710017204284668,
+      "learning_rate": 9.349816428542872e-05,
+      "loss": 0.0458,
+      "step": 2290
+    },
+    {
+      "epoch": 4.136690647482014,
+      "grad_norm": 0.636737585067749,
+      "learning_rate": 9.34246489135464e-05,
+      "loss": 0.0372,
+      "step": 2300
+    },
+    {
+      "epoch": 4.154676258992806,
+      "grad_norm": 0.5894988179206848,
+      "learning_rate": 9.335074949945767e-05,
+      "loss": 0.0462,
+      "step": 2310
+    },
+    {
+      "epoch": 4.172661870503597,
+      "grad_norm": 0.7972150444984436,
+      "learning_rate": 9.327646669671984e-05,
+      "loss": 0.0328,
+      "step": 2320
+    },
+    {
+      "epoch": 4.190647482014389,
+      "grad_norm": 0.7291680574417114,
+      "learning_rate": 9.320180116228073e-05,
+      "loss": 0.0467,
+      "step": 2330
+    },
+    {
+      "epoch": 4.2086330935251794,
+      "grad_norm": 0.5354090332984924,
+      "learning_rate": 9.312675355647307e-05,
+      "loss": 0.0417,
+      "step": 2340
+    },
+    {
+      "epoch": 4.226618705035971,
+      "grad_norm": 0.5671231150627136,
+      "learning_rate": 9.30513245430086e-05,
+      "loss": 0.0432,
+      "step": 2350
+    },
+    {
+      "epoch": 4.244604316546763,
+      "grad_norm": 0.6749477982521057,
+      "learning_rate": 9.297551478897215e-05,
+      "loss": 0.0469,
+      "step": 2360
+    },
+    {
+      "epoch": 4.262589928057554,
+      "grad_norm": 0.8829865455627441,
+      "learning_rate": 9.289932496481576e-05,
+      "loss": 0.0408,
+      "step": 2370
+    },
+    {
+      "epoch": 4.280575539568345,
+      "grad_norm": 0.8449037075042725,
+      "learning_rate": 9.282275574435281e-05,
+      "loss": 0.037,
+      "step": 2380
+    },
+    {
+      "epoch": 4.298561151079137,
+      "grad_norm": 0.5600851774215698,
+      "learning_rate": 9.274580780475194e-05,
+      "loss": 0.0426,
+      "step": 2390
+    },
+    {
+      "epoch": 4.316546762589928,
+      "grad_norm": 0.5598462820053101,
+      "learning_rate": 9.266848182653124e-05,
+      "loss": 0.0428,
+      "step": 2400
+    },
+    {
+      "epoch": 4.33453237410072,
+      "grad_norm": 0.9153916835784912,
+      "learning_rate": 9.2590778493552e-05,
+      "loss": 0.0393,
+      "step": 2410
+    },
+    {
+      "epoch": 4.35251798561151,
+      "grad_norm": 0.48298609256744385,
+      "learning_rate": 9.251269849301291e-05,
+      "loss": 0.0472,
+      "step": 2420
+    },
+    {
+      "epoch": 4.370503597122302,
+      "grad_norm": 0.3913557231426239,
+      "learning_rate": 9.243424251544377e-05,
+      "loss": 0.0315,
+      "step": 2430
+    },
+    {
+      "epoch": 4.388489208633094,
+      "grad_norm": 0.5905225872993469,
+      "learning_rate": 9.235541125469952e-05,
+      "loss": 0.0382,
+      "step": 2440
+    },
+    {
+      "epoch": 4.406474820143885,
+      "grad_norm": 0.5525035858154297,
+      "learning_rate": 9.227620540795406e-05,
+      "loss": 0.038,
+      "step": 2450
+    },
+    {
+      "epoch": 4.424460431654676,
+      "grad_norm": 0.8970173001289368,
+      "learning_rate": 9.219662567569404e-05,
+      "loss": 0.0382,
+      "step": 2460
+    },
+    {
+      "epoch": 4.442446043165468,
+      "grad_norm": 0.8079715371131897,
+      "learning_rate": 9.211667276171278e-05,
+      "loss": 0.0409,
+      "step": 2470
+    },
+    {
+      "epoch": 4.460431654676259,
+      "grad_norm": 0.7845515608787537,
+      "learning_rate": 9.20363473731039e-05,
+      "loss": 0.0382,
+      "step": 2480
+    },
+    {
+      "epoch": 4.4784172661870505,
+      "grad_norm": 0.51947021484375,
+      "learning_rate": 9.195565022025516e-05,
+      "loss": 0.0371,
+      "step": 2490
+    },
+    {
+      "epoch": 4.496402877697841,
+      "grad_norm": 0.7945122122764587,
+      "learning_rate": 9.187458201684219e-05,
+      "loss": 0.0467,
+      "step": 2500
+    },
+    {
+      "epoch": 4.514388489208633,
+      "grad_norm": 0.7294409275054932,
+      "learning_rate": 9.17931434798221e-05,
+      "loss": 0.0412,
+      "step": 2510
+    },
+    {
+      "epoch": 4.532374100719425,
+      "grad_norm": 0.5720701813697815,
+      "learning_rate": 9.171133532942725e-05,
+      "loss": 0.0387,
+      "step": 2520
+    },
+    {
+      "epoch": 4.5503597122302155,
+      "grad_norm": 1.2283540964126587,
+      "learning_rate": 9.162915828915872e-05,
+      "loss": 0.0439,
+      "step": 2530
+    },
+    {
+      "epoch": 4.568345323741007,
+      "grad_norm": 0.6381161212921143,
+      "learning_rate": 9.154661308578008e-05,
+      "loss": 0.0432,
+      "step": 2540
+    },
+    {
+      "epoch": 4.586330935251799,
+      "grad_norm": 0.6119482517242432,
+      "learning_rate": 9.146370044931085e-05,
+      "loss": 0.046,
+      "step": 2550
+    },
+    {
+      "epoch": 4.60431654676259,
+      "grad_norm": 0.9606117606163025,
+      "learning_rate": 9.138042111302012e-05,
+      "loss": 0.0446,
+      "step": 2560
+    },
+    {
+      "epoch": 4.622302158273381,
+      "grad_norm": 0.576403796672821,
+      "learning_rate": 9.129677581342e-05,
+      "loss": 0.0441,
+      "step": 2570
+    },
+    {
+      "epoch": 4.640287769784173,
+      "grad_norm": 0.5738866925239563,
+      "learning_rate": 9.121276529025912e-05,
+      "loss": 0.0389,
+      "step": 2580
+    },
+    {
+      "epoch": 4.658273381294964,
+      "grad_norm": 0.606208324432373,
+      "learning_rate": 9.112839028651616e-05,
+      "loss": 0.0467,
+      "step": 2590
+    },
+    {
+      "epoch": 4.676258992805756,
+      "grad_norm": 0.6487197875976562,
+      "learning_rate": 9.104365154839313e-05,
+      "loss": 0.0416,
+      "step": 2600
+    },
+    {
+      "epoch": 4.694244604316546,
+      "grad_norm": 0.8250986337661743,
+      "learning_rate": 9.095854982530896e-05,
+      "loss": 0.0354,
+      "step": 2610
+    },
+    {
+      "epoch": 4.712230215827338,
+      "grad_norm": 0.6447311043739319,
+      "learning_rate": 9.087308586989272e-05,
+      "loss": 0.0465,
+      "step": 2620
+    },
+    {
+      "epoch": 4.73021582733813,
+      "grad_norm": 0.5524486303329468,
+      "learning_rate": 9.078726043797701e-05,
+      "loss": 0.0393,
+      "step": 2630
+    },
+    {
+      "epoch": 4.748201438848921,
+      "grad_norm": 0.5382457375526428,
+      "learning_rate": 9.070107428859131e-05,
+      "loss": 0.0369,
+      "step": 2640
+    },
+    {
+      "epoch": 4.766187050359712,
+      "grad_norm": 0.5520634651184082,
+      "learning_rate": 9.061452818395524e-05,
+      "loss": 0.0396,
+      "step": 2650
+    },
+    {
+      "epoch": 4.784172661870503,
+      "grad_norm": 0.7519980669021606,
+      "learning_rate": 9.052762288947178e-05,
+      "loss": 0.0489,
+      "step": 2660
+    },
+    {
+      "epoch": 4.802158273381295,
+      "grad_norm": 0.6335961222648621,
+      "learning_rate": 9.044035917372063e-05,
+      "loss": 0.0375,
+      "step": 2670
+    },
+    {
+      "epoch": 4.820143884892087,
+      "grad_norm": 0.6417689919471741,
+      "learning_rate": 9.035273780845118e-05,
+      "loss": 0.0411,
+      "step": 2680
+    },
+    {
+      "epoch": 4.838129496402877,
+      "grad_norm": 0.4572564363479614,
+      "learning_rate": 9.026475956857598e-05,
+      "loss": 0.0379,
+      "step": 2690
+    },
+    {
+      "epoch": 4.856115107913669,
+      "grad_norm": 0.6230306625366211,
+      "learning_rate": 9.017642523216362e-05,
+      "loss": 0.0408,
+      "step": 2700
+    },
+    {
+      "epoch": 4.874100719424461,
+      "grad_norm": 0.5241543650627136,
+      "learning_rate": 9.008773558043205e-05,
+      "loss": 0.0413,
+      "step": 2710
+    },
+    {
+      "epoch": 4.892086330935252,
+      "grad_norm": 0.6646131873130798,
+      "learning_rate": 8.999869139774151e-05,
+      "loss": 0.038,
+      "step": 2720
+    },
+    {
+      "epoch": 4.910071942446043,
+      "grad_norm": 0.4854746162891388,
+      "learning_rate": 8.990929347158773e-05,
+      "loss": 0.0404,
+      "step": 2730
+    },
+    {
+      "epoch": 4.928057553956835,
+      "grad_norm": 0.6121888160705566,
+      "learning_rate": 8.981954259259487e-05,
+      "loss": 0.0476,
+      "step": 2740
+    },
+    {
+      "epoch": 4.946043165467626,
+      "grad_norm": 0.5747014880180359,
+      "learning_rate": 8.972943955450854e-05,
+      "loss": 0.0353,
+      "step": 2750
+    },
+    {
+      "epoch": 4.9640287769784175,
+      "grad_norm": 0.5273730754852295,
+      "learning_rate": 8.963898515418884e-05,
+      "loss": 0.0382,
+      "step": 2760
+    },
+    {
+      "epoch": 4.982014388489208,
+      "grad_norm": 0.37167927622795105,
+      "learning_rate": 8.954818019160329e-05,
+      "loss": 0.0362,
+      "step": 2770
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 0.5870368480682373,
+      "learning_rate": 8.945702546981969e-05,
+      "loss": 0.0391,
+      "step": 2780
+    },
+    {
+      "epoch": 5.017985611510792,
+      "grad_norm": 0.5314123630523682,
+      "learning_rate": 8.936552179499909e-05,
+      "loss": 0.0348,
+      "step": 2790
+    },
+    {
+      "epoch": 5.0359712230215825,
+      "grad_norm": 0.42085975408554077,
+      "learning_rate": 8.927366997638861e-05,
+      "loss": 0.0402,
+      "step": 2800
+    },
+    {
+      "epoch": 5.053956834532374,
+      "grad_norm": 0.43196380138397217,
+      "learning_rate": 8.918147082631439e-05,
+      "loss": 0.0375,
+      "step": 2810
+    },
+    {
+      "epoch": 5.071942446043165,
+      "grad_norm": 0.5957926511764526,
+      "learning_rate": 8.90889251601742e-05,
+      "loss": 0.0328,
+      "step": 2820
+    },
+    {
+      "epoch": 5.089928057553957,
+      "grad_norm": 0.698459267616272,
+      "learning_rate": 8.899603379643051e-05,
+      "loss": 0.0413,
+      "step": 2830
+    },
+    {
+      "epoch": 5.107913669064748,
+      "grad_norm": 0.5308762788772583,
+      "learning_rate": 8.890279755660295e-05,
+      "loss": 0.036,
+      "step": 2840
+    },
+    {
+      "epoch": 5.125899280575539,
+      "grad_norm": 0.8266509771347046,
+      "learning_rate": 8.880921726526129e-05,
+      "loss": 0.0319,
+      "step": 2850
+    },
+    {
+      "epoch": 5.143884892086331,
+      "grad_norm": 0.36493539810180664,
+      "learning_rate": 8.8715293750018e-05,
+      "loss": 0.0376,
+      "step": 2860
+    },
+    {
+      "epoch": 5.161870503597123,
+      "grad_norm": 1.0405480861663818,
+      "learning_rate": 8.862102784152099e-05,
+      "loss": 0.0366,
+      "step": 2870
+    },
+    {
+      "epoch": 5.179856115107913,
+      "grad_norm": 0.5131875872612,
+      "learning_rate": 8.852642037344626e-05,
+      "loss": 0.0371,
+      "step": 2880
+    },
+    {
+      "epoch": 5.197841726618705,
+      "grad_norm": 0.5420644283294678,
+      "learning_rate": 8.843147218249053e-05,
+      "loss": 0.0352,
+      "step": 2890
+    },
+    {
+      "epoch": 5.215827338129497,
+      "grad_norm": 0.5955025553703308,
+      "learning_rate": 8.833618410836381e-05,
+      "loss": 0.037,
+      "step": 2900
+    },
+    {
+      "epoch": 5.233812949640288,
+      "grad_norm": 0.5233142971992493,
+      "learning_rate": 8.8240556993782e-05,
+      "loss": 0.0428,
+      "step": 2910
+    },
+    {
+      "epoch": 5.251798561151079,
+      "grad_norm": 0.6795997619628906,
+      "learning_rate": 8.814459168445943e-05,
+      "loss": 0.0359,
+      "step": 2920
+    },
+    {
+      "epoch": 5.26978417266187,
+      "grad_norm": 0.5889835357666016,
+      "learning_rate": 8.80482890291014e-05,
+      "loss": 0.0376,
+      "step": 2930
+    },
+    {
+      "epoch": 5.287769784172662,
+      "grad_norm": 0.4793168902397156,
+      "learning_rate": 8.795164987939665e-05,
+      "loss": 0.0356,
+      "step": 2940
+    },
+    {
+      "epoch": 5.305755395683454,
+      "grad_norm": 0.6167388558387756,
+      "learning_rate": 8.78546750900098e-05,
+      "loss": 0.0389,
+      "step": 2950
+    },
+    {
+      "epoch": 5.323741007194244,
+      "grad_norm": 0.6507744789123535,
+      "learning_rate": 8.775736551857387e-05,
+      "loss": 0.0417,
+      "step": 2960
+    },
+    {
+      "epoch": 5.341726618705036,
+      "grad_norm": 0.4230986535549164,
+      "learning_rate": 8.765972202568265e-05,
+      "loss": 0.0377,
+      "step": 2970
+    },
+    {
+      "epoch": 5.359712230215827,
+      "grad_norm": 0.4343918263912201,
+      "learning_rate": 8.756174547488305e-05,
+      "loss": 0.0343,
+      "step": 2980
+    },
+    {
+      "epoch": 5.377697841726619,
+      "grad_norm": 0.4428558647632599,
+      "learning_rate": 8.746343673266755e-05,
+      "loss": 0.0324,
+      "step": 2990
+    },
+    {
+      "epoch": 5.39568345323741,
+      "grad_norm": 0.4546818435192108,
+      "learning_rate": 8.736479666846642e-05,
+      "loss": 0.0446,
+      "step": 3000
+    },
+    {
+      "epoch": 5.413669064748201,
+      "grad_norm": 0.40336257219314575,
+      "learning_rate": 8.726582615464021e-05,
+      "loss": 0.0389,
+      "step": 3010
+    },
+    {
+      "epoch": 5.431654676258993,
+      "grad_norm": 0.407710999250412,
+      "learning_rate": 8.716652606647182e-05,
+      "loss": 0.0399,
+      "step": 3020
+    },
+    {
+      "epoch": 5.4496402877697845,
+      "grad_norm": 0.5834148526191711,
+      "learning_rate": 8.706689728215893e-05,
+      "loss": 0.0328,
+      "step": 3030
+    },
+    {
+      "epoch": 5.467625899280575,
+      "grad_norm": 0.35896340012550354,
+      "learning_rate": 8.696694068280614e-05,
+      "loss": 0.0356,
+      "step": 3040
+    },
+    {
+      "epoch": 5.485611510791367,
+      "grad_norm": 0.4738776981830597,
+      "learning_rate": 8.686665715241719e-05,
+      "loss": 0.0376,
+      "step": 3050
+    },
+    {
+      "epoch": 5.503597122302159,
+      "grad_norm": 0.5131441354751587,
+      "learning_rate": 8.676604757788722e-05,
+      "loss": 0.0374,
+      "step": 3060
+    },
+    {
+      "epoch": 5.5215827338129495,
+      "grad_norm": 0.643847644329071,
+      "learning_rate": 8.66651128489948e-05,
+      "loss": 0.0433,
+      "step": 3070
+    },
+    {
+      "epoch": 5.539568345323741,
+      "grad_norm": 0.6620198488235474,
+      "learning_rate": 8.656385385839412e-05,
+      "loss": 0.0333,
+      "step": 3080
+    },
+    {
+      "epoch": 5.557553956834532,
+      "grad_norm": 0.6396639943122864,
+      "learning_rate": 8.646227150160717e-05,
+      "loss": 0.0374,
+      "step": 3090
+    },
+    {
+      "epoch": 5.575539568345324,
+      "grad_norm": 0.6719627976417542,
+      "learning_rate": 8.636036667701568e-05,
+      "loss": 0.0429,
+      "step": 3100
+    },
+    {
+      "epoch": 5.593525179856115,
+      "grad_norm": 0.8319839239120483,
+      "learning_rate": 8.62581402858533e-05,
+      "loss": 0.0447,
+      "step": 3110
+    },
+    {
+      "epoch": 5.611510791366906,
+      "grad_norm": 0.4461003839969635,
+      "learning_rate": 8.615559323219753e-05,
+      "loss": 0.0325,
+      "step": 3120
+    },
+    {
+      "epoch": 5.629496402877698,
+      "grad_norm": 0.7610004544258118,
+      "learning_rate": 8.605272642296178e-05,
+      "loss": 0.0373,
+      "step": 3130
+    },
+    {
+      "epoch": 5.647482014388489,
+      "grad_norm": 0.5503531694412231,
+      "learning_rate": 8.594954076788736e-05,
+      "loss": 0.0423,
+      "step": 3140
+    },
+    {
+      "epoch": 5.66546762589928,
+      "grad_norm": 0.5198085904121399,
+      "learning_rate": 8.58460371795354e-05,
+      "loss": 0.0429,
+      "step": 3150
+    },
+    {
+      "epoch": 5.683453237410072,
+      "grad_norm": 0.6385695338249207,
+      "learning_rate": 8.57422165732788e-05,
+      "loss": 0.0322,
+      "step": 3160
+    },
+    {
+      "epoch": 5.701438848920863,
+      "grad_norm": 0.9159457683563232,
+      "learning_rate": 8.563807986729408e-05,
+      "loss": 0.0378,
+      "step": 3170
+    },
+    {
+      "epoch": 5.719424460431655,
+      "grad_norm": 0.4671606421470642,
+      "learning_rate": 8.553362798255338e-05,
+      "loss": 0.0368,
+      "step": 3180
+    },
+    {
+      "epoch": 5.737410071942446,
+      "grad_norm": 0.7837393879890442,
+      "learning_rate": 8.54288618428162e-05,
+      "loss": 0.0392,
+      "step": 3190
+    },
+    {
+      "epoch": 5.755395683453237,
+      "grad_norm": 0.8009852766990662,
+      "learning_rate": 8.532378237462128e-05,
+      "loss": 0.0425,
+      "step": 3200
+    },
+    {
+      "epoch": 5.773381294964029,
+      "grad_norm": 0.6494861841201782,
+      "learning_rate": 8.521839050727838e-05,
+      "loss": 0.039,
+      "step": 3210
+    },
+    {
+      "epoch": 5.7913669064748206,
+      "grad_norm": 0.5232138633728027,
+      "learning_rate": 8.511268717286013e-05,
+      "loss": 0.0341,
+      "step": 3220
+    },
+    {
+      "epoch": 5.809352517985611,
+      "grad_norm": 0.5768583416938782,
+      "learning_rate": 8.500667330619367e-05,
+      "loss": 0.0403,
+      "step": 3230
+    },
+    {
+      "epoch": 5.827338129496403,
+      "grad_norm": 0.4913746118545532,
+      "learning_rate": 8.490034984485251e-05,
+      "loss": 0.0316,
+      "step": 3240
+    },
+    {
+      "epoch": 5.845323741007194,
+      "grad_norm": 0.7548671364784241,
+      "learning_rate": 8.479371772914813e-05,
+      "loss": 0.0329,
+      "step": 3250
+    },
+    {
+      "epoch": 5.863309352517986,
+      "grad_norm": 0.810468316078186,
+      "learning_rate": 8.468677790212173e-05,
+      "loss": 0.0403,
+      "step": 3260
+    },
+    {
+      "epoch": 5.881294964028777,
+      "grad_norm": 0.38834044337272644,
+      "learning_rate": 8.457953130953587e-05,
+      "loss": 0.0387,
+      "step": 3270
+    },
+    {
+      "epoch": 5.899280575539568,
+      "grad_norm": 0.40479061007499695,
+      "learning_rate": 8.447197889986611e-05,
+      "loss": 0.0296,
+      "step": 3280
+    },
+    {
+      "epoch": 5.91726618705036,
+      "grad_norm": 0.30964717268943787,
+      "learning_rate": 8.436412162429261e-05,
+      "loss": 0.0328,
+      "step": 3290
+    },
+    {
+      "epoch": 5.935251798561151,
+      "grad_norm": 0.6192933917045593,
+      "learning_rate": 8.425596043669175e-05,
+      "loss": 0.0309,
+      "step": 3300
+    },
+    {
+      "epoch": 5.953237410071942,
+      "grad_norm": 0.4001467525959015,
+      "learning_rate": 8.414749629362762e-05,
+      "loss": 0.0339,
+      "step": 3310
+    },
+    {
+      "epoch": 5.971223021582734,
+      "grad_norm": 0.6769677996635437,
+      "learning_rate": 8.403873015434367e-05,
+      "loss": 0.0324,
+      "step": 3320
+    },
+    {
+      "epoch": 5.989208633093525,
+      "grad_norm": 0.5511227250099182,
+      "learning_rate": 8.392966298075413e-05,
+      "loss": 0.0341,
+      "step": 3330
+    },
+    {
+      "epoch": 6.0071942446043165,
+      "grad_norm": 1.1561652421951294,
+      "learning_rate": 8.382029573743552e-05,
+      "loss": 0.034,
+      "step": 3340
+    },
+    {
+      "epoch": 6.025179856115108,
+      "grad_norm": 0.7122268080711365,
+      "learning_rate": 8.371062939161823e-05,
+      "loss": 0.0344,
+      "step": 3350
+    },
+    {
+      "epoch": 6.043165467625899,
+      "grad_norm": 0.577812135219574,
+      "learning_rate": 8.360066491317777e-05,
+      "loss": 0.0367,
+      "step": 3360
+    },
+    {
+      "epoch": 6.061151079136691,
+      "grad_norm": 0.5453331470489502,
+      "learning_rate": 8.349040327462634e-05,
+      "loss": 0.0387,
+      "step": 3370
+    },
+    {
+      "epoch": 6.079136690647482,
+      "grad_norm": 0.4377831518650055,
+      "learning_rate": 8.33798454511042e-05,
+      "loss": 0.0396,
+      "step": 3380
+    },
+    {
+      "epoch": 6.097122302158273,
+      "grad_norm": 0.7018815279006958,
+      "learning_rate": 8.326899242037102e-05,
+      "loss": 0.0441,
+      "step": 3390
+    },
+    {
+      "epoch": 6.115107913669065,
+      "grad_norm": 0.34575697779655457,
+      "learning_rate": 8.315784516279728e-05,
+      "loss": 0.0387,
+      "step": 3400
+    },
+    {
+      "epoch": 6.133093525179856,
+      "grad_norm": 0.6859731674194336,
+      "learning_rate": 8.304640466135547e-05,
+      "loss": 0.0389,
+      "step": 3410
+    },
+    {
+      "epoch": 6.151079136690647,
+      "grad_norm": 0.7379283308982849,
+      "learning_rate": 8.29346719016116e-05,
+      "loss": 0.0401,
+      "step": 3420
+    },
+    {
+      "epoch": 6.169064748201439,
+      "grad_norm": 0.9674530029296875,
+      "learning_rate": 8.282264787171632e-05,
+      "loss": 0.0361,
+      "step": 3430
+    },
+    {
+      "epoch": 6.18705035971223,
+      "grad_norm": 0.6345366835594177,
+      "learning_rate": 8.271033356239625e-05,
+      "loss": 0.0313,
+      "step": 3440
+    },
+    {
+      "epoch": 6.205035971223022,
+      "grad_norm": 0.6643152236938477,
+      "learning_rate": 8.259772996694522e-05,
+      "loss": 0.0387,
+      "step": 3450
+    },
+    {
+      "epoch": 6.223021582733813,
+      "grad_norm": 0.4115314781665802,
+      "learning_rate": 8.248483808121541e-05,
+      "loss": 0.0407,
+      "step": 3460
+    },
+    {
+      "epoch": 6.241007194244604,
+      "grad_norm": 0.5159537196159363,
+      "learning_rate": 8.237165890360869e-05,
+      "loss": 0.0366,
+      "step": 3470
+    },
+    {
+      "epoch": 6.258992805755396,
+      "grad_norm": 0.6857429146766663,
+      "learning_rate": 8.225819343506767e-05,
+      "loss": 0.0352,
+      "step": 3480
+    },
+    {
+      "epoch": 6.276978417266187,
+      "grad_norm": 1.1055829524993896,
+      "learning_rate": 8.214444267906683e-05,
+      "loss": 0.0329,
+      "step": 3490
+    },
+    {
+      "epoch": 6.294964028776978,
+      "grad_norm": 0.8197521567344666,
+      "learning_rate": 8.203040764160378e-05,
+      "loss": 0.0405,
+      "step": 3500
+    },
+    {
+      "epoch": 6.31294964028777,
+      "grad_norm": 0.5625283718109131,
+      "learning_rate": 8.191608933119021e-05,
+      "loss": 0.0374,
+      "step": 3510
+    },
+    {
+      "epoch": 6.330935251798561,
+      "grad_norm": 0.7339150309562683,
+      "learning_rate": 8.18014887588431e-05,
+      "loss": 0.0406,
+      "step": 3520
+    },
+    {
+      "epoch": 6.348920863309353,
+      "grad_norm": 0.5098793506622314,
+      "learning_rate": 8.168660693807569e-05,
+      "loss": 0.0363,
+      "step": 3530
+    },
+    {
+      "epoch": 6.366906474820144,
+      "grad_norm": 0.5724267959594727,
+      "learning_rate": 8.157144488488853e-05,
+      "loss": 0.0373,
+      "step": 3540
+    },
+    {
+      "epoch": 6.384892086330935,
+      "grad_norm": 0.6223220825195312,
+      "learning_rate": 8.145600361776051e-05,
+      "loss": 0.0382,
+      "step": 3550
+    },
+    {
+      "epoch": 6.402877697841727,
+      "grad_norm": 0.41262897849082947,
+      "learning_rate": 8.134028415763993e-05,
+      "loss": 0.0381,
+      "step": 3560
+    },
+    {
+      "epoch": 6.420863309352518,
+      "grad_norm": 0.4510844647884369,
+      "learning_rate": 8.122428752793528e-05,
+      "loss": 0.0408,
+      "step": 3570
+    },
+    {
+      "epoch": 6.438848920863309,
+      "grad_norm": 0.47913122177124023,
+      "learning_rate": 8.110801475450638e-05,
+      "loss": 0.04,
+      "step": 3580
+    },
+    {
+      "epoch": 6.456834532374101,
+      "grad_norm": 0.9318785667419434,
+      "learning_rate": 8.09914668656552e-05,
+      "loss": 0.0332,
+      "step": 3590
+    },
+    {
+      "epoch": 6.474820143884892,
+      "grad_norm": 0.4933910071849823,
+      "learning_rate": 8.087464489211682e-05,
+      "loss": 0.0356,
+      "step": 3600
+    },
+    {
+      "epoch": 6.4928057553956835,
+      "grad_norm": 0.5342298150062561,
+      "learning_rate": 8.075754986705029e-05,
+      "loss": 0.0308,
+      "step": 3610
+    },
+    {
+      "epoch": 6.510791366906475,
+      "grad_norm": 0.5947268605232239,
+      "learning_rate": 8.064018282602945e-05,
+      "loss": 0.0352,
+      "step": 3620
+    },
+    {
+      "epoch": 6.528776978417266,
+      "grad_norm": 0.39868468046188354,
+      "learning_rate": 8.052254480703385e-05,
+      "loss": 0.0315,
+      "step": 3630
+    },
+    {
+      "epoch": 6.546762589928058,
+      "grad_norm": 0.7511667013168335,
+      "learning_rate": 8.040463685043957e-05,
+      "loss": 0.0386,
+      "step": 3640
+    },
+    {
+      "epoch": 6.564748201438849,
+      "grad_norm": 0.6233493685722351,
+      "learning_rate": 8.02864599990099e-05,
+      "loss": 0.0367,
+      "step": 3650
+    },
+    {
+      "epoch": 6.58273381294964,
+      "grad_norm": 0.6306430697441101,
+      "learning_rate": 8.016801529788628e-05,
+      "loss": 0.0287,
+      "step": 3660
+    },
+    {
+      "epoch": 6.600719424460432,
+      "grad_norm": 0.5310491323471069,
+      "learning_rate": 8.004930379457892e-05,
+      "loss": 0.0375,
+      "step": 3670
+    },
+    {
+      "epoch": 6.618705035971223,
+      "grad_norm": 0.5883777141571045,
+      "learning_rate": 7.993032653895767e-05,
+      "loss": 0.0383,
+      "step": 3680
+    },
+    {
+      "epoch": 6.636690647482014,
+      "grad_norm": 0.9777166843414307,
+      "learning_rate": 7.981108458324256e-05,
+      "loss": 0.039,
+      "step": 3690
+    },
+    {
+      "epoch": 6.654676258992806,
+      "grad_norm": 0.8692882657051086,
+      "learning_rate": 7.969157898199469e-05,
+      "loss": 0.0393,
+      "step": 3700
+    },
+    {
+      "epoch": 6.672661870503597,
+      "grad_norm": 0.46635007858276367,
+      "learning_rate": 7.957181079210675e-05,
+      "loss": 0.0393,
+      "step": 3710
+    },
+    {
+      "epoch": 6.690647482014389,
+      "grad_norm": 0.8771538734436035,
+      "learning_rate": 7.945178107279376e-05,
+      "loss": 0.035,
+      "step": 3720
+    },
+    {
+      "epoch": 6.7086330935251794,
+      "grad_norm": 0.39784759283065796,
+      "learning_rate": 7.933149088558368e-05,
+      "loss": 0.0332,
+      "step": 3730
+    },
+    {
+      "epoch": 6.726618705035971,
+      "grad_norm": 1.5518379211425781,
+      "learning_rate": 7.921094129430793e-05,
+      "loss": 0.0367,
+      "step": 3740
+    },
+    {
+      "epoch": 6.744604316546763,
+      "grad_norm": 0.6117278933525085,
+      "learning_rate": 7.909013336509221e-05,
+      "loss": 0.0352,
+      "step": 3750
+    },
+    {
+      "epoch": 6.762589928057554,
+      "grad_norm": 0.9333751201629639,
+      "learning_rate": 7.896906816634682e-05,
+      "loss": 0.0375,
+      "step": 3760
+    },
+    {
+      "epoch": 6.780575539568345,
+      "grad_norm": 0.8926209211349487,
+      "learning_rate": 7.884774676875734e-05,
+      "loss": 0.0362,
+      "step": 3770
+    },
+    {
+      "epoch": 6.798561151079137,
+      "grad_norm": 0.6408644914627075,
+      "learning_rate": 7.872617024527521e-05,
+      "loss": 0.0402,
+      "step": 3780
+    },
+    {
+      "epoch": 6.816546762589928,
+      "grad_norm": 0.40880292654037476,
+      "learning_rate": 7.860433967110808e-05,
+      "loss": 0.0382,
+      "step": 3790
+    },
+    {
+      "epoch": 6.83453237410072,
+      "grad_norm": 0.6417644023895264,
+      "learning_rate": 7.84822561237104e-05,
+      "loss": 0.038,
+      "step": 3800
+    },
+    {
+      "epoch": 6.852517985611511,
+      "grad_norm": 0.5577982664108276,
+      "learning_rate": 7.835992068277398e-05,
+      "loss": 0.031,
+      "step": 3810
+    },
+    {
+      "epoch": 6.870503597122302,
+      "grad_norm": 0.5106334090232849,
+      "learning_rate": 7.823733443021824e-05,
+      "loss": 0.0438,
+      "step": 3820
+    },
+    {
+      "epoch": 6.888489208633094,
+      "grad_norm": 0.5679177641868591,
+      "learning_rate": 7.811449845018078e-05,
+      "loss": 0.0378,
+      "step": 3830
+    },
+    {
+      "epoch": 6.906474820143885,
+      "grad_norm": 0.5835026502609253,
+      "learning_rate": 7.799141382900778e-05,
+      "loss": 0.0289,
+      "step": 3840
+    },
+    {
+      "epoch": 6.924460431654676,
+      "grad_norm": 0.5125241279602051,
+      "learning_rate": 7.786808165524437e-05,
+      "loss": 0.0312,
+      "step": 3850
+    },
+    {
+      "epoch": 6.942446043165468,
+      "grad_norm": 0.6633017063140869,
+      "learning_rate": 7.774450301962496e-05,
+      "loss": 0.0405,
+      "step": 3860
+    },
+    {
+      "epoch": 6.960431654676259,
+      "grad_norm": 0.5421878695487976,
+      "learning_rate": 7.762067901506369e-05,
+      "loss": 0.0328,
+      "step": 3870
+    },
+    {
+      "epoch": 6.9784172661870505,
+      "grad_norm": 0.6832766532897949,
+      "learning_rate": 7.749661073664467e-05,
+      "loss": 0.0371,
+      "step": 3880
+    },
+    {
+      "epoch": 6.996402877697841,
+      "grad_norm": 0.723677933216095,
+      "learning_rate": 7.737229928161238e-05,
+      "loss": 0.0297,
+      "step": 3890
+    },
+    {
+      "epoch": 7.014388489208633,
+      "grad_norm": 1.1337522268295288,
+      "learning_rate": 7.724774574936188e-05,
+      "loss": 0.0379,
+      "step": 3900
+    },
+    {
+      "epoch": 7.032374100719425,
+      "grad_norm": 0.7429106831550598,
+      "learning_rate": 7.712295124142918e-05,
+      "loss": 0.0354,
+      "step": 3910
+    },
+    {
+      "epoch": 7.0503597122302155,
+      "grad_norm": 0.4767836630344391,
+      "learning_rate": 7.699791686148141e-05,
+      "loss": 0.0338,
+      "step": 3920
+    },
+    {
+      "epoch": 7.068345323741007,
+      "grad_norm": 0.48121386766433716,
+      "learning_rate": 7.68726437153071e-05,
+      "loss": 0.035,
+      "step": 3930
+    },
+    {
+      "epoch": 7.086330935251799,
+      "grad_norm": 0.4909757375717163,
+      "learning_rate": 7.674713291080644e-05,
+      "loss": 0.0397,
+      "step": 3940
+    },
+    {
+      "epoch": 7.10431654676259,
+      "grad_norm": 0.6398098468780518,
+      "learning_rate": 7.662138555798134e-05,
+      "loss": 0.0404,
+      "step": 3950
+    },
+    {
+      "epoch": 7.122302158273381,
+      "grad_norm": 0.5961847305297852,
+      "learning_rate": 7.649540276892588e-05,
+      "loss": 0.0294,
+      "step": 3960
+    },
+    {
+      "epoch": 7.140287769784172,
+      "grad_norm": 0.6063967347145081,
+      "learning_rate": 7.636918565781616e-05,
+      "loss": 0.0396,
+      "step": 3970
+    },
+    {
+      "epoch": 7.158273381294964,
+      "grad_norm": 0.4251496493816376,
+      "learning_rate": 7.624273534090064e-05,
+      "loss": 0.0302,
+      "step": 3980
+    },
+    {
+      "epoch": 7.176258992805756,
+      "grad_norm": 0.54291170835495,
+      "learning_rate": 7.611605293649026e-05,
+      "loss": 0.035,
+      "step": 3990
+    },
+    {
+      "epoch": 7.194244604316546,
+      "grad_norm": 0.7140129804611206,
+      "learning_rate": 7.598913956494847e-05,
+      "loss": 0.034,
+      "step": 4000
+    },
+    {
+      "epoch": 7.212230215827338,
+      "grad_norm": 0.47695836424827576,
+      "learning_rate": 7.586199634868139e-05,
+      "loss": 0.0343,
+      "step": 4010
+    },
+    {
+      "epoch": 7.23021582733813,
+      "grad_norm": 0.9651088714599609,
+      "learning_rate": 7.573462441212781e-05,
+      "loss": 0.036,
+      "step": 4020
+    },
+    {
+      "epoch": 7.248201438848921,
+      "grad_norm": 0.7029420733451843,
+      "learning_rate": 7.560702488174938e-05,
+      "loss": 0.0388,
+      "step": 4030
+    },
+    {
+      "epoch": 7.266187050359712,
+      "grad_norm": 0.6637154221534729,
+      "learning_rate": 7.547919888602049e-05,
+      "loss": 0.0321,
+      "step": 4040
+    },
+    {
+      "epoch": 7.284172661870503,
+      "grad_norm": 0.5793517827987671,
+      "learning_rate": 7.535114755541838e-05,
+      "loss": 0.0311,
+      "step": 4050
+    },
+    {
+      "epoch": 7.302158273381295,
+      "grad_norm": 0.4631473422050476,
+      "learning_rate": 7.522287202241313e-05,
+      "loss": 0.0376,
+      "step": 4060
+    },
+    {
+      "epoch": 7.320143884892087,
+      "grad_norm": 0.3995862603187561,
+      "learning_rate": 7.509437342145764e-05,
+      "loss": 0.0384,
+      "step": 4070
+    },
+    {
+      "epoch": 7.338129496402877,
+      "grad_norm": 0.48294979333877563,
+      "learning_rate": 7.496565288897761e-05,
+      "loss": 0.0313,
+      "step": 4080
+    },
+    {
+      "epoch": 7.356115107913669,
+      "grad_norm": 0.5210274457931519,
+      "learning_rate": 7.483671156336141e-05,
+      "loss": 0.0322,
+      "step": 4090
+    },
+    {
+      "epoch": 7.374100719424461,
+      "grad_norm": 0.6423503160476685,
+      "learning_rate": 7.470755058495018e-05,
+      "loss": 0.0344,
+      "step": 4100
+    },
+    {
+      "epoch": 7.392086330935252,
+      "grad_norm": 0.37342312932014465,
+      "learning_rate": 7.457817109602754e-05,
+      "loss": 0.0348,
+      "step": 4110
+    },
+    {
+      "epoch": 7.410071942446043,
+      "grad_norm": 0.5671528577804565,
+      "learning_rate": 7.444857424080964e-05,
+      "loss": 0.0299,
+      "step": 4120
+    },
+    {
+      "epoch": 7.428057553956835,
+      "grad_norm": 0.5335654020309448,
+      "learning_rate": 7.431876116543502e-05,
+      "loss": 0.0366,
+      "step": 4130
+    },
+    {
+      "epoch": 7.446043165467626,
+      "grad_norm": 0.6219868063926697,
+      "learning_rate": 7.418873301795437e-05,
+      "loss": 0.0367,
+      "step": 4140
+    },
+    {
+      "epoch": 7.4640287769784175,
+      "grad_norm": 0.7356789112091064,
+      "learning_rate": 7.405849094832052e-05,
+      "loss": 0.0307,
+      "step": 4150
+    },
+    {
+      "epoch": 7.482014388489208,
+      "grad_norm": 0.5447286367416382,
+      "learning_rate": 7.392803610837816e-05,
+      "loss": 0.028,
+      "step": 4160
+    },
+    {
+      "epoch": 7.5,
+      "grad_norm": 0.7504562735557556,
+      "learning_rate": 7.379736965185368e-05,
+      "loss": 0.0392,
+      "step": 4170
+    },
+    {
+      "epoch": 7.517985611510792,
+      "grad_norm": 0.48435309529304504,
+      "learning_rate": 7.366649273434506e-05,
+      "loss": 0.0374,
+      "step": 4180
+    },
+    {
+      "epoch": 7.5359712230215825,
+      "grad_norm": 0.6540862917900085,
+      "learning_rate": 7.353540651331148e-05,
+      "loss": 0.0351,
+      "step": 4190
+    },
+    {
+      "epoch": 7.553956834532374,
+      "grad_norm": 0.5130005478858948,
+      "learning_rate": 7.340411214806322e-05,
+      "loss": 0.029,
+      "step": 4200
+    },
+    {
+      "epoch": 7.571942446043165,
+      "grad_norm": 1.5337539911270142,
+      "learning_rate": 7.327261079975137e-05,
+      "loss": 0.0371,
+      "step": 4210
+    },
+    {
+      "epoch": 7.589928057553957,
+      "grad_norm": 0.7431342601776123,
+      "learning_rate": 7.314090363135752e-05,
+      "loss": 0.0311,
+      "step": 4220
+    },
+    {
+      "epoch": 7.607913669064748,
+      "grad_norm": 0.47893115878105164,
+      "learning_rate": 7.300899180768352e-05,
+      "loss": 0.029,
+      "step": 4230
+    },
+    {
+      "epoch": 7.625899280575539,
+      "grad_norm": 0.6539574265480042,
+      "learning_rate": 7.287687649534118e-05,
+      "loss": 0.0321,
+      "step": 4240
+    },
+    {
+      "epoch": 7.643884892086331,
+      "grad_norm": 0.43070584535598755,
+      "learning_rate": 7.274455886274193e-05,
+      "loss": 0.0343,
+      "step": 4250
+    },
+    {
+      "epoch": 7.661870503597123,
+      "grad_norm": 0.471818208694458,
+      "learning_rate": 7.261204008008648e-05,
+      "loss": 0.0371,
+      "step": 4260
+    },
+    {
+      "epoch": 7.679856115107913,
+      "grad_norm": 0.5291281342506409,
+      "learning_rate": 7.247932131935451e-05,
+      "loss": 0.0315,
+      "step": 4270
+    },
+    {
+      "epoch": 7.697841726618705,
+      "grad_norm": 0.7470479607582092,
+      "learning_rate": 7.234640375429427e-05,
+      "loss": 0.032,
+      "step": 4280
+    },
+    {
+      "epoch": 7.715827338129497,
+      "grad_norm": 0.7638947367668152,
+      "learning_rate": 7.221328856041221e-05,
+      "loss": 0.0283,
+      "step": 4290
+    },
+    {
+      "epoch": 7.733812949640288,
+      "grad_norm": 0.48779767751693726,
+      "learning_rate": 7.207997691496259e-05,
+      "loss": 0.031,
+      "step": 4300
+    },
+    {
+      "epoch": 7.751798561151079,
+      "grad_norm": 0.8140793442726135,
+      "learning_rate": 7.194646999693703e-05,
+      "loss": 0.0315,
+      "step": 4310
+    },
+    {
+      "epoch": 7.76978417266187,
+      "grad_norm": 0.5031821727752686,
+      "learning_rate": 7.18127689870542e-05,
+      "loss": 0.0357,
+      "step": 4320
+    },
+    {
+      "epoch": 7.787769784172662,
+      "grad_norm": 0.5815337896347046,
+      "learning_rate": 7.167887506774922e-05,
+      "loss": 0.0308,
+      "step": 4330
+    },
+    {
+      "epoch": 7.805755395683454,
+      "grad_norm": 0.7105207443237305,
+      "learning_rate": 7.154478942316328e-05,
+      "loss": 0.0327,
+      "step": 4340
+    },
+    {
+      "epoch": 7.823741007194244,
+      "grad_norm": 1.1678744554519653,
+      "learning_rate": 7.141051323913317e-05,
+      "loss": 0.0363,
+      "step": 4350
+    },
+    {
+      "epoch": 7.841726618705036,
+      "grad_norm": 0.7489168643951416,
+      "learning_rate": 7.127604770318084e-05,
+      "loss": 0.0318,
+      "step": 4360
+    },
+    {
+      "epoch": 7.859712230215827,
+      "grad_norm": 0.5358786582946777,
+      "learning_rate": 7.114139400450279e-05,
+      "loss": 0.036,
+      "step": 4370
+    },
+    {
+      "epoch": 7.877697841726619,
+      "grad_norm": 0.6793840527534485,
+      "learning_rate": 7.100655333395959e-05,
+      "loss": 0.0293,
+      "step": 4380
+    },
+    {
+      "epoch": 7.89568345323741,
+      "grad_norm": 0.37476611137390137,
+      "learning_rate": 7.087152688406544e-05,
+      "loss": 0.0349,
+      "step": 4390
+    },
+    {
+      "epoch": 7.913669064748201,
+      "grad_norm": 0.5735613107681274,
+      "learning_rate": 7.07363158489775e-05,
+      "loss": 0.0422,
+      "step": 4400
+    },
+    {
+      "epoch": 7.931654676258993,
+      "grad_norm": 0.66266268491745,
+      "learning_rate": 7.060092142448537e-05,
+      "loss": 0.0377,
+      "step": 4410
+    },
+    {
+      "epoch": 7.9496402877697845,
+      "grad_norm": 0.5422354936599731,
+      "learning_rate": 7.046534480800055e-05,
+      "loss": 0.0359,
+      "step": 4420
+    },
+    {
+      "epoch": 7.967625899280575,
+      "grad_norm": 0.6018063426017761,
+      "learning_rate": 7.032958719854579e-05,
+      "loss": 0.0335,
+      "step": 4430
+    },
+    {
+      "epoch": 7.985611510791367,
+      "grad_norm": 0.526896595954895,
+      "learning_rate": 7.019364979674458e-05,
+      "loss": 0.0327,
+      "step": 4440
+    },
+    {
+      "epoch": 8.003597122302159,
+      "grad_norm": 0.7307107448577881,
+      "learning_rate": 7.005753380481041e-05,
+      "loss": 0.0348,
+      "step": 4450
+    },
+    {
+      "epoch": 8.02158273381295,
+      "grad_norm": 0.42269080877304077,
+      "learning_rate": 6.992124042653623e-05,
+      "loss": 0.0383,
+      "step": 4460
+    },
+    {
+      "epoch": 8.03956834532374,
+      "grad_norm": 0.5988796949386597,
+      "learning_rate": 6.978477086728374e-05,
+      "loss": 0.0359,
+      "step": 4470
+    },
+    {
+      "epoch": 8.057553956834532,
+      "grad_norm": 0.5922524929046631,
+      "learning_rate": 6.964812633397283e-05,
+      "loss": 0.0344,
+      "step": 4480
+    },
+    {
+      "epoch": 8.075539568345324,
+      "grad_norm": 0.4443754255771637,
+      "learning_rate": 6.951130803507076e-05,
+      "loss": 0.0368,
+      "step": 4490
+    },
+    {
+      "epoch": 8.093525179856115,
+      "grad_norm": 0.42000144720077515,
+      "learning_rate": 6.937431718058161e-05,
+      "loss": 0.033,
+      "step": 4500
+    },
+    {
+      "epoch": 8.111510791366907,
+      "grad_norm": 0.6107466816902161,
+      "learning_rate": 6.923715498203548e-05,
+      "loss": 0.0401,
+      "step": 4510
+    },
+    {
+      "epoch": 8.129496402877697,
+      "grad_norm": 0.5843360424041748,
+      "learning_rate": 6.909982265247785e-05,
+      "loss": 0.0292,
+      "step": 4520
+    },
+    {
+      "epoch": 8.147482014388489,
+      "grad_norm": 0.6366044282913208,
+      "learning_rate": 6.89623214064588e-05,
+      "loss": 0.0311,
+      "step": 4530
+    },
+    {
+      "epoch": 8.16546762589928,
+      "grad_norm": 0.6470826268196106,
+      "learning_rate": 6.882465246002227e-05,
+      "loss": 0.0341,
+      "step": 4540
+    },
+    {
+      "epoch": 8.183453237410072,
+      "grad_norm": 2.338733196258545,
+      "learning_rate": 6.868681703069534e-05,
+      "loss": 0.0333,
+      "step": 4550
+    },
+    {
+      "epoch": 8.201438848920864,
+      "grad_norm": 0.518305778503418,
+      "learning_rate": 6.854881633747747e-05,
+      "loss": 0.03,
+      "step": 4560
+    },
+    {
+      "epoch": 8.219424460431656,
+      "grad_norm": 0.49466755986213684,
+      "learning_rate": 6.841065160082964e-05,
+      "loss": 0.034,
+      "step": 4570
+    },
+    {
+      "epoch": 8.237410071942445,
+      "grad_norm": 0.6237930059432983,
+      "learning_rate": 6.827232404266365e-05,
+      "loss": 0.0368,
+      "step": 4580
+    },
+    {
+      "epoch": 8.255395683453237,
+      "grad_norm": 0.6763109564781189,
+      "learning_rate": 6.813383488633123e-05,
+      "loss": 0.0354,
+      "step": 4590
+    },
+    {
+      "epoch": 8.273381294964029,
+      "grad_norm": 0.48339012265205383,
+      "learning_rate": 6.799518535661336e-05,
+      "loss": 0.0302,
+      "step": 4600
+    },
+    {
+      "epoch": 8.29136690647482,
+      "grad_norm": 0.7084135413169861,
+      "learning_rate": 6.785637667970922e-05,
+      "loss": 0.0341,
+      "step": 4610
+    },
+    {
+      "epoch": 8.309352517985612,
+      "grad_norm": 0.7776214480400085,
+      "learning_rate": 6.77174100832255e-05,
+      "loss": 0.0346,
+      "step": 4620
+    },
+    {
+      "epoch": 8.327338129496402,
+      "grad_norm": 0.6965737342834473,
+      "learning_rate": 6.75782867961656e-05,
+      "loss": 0.0322,
+      "step": 4630
+    },
+    {
+      "epoch": 8.345323741007194,
+      "grad_norm": 0.448886513710022,
+      "learning_rate": 6.743900804891859e-05,
+      "loss": 0.0333,
+      "step": 4640
+    },
+    {
+      "epoch": 8.363309352517986,
+      "grad_norm": 0.6683182120323181,
+      "learning_rate": 6.72995750732484e-05,
+      "loss": 0.0361,
+      "step": 4650
+    },
+    {
+      "epoch": 8.381294964028777,
+      "grad_norm": 0.4467858076095581,
+      "learning_rate": 6.715998910228296e-05,
+      "loss": 0.0373,
+      "step": 4660
+    },
+    {
+      "epoch": 8.399280575539569,
+      "grad_norm": 0.7671613097190857,
+      "learning_rate": 6.702025137050332e-05,
+      "loss": 0.0349,
+      "step": 4670
+    },
+    {
+      "epoch": 8.417266187050359,
+      "grad_norm": 0.4649515450000763,
+      "learning_rate": 6.688036311373263e-05,
+      "loss": 0.0305,
+      "step": 4680
+    },
+    {
+      "epoch": 8.43525179856115,
+      "grad_norm": 0.5214066505432129,
+      "learning_rate": 6.674032556912522e-05,
+      "loss": 0.0323,
+      "step": 4690
+    },
+    {
+      "epoch": 8.453237410071942,
+      "grad_norm": 0.46762508153915405,
+      "learning_rate": 6.660013997515583e-05,
+      "loss": 0.0385,
+      "step": 4700
+    },
+    {
+      "epoch": 8.471223021582734,
+      "grad_norm": 0.8438305854797363,
+      "learning_rate": 6.645980757160842e-05,
+      "loss": 0.0353,
+      "step": 4710
+    },
+    {
+      "epoch": 8.489208633093526,
+      "grad_norm": 0.5806241035461426,
+      "learning_rate": 6.631932959956535e-05,
+      "loss": 0.0331,
+      "step": 4720
+    },
+    {
+      "epoch": 8.507194244604317,
+      "grad_norm": 0.38535448908805847,
+      "learning_rate": 6.617870730139638e-05,
+      "loss": 0.0329,
+      "step": 4730
+    },
+    {
+      "epoch": 8.525179856115107,
+      "grad_norm": 0.3644998371601105,
+      "learning_rate": 6.603794192074766e-05,
+      "loss": 0.031,
+      "step": 4740
+    },
+    {
+      "epoch": 8.543165467625899,
+      "grad_norm": 0.39075425267219543,
+      "learning_rate": 6.589703470253076e-05,
+      "loss": 0.0337,
+      "step": 4750
+    },
+    {
+      "epoch": 8.56115107913669,
+      "grad_norm": 0.6422626972198486,
+      "learning_rate": 6.575598689291163e-05,
+      "loss": 0.035,
+      "step": 4760
+    },
+    {
+      "epoch": 8.579136690647482,
+      "grad_norm": 0.6469767689704895,
+      "learning_rate": 6.561479973929958e-05,
+      "loss": 0.0316,
+      "step": 4770
+    },
+    {
+      "epoch": 8.597122302158274,
+      "grad_norm": 0.3999459445476532,
+      "learning_rate": 6.547347449033628e-05,
+      "loss": 0.0334,
+      "step": 4780
+    },
+    {
+      "epoch": 8.615107913669064,
+      "grad_norm": 0.46328574419021606,
+      "learning_rate": 6.533201239588474e-05,
+      "loss": 0.0327,
+      "step": 4790
+    },
+    {
+      "epoch": 8.633093525179856,
+      "grad_norm": 0.3814522922039032,
+      "learning_rate": 6.51904147070181e-05,
+      "loss": 0.032,
+      "step": 4800
+    },
+    {
+      "epoch": 8.651079136690647,
+      "grad_norm": 0.4845075309276581,
+      "learning_rate": 6.504868267600879e-05,
+      "loss": 0.0304,
+      "step": 4810
+    },
+    {
+      "epoch": 8.66906474820144,
+      "grad_norm": 0.4320191740989685,
+      "learning_rate": 6.49068175563173e-05,
+      "loss": 0.0346,
+      "step": 4820
+    },
+    {
+      "epoch": 8.68705035971223,
+      "grad_norm": 0.7351819276809692,
+      "learning_rate": 6.476482060258113e-05,
+      "loss": 0.0345,
+      "step": 4830
+    },
+    {
+      "epoch": 8.70503597122302,
+      "grad_norm": 0.6102946996688843,
+      "learning_rate": 6.462269307060375e-05,
+      "loss": 0.0407,
+      "step": 4840
+    },
+    {
+      "epoch": 8.723021582733812,
+      "grad_norm": 0.45671916007995605,
+      "learning_rate": 6.448043621734337e-05,
+      "loss": 0.0301,
+      "step": 4850
+    },
+    {
+      "epoch": 8.741007194244604,
+      "grad_norm": 0.45977166295051575,
+      "learning_rate": 6.433805130090198e-05,
+      "loss": 0.0309,
+      "step": 4860
+    },
+    {
+      "epoch": 8.758992805755396,
+      "grad_norm": 0.6591131091117859,
+      "learning_rate": 6.419553958051413e-05,
+      "loss": 0.0331,
+      "step": 4870
+    },
+    {
+      "epoch": 8.776978417266188,
+      "grad_norm": 0.3430440127849579,
+      "learning_rate": 6.405290231653573e-05,
+      "loss": 0.0307,
+      "step": 4880
+    },
+    {
+      "epoch": 8.79496402877698,
+      "grad_norm": 0.6970073580741882,
+      "learning_rate": 6.39101407704331e-05,
+      "loss": 0.0352,
+      "step": 4890
+    },
+    {
+      "epoch": 8.81294964028777,
+      "grad_norm": 0.5831676721572876,
+      "learning_rate": 6.376725620477159e-05,
+      "loss": 0.0331,
+      "step": 4900
+    },
+    {
+      "epoch": 8.83093525179856,
+      "grad_norm": 0.6279450058937073,
+      "learning_rate": 6.362424988320464e-05,
+      "loss": 0.0382,
+      "step": 4910
+    },
+    {
+      "epoch": 8.848920863309353,
+      "grad_norm": 0.8837076425552368,
+      "learning_rate": 6.348112307046235e-05,
+      "loss": 0.0331,
+      "step": 4920
+    },
+    {
+      "epoch": 8.866906474820144,
+      "grad_norm": 0.4263574481010437,
+      "learning_rate": 6.333787703234048e-05,
+      "loss": 0.0338,
+      "step": 4930
+    },
+    {
+      "epoch": 8.884892086330936,
+      "grad_norm": 0.9537444710731506,
+      "learning_rate": 6.319451303568927e-05,
+      "loss": 0.0321,
+      "step": 4940
+    },
+    {
+      "epoch": 8.902877697841726,
+      "grad_norm": 0.6129856705665588,
+      "learning_rate": 6.305103234840208e-05,
+      "loss": 0.0341,
+      "step": 4950
+    },
+    {
+      "epoch": 8.920863309352518,
+      "grad_norm": 0.5611755847930908,
+      "learning_rate": 6.290743623940431e-05,
+      "loss": 0.0281,
+      "step": 4960
+    },
+    {
+      "epoch": 8.93884892086331,
+      "grad_norm": 0.7212589383125305,
+      "learning_rate": 6.276372597864213e-05,
+      "loss": 0.0398,
+      "step": 4970
+    },
+    {
+      "epoch": 8.956834532374101,
+      "grad_norm": 0.6979126930236816,
+      "learning_rate": 6.261990283707126e-05,
+      "loss": 0.0318,
+      "step": 4980
+    },
+    {
+      "epoch": 8.974820143884893,
+      "grad_norm": 0.5953817367553711,
+      "learning_rate": 6.247596808664575e-05,
+      "loss": 0.0329,
+      "step": 4990
+    },
+    {
+      "epoch": 8.992805755395683,
+      "grad_norm": 0.5423066020011902,
+      "learning_rate": 6.233192300030658e-05,
+      "loss": 0.0331,
+      "step": 5000
+    },
+    {
+      "epoch": 9.010791366906474,
+      "grad_norm": 0.4733891189098358,
+      "learning_rate": 6.21877688519707e-05,
+      "loss": 0.0356,
+      "step": 5010
+    },
+    {
+      "epoch": 9.028776978417266,
+      "grad_norm": 0.4049428105354309,
+      "learning_rate": 6.204350691651948e-05,
+      "loss": 0.0341,
+      "step": 5020
+    },
+    {
+      "epoch": 9.046762589928058,
+      "grad_norm": 0.5135229229927063,
+      "learning_rate": 6.189913846978758e-05,
+      "loss": 0.031,
+      "step": 5030
+    },
+    {
+      "epoch": 9.06474820143885,
+      "grad_norm": 0.6415161490440369,
+      "learning_rate": 6.17546647885516e-05,
+      "loss": 0.0327,
+      "step": 5040
+    },
+    {
+      "epoch": 9.082733812949641,
+      "grad_norm": 1.0162811279296875,
+      "learning_rate": 6.16100871505189e-05,
+      "loss": 0.0353,
+      "step": 5050
+    },
+    {
+      "epoch": 9.100719424460431,
+      "grad_norm": 0.9004434943199158,
+      "learning_rate": 6.146540683431615e-05,
+      "loss": 0.0343,
+      "step": 5060
+    },
+    {
+      "epoch": 9.118705035971223,
+      "grad_norm": 0.8138094544410706,
+      "learning_rate": 6.13206251194781e-05,
+      "loss": 0.03,
+      "step": 5070
+    },
+    {
+      "epoch": 9.136690647482014,
+      "grad_norm": 1.1680430173873901,
+      "learning_rate": 6.117574328643629e-05,
+      "loss": 0.0316,
+      "step": 5080
+    },
+    {
+      "epoch": 9.154676258992806,
+      "grad_norm": 0.6252637505531311,
+      "learning_rate": 6.103076261650764e-05,
+      "loss": 0.0358,
+      "step": 5090
+    },
+    {
+      "epoch": 9.172661870503598,
+      "grad_norm": 0.5080680847167969,
+      "learning_rate": 6.0885684391883244e-05,
+      "loss": 0.0352,
+      "step": 5100
+    },
+    {
+      "epoch": 9.190647482014388,
+      "grad_norm": 0.8690853714942932,
+      "learning_rate": 6.0740509895616904e-05,
+      "loss": 0.0312,
+      "step": 5110
+    },
+    {
+      "epoch": 9.20863309352518,
+      "grad_norm": 0.4944112002849579,
+      "learning_rate": 6.059524041161382e-05,
+      "loss": 0.0323,
+      "step": 5120
+    },
+    {
+      "epoch": 9.226618705035971,
+      "grad_norm": 0.4707288146018982,
+      "learning_rate": 6.044987722461932e-05,
+      "loss": 0.0279,
+      "step": 5130
+    },
+    {
+      "epoch": 9.244604316546763,
+      "grad_norm": 0.5041756629943848,
+      "learning_rate": 6.0304421620207386e-05,
+      "loss": 0.0313,
+      "step": 5140
+    },
+    {
+      "epoch": 9.262589928057555,
+      "grad_norm": 0.4900687336921692,
+      "learning_rate": 6.015887488476932e-05,
+      "loss": 0.032,
+      "step": 5150
+    },
+    {
+      "epoch": 9.280575539568344,
+      "grad_norm": 0.6152610778808594,
+      "learning_rate": 6.0013238305502396e-05,
+      "loss": 0.034,
+      "step": 5160
+    },
+    {
+      "epoch": 9.298561151079136,
+      "grad_norm": 0.9467570185661316,
+      "learning_rate": 5.986751317039845e-05,
+      "loss": 0.0306,
+      "step": 5170
+    },
+    {
+      "epoch": 9.316546762589928,
+      "grad_norm": 0.47657251358032227,
+      "learning_rate": 5.9721700768232504e-05,
+      "loss": 0.0298,
+      "step": 5180
+    },
+    {
+      "epoch": 9.33453237410072,
+      "grad_norm": 0.5181137323379517,
+      "learning_rate": 5.957580238855132e-05,
+      "loss": 0.0358,
+      "step": 5190
+    },
+    {
+      "epoch": 9.352517985611511,
+      "grad_norm": 0.6867415308952332,
+      "learning_rate": 5.9429819321662087e-05,
+      "loss": 0.0361,
+      "step": 5200
+    },
+    {
+      "epoch": 9.370503597122303,
+      "grad_norm": 0.6344670057296753,
+      "learning_rate": 5.928375285862091e-05,
+      "loss": 0.0375,
+      "step": 5210
+    },
+    {
+      "epoch": 9.388489208633093,
+      "grad_norm": 0.5641818046569824,
+      "learning_rate": 5.913760429122151e-05,
+      "loss": 0.0341,
+      "step": 5220
+    },
+    {
+      "epoch": 9.406474820143885,
+      "grad_norm": 0.5127703547477722,
+      "learning_rate": 5.899137491198363e-05,
+      "loss": 0.033,
+      "step": 5230
+    },
+    {
+      "epoch": 9.424460431654676,
+      "grad_norm": 0.7453180551528931,
+      "learning_rate": 5.884506601414178e-05,
+      "loss": 0.0306,
+      "step": 5240
+    },
+    {
+      "epoch": 9.442446043165468,
+      "grad_norm": 0.3943975567817688,
+      "learning_rate": 5.86986788916337e-05,
+      "loss": 0.0315,
+      "step": 5250
+    },
+    {
+      "epoch": 9.46043165467626,
+      "grad_norm": 0.4050646424293518,
+      "learning_rate": 5.855221483908893e-05,
+      "loss": 0.0341,
+      "step": 5260
+    },
+    {
+      "epoch": 9.47841726618705,
+      "grad_norm": 0.35250478982925415,
+      "learning_rate": 5.840567515181736e-05,
+      "loss": 0.0309,
+      "step": 5270
+    },
+    {
+      "epoch": 9.496402877697841,
+      "grad_norm": 0.37676069140434265,
+      "learning_rate": 5.825906112579781e-05,
+      "loss": 0.0329,
+      "step": 5280
+    },
+    {
+      "epoch": 9.514388489208633,
+      "grad_norm": 0.7634493112564087,
+      "learning_rate": 5.811237405766655e-05,
+      "loss": 0.0341,
+      "step": 5290
+    },
+    {
+      "epoch": 9.532374100719425,
+      "grad_norm": 0.5277857184410095,
+      "learning_rate": 5.796561524470579e-05,
+      "loss": 0.0359,
+      "step": 5300
+    },
+    {
+      "epoch": 9.550359712230216,
+      "grad_norm": 0.7182460427284241,
+      "learning_rate": 5.781878598483227e-05,
+      "loss": 0.0299,
+      "step": 5310
+    },
+    {
+      "epoch": 9.568345323741006,
+      "grad_norm": 0.6671009063720703,
+      "learning_rate": 5.767188757658574e-05,
+      "loss": 0.0343,
+      "step": 5320
+    },
+    {
+      "epoch": 9.586330935251798,
+      "grad_norm": 0.35860568284988403,
+      "learning_rate": 5.752492131911749e-05,
+      "loss": 0.0305,
+      "step": 5330
+    },
+    {
+      "epoch": 9.60431654676259,
+      "grad_norm": 0.33959710597991943,
+      "learning_rate": 5.737788851217887e-05,
+      "loss": 0.0333,
+      "step": 5340
+    },
+    {
+      "epoch": 9.622302158273381,
+      "grad_norm": 0.4643886387348175,
+      "learning_rate": 5.723079045610976e-05,
+      "loss": 0.0327,
+      "step": 5350
+    },
+    {
+      "epoch": 9.640287769784173,
+      "grad_norm": 0.6244954466819763,
+      "learning_rate": 5.7083628451827154e-05,
+      "loss": 0.0347,
+      "step": 5360
+    },
+    {
+      "epoch": 9.658273381294965,
+      "grad_norm": 0.4591357409954071,
+      "learning_rate": 5.693640380081352e-05,
+      "loss": 0.0343,
+      "step": 5370
+    },
+    {
+      "epoch": 9.676258992805755,
+      "grad_norm": 0.4514084756374359,
+      "learning_rate": 5.678911780510542e-05,
+      "loss": 0.0309,
+      "step": 5380
+    },
+    {
+      "epoch": 9.694244604316546,
+      "grad_norm": 0.36487889289855957,
+      "learning_rate": 5.664177176728191e-05,
+      "loss": 0.0309,
+      "step": 5390
+    },
+    {
+      "epoch": 9.712230215827338,
+      "grad_norm": 0.38746705651283264,
+      "learning_rate": 5.6494366990453073e-05,
+      "loss": 0.0348,
+      "step": 5400
+    },
+    {
+      "epoch": 9.73021582733813,
+      "grad_norm": 0.6331703066825867,
+      "learning_rate": 5.634690477824849e-05,
+      "loss": 0.0331,
+      "step": 5410
+    },
+    {
+      "epoch": 9.748201438848922,
+      "grad_norm": 0.4631873071193695,
+      "learning_rate": 5.619938643480561e-05,
+      "loss": 0.032,
+      "step": 5420
+    },
+    {
+      "epoch": 9.766187050359711,
+      "grad_norm": 0.4932483434677124,
+      "learning_rate": 5.60518132647584e-05,
+      "loss": 0.0322,
+      "step": 5430
+    },
+    {
+      "epoch": 9.784172661870503,
+      "grad_norm": 0.5406609177589417,
+      "learning_rate": 5.5904186573225645e-05,
+      "loss": 0.0296,
+      "step": 5440
+    },
+    {
+      "epoch": 9.802158273381295,
+      "grad_norm": 0.42935553193092346,
+      "learning_rate": 5.5756507665799494e-05,
+      "loss": 0.0334,
+      "step": 5450
+    },
+    {
+      "epoch": 9.820143884892087,
+      "grad_norm": 2.547171115875244,
+      "learning_rate": 5.560877784853389e-05,
+      "loss": 0.035,
+      "step": 5460
+    },
+    {
+      "epoch": 9.838129496402878,
+      "grad_norm": 0.7417351603507996,
+      "learning_rate": 5.546099842793296e-05,
+      "loss": 0.0304,
+      "step": 5470
+    },
+    {
+      "epoch": 9.85611510791367,
+      "grad_norm": 0.5341687798500061,
+      "learning_rate": 5.5313170710939624e-05,
+      "loss": 0.0322,
+      "step": 5480
+    },
+    {
+      "epoch": 9.87410071942446,
+      "grad_norm": 0.5200015902519226,
+      "learning_rate": 5.516529600492382e-05,
+      "loss": 0.0266,
+      "step": 5490
+    },
+    {
+      "epoch": 9.892086330935252,
+      "grad_norm": 0.4935990273952484,
+      "learning_rate": 5.501737561767112e-05,
+      "loss": 0.029,
+      "step": 5500
+    },
+    {
+      "epoch": 9.910071942446043,
+      "grad_norm": 0.6837783455848694,
+      "learning_rate": 5.486941085737106e-05,
+      "loss": 0.0323,
+      "step": 5510
+    },
+    {
+      "epoch": 9.928057553956835,
+      "grad_norm": 0.3995627164840698,
+      "learning_rate": 5.472140303260562e-05,
+      "loss": 0.0298,
+      "step": 5520
+    },
+    {
+      "epoch": 9.946043165467627,
+      "grad_norm": 0.7096236348152161,
+      "learning_rate": 5.457335345233767e-05,
+      "loss": 0.0365,
+      "step": 5530
+    },
+    {
+      "epoch": 9.964028776978417,
+      "grad_norm": 0.45096665620803833,
+      "learning_rate": 5.442526342589927e-05,
+      "loss": 0.0273,
+      "step": 5540
+    },
+    {
+      "epoch": 9.982014388489208,
+      "grad_norm": 0.4465189576148987,
+      "learning_rate": 5.427713426298028e-05,
+      "loss": 0.0287,
+      "step": 5550
+    },
+    {
+      "epoch": 10.0,
+      "grad_norm": 0.5832507014274597,
+      "learning_rate": 5.4128967273616625e-05,
+      "loss": 0.0338,
+      "step": 5560
+    },
+    {
+      "epoch": 10.017985611510792,
+      "grad_norm": 0.48226141929626465,
+      "learning_rate": 5.398076376817876e-05,
+      "loss": 0.03,
+      "step": 5570
+    },
+    {
+      "epoch": 10.035971223021583,
+      "grad_norm": 0.5855209231376648,
+      "learning_rate": 5.3832525057360114e-05,
+      "loss": 0.0314,
+      "step": 5580
+    },
+    {
+      "epoch": 10.053956834532373,
+      "grad_norm": 0.41245484352111816,
+      "learning_rate": 5.368425245216543e-05,
+      "loss": 0.0332,
+      "step": 5590
+    },
+    {
+      "epoch": 10.071942446043165,
+      "grad_norm": 0.49459171295166016,
+      "learning_rate": 5.3535947263899246e-05,
+      "loss": 0.0331,
+      "step": 5600
+    },
+    {
+      "epoch": 10.089928057553957,
+      "grad_norm": 0.43832942843437195,
+      "learning_rate": 5.338761080415424e-05,
+      "loss": 0.0374,
+      "step": 5610
+    },
+    {
+      "epoch": 10.107913669064748,
+      "grad_norm": 0.3302553594112396,
+      "learning_rate": 5.3239244384799645e-05,
+      "loss": 0.0308,
+      "step": 5620
+    },
+    {
+      "epoch": 10.12589928057554,
+      "grad_norm": 0.603073000907898,
+      "learning_rate": 5.309084931796968e-05,
+      "loss": 0.034,
+      "step": 5630
+    },
+    {
+      "epoch": 10.14388489208633,
+      "grad_norm": 0.2612258195877075,
+      "learning_rate": 5.29424269160519e-05,
+      "loss": 0.033,
+      "step": 5640
+    },
+    {
+      "epoch": 10.161870503597122,
+      "grad_norm": 0.41962552070617676,
+      "learning_rate": 5.27939784916756e-05,
+      "loss": 0.0305,
+      "step": 5650
+    },
+    {
+      "epoch": 10.179856115107913,
+      "grad_norm": 1.393454909324646,
+      "learning_rate": 5.264550535770021e-05,
+      "loss": 0.0309,
+      "step": 5660
+    },
+    {
+      "epoch": 10.197841726618705,
+      "grad_norm": 0.5723445415496826,
+      "learning_rate": 5.2497008827203734e-05,
+      "loss": 0.0316,
+      "step": 5670
+    },
+    {
+      "epoch": 10.215827338129497,
+      "grad_norm": 0.3324255049228668,
+      "learning_rate": 5.2348490213471034e-05,
+      "loss": 0.0316,
+      "step": 5680
+    },
+    {
+      "epoch": 10.233812949640289,
+      "grad_norm": 0.5484040379524231,
+      "learning_rate": 5.219995082998233e-05,
+      "loss": 0.031,
+      "step": 5690
+    },
+    {
+      "epoch": 10.251798561151078,
+      "grad_norm": 0.4781665503978729,
+      "learning_rate": 5.205139199040146e-05,
+      "loss": 0.0317,
+      "step": 5700
+    },
+    {
+      "epoch": 10.26978417266187,
+      "grad_norm": 0.371336430311203,
+      "learning_rate": 5.1902815008564386e-05,
+      "loss": 0.0321,
+      "step": 5710
+    },
+    {
+      "epoch": 10.287769784172662,
+      "grad_norm": 0.5616414546966553,
+      "learning_rate": 5.175422119846752e-05,
+      "loss": 0.033,
+      "step": 5720
+    },
+    {
+      "epoch": 10.305755395683454,
+      "grad_norm": 0.5183143615722656,
+      "learning_rate": 5.160561187425604e-05,
+      "loss": 0.0352,
+      "step": 5730
+    },
+    {
+      "epoch": 10.323741007194245,
+      "grad_norm": 0.37317168712615967,
+      "learning_rate": 5.1456988350212385e-05,
+      "loss": 0.0291,
+      "step": 5740
+    },
+    {
+      "epoch": 10.341726618705035,
+      "grad_norm": 0.49774816632270813,
+      "learning_rate": 5.130835194074457e-05,
+      "loss": 0.0265,
+      "step": 5750
+    },
+    {
+      "epoch": 10.359712230215827,
+      "grad_norm": 0.7472432851791382,
+      "learning_rate": 5.115970396037455e-05,
+      "loss": 0.0274,
+      "step": 5760
+    },
+    {
+      "epoch": 10.377697841726619,
+      "grad_norm": 0.3827255368232727,
+      "learning_rate": 5.10110457237266e-05,
+      "loss": 0.0319,
+      "step": 5770
+    },
+    {
+      "epoch": 10.39568345323741,
+      "grad_norm": 0.7142488956451416,
+      "learning_rate": 5.086237854551572e-05,
+      "loss": 0.0307,
+      "step": 5780
+    },
+    {
+      "epoch": 10.413669064748202,
+      "grad_norm": 0.48390504717826843,
+      "learning_rate": 5.071370374053601e-05,
+      "loss": 0.0312,
+      "step": 5790
+    },
+    {
+      "epoch": 10.431654676258994,
+      "grad_norm": 0.5775423049926758,
+      "learning_rate": 5.0565022623648995e-05,
+      "loss": 0.0288,
+      "step": 5800
+    },
+    {
+      "epoch": 10.449640287769784,
+      "grad_norm": 0.2882794141769409,
+      "learning_rate": 5.0416336509772e-05,
+      "loss": 0.0334,
+      "step": 5810
+    },
+    {
+      "epoch": 10.467625899280575,
+      "grad_norm": 0.4590305984020233,
+      "learning_rate": 5.026764671386658e-05,
+      "loss": 0.0295,
+      "step": 5820
+    },
+    {
+      "epoch": 10.485611510791367,
+      "grad_norm": 0.5438030958175659,
+      "learning_rate": 5.011895455092681e-05,
+      "loss": 0.033,
+      "step": 5830
+    },
+    {
+      "epoch": 10.503597122302159,
+      "grad_norm": 0.5108977556228638,
+      "learning_rate": 4.997026133596778e-05,
+      "loss": 0.0323,
+      "step": 5840
+    },
+    {
+      "epoch": 10.52158273381295,
+      "grad_norm": 0.2965015470981598,
+      "learning_rate": 4.98215683840138e-05,
+      "loss": 0.0372,
+      "step": 5850
+    },
+    {
+      "epoch": 10.53956834532374,
+      "grad_norm": 0.7534518837928772,
+      "learning_rate": 4.967287701008689e-05,
+      "loss": 0.0315,
+      "step": 5860
+    },
+    {
+      "epoch": 10.557553956834532,
+      "grad_norm": 0.5439291596412659,
+      "learning_rate": 4.9524188529195106e-05,
+      "loss": 0.0302,
+      "step": 5870
+    },
+    {
+      "epoch": 10.575539568345324,
+      "grad_norm": 0.5795459151268005,
+      "learning_rate": 4.937550425632093e-05,
+      "loss": 0.0365,
+      "step": 5880
+    },
+    {
+      "epoch": 10.593525179856115,
+      "grad_norm": 0.46216797828674316,
+      "learning_rate": 4.92268255064096e-05,
+      "loss": 0.0305,
+      "step": 5890
+    },
+    {
+      "epoch": 10.611510791366907,
+      "grad_norm": 0.6925672292709351,
+      "learning_rate": 4.907815359435757e-05,
+      "loss": 0.0329,
+      "step": 5900
+    },
+    {
+      "epoch": 10.629496402877697,
+      "grad_norm": 0.664426863193512,
+      "learning_rate": 4.8929489835000746e-05,
+      "loss": 0.0261,
+      "step": 5910
+    },
+    {
+      "epoch": 10.647482014388489,
+      "grad_norm": 0.4442538321018219,
+      "learning_rate": 4.8780835543103e-05,
+      "loss": 0.0323,
+      "step": 5920
+    },
+    {
+      "epoch": 10.66546762589928,
+      "grad_norm": 0.34549370408058167,
+      "learning_rate": 4.863219203334439e-05,
+      "loss": 0.0282,
+      "step": 5930
+    },
+    {
+      "epoch": 10.683453237410072,
+      "grad_norm": 0.6068468689918518,
+      "learning_rate": 4.848356062030973e-05,
+      "loss": 0.0268,
+      "step": 5940
+    },
+    {
+      "epoch": 10.701438848920864,
+      "grad_norm": 0.4484556019306183,
+      "learning_rate": 4.833494261847677e-05,
+      "loss": 0.026,
+      "step": 5950
+    },
+    {
+      "epoch": 10.719424460431654,
+      "grad_norm": 0.9311725497245789,
+      "learning_rate": 4.818633934220467e-05,
+      "loss": 0.0295,
+      "step": 5960
+    },
+    {
+      "epoch": 10.737410071942445,
+      "grad_norm": 0.3421158194541931,
+      "learning_rate": 4.803775210572236e-05,
+      "loss": 0.0269,
+      "step": 5970
+    },
+    {
+      "epoch": 10.755395683453237,
+      "grad_norm": 0.4824845790863037,
+      "learning_rate": 4.788918222311692e-05,
+      "loss": 0.0274,
+      "step": 5980
+    },
+    {
+      "epoch": 10.773381294964029,
+      "grad_norm": 0.5051682591438293,
+      "learning_rate": 4.774063100832199e-05,
+      "loss": 0.033,
+      "step": 5990
+    },
+    {
+      "epoch": 10.79136690647482,
+      "grad_norm": 0.47549042105674744,
+      "learning_rate": 4.7592099775106045e-05,
+      "loss": 0.0288,
+      "step": 6000
+    },
+    {
+      "epoch": 10.809352517985612,
+      "grad_norm": 0.7458439469337463,
+      "learning_rate": 4.74435898370609e-05,
+      "loss": 0.0223,
+      "step": 6010
+    },
+    {
+      "epoch": 10.827338129496402,
+      "grad_norm": 0.781338095664978,
+      "learning_rate": 4.7295102507589995e-05,
+      "loss": 0.0357,
+      "step": 6020
+    },
+    {
+      "epoch": 10.845323741007194,
+      "grad_norm": 0.2663012742996216,
+      "learning_rate": 4.7146639099896864e-05,
+      "loss": 0.0292,
+      "step": 6030
+    },
+    {
+      "epoch": 10.863309352517986,
+      "grad_norm": 0.6557360887527466,
+      "learning_rate": 4.699820092697346e-05,
+      "loss": 0.0374,
+      "step": 6040
+    },
+    {
+      "epoch": 10.881294964028777,
+      "grad_norm": 0.37071770429611206,
+      "learning_rate": 4.684978930158856e-05,
+      "loss": 0.0266,
+      "step": 6050
+    },
+    {
+      "epoch": 10.899280575539569,
+      "grad_norm": 0.3645109534263611,
+      "learning_rate": 4.670140553627616e-05,
+      "loss": 0.0299,
+      "step": 6060
+    },
+    {
+      "epoch": 10.917266187050359,
+      "grad_norm": 0.3512324094772339,
+      "learning_rate": 4.6553050943323886e-05,
+      "loss": 0.0253,
+      "step": 6070
+    },
+    {
+      "epoch": 10.93525179856115,
+      "grad_norm": 0.45498642325401306,
+      "learning_rate": 4.640472683476132e-05,
+      "loss": 0.0293,
+      "step": 6080
+    },
+    {
+      "epoch": 10.953237410071942,
+      "grad_norm": 0.5903925895690918,
+      "learning_rate": 4.6256434522348494e-05,
+      "loss": 0.0281,
+      "step": 6090
+    },
+    {
+      "epoch": 10.971223021582734,
+      "grad_norm": 0.5731881856918335,
+      "learning_rate": 4.610817531756421e-05,
+      "loss": 0.0274,
+      "step": 6100
+    },
+    {
+      "epoch": 10.989208633093526,
+      "grad_norm": 0.4180248975753784,
+      "learning_rate": 4.595995053159448e-05,
+      "loss": 0.0269,
+      "step": 6110
+    },
+    {
+      "epoch": 11.007194244604317,
+      "grad_norm": 0.4542294442653656,
+      "learning_rate": 4.581176147532091e-05,
+      "loss": 0.0306,
+      "step": 6120
+    },
+    {
+      "epoch": 11.025179856115107,
+      "grad_norm": 0.548284113407135,
+      "learning_rate": 4.566360945930913e-05,
+      "loss": 0.0252,
+      "step": 6130
+    },
+    {
+      "epoch": 11.043165467625899,
+      "grad_norm": 0.5517229437828064,
+      "learning_rate": 4.55154957937972e-05,
+      "loss": 0.028,
+      "step": 6140
+    },
+    {
+      "epoch": 11.06115107913669,
+      "grad_norm": 0.41214630007743835,
+      "learning_rate": 4.5367421788683984e-05,
+      "loss": 0.0254,
+      "step": 6150
+    },
+    {
+      "epoch": 11.079136690647482,
+      "grad_norm": 0.9225776791572571,
+      "learning_rate": 4.52193887535176e-05,
+      "loss": 0.0309,
+      "step": 6160
+    },
+    {
+      "epoch": 11.097122302158274,
+      "grad_norm": 0.5597783327102661,
+      "learning_rate": 4.507139799748383e-05,
+      "loss": 0.0298,
+      "step": 6170
+    },
+    {
+      "epoch": 11.115107913669064,
+      "grad_norm": 0.5121003985404968,
+      "learning_rate": 4.4923450829394605e-05,
+      "loss": 0.0292,
+      "step": 6180
+    },
+    {
+      "epoch": 11.133093525179856,
+      "grad_norm": 0.4672529995441437,
+      "learning_rate": 4.4775548557676284e-05,
+      "loss": 0.0295,
+      "step": 6190
+    },
+    {
+      "epoch": 11.151079136690647,
+      "grad_norm": 0.6159136295318604,
+      "learning_rate": 4.462769249035821e-05,
+      "loss": 0.0313,
+      "step": 6200
+    },
+    {
+      "epoch": 11.16906474820144,
+      "grad_norm": 0.3899865746498108,
+      "learning_rate": 4.44798839350611e-05,
+      "loss": 0.0278,
+      "step": 6210
+    },
+    {
+      "epoch": 11.18705035971223,
+      "grad_norm": 0.8864157199859619,
+      "learning_rate": 4.433212419898546e-05,
+      "loss": 0.0316,
+      "step": 6220
+    },
+    {
+      "epoch": 11.20503597122302,
+      "grad_norm": 0.5556218028068542,
+      "learning_rate": 4.418441458890009e-05,
+      "loss": 0.0318,
+      "step": 6230
+    },
+    {
+      "epoch": 11.223021582733812,
+      "grad_norm": 0.6662029027938843,
+      "learning_rate": 4.403675641113043e-05,
+      "loss": 0.0288,
+      "step": 6240
+    },
+    {
+      "epoch": 11.241007194244604,
+      "grad_norm": 0.6300651431083679,
+      "learning_rate": 4.3889150971547085e-05,
+      "loss": 0.0294,
+      "step": 6250
+    },
+    {
+      "epoch": 11.258992805755396,
+      "grad_norm": 0.28359514474868774,
+      "learning_rate": 4.3741599575554245e-05,
+      "loss": 0.039,
+      "step": 6260
+    },
+    {
+      "epoch": 11.276978417266188,
+      "grad_norm": 0.49965447187423706,
+      "learning_rate": 4.359410352807815e-05,
+      "loss": 0.0308,
+      "step": 6270
+    },
+    {
+      "epoch": 11.29496402877698,
+      "grad_norm": 0.6804771423339844,
+      "learning_rate": 4.344666413355554e-05,
+      "loss": 0.0338,
+      "step": 6280
+    },
+    {
+      "epoch": 11.31294964028777,
+      "grad_norm": 0.5171783566474915,
+      "learning_rate": 4.329928269592213e-05,
+      "loss": 0.0358,
+      "step": 6290
+    },
+    {
+      "epoch": 11.33093525179856,
+      "grad_norm": 0.7342168688774109,
+      "learning_rate": 4.315196051860106e-05,
+      "loss": 0.0324,
+      "step": 6300
+    },
+    {
+      "epoch": 11.348920863309353,
+      "grad_norm": 0.4210442006587982,
+      "learning_rate": 4.300469890449137e-05,
+      "loss": 0.0281,
+      "step": 6310
+    },
+    {
+      "epoch": 11.366906474820144,
+      "grad_norm": 0.5872474312782288,
+      "learning_rate": 4.285749915595653e-05,
+      "loss": 0.024,
+      "step": 6320
+    },
+    {
+      "epoch": 11.384892086330936,
+      "grad_norm": 0.5810453295707703,
+      "learning_rate": 4.2710362574812834e-05,
+      "loss": 0.0277,
+      "step": 6330
+    },
+    {
+      "epoch": 11.402877697841726,
+      "grad_norm": 0.4863584041595459,
+      "learning_rate": 4.2563290462317954e-05,
+      "loss": 0.0284,
+      "step": 6340
+    },
+    {
+      "epoch": 11.420863309352518,
+      "grad_norm": 0.3982340395450592,
+      "learning_rate": 4.241628411915941e-05,
+      "loss": 0.0323,
+      "step": 6350
+    },
+    {
+      "epoch": 11.43884892086331,
+      "grad_norm": 0.5989066362380981,
+      "learning_rate": 4.2269344845443004e-05,
+      "loss": 0.0278,
+      "step": 6360
+    },
+    {
+      "epoch": 11.456834532374101,
+      "grad_norm": 0.3513637185096741,
+      "learning_rate": 4.2122473940681514e-05,
+      "loss": 0.0296,
+      "step": 6370
+    },
+    {
+      "epoch": 11.474820143884893,
+      "grad_norm": 0.721295177936554,
+      "learning_rate": 4.197567270378297e-05,
+      "loss": 0.0263,
+      "step": 6380
+    },
+    {
+      "epoch": 11.492805755395683,
+      "grad_norm": 0.37077784538269043,
+      "learning_rate": 4.1828942433039276e-05,
+      "loss": 0.0284,
+      "step": 6390
+    },
+    {
+      "epoch": 11.510791366906474,
+      "grad_norm": 0.46234238147735596,
+      "learning_rate": 4.168228442611476e-05,
+      "loss": 0.0277,
+      "step": 6400
+    },
+    {
+      "epoch": 11.528776978417266,
+      "grad_norm": 0.41237565875053406,
+      "learning_rate": 4.153569998003461e-05,
+      "loss": 0.0317,
+      "step": 6410
+    },
+    {
+      "epoch": 11.546762589928058,
+      "grad_norm": 0.28192874789237976,
+      "learning_rate": 4.13891903911735e-05,
+      "loss": 0.028,
+      "step": 6420
+    },
+    {
+      "epoch": 11.56474820143885,
+      "grad_norm": 0.4698995351791382,
+      "learning_rate": 4.124275695524406e-05,
+      "loss": 0.0278,
+      "step": 6430
+    },
+    {
+      "epoch": 11.582733812949641,
+      "grad_norm": 0.41778913140296936,
+      "learning_rate": 4.109640096728541e-05,
+      "loss": 0.0275,
+      "step": 6440
+    },
+    {
+      "epoch": 11.600719424460431,
+      "grad_norm": 0.7228760719299316,
+      "learning_rate": 4.095012372165174e-05,
+      "loss": 0.0266,
+      "step": 6450
+    },
+    {
+      "epoch": 11.618705035971223,
+      "grad_norm": 0.6006641387939453,
+      "learning_rate": 4.080392651200088e-05,
+      "loss": 0.0262,
+      "step": 6460
+    },
+    {
+      "epoch": 11.636690647482014,
+      "grad_norm": 0.4687000513076782,
+      "learning_rate": 4.06578106312828e-05,
+      "loss": 0.0296,
+      "step": 6470
+    },
+    {
+      "epoch": 11.654676258992806,
+      "grad_norm": 0.420821875333786,
+      "learning_rate": 4.051177737172821e-05,
+      "loss": 0.0264,
+      "step": 6480
+    },
+    {
+      "epoch": 11.672661870503598,
+      "grad_norm": 0.5298236012458801,
+      "learning_rate": 4.036582802483716e-05,
+      "loss": 0.0269,
+      "step": 6490
+    },
+    {
+      "epoch": 11.690647482014388,
+      "grad_norm": 0.461881548166275,
+      "learning_rate": 4.0219963881367526e-05,
+      "loss": 0.0261,
+      "step": 6500
+    },
+    {
+      "epoch": 11.70863309352518,
+      "grad_norm": 0.6257002353668213,
+      "learning_rate": 4.007418623132372e-05,
+      "loss": 0.0322,
+      "step": 6510
+    },
+    {
+      "epoch": 11.726618705035971,
+      "grad_norm": 0.664402425289154,
+      "learning_rate": 3.9928496363945187e-05,
+      "loss": 0.0315,
+      "step": 6520
+    },
+    {
+      "epoch": 11.744604316546763,
+      "grad_norm": 0.8453108668327332,
+      "learning_rate": 3.9782895567695035e-05,
+      "loss": 0.0292,
+      "step": 6530
+    },
+    {
+      "epoch": 11.762589928057555,
+      "grad_norm": 0.5592852234840393,
+      "learning_rate": 3.9637385130248636e-05,
+      "loss": 0.0266,
+      "step": 6540
+    },
+    {
+      "epoch": 11.780575539568346,
+      "grad_norm": 0.5482608079910278,
+      "learning_rate": 3.9491966338482214e-05,
+      "loss": 0.0297,
+      "step": 6550
+    },
+    {
+      "epoch": 11.798561151079136,
+      "grad_norm": 0.6847444772720337,
+      "learning_rate": 3.934664047846156e-05,
+      "loss": 0.0297,
+      "step": 6560
+    },
+    {
+      "epoch": 11.816546762589928,
+      "grad_norm": 0.6780993342399597,
+      "learning_rate": 3.920140883543053e-05,
+      "loss": 0.0329,
+      "step": 6570
+    },
+    {
+      "epoch": 11.83453237410072,
+      "grad_norm": 0.41092994809150696,
+      "learning_rate": 3.9056272693799765e-05,
+      "loss": 0.0357,
+      "step": 6580
+    },
+    {
+      "epoch": 11.852517985611511,
+      "grad_norm": 0.35687899589538574,
+      "learning_rate": 3.891123333713528e-05,
+      "loss": 0.0242,
+      "step": 6590
+    },
+    {
+      "epoch": 11.870503597122303,
+      "grad_norm": 0.35618337988853455,
+      "learning_rate": 3.876629204814712e-05,
+      "loss": 0.0273,
+      "step": 6600
+    },
+    {
+      "epoch": 11.888489208633093,
+      "grad_norm": 0.423078715801239,
+      "learning_rate": 3.8621450108678114e-05,
+      "loss": 0.027,
+      "step": 6610
+    },
+    {
+      "epoch": 11.906474820143885,
+      "grad_norm": 0.2996216118335724,
+      "learning_rate": 3.847670879969237e-05,
+      "loss": 0.0244,
+      "step": 6620
+    },
+    {
+      "epoch": 11.924460431654676,
+      "grad_norm": 0.9744521975517273,
+      "learning_rate": 3.833206940126406e-05,
+      "loss": 0.0324,
+      "step": 6630
+    },
+    {
+      "epoch": 11.942446043165468,
+      "grad_norm": 0.6271923184394836,
+      "learning_rate": 3.818753319256608e-05,
+      "loss": 0.0275,
+      "step": 6640
+    },
+    {
+      "epoch": 11.96043165467626,
+      "grad_norm": 0.39131006598472595,
+      "learning_rate": 3.804310145185873e-05,
+      "loss": 0.0247,
+      "step": 6650
+    },
+    {
+      "epoch": 11.97841726618705,
+      "grad_norm": 0.42249709367752075,
+      "learning_rate": 3.7898775456478404e-05,
+      "loss": 0.0293,
+      "step": 6660
+    },
+    {
+      "epoch": 11.996402877697841,
+      "grad_norm": 0.5441145896911621,
+      "learning_rate": 3.775455648282629e-05,
+      "loss": 0.025,
+      "step": 6670
+    },
+    {
+      "epoch": 12.014388489208633,
+      "grad_norm": 0.4512350559234619,
+      "learning_rate": 3.7610445806357095e-05,
+      "loss": 0.0304,
+      "step": 6680
+    },
+    {
+      "epoch": 12.032374100719425,
+      "grad_norm": 0.7702916264533997,
+      "learning_rate": 3.746644470156775e-05,
+      "loss": 0.0276,
+      "step": 6690
+    },
+    {
+      "epoch": 12.050359712230216,
+      "grad_norm": 1.137279987335205,
+      "learning_rate": 3.732255444198617e-05,
+      "loss": 0.0278,
+      "step": 6700
+    },
+    {
+      "epoch": 12.068345323741006,
+      "grad_norm": 0.5348918437957764,
+      "learning_rate": 3.7178776300159953e-05,
+      "loss": 0.0276,
+      "step": 6710
+    },
+    {
+      "epoch": 12.086330935251798,
+      "grad_norm": 0.7446044087409973,
+      "learning_rate": 3.703511154764514e-05,
+      "loss": 0.0324,
+      "step": 6720
+    },
+    {
+      "epoch": 12.10431654676259,
+      "grad_norm": 0.5612084269523621,
+      "learning_rate": 3.689156145499498e-05,
+      "loss": 0.0301,
+      "step": 6730
+    },
+    {
+      "epoch": 12.122302158273381,
+      "grad_norm": 0.3164941966533661,
+      "learning_rate": 3.674812729174867e-05,
+      "loss": 0.0305,
+      "step": 6740
+    },
+    {
+      "epoch": 12.140287769784173,
+      "grad_norm": 0.2882361114025116,
+      "learning_rate": 3.660481032642016e-05,
+      "loss": 0.0286,
+      "step": 6750
+    },
+    {
+      "epoch": 12.158273381294965,
+      "grad_norm": 0.8244530558586121,
+      "learning_rate": 3.646161182648691e-05,
+      "loss": 0.0244,
+      "step": 6760
+    },
+    {
+      "epoch": 12.176258992805755,
+      "grad_norm": 0.4218905568122864,
+      "learning_rate": 3.631853305837868e-05,
+      "loss": 0.0265,
+      "step": 6770
+    },
+    {
+      "epoch": 12.194244604316546,
+      "grad_norm": 1.2407538890838623,
+      "learning_rate": 3.617557528746633e-05,
+      "loss": 0.0273,
+      "step": 6780
+    },
+    {
+      "epoch": 12.212230215827338,
+      "grad_norm": 0.6009521484375,
+      "learning_rate": 3.6032739778050636e-05,
+      "loss": 0.0296,
+      "step": 6790
+    },
+    {
+      "epoch": 12.23021582733813,
+      "grad_norm": 0.6228532195091248,
+      "learning_rate": 3.5890027793351154e-05,
+      "loss": 0.034,
+      "step": 6800
+    },
+    {
+      "epoch": 12.248201438848922,
+      "grad_norm": 0.6862697601318359,
+      "learning_rate": 3.5747440595494966e-05,
+      "loss": 0.0279,
+      "step": 6810
+    },
+    {
+      "epoch": 12.266187050359711,
+      "grad_norm": 0.6849074363708496,
+      "learning_rate": 3.560497944550552e-05,
+      "loss": 0.0266,
+      "step": 6820
+    },
+    {
+      "epoch": 12.284172661870503,
+      "grad_norm": 0.4879355728626251,
+      "learning_rate": 3.546264560329156e-05,
+      "loss": 0.0273,
+      "step": 6830
+    },
+    {
+      "epoch": 12.302158273381295,
+      "grad_norm": 0.34242185950279236,
+      "learning_rate": 3.532044032763595e-05,
+      "loss": 0.0256,
+      "step": 6840
+    },
+    {
+      "epoch": 12.320143884892087,
+      "grad_norm": 0.3276647925376892,
+      "learning_rate": 3.5178364876184484e-05,
+      "loss": 0.0263,
+      "step": 6850
+    },
+    {
+      "epoch": 12.338129496402878,
+      "grad_norm": 0.7355037927627563,
+      "learning_rate": 3.503642050543482e-05,
+      "loss": 0.0264,
+      "step": 6860
+    },
+    {
+      "epoch": 12.35611510791367,
+      "grad_norm": 0.4725399911403656,
+      "learning_rate": 3.4894608470725365e-05,
+      "loss": 0.0328,
+      "step": 6870
+    },
+    {
+      "epoch": 12.37410071942446,
+      "grad_norm": 0.3546625077724457,
+      "learning_rate": 3.475293002622415e-05,
+      "loss": 0.0296,
+      "step": 6880
+    },
+    {
+      "epoch": 12.392086330935252,
+      "grad_norm": 0.5498871803283691,
+      "learning_rate": 3.461138642491777e-05,
+      "loss": 0.0339,
+      "step": 6890
+    },
+    {
+      "epoch": 12.410071942446043,
+      "grad_norm": 0.6432085037231445,
+      "learning_rate": 3.446997891860025e-05,
+      "loss": 0.0268,
+      "step": 6900
+    },
+    {
+      "epoch": 12.428057553956835,
+      "grad_norm": 0.46491846442222595,
+      "learning_rate": 3.432870875786206e-05,
+      "loss": 0.0314,
+      "step": 6910
+    },
+    {
+      "epoch": 12.446043165467627,
+      "grad_norm": 0.3839486539363861,
+      "learning_rate": 3.4187577192078946e-05,
+      "loss": 0.0322,
+      "step": 6920
+    },
+    {
+      "epoch": 12.464028776978417,
+      "grad_norm": 1.3864015340805054,
+      "learning_rate": 3.404658546940095e-05,
+      "loss": 0.0272,
+      "step": 6930
+    },
+    {
+      "epoch": 12.482014388489208,
+      "grad_norm": 0.6126661896705627,
+      "learning_rate": 3.390573483674142e-05,
+      "loss": 0.0274,
+      "step": 6940
+    },
+    {
+      "epoch": 12.5,
+      "grad_norm": 0.5150448083877563,
+      "learning_rate": 3.3765026539765834e-05,
+      "loss": 0.0274,
+      "step": 6950
+    },
+    {
+      "epoch": 12.517985611510792,
+      "grad_norm": 0.3786114454269409,
+      "learning_rate": 3.362446182288092e-05,
+      "loss": 0.0263,
+      "step": 6960
+    },
+    {
+      "epoch": 12.535971223021583,
+      "grad_norm": 0.5194931626319885,
+      "learning_rate": 3.3484041929223595e-05,
+      "loss": 0.0244,
+      "step": 6970
+    },
+    {
+      "epoch": 12.553956834532373,
+      "grad_norm": 0.3609524965286255,
+      "learning_rate": 3.334376810064994e-05,
+      "loss": 0.0295,
+      "step": 6980
+    },
+    {
+      "epoch": 12.571942446043165,
+      "grad_norm": 0.5378658175468445,
+      "learning_rate": 3.3203641577724345e-05,
+      "loss": 0.03,
+      "step": 6990
+    },
+    {
+      "epoch": 12.589928057553957,
+      "grad_norm": 1.0792741775512695,
+      "learning_rate": 3.306366359970837e-05,
+      "loss": 0.0295,
+      "step": 7000
+    },
+    {
+      "epoch": 12.607913669064748,
+      "grad_norm": 0.34386754035949707,
+      "learning_rate": 3.2923835404549875e-05,
+      "loss": 0.0303,
+      "step": 7010
+    },
+    {
+      "epoch": 12.62589928057554,
+      "grad_norm": 0.7501177191734314,
+      "learning_rate": 3.278415822887203e-05,
+      "loss": 0.0228,
+      "step": 7020
+    },
+    {
+      "epoch": 12.64388489208633,
+      "grad_norm": 1.202082872390747,
+      "learning_rate": 3.264463330796251e-05,
+      "loss": 0.0278,
+      "step": 7030
+    },
+    {
+      "epoch": 12.661870503597122,
+      "grad_norm": 0.5245298743247986,
+      "learning_rate": 3.2505261875762396e-05,
+      "loss": 0.0336,
+      "step": 7040
+    },
+    {
+      "epoch": 12.679856115107913,
+      "grad_norm": 0.41505035758018494,
+      "learning_rate": 3.236604516485529e-05,
+      "loss": 0.0343,
+      "step": 7050
+    },
+    {
+      "epoch": 12.697841726618705,
+      "grad_norm": 0.7059419751167297,
+      "learning_rate": 3.222698440645654e-05,
+      "loss": 0.0313,
+      "step": 7060
+    },
+    {
+      "epoch": 12.715827338129497,
+      "grad_norm": 0.5493305921554565,
+      "learning_rate": 3.208808083040222e-05,
+      "loss": 0.0289,
+      "step": 7070
+    },
+    {
+      "epoch": 12.733812949640289,
+      "grad_norm": 0.5413309931755066,
+      "learning_rate": 3.1949335665138346e-05,
+      "loss": 0.0256,
+      "step": 7080
+    },
+    {
+      "epoch": 12.751798561151078,
+      "grad_norm": 0.5997304916381836,
+      "learning_rate": 3.1810750137709926e-05,
+      "loss": 0.0268,
+      "step": 7090
+    },
+    {
+      "epoch": 12.76978417266187,
+      "grad_norm": 0.5178503394126892,
+      "learning_rate": 3.1672325473750176e-05,
+      "loss": 0.0244,
+      "step": 7100
+    },
+    {
+      "epoch": 12.787769784172662,
+      "grad_norm": 0.9184396266937256,
+      "learning_rate": 3.153406289746963e-05,
+      "loss": 0.0257,
+      "step": 7110
+    },
+    {
+      "epoch": 12.805755395683454,
+      "grad_norm": 0.7705658078193665,
+      "learning_rate": 3.1395963631645365e-05,
+      "loss": 0.0263,
+      "step": 7120
+    },
+    {
+      "epoch": 12.823741007194245,
+      "grad_norm": 0.5064858794212341,
+      "learning_rate": 3.125802889761016e-05,
+      "loss": 0.0288,
+      "step": 7130
+    },
+    {
+      "epoch": 12.841726618705035,
+      "grad_norm": 0.3358422815799713,
+      "learning_rate": 3.112025991524167e-05,
+      "loss": 0.0307,
+      "step": 7140
+    },
+    {
+      "epoch": 12.859712230215827,
+      "grad_norm": 0.38877683877944946,
+      "learning_rate": 3.09826579029517e-05,
+      "loss": 0.028,
+      "step": 7150
+    },
+    {
+      "epoch": 12.877697841726619,
+      "grad_norm": 0.3808923661708832,
+      "learning_rate": 3.084522407767536e-05,
+      "loss": 0.0275,
+      "step": 7160
+    },
+    {
+      "epoch": 12.89568345323741,
+      "grad_norm": 0.37734875082969666,
+      "learning_rate": 3.070795965486033e-05,
+      "loss": 0.0257,
+      "step": 7170
+    },
+    {
+      "epoch": 12.913669064748202,
+      "grad_norm": 0.9107422828674316,
+      "learning_rate": 3.057086584845615e-05,
+      "loss": 0.0306,
+      "step": 7180
+    },
+    {
+      "epoch": 12.931654676258994,
+      "grad_norm": 0.5136073231697083,
+      "learning_rate": 3.043394387090344e-05,
+      "loss": 0.0309,
+      "step": 7190
+    },
+    {
+      "epoch": 12.949640287769784,
+      "grad_norm": 0.4331435263156891,
+      "learning_rate": 3.029719493312318e-05,
+      "loss": 0.0262,
+      "step": 7200
+    },
+    {
+      "epoch": 12.967625899280575,
+      "grad_norm": 0.4082922041416168,
+      "learning_rate": 3.016062024450596e-05,
+      "loss": 0.0223,
+      "step": 7210
+    },
+    {
+      "epoch": 12.985611510791367,
+      "grad_norm": 0.6174339056015015,
+      "learning_rate": 3.0024221012901434e-05,
+      "loss": 0.0237,
+      "step": 7220
+    },
+    {
+      "epoch": 13.003597122302159,
+      "grad_norm": 0.34224146604537964,
+      "learning_rate": 2.9887998444607472e-05,
+      "loss": 0.0286,
+      "step": 7230
+    },
+    {
+      "epoch": 13.02158273381295,
+      "grad_norm": 0.4418945014476776,
+      "learning_rate": 2.975195374435954e-05,
+      "loss": 0.0243,
+      "step": 7240
+    },
+    {
+      "epoch": 13.03956834532374,
+      "grad_norm": 0.5699992179870605,
+      "learning_rate": 2.96160881153201e-05,
+      "loss": 0.0247,
+      "step": 7250
+    },
+    {
+      "epoch": 13.057553956834532,
+      "grad_norm": 0.33255311846733093,
+      "learning_rate": 2.9480402759067893e-05,
+      "loss": 0.0271,
+      "step": 7260
+    },
+    {
+      "epoch": 13.075539568345324,
+      "grad_norm": 0.4527534544467926,
+      "learning_rate": 2.934489887558742e-05,
+      "loss": 0.0268,
+      "step": 7270
+    },
+    {
+      "epoch": 13.093525179856115,
+      "grad_norm": 0.5959901213645935,
+      "learning_rate": 2.920957766325817e-05,
+      "loss": 0.0265,
+      "step": 7280
+    },
+    {
+      "epoch": 13.111510791366907,
+      "grad_norm": 0.6262191534042358,
+      "learning_rate": 2.907444031884416e-05,
+      "loss": 0.0274,
+      "step": 7290
+    },
+    {
+      "epoch": 13.129496402877697,
+      "grad_norm": 0.38675346970558167,
+      "learning_rate": 2.8939488037483332e-05,
+      "loss": 0.0314,
+      "step": 7300
+    },
+    {
+      "epoch": 13.147482014388489,
+      "grad_norm": 0.6364177465438843,
+      "learning_rate": 2.8804722012676843e-05,
+      "loss": 0.0322,
+      "step": 7310
+    },
+    {
+      "epoch": 13.16546762589928,
+      "grad_norm": 0.5858877897262573,
+      "learning_rate": 2.8670143436278756e-05,
+      "loss": 0.0243,
+      "step": 7320
+    },
+    {
+      "epoch": 13.183453237410072,
+      "grad_norm": 0.4976436495780945,
+      "learning_rate": 2.853575349848528e-05,
+      "loss": 0.0272,
+      "step": 7330
+    },
+    {
+      "epoch": 13.201438848920864,
+      "grad_norm": 0.40862083435058594,
+      "learning_rate": 2.840155338782435e-05,
+      "loss": 0.0295,
+      "step": 7340
+    },
+    {
+      "epoch": 13.219424460431656,
+      "grad_norm": 0.3103615343570709,
+      "learning_rate": 2.8267544291145086e-05,
+      "loss": 0.0232,
+      "step": 7350
+    },
+    {
+      "epoch": 13.237410071942445,
+      "grad_norm": 0.6644999980926514,
+      "learning_rate": 2.8133727393607312e-05,
+      "loss": 0.0247,
+      "step": 7360
+    },
+    {
+      "epoch": 13.255395683453237,
+      "grad_norm": 0.5488556027412415,
+      "learning_rate": 2.8000103878671076e-05,
+      "loss": 0.0229,
+      "step": 7370
+    },
+    {
+      "epoch": 13.273381294964029,
+      "grad_norm": 0.3096403479576111,
+      "learning_rate": 2.7866674928086144e-05,
+      "loss": 0.0289,
+      "step": 7380
+    },
+    {
+      "epoch": 13.29136690647482,
+      "grad_norm": 1.0209321975708008,
+      "learning_rate": 2.773344172188161e-05,
+      "loss": 0.0306,
+      "step": 7390
+    },
+    {
+      "epoch": 13.309352517985612,
+      "grad_norm": 0.8741286396980286,
+      "learning_rate": 2.760040543835539e-05,
+      "loss": 0.028,
+      "step": 7400
+    },
+    {
+      "epoch": 13.327338129496402,
+      "grad_norm": 0.34399858117103577,
+      "learning_rate": 2.7467567254063915e-05,
+      "loss": 0.025,
+      "step": 7410
+    },
+    {
+      "epoch": 13.345323741007194,
+      "grad_norm": 0.5274016857147217,
+      "learning_rate": 2.7334928343811573e-05,
+      "loss": 0.0282,
+      "step": 7420
+    },
+    {
+      "epoch": 13.363309352517986,
+      "grad_norm": 0.5535068511962891,
+      "learning_rate": 2.7202489880640435e-05,
+      "loss": 0.0265,
+      "step": 7430
+    },
+    {
+      "epoch": 13.381294964028777,
+      "grad_norm": 0.3902623951435089,
+      "learning_rate": 2.7070253035819833e-05,
+      "loss": 0.0305,
+      "step": 7440
+    },
+    {
+      "epoch": 13.399280575539569,
+      "grad_norm": 0.4680734872817993,
+      "learning_rate": 2.6938218978836005e-05,
+      "loss": 0.0273,
+      "step": 7450
+    },
+    {
+      "epoch": 13.417266187050359,
+      "grad_norm": 0.5480627417564392,
+      "learning_rate": 2.6806388877381784e-05,
+      "loss": 0.0274,
+      "step": 7460
+    },
+    {
+      "epoch": 13.43525179856115,
+      "grad_norm": 0.48696663975715637,
+      "learning_rate": 2.6674763897346213e-05,
+      "loss": 0.0279,
+      "step": 7470
+    },
+    {
+      "epoch": 13.453237410071942,
+      "grad_norm": 1.0533092021942139,
+      "learning_rate": 2.6543345202804282e-05,
+      "loss": 0.0294,
+      "step": 7480
+    },
+    {
+      "epoch": 13.471223021582734,
+      "grad_norm": 0.762681782245636,
+      "learning_rate": 2.6412133956006624e-05,
+      "loss": 0.0285,
+      "step": 7490
+    },
+    {
+      "epoch": 13.489208633093526,
+      "grad_norm": 0.41194671392440796,
+      "learning_rate": 2.628113131736921e-05,
+      "loss": 0.0277,
+      "step": 7500
+    },
+    {
+      "epoch": 13.507194244604317,
+      "grad_norm": 0.7451456785202026,
+      "learning_rate": 2.6150338445463145e-05,
+      "loss": 0.0285,
+      "step": 7510
+    },
+    {
+      "epoch": 13.525179856115107,
+      "grad_norm": 0.7570236921310425,
+      "learning_rate": 2.6019756497004365e-05,
+      "loss": 0.029,
+      "step": 7520
+    },
+    {
+      "epoch": 13.543165467625899,
+      "grad_norm": 0.4345681965351105,
+      "learning_rate": 2.588938662684338e-05,
+      "loss": 0.0297,
+      "step": 7530
+    },
+    {
+      "epoch": 13.56115107913669,
+      "grad_norm": 0.299492746591568,
+      "learning_rate": 2.575922998795517e-05,
+      "loss": 0.0236,
+      "step": 7540
+    },
+    {
+      "epoch": 13.579136690647482,
+      "grad_norm": 0.5992304682731628,
+      "learning_rate": 2.5629287731428864e-05,
+      "loss": 0.0257,
+      "step": 7550
+    },
+    {
+      "epoch": 13.597122302158274,
+      "grad_norm": 0.40048569440841675,
+      "learning_rate": 2.5499561006457693e-05,
+      "loss": 0.0233,
+      "step": 7560
+    },
+    {
+      "epoch": 13.615107913669064,
+      "grad_norm": 0.500738799571991,
+      "learning_rate": 2.5370050960328716e-05,
+      "loss": 0.0275,
+      "step": 7570
+    },
+    {
+      "epoch": 13.633093525179856,
+      "grad_norm": 0.3336072564125061,
+      "learning_rate": 2.524075873841269e-05,
+      "loss": 0.0328,
+      "step": 7580
+    },
+    {
+      "epoch": 13.651079136690647,
+      "grad_norm": 0.6274125576019287,
+      "learning_rate": 2.511168548415399e-05,
+      "loss": 0.0286,
+      "step": 7590
+    },
+    {
+      "epoch": 13.66906474820144,
+      "grad_norm": 0.8142151236534119,
+      "learning_rate": 2.4982832339060453e-05,
+      "loss": 0.0297,
+      "step": 7600
+    },
+    {
+      "epoch": 13.68705035971223,
+      "grad_norm": 0.5592523813247681,
+      "learning_rate": 2.485420044269331e-05,
+      "loss": 0.0218,
+      "step": 7610
+    },
+    {
+      "epoch": 13.70503597122302,
+      "grad_norm": 0.4702487587928772,
+      "learning_rate": 2.4725790932657094e-05,
+      "loss": 0.0239,
+      "step": 7620
+    },
+    {
+      "epoch": 13.723021582733812,
+      "grad_norm": 0.47041937708854675,
+      "learning_rate": 2.4597604944589568e-05,
+      "loss": 0.0246,
+      "step": 7630
+    },
+    {
+      "epoch": 13.741007194244604,
+      "grad_norm": 0.45115765929222107,
+      "learning_rate": 2.4469643612151695e-05,
+      "loss": 0.0249,
+      "step": 7640
+    },
+    {
+      "epoch": 13.758992805755396,
+      "grad_norm": 0.49744996428489685,
+      "learning_rate": 2.434190806701765e-05,
+      "loss": 0.0249,
+      "step": 7650
+    },
+    {
+      "epoch": 13.776978417266188,
+      "grad_norm": 0.49035778641700745,
+      "learning_rate": 2.421439943886473e-05,
+      "loss": 0.0268,
+      "step": 7660
+    },
+    {
+      "epoch": 13.79496402877698,
+      "grad_norm": 0.8184385299682617,
+      "learning_rate": 2.4087118855363416e-05,
+      "loss": 0.0281,
+      "step": 7670
+    },
+    {
+      "epoch": 13.81294964028777,
+      "grad_norm": 0.43353429436683655,
+      "learning_rate": 2.3960067442167395e-05,
+      "loss": 0.0224,
+      "step": 7680
+    },
+    {
+      "epoch": 13.83093525179856,
+      "grad_norm": 0.29570966958999634,
+      "learning_rate": 2.3833246322903603e-05,
+      "loss": 0.0264,
+      "step": 7690
+    },
+    {
+      "epoch": 13.848920863309353,
+      "grad_norm": 0.3498859405517578,
+      "learning_rate": 2.3706656619162278e-05,
+      "loss": 0.0246,
+      "step": 7700
+    },
+    {
+      "epoch": 13.866906474820144,
+      "grad_norm": 0.2970014214515686,
+      "learning_rate": 2.3580299450487042e-05,
+      "loss": 0.025,
+      "step": 7710
+    },
+    {
+      "epoch": 13.884892086330936,
+      "grad_norm": 0.863670289516449,
+      "learning_rate": 2.3454175934365037e-05,
+      "loss": 0.0267,
+      "step": 7720
+    },
+    {
+      "epoch": 13.902877697841726,
+      "grad_norm": 0.5452934503555298,
+      "learning_rate": 2.332828718621695e-05,
+      "loss": 0.032,
+      "step": 7730
+    },
+    {
+      "epoch": 13.920863309352518,
+      "grad_norm": 0.43150079250335693,
+      "learning_rate": 2.3202634319387295e-05,
+      "loss": 0.0267,
+      "step": 7740
+    },
+    {
+      "epoch": 13.93884892086331,
+      "grad_norm": 0.6762245297431946,
+      "learning_rate": 2.3077218445134434e-05,
+      "loss": 0.026,
+      "step": 7750
+    },
+    {
+      "epoch": 13.956834532374101,
+      "grad_norm": 0.47107309103012085,
+      "learning_rate": 2.2952040672620755e-05,
+      "loss": 0.0254,
+      "step": 7760
+    },
+    {
+      "epoch": 13.974820143884893,
+      "grad_norm": 0.6174061894416809,
+      "learning_rate": 2.2827102108902964e-05,
+      "loss": 0.0206,
+      "step": 7770
+    },
+    {
+      "epoch": 13.992805755395683,
+      "grad_norm": 1.5655021667480469,
+      "learning_rate": 2.27024038589222e-05,
+      "loss": 0.0232,
+      "step": 7780
+    },
+    {
+      "epoch": 14.010791366906474,
+      "grad_norm": 0.30220934748649597,
+      "learning_rate": 2.2577947025494345e-05,
+      "loss": 0.0248,
+      "step": 7790
+    },
+    {
+      "epoch": 14.028776978417266,
+      "grad_norm": 0.22649778425693512,
+      "learning_rate": 2.2453732709300168e-05,
+      "loss": 0.0258,
+      "step": 7800
+    },
+    {
+      "epoch": 14.046762589928058,
+      "grad_norm": 0.40873950719833374,
+      "learning_rate": 2.2329762008875676e-05,
+      "loss": 0.0248,
+      "step": 7810
+    },
+    {
+      "epoch": 14.06474820143885,
+      "grad_norm": 0.298087477684021,
+      "learning_rate": 2.220603602060236e-05,
+      "loss": 0.0292,
+      "step": 7820
+    },
+    {
+      "epoch": 14.082733812949641,
+      "grad_norm": 0.664714515209198,
+      "learning_rate": 2.20825558386975e-05,
+      "loss": 0.0333,
+      "step": 7830
+    },
+    {
+      "epoch": 14.100719424460431,
+      "grad_norm": 0.5696831345558167,
+      "learning_rate": 2.1959322555204505e-05,
+      "loss": 0.0251,
+      "step": 7840
+    },
+    {
+      "epoch": 14.118705035971223,
+      "grad_norm": 0.684173583984375,
+      "learning_rate": 2.1836337259983235e-05,
+      "loss": 0.0301,
+      "step": 7850
+    },
+    {
+      "epoch": 14.136690647482014,
+      "grad_norm": 0.7715722322463989,
+      "learning_rate": 2.17136010407004e-05,
+      "loss": 0.0228,
+      "step": 7860
+    },
+    {
+      "epoch": 14.154676258992806,
+      "grad_norm": 0.8577022552490234,
+      "learning_rate": 2.159111498281987e-05,
+      "loss": 0.0257,
+      "step": 7870
+    },
+    {
+      "epoch": 14.172661870503598,
+      "grad_norm": 0.4395459294319153,
+      "learning_rate": 2.1468880169593136e-05,
+      "loss": 0.0324,
+      "step": 7880
+    },
+    {
+      "epoch": 14.190647482014388,
+      "grad_norm": 0.35227638483047485,
+      "learning_rate": 2.134689768204975e-05,
+      "loss": 0.0272,
+      "step": 7890
+    },
+    {
+      "epoch": 14.20863309352518,
+      "grad_norm": 0.59871906042099,
+      "learning_rate": 2.1225168598987677e-05,
+      "loss": 0.0265,
+      "step": 7900
+    },
+    {
+      "epoch": 14.226618705035971,
+      "grad_norm": 0.48534005880355835,
+      "learning_rate": 2.1103693996963823e-05,
+      "loss": 0.0273,
+      "step": 7910
+    },
+    {
+      "epoch": 14.244604316546763,
+      "grad_norm": 0.3349665105342865,
+      "learning_rate": 2.09824749502845e-05,
+      "loss": 0.029,
+      "step": 7920
+    },
+    {
+      "epoch": 14.262589928057555,
+      "grad_norm": 0.35108402371406555,
+      "learning_rate": 2.0861512530995908e-05,
+      "loss": 0.022,
+      "step": 7930
+    },
+    {
+      "epoch": 14.280575539568344,
+      "grad_norm": 0.8136747479438782,
+      "learning_rate": 2.0740807808874674e-05,
+      "loss": 0.026,
+      "step": 7940
+    },
+    {
+      "epoch": 14.298561151079136,
+      "grad_norm": 0.6949844360351562,
+      "learning_rate": 2.062036185141838e-05,
+      "loss": 0.0342,
+      "step": 7950
+    },
+    {
+      "epoch": 14.316546762589928,
+      "grad_norm": 0.3748473525047302,
+      "learning_rate": 2.050017572383613e-05,
+      "loss": 0.0276,
+      "step": 7960
+    },
+    {
+      "epoch": 14.33453237410072,
+      "grad_norm": 0.5470220446586609,
+      "learning_rate": 2.0380250489039106e-05,
+      "loss": 0.0304,
+      "step": 7970
+    },
+    {
+      "epoch": 14.352517985611511,
+      "grad_norm": 0.3599213659763336,
+      "learning_rate": 2.0260587207631253e-05,
+      "loss": 0.0318,
+      "step": 7980
+    },
+    {
+      "epoch": 14.370503597122303,
+      "grad_norm": 0.608016848564148,
+      "learning_rate": 2.014118693789973e-05,
+      "loss": 0.0263,
+      "step": 7990
+    },
+    {
+      "epoch": 14.388489208633093,
+      "grad_norm": 0.3938364088535309,
+      "learning_rate": 2.0022050735805726e-05,
+      "loss": 0.0274,
+      "step": 8000
+    },
+    {
+      "epoch": 14.406474820143885,
+      "grad_norm": 0.4259718954563141,
+      "learning_rate": 1.990317965497503e-05,
+      "loss": 0.0231,
+      "step": 8010
+    },
+    {
+      "epoch": 14.424460431654676,
+      "grad_norm": 0.7867485880851746,
+      "learning_rate": 1.9784574746688704e-05,
+      "loss": 0.031,
+      "step": 8020
+    },
+    {
+      "epoch": 14.442446043165468,
+      "grad_norm": 0.5324791073799133,
+      "learning_rate": 1.9666237059873882e-05,
+      "loss": 0.0241,
+      "step": 8030
+    },
+    {
+      "epoch": 14.46043165467626,
+      "grad_norm": 0.8580402135848999,
+      "learning_rate": 1.9548167641094366e-05,
+      "loss": 0.0264,
+      "step": 8040
+    },
+    {
+      "epoch": 14.47841726618705,
+      "grad_norm": 0.8546720147132874,
+      "learning_rate": 1.943036753454144e-05,
+      "loss": 0.0233,
+      "step": 8050
+    },
+    {
+      "epoch": 14.496402877697841,
+      "grad_norm": 0.3992490768432617,
+      "learning_rate": 1.9312837782024622e-05,
+      "loss": 0.0255,
+      "step": 8060
+    },
+    {
+      "epoch": 14.514388489208633,
+      "grad_norm": 0.6360202431678772,
+      "learning_rate": 1.9195579422962472e-05,
+      "loss": 0.0255,
+      "step": 8070
+    },
+    {
+      "epoch": 14.532374100719425,
+      "grad_norm": 0.3448922336101532,
+      "learning_rate": 1.907859349437336e-05,
+      "loss": 0.0232,
+      "step": 8080
+    },
+    {
+      "epoch": 14.550359712230216,
+      "grad_norm": 0.5019221305847168,
+      "learning_rate": 1.8961881030866325e-05,
+      "loss": 0.0236,
+      "step": 8090
+    },
+    {
+      "epoch": 14.568345323741006,
+      "grad_norm": 0.2645845413208008,
+      "learning_rate": 1.8845443064631902e-05,
+      "loss": 0.0249,
+      "step": 8100
+    },
+    {
+      "epoch": 14.586330935251798,
+      "grad_norm": 0.5084060430526733,
+      "learning_rate": 1.872928062543301e-05,
+      "loss": 0.0246,
+      "step": 8110
+    },
+    {
+      "epoch": 14.60431654676259,
+      "grad_norm": 0.4594169557094574,
+      "learning_rate": 1.8613394740595875e-05,
+      "loss": 0.0265,
+      "step": 8120
+    },
+    {
+      "epoch": 14.622302158273381,
+      "grad_norm": 0.4073614478111267,
+      "learning_rate": 1.8497786435000864e-05,
+      "loss": 0.0212,
+      "step": 8130
+    },
+    {
+      "epoch": 14.640287769784173,
+      "grad_norm": 0.48052847385406494,
+      "learning_rate": 1.8382456731073488e-05,
+      "loss": 0.0256,
+      "step": 8140
+    },
+    {
+      "epoch": 14.658273381294965,
+      "grad_norm": 0.524886965751648,
+      "learning_rate": 1.8267406648775352e-05,
+      "loss": 0.0217,
+      "step": 8150
+    },
+    {
+      "epoch": 14.676258992805755,
+      "grad_norm": 0.633929431438446,
+      "learning_rate": 1.8152637205595063e-05,
+      "loss": 0.0264,
+      "step": 8160
+    },
+    {
+      "epoch": 14.694244604316546,
+      "grad_norm": 0.33972084522247314,
+      "learning_rate": 1.8038149416539375e-05,
+      "loss": 0.0238,
+      "step": 8170
+    },
+    {
+      "epoch": 14.712230215827338,
+      "grad_norm": 0.4787234663963318,
+      "learning_rate": 1.792394429412408e-05,
+      "loss": 0.0284,
+      "step": 8180
+    },
+    {
+      "epoch": 14.73021582733813,
+      "grad_norm": 0.7825778722763062,
+      "learning_rate": 1.7810022848365104e-05,
+      "loss": 0.026,
+      "step": 8190
+    },
+    {
+      "epoch": 14.748201438848922,
+      "grad_norm": 0.9001017212867737,
+      "learning_rate": 1.7696386086769575e-05,
+      "loss": 0.027,
+      "step": 8200
+    },
+    {
+      "epoch": 14.766187050359711,
+      "grad_norm": 1.1512305736541748,
+      "learning_rate": 1.7583035014326915e-05,
+      "loss": 0.0272,
+      "step": 8210
+    },
+    {
+      "epoch": 14.784172661870503,
+      "grad_norm": 0.6733930706977844,
+      "learning_rate": 1.7469970633499944e-05,
+      "loss": 0.0252,
+      "step": 8220
+    },
+    {
+      "epoch": 14.802158273381295,
+      "grad_norm": 0.5925114750862122,
+      "learning_rate": 1.735719394421601e-05,
+      "loss": 0.0267,
+      "step": 8230
+    },
+    {
+      "epoch": 14.820143884892087,
+      "grad_norm": 0.4901937246322632,
+      "learning_rate": 1.7244705943858162e-05,
+      "loss": 0.0273,
+      "step": 8240
+    },
+    {
+      "epoch": 14.838129496402878,
+      "grad_norm": 1.2958598136901855,
+      "learning_rate": 1.7132507627256322e-05,
+      "loss": 0.0251,
+      "step": 8250
+    },
+    {
+      "epoch": 14.85611510791367,
+      "grad_norm": 0.45124995708465576,
+      "learning_rate": 1.7020599986678453e-05,
+      "loss": 0.0239,
+      "step": 8260
+    },
+    {
+      "epoch": 14.87410071942446,
+      "grad_norm": 0.4225548803806305,
+      "learning_rate": 1.6908984011821883e-05,
+      "loss": 0.0282,
+      "step": 8270
+    },
+    {
+      "epoch": 14.892086330935252,
+      "grad_norm": 0.48368096351623535,
+      "learning_rate": 1.679766068980442e-05,
+      "loss": 0.0296,
+      "step": 8280
+    },
+    {
+      "epoch": 14.910071942446043,
+      "grad_norm": 0.4384301006793976,
+      "learning_rate": 1.6686631005155705e-05,
+      "loss": 0.0238,
+      "step": 8290
+    },
+    {
+      "epoch": 14.928057553956835,
+      "grad_norm": 0.6683114171028137,
+      "learning_rate": 1.6575895939808505e-05,
+      "loss": 0.0219,
+      "step": 8300
+    },
+    {
+      "epoch": 14.946043165467627,
+      "grad_norm": 0.721226155757904,
+      "learning_rate": 1.646545647308999e-05,
+      "loss": 0.028,
+      "step": 8310
+    },
+    {
+      "epoch": 14.964028776978417,
+      "grad_norm": 0.28487199544906616,
+      "learning_rate": 1.63553135817131e-05,
+      "loss": 0.0265,
+      "step": 8320
+    },
+    {
+      "epoch": 14.982014388489208,
+      "grad_norm": 1.0410717725753784,
+      "learning_rate": 1.6245468239767897e-05,
+      "loss": 0.03,
+      "step": 8330
+    },
+    {
+      "epoch": 15.0,
+      "grad_norm": 0.4688461720943451,
+      "learning_rate": 1.6135921418712956e-05,
+      "loss": 0.0293,
+      "step": 8340
+    },
+    {
+      "epoch": 15.017985611510792,
+      "grad_norm": 0.6519284248352051,
+      "learning_rate": 1.6026674087366748e-05,
+      "loss": 0.0259,
+      "step": 8350
+    },
+    {
+      "epoch": 15.035971223021583,
+      "grad_norm": 0.35547423362731934,
+      "learning_rate": 1.591772721189915e-05,
+      "loss": 0.0217,
+      "step": 8360
+    },
+    {
+      "epoch": 15.053956834532373,
+      "grad_norm": 0.6237521171569824,
+      "learning_rate": 1.5809081755822802e-05,
+      "loss": 0.0209,
+      "step": 8370
+    },
+    {
+      "epoch": 15.071942446043165,
+      "grad_norm": 1.209920048713684,
+      "learning_rate": 1.5700738679984637e-05,
+      "loss": 0.0272,
+      "step": 8380
+    },
+    {
+      "epoch": 15.089928057553957,
+      "grad_norm": 0.7148857116699219,
+      "learning_rate": 1.5592698942557354e-05,
+      "loss": 0.0241,
+      "step": 8390
+    },
+    {
+      "epoch": 15.107913669064748,
+      "grad_norm": 0.6391850113868713,
+      "learning_rate": 1.5484963499030976e-05,
+      "loss": 0.028,
+      "step": 8400
+    },
+    {
+      "epoch": 15.12589928057554,
+      "grad_norm": 0.5094900131225586,
+      "learning_rate": 1.537753330220443e-05,
+      "loss": 0.0236,
+      "step": 8410
+    },
+    {
+      "epoch": 15.14388489208633,
+      "grad_norm": 0.4152078926563263,
+      "learning_rate": 1.5270409302177025e-05,
+      "loss": 0.0237,
+      "step": 8420
+    },
+    {
+      "epoch": 15.161870503597122,
+      "grad_norm": 0.6238788366317749,
+      "learning_rate": 1.5163592446340135e-05,
+      "loss": 0.0246,
+      "step": 8430
+    },
+    {
+      "epoch": 15.179856115107913,
+      "grad_norm": 0.4302082657814026,
+      "learning_rate": 1.5057083679368773e-05,
+      "loss": 0.025,
+      "step": 8440
+    },
+    {
+      "epoch": 15.197841726618705,
+      "grad_norm": 0.4804670214653015,
+      "learning_rate": 1.4950883943213257e-05,
+      "loss": 0.025,
+      "step": 8450
+    },
+    {
+      "epoch": 15.215827338129497,
+      "grad_norm": 0.6037356853485107,
+      "learning_rate": 1.484499417709087e-05,
+      "loss": 0.0218,
+      "step": 8460
+    },
+    {
+      "epoch": 15.233812949640289,
+      "grad_norm": 0.3752347230911255,
+      "learning_rate": 1.4739415317477574e-05,
+      "loss": 0.0264,
+      "step": 8470
+    },
+    {
+      "epoch": 15.251798561151078,
+      "grad_norm": 0.5201825499534607,
+      "learning_rate": 1.4634148298099692e-05,
+      "loss": 0.0239,
+      "step": 8480
+    },
+    {
+      "epoch": 15.26978417266187,
+      "grad_norm": 0.6259966492652893,
+      "learning_rate": 1.4529194049925677e-05,
+      "loss": 0.0314,
+      "step": 8490
+    },
+    {
+      "epoch": 15.287769784172662,
+      "grad_norm": 0.6927889585494995,
+      "learning_rate": 1.4424553501157894e-05,
+      "loss": 0.0224,
+      "step": 8500
+    },
+    {
+      "epoch": 15.305755395683454,
+      "grad_norm": 0.7196037173271179,
+      "learning_rate": 1.432022757722436e-05,
+      "loss": 0.0271,
+      "step": 8510
+    },
+    {
+      "epoch": 15.323741007194245,
+      "grad_norm": 0.33717089891433716,
+      "learning_rate": 1.4216217200770604e-05,
+      "loss": 0.0254,
+      "step": 8520
+    },
+    {
+      "epoch": 15.341726618705035,
+      "grad_norm": 0.3733397126197815,
+      "learning_rate": 1.4112523291651491e-05,
+      "loss": 0.0216,
+      "step": 8530
+    },
+    {
+      "epoch": 15.359712230215827,
+      "grad_norm": 0.4104047417640686,
+      "learning_rate": 1.4009146766923082e-05,
+      "loss": 0.0238,
+      "step": 8540
+    },
+    {
+      "epoch": 15.377697841726619,
+      "grad_norm": 0.49362972378730774,
+      "learning_rate": 1.3906088540834527e-05,
+      "loss": 0.0267,
+      "step": 8550
+    },
+    {
+      "epoch": 15.39568345323741,
+      "grad_norm": 0.9763308763504028,
+      "learning_rate": 1.3803349524819997e-05,
+      "loss": 0.0241,
+      "step": 8560
+    },
+    {
+      "epoch": 15.413669064748202,
+      "grad_norm": 0.7839420437812805,
+      "learning_rate": 1.37009306274906e-05,
+      "loss": 0.0257,
+      "step": 8570
+    },
+    {
+      "epoch": 15.431654676258994,
+      "grad_norm": 0.5211053490638733,
+      "learning_rate": 1.3598832754626356e-05,
+      "loss": 0.022,
+      "step": 8580
+    },
+    {
+      "epoch": 15.449640287769784,
+      "grad_norm": 0.6030896306037903,
+      "learning_rate": 1.3497056809168163e-05,
+      "loss": 0.0235,
+      "step": 8590
+    },
+    {
+      "epoch": 15.467625899280575,
+      "grad_norm": 0.28349077701568604,
+      "learning_rate": 1.3395603691209886e-05,
+      "loss": 0.0278,
+      "step": 8600
+    },
+    {
+      "epoch": 15.485611510791367,
+      "grad_norm": 0.6997207403182983,
+      "learning_rate": 1.3294474297990311e-05,
+      "loss": 0.0261,
+      "step": 8610
+    },
+    {
+      "epoch": 15.503597122302159,
+      "grad_norm": 0.5309312343597412,
+      "learning_rate": 1.319366952388521e-05,
+      "loss": 0.0287,
+      "step": 8620
+    },
+    {
+      "epoch": 15.52158273381295,
+      "grad_norm": 0.4767550528049469,
+      "learning_rate": 1.3093190260399518e-05,
+      "loss": 0.0208,
+      "step": 8630
+    },
+    {
+      "epoch": 15.53956834532374,
+      "grad_norm": 0.32474377751350403,
+      "learning_rate": 1.2993037396159346e-05,
+      "loss": 0.0222,
+      "step": 8640
+    },
+    {
+      "epoch": 15.557553956834532,
+      "grad_norm": 0.7856822609901428,
+      "learning_rate": 1.2893211816904243e-05,
+      "loss": 0.0251,
+      "step": 8650
+    },
+    {
+      "epoch": 15.575539568345324,
+      "grad_norm": 0.2919895648956299,
+      "learning_rate": 1.2793714405479218e-05,
+      "loss": 0.021,
+      "step": 8660
+    },
+    {
+      "epoch": 15.593525179856115,
+      "grad_norm": 0.5322901010513306,
+      "learning_rate": 1.2694546041827032e-05,
+      "loss": 0.0227,
+      "step": 8670
+    },
+    {
+      "epoch": 15.611510791366907,
+      "grad_norm": 0.5810285806655884,
+      "learning_rate": 1.259570760298039e-05,
+      "loss": 0.0311,
+      "step": 8680
+    },
+    {
+      "epoch": 15.629496402877697,
+      "grad_norm": 1.2920984029769897,
+      "learning_rate": 1.2497199963054179e-05,
+      "loss": 0.0294,
+      "step": 8690
+    },
+    {
+      "epoch": 15.647482014388489,
+      "grad_norm": 0.2701358199119568,
+      "learning_rate": 1.2399023993237741e-05,
+      "loss": 0.0244,
+      "step": 8700
+    },
+    {
+      "epoch": 15.66546762589928,
+      "grad_norm": 0.4531395435333252,
+      "learning_rate": 1.2301180561787174e-05,
+      "loss": 0.0205,
+      "step": 8710
+    },
+    {
+      "epoch": 15.683453237410072,
+      "grad_norm": 0.33455538749694824,
+      "learning_rate": 1.2203670534017647e-05,
+      "loss": 0.0292,
+      "step": 8720
+    },
+    {
+      "epoch": 15.701438848920864,
+      "grad_norm": 0.8127555847167969,
+      "learning_rate": 1.2106494772295717e-05,
+      "loss": 0.0253,
+      "step": 8730
+    },
+    {
+      "epoch": 15.719424460431654,
+      "grad_norm": 0.7903777956962585,
+      "learning_rate": 1.2009654136031811e-05,
+      "loss": 0.0243,
+      "step": 8740
+    },
+    {
+      "epoch": 15.737410071942445,
+      "grad_norm": 0.5273742079734802,
+      "learning_rate": 1.191314948167247e-05,
+      "loss": 0.0286,
+      "step": 8750
+    },
+    {
+      "epoch": 15.755395683453237,
+      "grad_norm": 0.7798042893409729,
+      "learning_rate": 1.1816981662692883e-05,
+      "loss": 0.0225,
+      "step": 8760
+    },
+    {
+      "epoch": 15.773381294964029,
+      "grad_norm": 0.6813958287239075,
+      "learning_rate": 1.172115152958933e-05,
+      "loss": 0.0234,
+      "step": 8770
+    },
+    {
+      "epoch": 15.79136690647482,
+      "grad_norm": 0.6172210574150085,
+      "learning_rate": 1.1625659929871562e-05,
+      "loss": 0.0255,
+      "step": 8780
+    },
+    {
+      "epoch": 15.809352517985612,
+      "grad_norm": 0.6527241468429565,
+      "learning_rate": 1.1530507708055493e-05,
+      "loss": 0.0261,
+      "step": 8790
+    },
+    {
+      "epoch": 15.827338129496402,
+      "grad_norm": 0.7393059730529785,
+      "learning_rate": 1.1435695705655541e-05,
+      "loss": 0.0258,
+      "step": 8800
+    },
+    {
+      "epoch": 15.845323741007194,
+      "grad_norm": 0.44266852736473083,
+      "learning_rate": 1.1341224761177294e-05,
+      "loss": 0.0253,
+      "step": 8810
+    },
+    {
+      "epoch": 15.863309352517986,
+      "grad_norm": 0.8522095084190369,
+      "learning_rate": 1.1247095710110062e-05,
+      "loss": 0.0228,
+      "step": 8820
+    },
+    {
+      "epoch": 15.881294964028777,
+      "grad_norm": 0.40083515644073486,
+      "learning_rate": 1.1153309384919481e-05,
+      "loss": 0.0275,
+      "step": 8830
+    },
+    {
+      "epoch": 15.899280575539569,
+      "grad_norm": 0.8442035913467407,
+      "learning_rate": 1.1059866615040204e-05,
+      "loss": 0.0231,
+      "step": 8840
+    },
+    {
+      "epoch": 15.917266187050359,
+      "grad_norm": 0.44408664107322693,
+      "learning_rate": 1.0966768226868463e-05,
+      "loss": 0.0221,
+      "step": 8850
+    },
+    {
+      "epoch": 15.93525179856115,
+      "grad_norm": 0.8646157383918762,
+      "learning_rate": 1.087401504375486e-05,
+      "loss": 0.0297,
+      "step": 8860
+    },
+    {
+      "epoch": 15.953237410071942,
+      "grad_norm": 0.33869028091430664,
+      "learning_rate": 1.078160788599702e-05,
+      "loss": 0.0222,
+      "step": 8870
+    },
+    {
+      "epoch": 15.971223021582734,
+      "grad_norm": 0.386554479598999,
+      "learning_rate": 1.0689547570832408e-05,
+      "loss": 0.0222,
+      "step": 8880
+    },
+    {
+      "epoch": 15.989208633093526,
+      "grad_norm": 0.46215781569480896,
+      "learning_rate": 1.0597834912431014e-05,
+      "loss": 0.0263,
+      "step": 8890
+    },
+    {
+      "epoch": 16.007194244604317,
+      "grad_norm": 0.7558487057685852,
+      "learning_rate": 1.0506470721888212e-05,
+      "loss": 0.0261,
+      "step": 8900
+    },
+    {
+      "epoch": 16.02517985611511,
+      "grad_norm": 0.8021785020828247,
+      "learning_rate": 1.0415455807217566e-05,
+      "loss": 0.0239,
+      "step": 8910
+    },
+    {
+      "epoch": 16.0431654676259,
+      "grad_norm": 1.1028897762298584,
+      "learning_rate": 1.0324790973343679e-05,
+      "loss": 0.0256,
+      "step": 8920
+    },
+    {
+      "epoch": 16.06115107913669,
+      "grad_norm": 0.9204623103141785,
+      "learning_rate": 1.0234477022095095e-05,
+      "loss": 0.0217,
+      "step": 8930
+    },
+    {
+      "epoch": 16.07913669064748,
+      "grad_norm": 0.9754385352134705,
+      "learning_rate": 1.0144514752197198e-05,
+      "loss": 0.0225,
+      "step": 8940
+    },
+    {
+      "epoch": 16.097122302158272,
+      "grad_norm": 0.7273303270339966,
+      "learning_rate": 1.0054904959265126e-05,
+      "loss": 0.0247,
+      "step": 8950
+    },
+    {
+      "epoch": 16.115107913669064,
+      "grad_norm": 1.0062978267669678,
+      "learning_rate": 9.965648435796782e-06,
+      "loss": 0.0247,
+      "step": 8960
+    },
+    {
+      "epoch": 16.133093525179856,
+      "grad_norm": 0.4822646677494049,
+      "learning_rate": 9.876745971165762e-06,
+      "loss": 0.0213,
+      "step": 8970
+    },
+    {
+      "epoch": 16.151079136690647,
+      "grad_norm": 0.251082181930542,
+      "learning_rate": 9.788198351614464e-06,
+      "loss": 0.0289,
+      "step": 8980
+    },
+    {
+      "epoch": 16.16906474820144,
+      "grad_norm": 1.4881809949874878,
+      "learning_rate": 9.700006360247027e-06,
+      "loss": 0.0274,
+      "step": 8990
+    },
+    {
+      "epoch": 16.18705035971223,
+      "grad_norm": 0.41107016801834106,
+      "learning_rate": 9.6121707770225e-06,
+      "loss": 0.0253,
+      "step": 9000
+    },
+    {
+      "epoch": 16.205035971223023,
+      "grad_norm": 0.3931741416454315,
+      "learning_rate": 9.524692378747845e-06,
+      "loss": 0.0224,
+      "step": 9010
+    },
+    {
+      "epoch": 16.223021582733814,
+      "grad_norm": 0.5247843265533447,
+      "learning_rate": 9.437571939071171e-06,
+      "loss": 0.0208,
+      "step": 9020
+    },
+    {
+      "epoch": 16.241007194244606,
+      "grad_norm": 0.5927461385726929,
+      "learning_rate": 9.350810228474855e-06,
+      "loss": 0.0255,
+      "step": 9030
+    },
+    {
+      "epoch": 16.258992805755394,
+      "grad_norm": 0.5177347660064697,
+      "learning_rate": 9.26440801426869e-06,
+      "loss": 0.0194,
+      "step": 9040
+    },
+    {
+      "epoch": 16.276978417266186,
+      "grad_norm": 0.6080642938613892,
+      "learning_rate": 9.178366060583132e-06,
+      "loss": 0.0237,
+      "step": 9050
+    },
+    {
+      "epoch": 16.294964028776977,
+      "grad_norm": 0.8624924421310425,
+      "learning_rate": 9.092685128362538e-06,
+      "loss": 0.0291,
+      "step": 9060
+    },
+    {
+      "epoch": 16.31294964028777,
+      "grad_norm": 0.5210402011871338,
+      "learning_rate": 9.007365975358466e-06,
+      "loss": 0.0211,
+      "step": 9070
+    },
+    {
+      "epoch": 16.33093525179856,
+      "grad_norm": 0.47911182045936584,
+      "learning_rate": 8.922409356122896e-06,
+      "loss": 0.0232,
+      "step": 9080
+    },
+    {
+      "epoch": 16.348920863309353,
+      "grad_norm": 0.4846671521663666,
+      "learning_rate": 8.837816022001627e-06,
+      "loss": 0.0268,
+      "step": 9090
+    },
+    {
+      "epoch": 16.366906474820144,
+      "grad_norm": 0.603743851184845,
+      "learning_rate": 8.75358672112761e-06,
+      "loss": 0.0278,
+      "step": 9100
+    },
+    {
+      "epoch": 16.384892086330936,
+      "grad_norm": 0.5260509252548218,
+      "learning_rate": 8.66972219841432e-06,
+      "loss": 0.0245,
+      "step": 9110
+    },
+    {
+      "epoch": 16.402877697841728,
+      "grad_norm": 1.3173530101776123,
+      "learning_rate": 8.586223195549215e-06,
+      "loss": 0.0247,
+      "step": 9120
+    },
+    {
+      "epoch": 16.42086330935252,
+      "grad_norm": 0.31255581974983215,
+      "learning_rate": 8.503090450987095e-06,
+      "loss": 0.0218,
+      "step": 9130
+    },
+    {
+      "epoch": 16.43884892086331,
+      "grad_norm": 0.3008738160133362,
+      "learning_rate": 8.420324699943648e-06,
+      "loss": 0.0219,
+      "step": 9140
+    },
+    {
+      "epoch": 16.4568345323741,
+      "grad_norm": 0.5180498361587524,
+      "learning_rate": 8.337926674388896e-06,
+      "loss": 0.0309,
+      "step": 9150
+    },
+    {
+      "epoch": 16.47482014388489,
+      "grad_norm": 1.292531132698059,
+      "learning_rate": 8.25589710304076e-06,
+      "loss": 0.0252,
+      "step": 9160
+    },
+    {
+      "epoch": 16.492805755395683,
+      "grad_norm": 0.4705999493598938,
+      "learning_rate": 8.174236711358585e-06,
+      "loss": 0.0269,
+      "step": 9170
+    },
+    {
+      "epoch": 16.510791366906474,
+      "grad_norm": 0.5129850506782532,
+      "learning_rate": 8.092946221536745e-06,
+      "loss": 0.0275,
+      "step": 9180
+    },
+    {
+      "epoch": 16.528776978417266,
+      "grad_norm": 0.90230393409729,
+      "learning_rate": 8.012026352498241e-06,
+      "loss": 0.0265,
+      "step": 9190
+    },
+    {
+      "epoch": 16.546762589928058,
+      "grad_norm": 0.3909169137477875,
+      "learning_rate": 7.931477819888344e-06,
+      "loss": 0.024,
+      "step": 9200
+    },
+    {
+      "epoch": 16.56474820143885,
+      "grad_norm": 0.5020588040351868,
+      "learning_rate": 7.8513013360683e-06,
+      "loss": 0.0242,
+      "step": 9210
+    },
+    {
+      "epoch": 16.58273381294964,
+      "grad_norm": 0.5076807141304016,
+      "learning_rate": 7.77149761010898e-06,
+      "loss": 0.0241,
+      "step": 9220
+    },
+    {
+      "epoch": 16.600719424460433,
+      "grad_norm": 0.23666317760944366,
+      "learning_rate": 7.692067347784637e-06,
+      "loss": 0.0258,
+      "step": 9230
+    },
+    {
+      "epoch": 16.618705035971225,
+      "grad_norm": 1.695730447769165,
+      "learning_rate": 7.613011251566637e-06,
+      "loss": 0.0264,
+      "step": 9240
+    },
+    {
+      "epoch": 16.636690647482013,
+      "grad_norm": 1.1068884134292603,
+      "learning_rate": 7.53433002061728e-06,
+      "loss": 0.0281,
+      "step": 9250
+    },
+    {
+      "epoch": 16.654676258992804,
+      "grad_norm": 0.28573542833328247,
+      "learning_rate": 7.456024350783636e-06,
+      "loss": 0.0206,
+      "step": 9260
+    },
+    {
+      "epoch": 16.672661870503596,
+      "grad_norm": 0.37045443058013916,
+      "learning_rate": 7.378094934591317e-06,
+      "loss": 0.0234,
+      "step": 9270
+    },
+    {
+      "epoch": 16.690647482014388,
+      "grad_norm": 0.6459355354309082,
+      "learning_rate": 7.300542461238408e-06,
+      "loss": 0.0186,
+      "step": 9280
+    },
+    {
+      "epoch": 16.70863309352518,
+      "grad_norm": 0.5449643135070801,
+      "learning_rate": 7.223367616589366e-06,
+      "loss": 0.0209,
+      "step": 9290
+    },
+    {
+      "epoch": 16.72661870503597,
+      "grad_norm": 0.3917773962020874,
+      "learning_rate": 7.1465710831689345e-06,
+      "loss": 0.024,
+      "step": 9300
+    },
+    {
+      "epoch": 16.744604316546763,
+      "grad_norm": 0.7854135632514954,
+      "learning_rate": 7.070153540156133e-06,
+      "loss": 0.0234,
+      "step": 9310
+    },
+    {
+      "epoch": 16.762589928057555,
+      "grad_norm": 0.5904101729393005,
+      "learning_rate": 6.99411566337822e-06,
+      "loss": 0.0255,
+      "step": 9320
+    },
+    {
+      "epoch": 16.780575539568346,
+      "grad_norm": 0.5345132946968079,
+      "learning_rate": 6.918458125304744e-06,
+      "loss": 0.0269,
+      "step": 9330
+    },
+    {
+      "epoch": 16.798561151079138,
+      "grad_norm": 0.4916456937789917,
+      "learning_rate": 6.843181595041592e-06,
+      "loss": 0.0277,
+      "step": 9340
+    },
+    {
+      "epoch": 16.81654676258993,
+      "grad_norm": 0.535793662071228,
+      "learning_rate": 6.768286738325036e-06,
+      "loss": 0.0225,
+      "step": 9350
+    },
+    {
+      "epoch": 16.834532374100718,
+      "grad_norm": 0.5153817534446716,
+      "learning_rate": 6.6937742175159194e-06,
+      "loss": 0.0254,
+      "step": 9360
+    },
+    {
+      "epoch": 16.85251798561151,
+      "grad_norm": 0.4768829345703125,
+      "learning_rate": 6.619644691593718e-06,
+      "loss": 0.0302,
+      "step": 9370
+    },
+    {
+      "epoch": 16.8705035971223,
+      "grad_norm": 0.4367595911026001,
+      "learning_rate": 6.545898816150764e-06,
+      "loss": 0.0247,
+      "step": 9380
+    },
+    {
+      "epoch": 16.888489208633093,
+      "grad_norm": 0.39912503957748413,
+      "learning_rate": 6.472537243386428e-06,
+      "loss": 0.0257,
+      "step": 9390
+    },
+    {
+      "epoch": 16.906474820143885,
+      "grad_norm": 0.7161368131637573,
+      "learning_rate": 6.399560622101353e-06,
+      "loss": 0.0231,
+      "step": 9400
+    },
+    {
+      "epoch": 16.924460431654676,
+      "grad_norm": 0.5543472170829773,
+      "learning_rate": 6.326969597691723e-06,
+      "loss": 0.0228,
+      "step": 9410
+    },
+    {
+      "epoch": 16.942446043165468,
+      "grad_norm": 0.5978730320930481,
+      "learning_rate": 6.254764812143554e-06,
+      "loss": 0.023,
+      "step": 9420
+    },
+    {
+      "epoch": 16.96043165467626,
+      "grad_norm": 0.23862099647521973,
+      "learning_rate": 6.182946904026998e-06,
+      "loss": 0.0218,
+      "step": 9430
+    },
+    {
+      "epoch": 16.97841726618705,
+      "grad_norm": 0.6031172871589661,
+      "learning_rate": 6.111516508490728e-06,
+      "loss": 0.0263,
+      "step": 9440
+    },
+    {
+      "epoch": 16.996402877697843,
+      "grad_norm": 0.42859840393066406,
+      "learning_rate": 6.040474257256306e-06,
+      "loss": 0.0267,
+      "step": 9450
+    },
+    {
+      "epoch": 17.014388489208635,
+      "grad_norm": 0.912424623966217,
+      "learning_rate": 5.969820778612595e-06,
+      "loss": 0.0245,
+      "step": 9460
+    },
+    {
+      "epoch": 17.032374100719423,
+      "grad_norm": 0.35150790214538574,
+      "learning_rate": 5.899556697410169e-06,
+      "loss": 0.0243,
+      "step": 9470
+    },
+    {
+      "epoch": 17.050359712230215,
+      "grad_norm": 0.7533866763114929,
+      "learning_rate": 5.82968263505585e-06,
+      "loss": 0.0262,
+      "step": 9480
+    },
+    {
+      "epoch": 17.068345323741006,
+      "grad_norm": 0.6731085181236267,
+      "learning_rate": 5.760199209507166e-06,
+      "loss": 0.0256,
+      "step": 9490
+    },
+    {
+      "epoch": 17.086330935251798,
+      "grad_norm": 1.1619879007339478,
+      "learning_rate": 5.691107035266935e-06,
+      "loss": 0.0217,
+      "step": 9500
+    },
+    {
+      "epoch": 17.10431654676259,
+      "grad_norm": 0.525164008140564,
+      "learning_rate": 5.622406723377754e-06,
+      "loss": 0.0285,
+      "step": 9510
+    },
+    {
+      "epoch": 17.12230215827338,
+      "grad_norm": 0.9267173409461975,
+      "learning_rate": 5.5540988814166635e-06,
+      "loss": 0.0237,
+      "step": 9520
+    },
+    {
+      "epoch": 17.140287769784173,
+      "grad_norm": 0.5514078736305237,
+      "learning_rate": 5.48618411348974e-06,
+      "loss": 0.0256,
+      "step": 9530
+    },
+    {
+      "epoch": 17.158273381294965,
+      "grad_norm": 0.6464403867721558,
+      "learning_rate": 5.418663020226766e-06,
+      "loss": 0.0263,
+      "step": 9540
+    },
+    {
+      "epoch": 17.176258992805757,
+      "grad_norm": 0.8582285642623901,
+      "learning_rate": 5.351536198775903e-06,
+      "loss": 0.0244,
+      "step": 9550
+    },
+    {
+      "epoch": 17.194244604316548,
+      "grad_norm": 0.447550892829895,
+      "learning_rate": 5.284804242798436e-06,
+      "loss": 0.0257,
+      "step": 9560
+    },
+    {
+      "epoch": 17.21223021582734,
+      "grad_norm": 0.9709569811820984,
+      "learning_rate": 5.218467742463501e-06,
+      "loss": 0.0258,
+      "step": 9570
+    },
+    {
+      "epoch": 17.230215827338128,
+      "grad_norm": 0.5255972146987915,
+      "learning_rate": 5.152527284442871e-06,
+      "loss": 0.0243,
+      "step": 9580
+    },
+    {
+      "epoch": 17.24820143884892,
+      "grad_norm": 0.3573249578475952,
+      "learning_rate": 5.086983451905786e-06,
+      "loss": 0.022,
+      "step": 9590
+    },
+    {
+      "epoch": 17.26618705035971,
+      "grad_norm": 0.5273535847663879,
+      "learning_rate": 5.021836824513759e-06,
+      "loss": 0.0224,
+      "step": 9600
+    },
+    {
+      "epoch": 17.284172661870503,
+      "grad_norm": 0.2634049952030182,
+      "learning_rate": 4.957087978415486e-06,
+      "loss": 0.027,
+      "step": 9610
+    },
+    {
+      "epoch": 17.302158273381295,
+      "grad_norm": 0.5875841379165649,
+      "learning_rate": 4.8927374862417275e-06,
+      "loss": 0.0251,
+      "step": 9620
+    },
+    {
+      "epoch": 17.320143884892087,
+      "grad_norm": 0.3263997733592987,
+      "learning_rate": 4.828785917100226e-06,
+      "loss": 0.0282,
+      "step": 9630
+    },
+    {
+      "epoch": 17.33812949640288,
+      "grad_norm": 0.39191120862960815,
+      "learning_rate": 4.765233836570754e-06,
+      "loss": 0.0227,
+      "step": 9640
+    },
+    {
+      "epoch": 17.35611510791367,
+      "grad_norm": 0.7046676874160767,
+      "learning_rate": 4.702081806700015e-06,
+      "loss": 0.0232,
+      "step": 9650
+    },
+    {
+      "epoch": 17.37410071942446,
+      "grad_norm": 0.5387694835662842,
+      "learning_rate": 4.639330385996721e-06,
+      "loss": 0.0287,
+      "step": 9660
+    },
+    {
+      "epoch": 17.392086330935253,
+      "grad_norm": 0.7632167339324951,
+      "learning_rate": 4.57698012942665e-06,
+      "loss": 0.0238,
+      "step": 9670
+    },
+    {
+      "epoch": 17.41007194244604,
+      "grad_norm": 0.6518047451972961,
+      "learning_rate": 4.515031588407731e-06,
+      "loss": 0.0256,
+      "step": 9680
+    },
+    {
+      "epoch": 17.428057553956833,
+      "grad_norm": 0.7861002087593079,
+      "learning_rate": 4.453485310805195e-06,
+      "loss": 0.0229,
+      "step": 9690
+    },
+    {
+      "epoch": 17.446043165467625,
+      "grad_norm": 0.7429606914520264,
+      "learning_rate": 4.392341840926661e-06,
+      "loss": 0.0217,
+      "step": 9700
+    },
+    {
+      "epoch": 17.464028776978417,
+      "grad_norm": 0.35120925307273865,
+      "learning_rate": 4.331601719517392e-06,
+      "loss": 0.0268,
+      "step": 9710
+    },
+    {
+      "epoch": 17.48201438848921,
+      "grad_norm": 0.9115272164344788,
+      "learning_rate": 4.271265483755488e-06,
+      "loss": 0.0258,
+      "step": 9720
+    },
+    {
+      "epoch": 17.5,
+      "grad_norm": 0.5279654860496521,
+      "learning_rate": 4.2113336672471245e-06,
+      "loss": 0.0238,
+      "step": 9730
+    },
+    {
+      "epoch": 17.51798561151079,
+      "grad_norm": 1.1599388122558594,
+      "learning_rate": 4.151806800021868e-06,
+      "loss": 0.0246,
+      "step": 9740
+    },
+    {
+      "epoch": 17.535971223021583,
+      "grad_norm": 0.5450431108474731,
+      "learning_rate": 4.09268540852793e-06,
+      "loss": 0.0236,
+      "step": 9750
+    },
+    {
+      "epoch": 17.553956834532375,
+      "grad_norm": 0.5093603730201721,
+      "learning_rate": 4.033970015627569e-06,
+      "loss": 0.0237,
+      "step": 9760
+    },
+    {
+      "epoch": 17.571942446043167,
+      "grad_norm": 0.5423790812492371,
+      "learning_rate": 3.975661140592424e-06,
+      "loss": 0.0207,
+      "step": 9770
+    },
+    {
+      "epoch": 17.58992805755396,
+      "grad_norm": 0.551464855670929,
+      "learning_rate": 3.917759299098955e-06,
+      "loss": 0.0236,
+      "step": 9780
+    },
+    {
+      "epoch": 17.607913669064747,
+      "grad_norm": 0.33895400166511536,
+      "learning_rate": 3.860265003223868e-06,
+      "loss": 0.0253,
+      "step": 9790
+    },
+    {
+      "epoch": 17.62589928057554,
+      "grad_norm": 0.7024912238121033,
+      "learning_rate": 3.8031787614395743e-06,
+      "loss": 0.025,
+      "step": 9800
+    },
+    {
+      "epoch": 17.64388489208633,
+      "grad_norm": 0.3595588803291321,
+      "learning_rate": 3.7465010786097154e-06,
+      "loss": 0.0226,
+      "step": 9810
+    },
+    {
+      "epoch": 17.66187050359712,
+      "grad_norm": 0.24189768731594086,
+      "learning_rate": 3.690232455984677e-06,
+      "loss": 0.0201,
+      "step": 9820
+    },
+    {
+      "epoch": 17.679856115107913,
+      "grad_norm": 0.738793134689331,
+      "learning_rate": 3.6343733911971876e-06,
+      "loss": 0.0291,
+      "step": 9830
+    },
+    {
+      "epoch": 17.697841726618705,
+      "grad_norm": 0.323628306388855,
+      "learning_rate": 3.5789243782578807e-06,
+      "loss": 0.0258,
+      "step": 9840
+    },
+    {
+      "epoch": 17.715827338129497,
+      "grad_norm": 0.30349797010421753,
+      "learning_rate": 3.5238859075509556e-06,
+      "loss": 0.0255,
+      "step": 9850
+    },
+    {
+      "epoch": 17.73381294964029,
+      "grad_norm": 0.5222554802894592,
+      "learning_rate": 3.4692584658297945e-06,
+      "loss": 0.0293,
+      "step": 9860
+    },
+    {
+      "epoch": 17.75179856115108,
+      "grad_norm": 0.5784351229667664,
+      "learning_rate": 3.41504253621272e-06,
+      "loss": 0.0208,
+      "step": 9870
+    },
+    {
+      "epoch": 17.769784172661872,
+      "grad_norm": 0.3522135019302368,
+      "learning_rate": 3.361238598178701e-06,
+      "loss": 0.0252,
+      "step": 9880
+    },
+    {
+      "epoch": 17.78776978417266,
+      "grad_norm": 0.803679883480072,
+      "learning_rate": 3.3078471275630794e-06,
+      "loss": 0.0253,
+      "step": 9890
+    },
+    {
+      "epoch": 17.805755395683452,
+      "grad_norm": 0.6999955177307129,
+      "learning_rate": 3.2548685965533954e-06,
+      "loss": 0.0224,
+      "step": 9900
+    },
+    {
+      "epoch": 17.823741007194243,
+      "grad_norm": 1.0477122068405151,
+      "learning_rate": 3.2023034736851976e-06,
+      "loss": 0.0212,
+      "step": 9910
+    },
+    {
+      "epoch": 17.841726618705035,
+      "grad_norm": 0.874548614025116,
+      "learning_rate": 3.150152223837921e-06,
+      "loss": 0.0283,
+      "step": 9920
+    },
+    {
+      "epoch": 17.859712230215827,
+      "grad_norm": 0.5578421354293823,
+      "learning_rate": 3.0984153082307456e-06,
+      "loss": 0.0209,
+      "step": 9930
+    },
+    {
+      "epoch": 17.87769784172662,
+      "grad_norm": 0.6624292135238647,
+      "learning_rate": 3.0470931844185256e-06,
+      "loss": 0.0253,
+      "step": 9940
+    },
+    {
+      "epoch": 17.89568345323741,
+      "grad_norm": 1.5036334991455078,
+      "learning_rate": 2.9961863062877573e-06,
+      "loss": 0.0255,
+      "step": 9950
+    },
+    {
+      "epoch": 17.913669064748202,
+      "grad_norm": 0.8845500946044922,
+      "learning_rate": 2.94569512405255e-06,
+      "loss": 0.0267,
+      "step": 9960
+    },
+    {
+      "epoch": 17.931654676258994,
+      "grad_norm": 0.5007345676422119,
+      "learning_rate": 2.895620084250672e-06,
+      "loss": 0.0248,
+      "step": 9970
+    },
+    {
+      "epoch": 17.949640287769785,
+      "grad_norm": 0.2975013554096222,
+      "learning_rate": 2.8459616297395466e-06,
+      "loss": 0.0245,
+      "step": 9980
+    },
+    {
+      "epoch": 17.967625899280577,
+      "grad_norm": 0.40914618968963623,
+      "learning_rate": 2.7967201996923875e-06,
+      "loss": 0.0265,
+      "step": 9990
+    },
+    {
+      "epoch": 17.985611510791365,
+      "grad_norm": 0.7163299918174744,
+      "learning_rate": 2.747896229594288e-06,
+      "loss": 0.0261,
+      "step": 10000
+    },
+    {
+      "epoch": 18.003597122302157,
+      "grad_norm": 0.5489468574523926,
+      "learning_rate": 2.6994901512383874e-06,
+      "loss": 0.0244,
+      "step": 10010
+    },
+    {
+      "epoch": 18.02158273381295,
+      "grad_norm": 0.5518163442611694,
+      "learning_rate": 2.651502392722027e-06,
+      "loss": 0.0258,
+      "step": 10020
+    },
+    {
+      "epoch": 18.03956834532374,
+      "grad_norm": 0.40058451890945435,
+      "learning_rate": 2.6039333784429965e-06,
+      "loss": 0.0229,
+      "step": 10030
+    },
+    {
+      "epoch": 18.057553956834532,
+      "grad_norm": 0.3614138066768646,
+      "learning_rate": 2.556783529095752e-06,
+      "loss": 0.0253,
+      "step": 10040
+    },
+    {
+      "epoch": 18.075539568345324,
+      "grad_norm": 0.3413621187210083,
+      "learning_rate": 2.510053261667705e-06,
+      "loss": 0.0249,
+      "step": 10050
+    },
+    {
+      "epoch": 18.093525179856115,
+      "grad_norm": 0.5375677943229675,
+      "learning_rate": 2.4637429894355467e-06,
+      "loss": 0.0234,
+      "step": 10060
+    },
+    {
+      "epoch": 18.111510791366907,
+      "grad_norm": 0.6946611404418945,
+      "learning_rate": 2.417853121961583e-06,
+      "loss": 0.0277,
+      "step": 10070
+    },
+    {
+      "epoch": 18.1294964028777,
+      "grad_norm": 0.3714732527732849,
+      "learning_rate": 2.372384065090111e-06,
+      "loss": 0.0207,
+      "step": 10080
+    },
+    {
+      "epoch": 18.14748201438849,
+      "grad_norm": 0.8689820170402527,
+      "learning_rate": 2.3273362209438097e-06,
+      "loss": 0.0217,
+      "step": 10090
+    },
+    {
+      "epoch": 18.165467625899282,
+      "grad_norm": 0.4403657019138336,
+      "learning_rate": 2.282709987920234e-06,
+      "loss": 0.0261,
+      "step": 10100
+    },
+    {
+      "epoch": 18.18345323741007,
+      "grad_norm": 0.2926042675971985,
+      "learning_rate": 2.238505760688242e-06,
+      "loss": 0.0229,
+      "step": 10110
+    },
+    {
+      "epoch": 18.201438848920862,
+      "grad_norm": 0.8282326459884644,
+      "learning_rate": 2.19472393018455e-06,
+      "loss": 0.0232,
+      "step": 10120
+    },
+    {
+      "epoch": 18.219424460431654,
+      "grad_norm": 0.7413493394851685,
+      "learning_rate": 2.151364883610224e-06,
+      "loss": 0.0255,
+      "step": 10130
+    },
+    {
+      "epoch": 18.237410071942445,
+      "grad_norm": 0.49940627813339233,
+      "learning_rate": 2.1084290044272923e-06,
+      "loss": 0.0221,
+      "step": 10140
+    },
+    {
+      "epoch": 18.255395683453237,
+      "grad_norm": 0.1549803763628006,
+      "learning_rate": 2.0659166723553446e-06,
+      "loss": 0.0214,
+      "step": 10150
+    },
+    {
+      "epoch": 18.27338129496403,
+      "grad_norm": 0.3592277467250824,
+      "learning_rate": 2.023828263368166e-06,
+      "loss": 0.0218,
+      "step": 10160
+    },
+    {
+      "epoch": 18.29136690647482,
+      "grad_norm": 0.5709612965583801,
+      "learning_rate": 1.98216414969043e-06,
+      "loss": 0.0242,
+      "step": 10170
+    },
+    {
+      "epoch": 18.309352517985612,
+      "grad_norm": 0.4012541174888611,
+      "learning_rate": 1.9409246997943774e-06,
+      "loss": 0.0178,
+      "step": 10180
+    },
+    {
+      "epoch": 18.327338129496404,
+      "grad_norm": 1.003873586654663,
+      "learning_rate": 1.900110278396594e-06,
+      "loss": 0.0255,
+      "step": 10190
+    },
+    {
+      "epoch": 18.345323741007196,
+      "grad_norm": 0.4295801520347595,
+      "learning_rate": 1.8597212464547541e-06,
+      "loss": 0.0227,
+      "step": 10200
+    },
+    {
+      "epoch": 18.363309352517987,
+      "grad_norm": 0.6608873009681702,
+      "learning_rate": 1.8197579611644478e-06,
+      "loss": 0.0201,
+      "step": 10210
+    },
+    {
+      "epoch": 18.381294964028775,
+      "grad_norm": 0.3636903464794159,
+      "learning_rate": 1.7802207759560042e-06,
+      "loss": 0.0209,
+      "step": 10220
+    },
+    {
+      "epoch": 18.399280575539567,
+      "grad_norm": 0.9239495992660522,
+      "learning_rate": 1.7411100404913949e-06,
+      "loss": 0.0273,
+      "step": 10230
+    },
+    {
+      "epoch": 18.41726618705036,
+      "grad_norm": 0.5481299757957458,
+      "learning_rate": 1.7024261006611086e-06,
+      "loss": 0.0185,
+      "step": 10240
+    },
+    {
+      "epoch": 18.43525179856115,
+      "grad_norm": 0.44658273458480835,
+      "learning_rate": 1.664169298581092e-06,
+      "loss": 0.0222,
+      "step": 10250
+    },
+    {
+      "epoch": 18.453237410071942,
+      "grad_norm": 0.5735198855400085,
+      "learning_rate": 1.6263399725897744e-06,
+      "loss": 0.0221,
+      "step": 10260
+    },
+    {
+      "epoch": 18.471223021582734,
+      "grad_norm": 0.40003567934036255,
+      "learning_rate": 1.5889384572450094e-06,
+      "loss": 0.0248,
+      "step": 10270
+    },
+    {
+      "epoch": 18.489208633093526,
+      "grad_norm": 0.8321783542633057,
+      "learning_rate": 1.5519650833211663e-06,
+      "loss": 0.0255,
+      "step": 10280
+    },
+    {
+      "epoch": 18.507194244604317,
+      "grad_norm": 0.3816922605037689,
+      "learning_rate": 1.5154201778061705e-06,
+      "loss": 0.0253,
+      "step": 10290
+    },
+    {
+      "epoch": 18.52517985611511,
+      "grad_norm": 1.833797574043274,
+      "learning_rate": 1.4793040638986289e-06,
+      "loss": 0.0251,
+      "step": 10300
+    },
+    {
+      "epoch": 18.5431654676259,
+      "grad_norm": 0.3318173289299011,
+      "learning_rate": 1.4436170610049816e-06,
+      "loss": 0.0237,
+      "step": 10310
+    },
+    {
+      "epoch": 18.56115107913669,
+      "grad_norm": 0.4023537039756775,
+      "learning_rate": 1.4083594847366488e-06,
+      "loss": 0.0254,
+      "step": 10320
+    },
+    {
+      "epoch": 18.57913669064748,
+      "grad_norm": 0.9918774962425232,
+      "learning_rate": 1.37353164690725e-06,
+      "loss": 0.0208,
+      "step": 10330
+    },
+    {
+      "epoch": 18.597122302158272,
+      "grad_norm": 0.47037509083747864,
+      "learning_rate": 1.3391338555298728e-06,
+      "loss": 0.0237,
+      "step": 10340
+    },
+    {
+      "epoch": 18.615107913669064,
+      "grad_norm": 0.6450695395469666,
+      "learning_rate": 1.3051664148143129e-06,
+      "loss": 0.0228,
+      "step": 10350
+    },
+    {
+      "epoch": 18.633093525179856,
+      "grad_norm": 0.8368081450462341,
+      "learning_rate": 1.2716296251644e-06,
+      "loss": 0.0216,
+      "step": 10360
+    },
+    {
+      "epoch": 18.651079136690647,
+      "grad_norm": 0.6411568522453308,
+      "learning_rate": 1.2385237831753382e-06,
+      "loss": 0.0239,
+      "step": 10370
+    },
+    {
+      "epoch": 18.66906474820144,
+      "grad_norm": 0.344478040933609,
+      "learning_rate": 1.2058491816310914e-06,
+      "loss": 0.0252,
+      "step": 10380
+    },
+    {
+      "epoch": 18.68705035971223,
+      "grad_norm": 0.6396554708480835,
+      "learning_rate": 1.1736061095017793e-06,
+      "loss": 0.0247,
+      "step": 10390
+    },
+    {
+      "epoch": 18.705035971223023,
+      "grad_norm": 0.4424918293952942,
+      "learning_rate": 1.1417948519411303e-06,
+      "loss": 0.0226,
+      "step": 10400
+    },
+    {
+      "epoch": 18.723021582733814,
+      "grad_norm": 0.860052764415741,
+      "learning_rate": 1.1104156902839557e-06,
+      "loss": 0.0224,
+      "step": 10410
+    },
+    {
+      "epoch": 18.741007194244606,
+      "grad_norm": 0.4110821485519409,
+      "learning_rate": 1.0794689020436678e-06,
+      "loss": 0.0223,
+      "step": 10420
+    },
+    {
+      "epoch": 18.758992805755394,
+      "grad_norm": 0.5921666622161865,
+      "learning_rate": 1.048954760909826e-06,
+      "loss": 0.0224,
+      "step": 10430
+    },
+    {
+      "epoch": 18.776978417266186,
+      "grad_norm": 0.5039754509925842,
+      "learning_rate": 1.0188735367456959e-06,
+      "loss": 0.0268,
+      "step": 10440
+    },
+    {
+      "epoch": 18.794964028776977,
+      "grad_norm": 0.5211593508720398,
+      "learning_rate": 9.892254955859103e-07,
+      "loss": 0.0208,
+      "step": 10450
+    },
+    {
+      "epoch": 18.81294964028777,
+      "grad_norm": 0.25734078884124756,
+      "learning_rate": 9.600108996340562e-07,
+      "loss": 0.021,
+      "step": 10460
+    },
+    {
+      "epoch": 18.83093525179856,
+      "grad_norm": 0.5847137570381165,
+      "learning_rate": 9.312300072603974e-07,
+      "loss": 0.0236,
+      "step": 10470
+    },
+    {
+      "epoch": 18.848920863309353,
+      "grad_norm": 0.40292730927467346,
+      "learning_rate": 9.028830729995608e-07,
+      "loss": 0.0227,
+      "step": 10480
+    },
+    {
+      "epoch": 18.866906474820144,
+      "grad_norm": 0.48411720991134644,
+      "learning_rate": 8.749703475483373e-07,
+      "loss": 0.0236,
+      "step": 10490
+    },
+    {
+      "epoch": 18.884892086330936,
+      "grad_norm": 0.6680406332015991,
+      "learning_rate": 8.474920777633954e-07,
+      "loss": 0.0237,
+      "step": 10500
+    },
+    {
+      "epoch": 18.902877697841728,
+      "grad_norm": 0.34248071908950806,
+      "learning_rate": 8.204485066591383e-07,
+      "loss": 0.0225,
+      "step": 10510
+    },
+    {
+      "epoch": 18.92086330935252,
+      "grad_norm": 0.33673006296157837,
+      "learning_rate": 7.938398734055663e-07,
+      "loss": 0.0249,
+      "step": 10520
+    },
+    {
+      "epoch": 18.93884892086331,
+      "grad_norm": 1.165103793144226,
+      "learning_rate": 7.67666413326118e-07,
+      "loss": 0.024,
+      "step": 10530
+    },
+    {
+      "epoch": 18.9568345323741,
+      "grad_norm": 0.32007914781570435,
+      "learning_rate": 7.419283578956382e-07,
+      "loss": 0.0224,
+      "step": 10540
+    },
+    {
+      "epoch": 18.97482014388489,
+      "grad_norm": 0.34726402163505554,
+      "learning_rate": 7.166259347382854e-07,
+      "loss": 0.0233,
+      "step": 10550
+    },
+    {
+      "epoch": 18.992805755395683,
+      "grad_norm": 0.3968420922756195,
+      "learning_rate": 6.917593676255496e-07,
+      "loss": 0.0223,
+      "step": 10560
+    },
+    {
+      "epoch": 19.010791366906474,
+      "grad_norm": 0.32797303795814514,
+      "learning_rate": 6.673288764742713e-07,
+      "loss": 0.0276,
+      "step": 10570
+    },
+    {
+      "epoch": 19.028776978417266,
+      "grad_norm": 0.6171910762786865,
+      "learning_rate": 6.433346773446758e-07,
+      "loss": 0.0226,
+      "step": 10580
+    },
+    {
+      "epoch": 19.046762589928058,
+      "grad_norm": 0.7115074396133423,
+      "learning_rate": 6.197769824384914e-07,
+      "loss": 0.0238,
+      "step": 10590
+    },
+    {
+      "epoch": 19.06474820143885,
+      "grad_norm": 0.37085476517677307,
+      "learning_rate": 5.966560000970456e-07,
+      "loss": 0.0203,
+      "step": 10600
+    },
+    {
+      "epoch": 19.08273381294964,
+      "grad_norm": 0.35244327783584595,
+      "learning_rate": 5.739719347994333e-07,
+      "loss": 0.0211,
+      "step": 10610
+    },
+    {
+      "epoch": 19.100719424460433,
+      "grad_norm": 0.6677242517471313,
+      "learning_rate": 5.517249871607122e-07,
+      "loss": 0.0249,
+      "step": 10620
+    },
+    {
+      "epoch": 19.118705035971225,
+      "grad_norm": 0.7113429307937622,
+      "learning_rate": 5.299153539301327e-07,
+      "loss": 0.0215,
+      "step": 10630
+    },
+    {
+      "epoch": 19.136690647482013,
+      "grad_norm": 0.673292875289917,
+      "learning_rate": 5.085432279893832e-07,
+      "loss": 0.0217,
+      "step": 10640
+    },
+    {
+      "epoch": 19.154676258992804,
+      "grad_norm": 0.790185809135437,
+      "learning_rate": 4.876087983508859e-07,
+      "loss": 0.0179,
+      "step": 10650
+    },
+    {
+      "epoch": 19.172661870503596,
+      "grad_norm": 0.30639782547950745,
+      "learning_rate": 4.671122501561487e-07,
+      "loss": 0.0248,
+      "step": 10660
+    },
+    {
+      "epoch": 19.190647482014388,
+      "grad_norm": 1.0782285928726196,
+      "learning_rate": 4.4705376467409356e-07,
+      "loss": 0.02,
+      "step": 10670
+    },
+    {
+      "epoch": 19.20863309352518,
+      "grad_norm": 0.6495368480682373,
+      "learning_rate": 4.274335192994805e-07,
+      "loss": 0.0227,
+      "step": 10680
+    },
+    {
+      "epoch": 19.22661870503597,
+      "grad_norm": 0.503280758857727,
+      "learning_rate": 4.082516875513254e-07,
+      "loss": 0.0228,
+      "step": 10690
+    },
+    {
+      "epoch": 19.244604316546763,
+      "grad_norm": 0.43824049830436707,
+      "learning_rate": 3.895084390713732e-07,
+      "loss": 0.0229,
+      "step": 10700
+    },
+    {
+      "epoch": 19.262589928057555,
+      "grad_norm": 0.5866854190826416,
+      "learning_rate": 3.7120393962257195e-07,
+      "loss": 0.0212,
+      "step": 10710
+    },
+    {
+      "epoch": 19.280575539568346,
+      "grad_norm": 0.741484522819519,
+      "learning_rate": 3.5333835108765644e-07,
+      "loss": 0.0236,
+      "step": 10720
+    },
+    {
+      "epoch": 19.298561151079138,
+      "grad_norm": 0.641202449798584,
+      "learning_rate": 3.359118314676668e-07,
+      "loss": 0.0217,
+      "step": 10730
+    },
+    {
+      "epoch": 19.31654676258993,
+      "grad_norm": 0.413107693195343,
+      "learning_rate": 3.1892453488058803e-07,
+      "loss": 0.0221,
+      "step": 10740
+    },
+    {
+      "epoch": 19.334532374100718,
+      "grad_norm": 0.6694846749305725,
+      "learning_rate": 3.023766115599569e-07,
+      "loss": 0.0239,
+      "step": 10750
+    },
+    {
+      "epoch": 19.35251798561151,
+      "grad_norm": 0.8946942687034607,
+      "learning_rate": 2.8626820785355726e-07,
+      "loss": 0.0217,
+      "step": 10760
+    },
+    {
+      "epoch": 19.3705035971223,
+      "grad_norm": 0.508875846862793,
+      "learning_rate": 2.705994662221156e-07,
+      "loss": 0.0245,
+      "step": 10770
+    },
+    {
+      "epoch": 19.388489208633093,
+      "grad_norm": 0.698316752910614,
+      "learning_rate": 2.5537052523803006e-07,
+      "loss": 0.0245,
+      "step": 10780
+    },
+    {
+      "epoch": 19.406474820143885,
+      "grad_norm": 0.3139958083629608,
+      "learning_rate": 2.4058151958417094e-07,
+      "loss": 0.0227,
+      "step": 10790
+    },
+    {
+      "epoch": 19.424460431654676,
+      "grad_norm": 0.4265716075897217,
+      "learning_rate": 2.2623258005266546e-07,
+      "loss": 0.025,
+      "step": 10800
+    },
+    {
+      "epoch": 19.442446043165468,
+      "grad_norm": 0.47745075821876526,
+      "learning_rate": 2.123238335437594e-07,
+      "loss": 0.0246,
+      "step": 10810
+    },
+    {
+      "epoch": 19.46043165467626,
+      "grad_norm": 0.5540276169776917,
+      "learning_rate": 1.9885540306467387e-07,
+      "loss": 0.0246,
+      "step": 10820
+    },
+    {
+      "epoch": 19.47841726618705,
+      "grad_norm": 0.40289539098739624,
+      "learning_rate": 1.8582740772853935e-07,
+      "loss": 0.0251,
+      "step": 10830
+    },
+    {
+      "epoch": 19.496402877697843,
+      "grad_norm": 0.6852353811264038,
+      "learning_rate": 1.7323996275333544e-07,
+      "loss": 0.0232,
+      "step": 10840
+    },
+    {
+      "epoch": 19.514388489208635,
+      "grad_norm": 0.42187145352363586,
+      "learning_rate": 1.6109317946085833e-07,
+      "loss": 0.024,
+      "step": 10850
+    },
+    {
+      "epoch": 19.532374100719423,
+      "grad_norm": 0.5259295105934143,
+      "learning_rate": 1.4938716527575502e-07,
+      "loss": 0.02,
+      "step": 10860
+    },
+    {
+      "epoch": 19.550359712230215,
+      "grad_norm": 1.0009793043136597,
+      "learning_rate": 1.381220237245573e-07,
+      "loss": 0.0227,
+      "step": 10870
+    },
+    {
+      "epoch": 19.568345323741006,
+      "grad_norm": 0.5160190463066101,
+      "learning_rate": 1.272978544347825e-07,
+      "loss": 0.0188,
+      "step": 10880
+    },
+    {
+      "epoch": 19.586330935251798,
+      "grad_norm": 0.43192073702812195,
+      "learning_rate": 1.1691475313403976e-07,
+      "loss": 0.0213,
+      "step": 10890
+    },
+    {
+      "epoch": 19.60431654676259,
+      "grad_norm": 0.4306440055370331,
+      "learning_rate": 1.0697281164919193e-07,
+      "loss": 0.023,
+      "step": 10900
+    },
+    {
+      "epoch": 19.62230215827338,
+      "grad_norm": 0.27822384238243103,
+      "learning_rate": 9.747211790553378e-08,
+      "loss": 0.0222,
+      "step": 10910
+    },
+    {
+      "epoch": 19.640287769784173,
+      "grad_norm": 0.3336507976055145,
+      "learning_rate": 8.841275592601506e-08,
+      "loss": 0.019,
+      "step": 10920
+    },
+    {
+      "epoch": 19.658273381294965,
+      "grad_norm": 0.45413902401924133,
+      "learning_rate": 7.979480583052424e-08,
+      "loss": 0.0254,
+      "step": 10930
+    },
+    {
+      "epoch": 19.676258992805757,
+      "grad_norm": 0.5479503870010376,
+      "learning_rate": 7.161834383512811e-08,
+      "loss": 0.022,
+      "step": 10940
+    },
+    {
+      "epoch": 19.694244604316548,
+      "grad_norm": 0.7516065239906311,
+      "learning_rate": 6.388344225144449e-08,
+      "loss": 0.0215,
+      "step": 10950
+    },
+    {
+      "epoch": 19.71223021582734,
+      "grad_norm": 0.6071555614471436,
+      "learning_rate": 5.6590169485992714e-08,
+      "loss": 0.0197,
+      "step": 10960
+    },
+    {
+      "epoch": 19.730215827338128,
+      "grad_norm": 0.5611714720726013,
+      "learning_rate": 4.973859003956083e-08,
+      "loss": 0.023,
+      "step": 10970
+    },
+    {
+      "epoch": 19.74820143884892,
+      "grad_norm": 0.7659204602241516,
+      "learning_rate": 4.332876450667267e-08,
+      "loss": 0.0224,
+      "step": 10980
+    },
+    {
+      "epoch": 19.76618705035971,
+      "grad_norm": 0.3969743549823761,
+      "learning_rate": 3.736074957502722e-08,
+      "loss": 0.0256,
+      "step": 10990
+    },
+    {
+      "epoch": 19.784172661870503,
+      "grad_norm": 0.614576518535614,
+      "learning_rate": 3.18345980250101e-08,
+      "loss": 0.0241,
+      "step": 11000
+    },
+    {
+      "epoch": 19.802158273381295,
+      "grad_norm": 0.43361860513687134,
+      "learning_rate": 2.6750358729221715e-08,
+      "loss": 0.023,
+      "step": 11010
+    },
+    {
+      "epoch": 19.820143884892087,
+      "grad_norm": 0.5586498379707336,
+      "learning_rate": 2.2108076652038733e-08,
+      "loss": 0.0209,
+      "step": 11020
+    },
+    {
+      "epoch": 19.83812949640288,
+      "grad_norm": 0.9000294208526611,
+      "learning_rate": 1.7907792849231054e-08,
+      "loss": 0.0328,
+      "step": 11030
+    },
+    {
+      "epoch": 19.85611510791367,
+      "grad_norm": 0.6076984405517578,
+      "learning_rate": 1.4149544467584318e-08,
+      "loss": 0.023,
+      "step": 11040
+    },
+    {
+      "epoch": 19.87410071942446,
+      "grad_norm": 0.3142702281475067,
+      "learning_rate": 1.0833364744583519e-08,
+      "loss": 0.023,
+      "step": 11050
+    },
+    {
+      "epoch": 19.892086330935253,
+      "grad_norm": 0.6960598230361938,
+      "learning_rate": 7.959283008113217e-09,
+      "loss": 0.0218,
+      "step": 11060
+    },
+    {
+      "epoch": 19.91007194244604,
+      "grad_norm": 0.3641016483306885,
+      "learning_rate": 5.527324676191104e-09,
+      "loss": 0.0229,
+      "step": 11070
+    },
+    {
+      "epoch": 19.928057553956833,
+      "grad_norm": 0.34496524930000305,
+      "learning_rate": 3.5375112567570532e-09,
+      "loss": 0.0254,
+      "step": 11080
+    },
+    {
+      "epoch": 19.946043165467625,
+      "grad_norm": 0.6787444949150085,
+      "learning_rate": 1.9898603474677314e-09,
+      "loss": 0.0241,
+      "step": 11090
+    },
+    {
+      "epoch": 19.964028776978417,
+      "grad_norm": 0.7640455365180969,
+      "learning_rate": 8.843856355689184e-10,
+      "loss": 0.024,
+      "step": 11100
+    },
+    {
+      "epoch": 19.98201438848921,
+      "grad_norm": 0.6631591320037842,
+      "learning_rate": 2.2109689772897846e-10,
+      "loss": 0.0269,
+      "step": 11110
+    },
+    {
+      "epoch": 20.0,
+      "grad_norm": 0.3174937963485718,
+      "learning_rate": 0.0,
+      "loss": 0.0222,
+      "step": 11120
+    },
+    {
+      "epoch": 20.0,
+      "step": 11120,
+      "total_flos": 7.619008418785206e+17,
+      "train_loss": 0.04595141168281758,
+      "train_runtime": 3437.5074,
+      "train_samples_per_second": 103.488,
+      "train_steps_per_second": 3.235
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 11120,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 20,
+  "save_steps": 20000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 7.619008418785206e+17,
+  "train_batch_size": 32,
+  "trial_name": null,
+  "trial_params": null
+}