diff --git a/blip3o_fast_vlm_3b_distillation_dit_learnable_512_model_1024_no_queries_frozen_mobile_condition_last_4_v2_blip3o_sharegpt_after_pretrain_3_cfm/trainer_state.json b/blip3o_fast_vlm_3b_distillation_dit_learnable_512_model_1024_no_queries_frozen_mobile_condition_last_4_v2_blip3o_sharegpt_after_pretrain_3_cfm/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..d98d626204bfdda2a79b37eeb4fb62ab32063933
--- /dev/null
+++ b/blip3o_fast_vlm_3b_distillation_dit_learnable_512_model_1024_no_queries_frozen_mobile_condition_last_4_v2_blip3o_sharegpt_after_pretrain_3_cfm/trainer_state.json
@@ -0,0 +1,3865 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 20.0,
+  "eval_steps": 500,
+  "global_step": 5460,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.03663003663003663,
+      "grad_norm": 0.21381588280200958,
+      "learning_rate": 6.5934065934065935e-06,
+      "loss": 0.6563,
+      "step": 10
+    },
+    {
+      "epoch": 0.07326007326007326,
+      "grad_norm": 0.16336338222026825,
+      "learning_rate": 1.391941391941392e-05,
+      "loss": 0.663,
+      "step": 20
+    },
+    {
+      "epoch": 0.10989010989010989,
+      "grad_norm": 0.13548393547534943,
+      "learning_rate": 2.1245421245421246e-05,
+      "loss": 0.6513,
+      "step": 30
+    },
+    {
+      "epoch": 0.14652014652014653,
+      "grad_norm": 0.12504926323890686,
+      "learning_rate": 2.857142857142857e-05,
+      "loss": 0.6433,
+      "step": 40
+    },
+    {
+      "epoch": 0.18315018315018314,
+      "grad_norm": 0.14509989321231842,
+      "learning_rate": 3.58974358974359e-05,
+      "loss": 0.627,
+      "step": 50
+    },
+    {
+      "epoch": 0.21978021978021978,
+      "grad_norm": 0.12336988747119904,
+      "learning_rate": 4.322344322344323e-05,
+      "loss": 0.6453,
+      "step": 60
+    },
+    {
+      "epoch": 0.2564102564102564,
+      "grad_norm": 0.11007435619831085,
+      "learning_rate": 5.054945054945055e-05,
+      "loss": 0.6319,
+      "step": 70
+    },
+    {
+      "epoch": 0.29304029304029305,
+      "grad_norm": 0.12032974511384964,
+      "learning_rate": 5.787545787545788e-05,
+      "loss": 0.6345,
+      "step": 80
+    },
+    {
+      "epoch": 0.32967032967032966,
+      "grad_norm": 0.0944727212190628,
+      "learning_rate": 6.52014652014652e-05,
+      "loss": 0.6347,
+      "step": 90
+    },
+    {
+      "epoch": 0.3663003663003663,
+      "grad_norm": 0.12763062119483948,
+      "learning_rate": 7.252747252747253e-05,
+      "loss": 0.6296,
+      "step": 100
+    },
+    {
+      "epoch": 0.40293040293040294,
+      "grad_norm": 0.1317506730556488,
+      "learning_rate": 7.985347985347986e-05,
+      "loss": 0.6458,
+      "step": 110
+    },
+    {
+      "epoch": 0.43956043956043955,
+      "grad_norm": 0.1032543033361435,
+      "learning_rate": 8.717948717948718e-05,
+      "loss": 0.6344,
+      "step": 120
+    },
+    {
+      "epoch": 0.47619047619047616,
+      "grad_norm": 0.1516275852918625,
+      "learning_rate": 9.450549450549451e-05,
+      "loss": 0.6285,
+      "step": 130
+    },
+    {
+      "epoch": 0.5128205128205128,
+      "grad_norm": 0.09366358071565628,
+      "learning_rate": 0.00010183150183150183,
+      "loss": 0.6356,
+      "step": 140
+    },
+    {
+      "epoch": 0.5494505494505495,
+      "grad_norm": 0.11660020053386688,
+      "learning_rate": 0.00010915750915750915,
+      "loss": 0.625,
+      "step": 150
+    },
+    {
+      "epoch": 0.5860805860805861,
+      "grad_norm": 0.1604832112789154,
+      "learning_rate": 0.0001164835164835165,
+      "loss": 0.6301,
+      "step": 160
+    },
+    {
+      "epoch": 0.6227106227106227,
+      "grad_norm": 0.11046287417411804,
+      "learning_rate": 0.0001238095238095238,
+      "loss": 0.6213,
+      "step": 170
+    },
+    {
+      "epoch": 0.6593406593406593,
+      "grad_norm": 0.13051177561283112,
+      "learning_rate": 0.00013113553113553114,
+      "loss": 0.6348,
+      "step": 180
+    },
+    {
+      "epoch": 0.6959706959706959,
+      "grad_norm": 0.15282383561134338,
+      "learning_rate": 0.00013846153846153847,
+      "loss": 0.6223,
+      "step": 190
+    },
+    {
+      "epoch": 0.7326007326007326,
+      "grad_norm": 0.09235327690839767,
+      "learning_rate": 0.00014578754578754578,
+      "loss": 0.6302,
+      "step": 200
+    },
+    {
+      "epoch": 0.7692307692307693,
+      "grad_norm": 0.15384027361869812,
+      "learning_rate": 0.0001531135531135531,
+      "loss": 0.6235,
+      "step": 210
+    },
+    {
+      "epoch": 0.8058608058608059,
+      "grad_norm": 0.1622333526611328,
+      "learning_rate": 0.00016043956043956044,
+      "loss": 0.6248,
+      "step": 220
+    },
+    {
+      "epoch": 0.8424908424908425,
+      "grad_norm": 0.1282617151737213,
+      "learning_rate": 0.00016776556776556777,
+      "loss": 0.6217,
+      "step": 230
+    },
+    {
+      "epoch": 0.8791208791208791,
+      "grad_norm": 0.15851661562919617,
+      "learning_rate": 0.0001750915750915751,
+      "loss": 0.6209,
+      "step": 240
+    },
+    {
+      "epoch": 0.9157509157509157,
+      "grad_norm": 0.11857573688030243,
+      "learning_rate": 0.0001824175824175824,
+      "loss": 0.6321,
+      "step": 250
+    },
+    {
+      "epoch": 0.9523809523809523,
+      "grad_norm": 0.16058525443077087,
+      "learning_rate": 0.00018974358974358974,
+      "loss": 0.6237,
+      "step": 260
+    },
+    {
+      "epoch": 0.989010989010989,
+      "grad_norm": 0.12434748560190201,
+      "learning_rate": 0.0001970695970695971,
+      "loss": 0.6194,
+      "step": 270
+    },
+    {
+      "epoch": 1.0256410256410255,
+      "grad_norm": 0.13378217816352844,
+      "learning_rate": 0.00019999934300455693,
+      "loss": 0.6161,
+      "step": 280
+    },
+    {
+      "epoch": 1.0622710622710623,
+      "grad_norm": 0.13045711815357208,
+      "learning_rate": 0.0001999953280638248,
+      "loss": 0.6309,
+      "step": 290
+    },
+    {
+      "epoch": 1.098901098901099,
+      "grad_norm": 0.13684824109077454,
+      "learning_rate": 0.00019998766332693084,
+      "loss": 0.6353,
+      "step": 300
+    },
+    {
+      "epoch": 1.1355311355311355,
+      "grad_norm": 0.13938014209270477,
+      "learning_rate": 0.0001999763490750413,
+      "loss": 0.6194,
+      "step": 310
+    },
+    {
+      "epoch": 1.1721611721611722,
+      "grad_norm": 0.1351129710674286,
+      "learning_rate": 0.00019996138572319787,
+      "loss": 0.6288,
+      "step": 320
+    },
+    {
+      "epoch": 1.2087912087912087,
+      "grad_norm": 0.14925244450569153,
+      "learning_rate": 0.00019994277382030259,
+      "loss": 0.6092,
+      "step": 330
+    },
+    {
+      "epoch": 1.2454212454212454,
+      "grad_norm": 0.10388221591711044,
+      "learning_rate": 0.00019992051404909758,
+      "loss": 0.6346,
+      "step": 340
+    },
+    {
+      "epoch": 1.282051282051282,
+      "grad_norm": 0.12104734778404236,
+      "learning_rate": 0.00019989460722614002,
+      "loss": 0.6315,
+      "step": 350
+    },
+    {
+      "epoch": 1.3186813186813187,
+      "grad_norm": 0.12242613732814789,
+      "learning_rate": 0.0001998650543017723,
+      "loss": 0.623,
+      "step": 360
+    },
+    {
+      "epoch": 1.3553113553113554,
+      "grad_norm": 0.12440565973520279,
+      "learning_rate": 0.000199831856360087,
+      "loss": 0.6134,
+      "step": 370
+    },
+    {
+      "epoch": 1.3919413919413919,
+      "grad_norm": 0.10807174444198608,
+      "learning_rate": 0.0001997950146188872,
+      "loss": 0.6262,
+      "step": 380
+    },
+    {
+      "epoch": 1.4285714285714286,
+      "grad_norm": 0.14467234909534454,
+      "learning_rate": 0.00019975453042964187,
+      "loss": 0.6146,
+      "step": 390
+    },
+    {
+      "epoch": 1.4652014652014653,
+      "grad_norm": 0.12609444558620453,
+      "learning_rate": 0.00019971040527743626,
+      "loss": 0.621,
+      "step": 400
+    },
+    {
+      "epoch": 1.5018315018315018,
+      "grad_norm": 0.09819615632295609,
+      "learning_rate": 0.00019966264078091723,
+      "loss": 0.6162,
+      "step": 410
+    },
+    {
+      "epoch": 1.5384615384615383,
+      "grad_norm": 0.11803926527500153,
+      "learning_rate": 0.00019961123869223423,
+      "loss": 0.6168,
+      "step": 420
+    },
+    {
+      "epoch": 1.575091575091575,
+      "grad_norm": 0.12449613213539124,
+      "learning_rate": 0.00019955620089697474,
+      "loss": 0.6221,
+      "step": 430
+    },
+    {
+      "epoch": 1.6117216117216118,
+      "grad_norm": 0.12128981202840805,
+      "learning_rate": 0.00019949752941409514,
+      "loss": 0.6229,
+      "step": 440
+    },
+    {
+      "epoch": 1.6483516483516483,
+      "grad_norm": 0.14718155562877655,
+      "learning_rate": 0.0001994352263958469,
+      "loss": 0.6173,
+      "step": 450
+    },
+    {
+      "epoch": 1.684981684981685,
+      "grad_norm": 0.14411243796348572,
+      "learning_rate": 0.00019936929412769713,
+      "loss": 0.6337,
+      "step": 460
+    },
+    {
+      "epoch": 1.7216117216117217,
+      "grad_norm": 0.1278337836265564,
+      "learning_rate": 0.00019929973502824533,
+      "loss": 0.6238,
+      "step": 470
+    },
+    {
+      "epoch": 1.7582417582417582,
+      "grad_norm": 0.13230876624584198,
+      "learning_rate": 0.00019922655164913427,
+      "loss": 0.6259,
+      "step": 480
+    },
+    {
+      "epoch": 1.7948717948717947,
+      "grad_norm": 0.12312595546245575,
+      "learning_rate": 0.00019914974667495646,
+      "loss": 0.6125,
+      "step": 490
+    },
+    {
+      "epoch": 1.8315018315018317,
+      "grad_norm": 0.09721134603023529,
+      "learning_rate": 0.00019906932292315573,
+      "loss": 0.6174,
+      "step": 500
+    },
+    {
+      "epoch": 1.8681318681318682,
+      "grad_norm": 0.12478078901767731,
+      "learning_rate": 0.0001989852833439239,
+      "loss": 0.6252,
+      "step": 510
+    },
+    {
+      "epoch": 1.9047619047619047,
+      "grad_norm": 0.11270100623369217,
+      "learning_rate": 0.0001988976310200925,
+      "loss": 0.6216,
+      "step": 520
+    },
+    {
+      "epoch": 1.9413919413919414,
+      "grad_norm": 0.11008603870868683,
+      "learning_rate": 0.0001988063691670197,
+      "loss": 0.6163,
+      "step": 530
+    },
+    {
+      "epoch": 1.978021978021978,
+      "grad_norm": 0.11471746116876602,
+      "learning_rate": 0.0001987115011324723,
+      "loss": 0.6328,
+      "step": 540
+    },
+    {
+      "epoch": 2.0146520146520146,
+      "grad_norm": 0.11322673410177231,
+      "learning_rate": 0.00019861303039650315,
+      "loss": 0.6099,
+      "step": 550
+    },
+    {
+      "epoch": 2.051282051282051,
+      "grad_norm": 0.11044928431510925,
+      "learning_rate": 0.00019851096057132308,
+      "loss": 0.619,
+      "step": 560
+    },
+    {
+      "epoch": 2.087912087912088,
+      "grad_norm": 0.09827021509408951,
+      "learning_rate": 0.0001984052954011688,
+      "loss": 0.6299,
+      "step": 570
+    },
+    {
+      "epoch": 2.1245421245421245,
+      "grad_norm": 0.12820111215114594,
+      "learning_rate": 0.00019829603876216537,
+      "loss": 0.6088,
+      "step": 580
+    },
+    {
+      "epoch": 2.161172161172161,
+      "grad_norm": 0.10461625456809998,
+      "learning_rate": 0.000198183194662184,
+      "loss": 0.6251,
+      "step": 590
+    },
+    {
+      "epoch": 2.197802197802198,
+      "grad_norm": 0.13258452713489532,
+      "learning_rate": 0.00019806676724069505,
+      "loss": 0.6124,
+      "step": 600
+    },
+    {
+      "epoch": 2.2344322344322345,
+      "grad_norm": 0.11297419667243958,
+      "learning_rate": 0.0001979467607686162,
+      "loss": 0.617,
+      "step": 610
+    },
+    {
+      "epoch": 2.271062271062271,
+      "grad_norm": 0.11036292463541031,
+      "learning_rate": 0.00019782317964815585,
+      "loss": 0.6274,
+      "step": 620
+    },
+    {
+      "epoch": 2.3076923076923075,
+      "grad_norm": 0.09046752005815506,
+      "learning_rate": 0.00019769602841265135,
+      "loss": 0.6175,
+      "step": 630
+    },
+    {
+      "epoch": 2.3443223443223444,
+      "grad_norm": 0.15168985724449158,
+      "learning_rate": 0.00019756531172640312,
+      "loss": 0.611,
+      "step": 640
+    },
+    {
+      "epoch": 2.380952380952381,
+      "grad_norm": 0.12152957171201706,
+      "learning_rate": 0.00019743103438450322,
+      "loss": 0.6206,
+      "step": 650
+    },
+    {
+      "epoch": 2.4175824175824174,
+      "grad_norm": 0.11831982433795929,
+      "learning_rate": 0.00019729320131265957,
+      "loss": 0.6092,
+      "step": 660
+    },
+    {
+      "epoch": 2.4542124542124544,
+      "grad_norm": 0.12286005914211273,
+      "learning_rate": 0.00019715181756701526,
+      "loss": 0.6131,
+      "step": 670
+    },
+    {
+      "epoch": 2.490842490842491,
+      "grad_norm": 0.14576148986816406,
+      "learning_rate": 0.00019700688833396314,
+      "loss": 0.6149,
+      "step": 680
+    },
+    {
+      "epoch": 2.5274725274725274,
+      "grad_norm": 0.13091808557510376,
+      "learning_rate": 0.00019685841892995537,
+      "loss": 0.6204,
+      "step": 690
+    },
+    {
+      "epoch": 2.564102564102564,
+      "grad_norm": 0.09313470125198364,
+      "learning_rate": 0.00019670641480130861,
+      "loss": 0.6148,
+      "step": 700
+    },
+    {
+      "epoch": 2.600732600732601,
+      "grad_norm": 0.12012402713298798,
+      "learning_rate": 0.00019655088152400416,
+      "loss": 0.6219,
+      "step": 710
+    },
+    {
+      "epoch": 2.6373626373626373,
+      "grad_norm": 0.13228215277194977,
+      "learning_rate": 0.00019639182480348332,
+      "loss": 0.6226,
+      "step": 720
+    },
+    {
+      "epoch": 2.6739926739926743,
+      "grad_norm": 0.12784317135810852,
+      "learning_rate": 0.00019622925047443825,
+      "loss": 0.6188,
+      "step": 730
+    },
+    {
+      "epoch": 2.7106227106227108,
+      "grad_norm": 0.12311346083879471,
+      "learning_rate": 0.00019606316450059786,
+      "loss": 0.6174,
+      "step": 740
+    },
+    {
+      "epoch": 2.7472527472527473,
+      "grad_norm": 0.11249978840351105,
+      "learning_rate": 0.00019589357297450897,
+      "loss": 0.6086,
+      "step": 750
+    },
+    {
+      "epoch": 2.7838827838827838,
+      "grad_norm": 0.0970926508307457,
+      "learning_rate": 0.00019572048211731295,
+      "loss": 0.6099,
+      "step": 760
+    },
+    {
+      "epoch": 2.8205128205128203,
+      "grad_norm": 0.1097983717918396,
+      "learning_rate": 0.00019554389827851752,
+      "loss": 0.6291,
+      "step": 770
+    },
+    {
+      "epoch": 2.857142857142857,
+      "grad_norm": 0.11450190097093582,
+      "learning_rate": 0.0001953638279357636,
+      "loss": 0.6158,
+      "step": 780
+    },
+    {
+      "epoch": 2.8937728937728937,
+      "grad_norm": 0.08861864358186722,
+      "learning_rate": 0.00019518027769458799,
+      "loss": 0.6255,
+      "step": 790
+    },
+    {
+      "epoch": 2.9304029304029307,
+      "grad_norm": 0.10655124485492706,
+      "learning_rate": 0.00019499325428818084,
+      "loss": 0.6149,
+      "step": 800
+    },
+    {
+      "epoch": 2.967032967032967,
+      "grad_norm": 0.1027015745639801,
+      "learning_rate": 0.00019480276457713874,
+      "loss": 0.6182,
+      "step": 810
+    },
+    {
+      "epoch": 3.0036630036630036,
+      "grad_norm": 0.12306101620197296,
+      "learning_rate": 0.00019460881554921315,
+      "loss": 0.6164,
+      "step": 820
+    },
+    {
+      "epoch": 3.04029304029304,
+      "grad_norm": 0.11898409575223923,
+      "learning_rate": 0.00019441141431905383,
+      "loss": 0.6062,
+      "step": 830
+    },
+    {
+      "epoch": 3.076923076923077,
+      "grad_norm": 0.1291988044977188,
+      "learning_rate": 0.0001942105681279481,
+      "loss": 0.6258,
+      "step": 840
+    },
+    {
+      "epoch": 3.1135531135531136,
+      "grad_norm": 0.11260422319173813,
+      "learning_rate": 0.00019400628434355495,
+      "loss": 0.6248,
+      "step": 850
+    },
+    {
+      "epoch": 3.15018315018315,
+      "grad_norm": 0.10207366943359375,
+      "learning_rate": 0.00019379857045963509,
+      "loss": 0.6159,
+      "step": 860
+    },
+    {
+      "epoch": 3.186813186813187,
+      "grad_norm": 0.10797443240880966,
+      "learning_rate": 0.00019358743409577577,
+      "loss": 0.6134,
+      "step": 870
+    },
+    {
+      "epoch": 3.2234432234432235,
+      "grad_norm": 0.09320534765720367,
+      "learning_rate": 0.00019337288299711144,
+      "loss": 0.6104,
+      "step": 880
+    },
+    {
+      "epoch": 3.26007326007326,
+      "grad_norm": 0.15332849323749542,
+      "learning_rate": 0.0001931549250340395,
+      "loss": 0.6178,
+      "step": 890
+    },
+    {
+      "epoch": 3.2967032967032965,
+      "grad_norm": 0.12163515388965607,
+      "learning_rate": 0.00019293356820193177,
+      "loss": 0.606,
+      "step": 900
+    },
+    {
+      "epoch": 3.3333333333333335,
+      "grad_norm": 0.12709611654281616,
+      "learning_rate": 0.00019270882062084099,
+      "loss": 0.6104,
+      "step": 910
+    },
+    {
+      "epoch": 3.36996336996337,
+      "grad_norm": 0.11496417224407196,
+      "learning_rate": 0.00019248069053520316,
+      "loss": 0.6177,
+      "step": 920
+    },
+    {
+      "epoch": 3.4065934065934065,
+      "grad_norm": 0.13935399055480957,
+      "learning_rate": 0.0001922491863135349,
+      "loss": 0.6159,
+      "step": 930
+    },
+    {
+      "epoch": 3.4432234432234434,
+      "grad_norm": 0.11748811602592468,
+      "learning_rate": 0.0001920143164481266,
+      "loss": 0.6223,
+      "step": 940
+    },
+    {
+      "epoch": 3.47985347985348,
+      "grad_norm": 0.11896082758903503,
+      "learning_rate": 0.00019177608955473096,
+      "loss": 0.6176,
+      "step": 950
+    },
+    {
+      "epoch": 3.5164835164835164,
+      "grad_norm": 0.12537701427936554,
+      "learning_rate": 0.0001915345143722467,
+      "loss": 0.6108,
+      "step": 960
+    },
+    {
+      "epoch": 3.553113553113553,
+      "grad_norm": 0.172188863158226,
+      "learning_rate": 0.0001912895997623982,
+      "loss": 0.6186,
+      "step": 970
+    },
+    {
+      "epoch": 3.58974358974359,
+      "grad_norm": 0.12360432744026184,
+      "learning_rate": 0.00019104135470941034,
+      "loss": 0.6219,
+      "step": 980
+    },
+    {
+      "epoch": 3.6263736263736264,
+      "grad_norm": 0.15684974193572998,
+      "learning_rate": 0.000190789788319679,
+      "loss": 0.6084,
+      "step": 990
+    },
+    {
+      "epoch": 3.663003663003663,
+      "grad_norm": 0.14988037943840027,
+      "learning_rate": 0.00019053490982143687,
+      "loss": 0.6028,
+      "step": 1000
+    },
+    {
+      "epoch": 3.6996336996337,
+      "grad_norm": 0.12309486418962479,
+      "learning_rate": 0.00019027672856441512,
+      "loss": 0.621,
+      "step": 1010
+    },
+    {
+      "epoch": 3.7362637362637363,
+      "grad_norm": 0.11249048262834549,
+      "learning_rate": 0.00019001525401950018,
+      "loss": 0.6088,
+      "step": 1020
+    },
+    {
+      "epoch": 3.772893772893773,
+      "grad_norm": 0.12147108465433121,
+      "learning_rate": 0.00018975049577838664,
+      "loss": 0.6169,
+      "step": 1030
+    },
+    {
+      "epoch": 3.8095238095238093,
+      "grad_norm": 0.11105571687221527,
+      "learning_rate": 0.0001894824635532251,
+      "loss": 0.6047,
+      "step": 1040
+    },
+    {
+      "epoch": 3.8461538461538463,
+      "grad_norm": 0.12829916179180145,
+      "learning_rate": 0.00018921116717626594,
+      "loss": 0.6197,
+      "step": 1050
+    },
+    {
+      "epoch": 3.8827838827838828,
+      "grad_norm": 0.1178424134850502,
+      "learning_rate": 0.00018893661659949895,
+      "loss": 0.602,
+      "step": 1060
+    },
+    {
+      "epoch": 3.9194139194139193,
+      "grad_norm": 0.09845280647277832,
+      "learning_rate": 0.00018865882189428778,
+      "loss": 0.61,
+      "step": 1070
+    },
+    {
+      "epoch": 3.956043956043956,
+      "grad_norm": 0.1087396964430809,
+      "learning_rate": 0.0001883777932510009,
+      "loss": 0.6275,
+      "step": 1080
+    },
+    {
+      "epoch": 3.9926739926739927,
+      "grad_norm": 0.15589557588100433,
+      "learning_rate": 0.0001880935409786376,
+      "loss": 0.6167,
+      "step": 1090
+    },
+    {
+      "epoch": 4.029304029304029,
+      "grad_norm": 0.0800589993596077,
+      "learning_rate": 0.00018780607550444978,
+      "loss": 0.6014,
+      "step": 1100
+    },
+    {
+      "epoch": 4.065934065934066,
+      "grad_norm": 0.11557458341121674,
+      "learning_rate": 0.00018751540737355958,
+      "loss": 0.6157,
+      "step": 1110
+    },
+    {
+      "epoch": 4.102564102564102,
+      "grad_norm": 0.11841552704572678,
+      "learning_rate": 0.00018722154724857241,
+      "loss": 0.6131,
+      "step": 1120
+    },
+    {
+      "epoch": 4.13919413919414,
+      "grad_norm": 0.09185516089200974,
+      "learning_rate": 0.00018692450590918598,
+      "loss": 0.6151,
+      "step": 1130
+    },
+    {
+      "epoch": 4.175824175824176,
+      "grad_norm": 0.1254640370607376,
+      "learning_rate": 0.00018662429425179476,
+      "loss": 0.6158,
+      "step": 1140
+    },
+    {
+      "epoch": 4.212454212454213,
+      "grad_norm": 0.08927269279956818,
+      "learning_rate": 0.00018632092328909022,
+      "loss": 0.6092,
+      "step": 1150
+    },
+    {
+      "epoch": 4.249084249084249,
+      "grad_norm": 0.09850620478391647,
+      "learning_rate": 0.000186014404149657,
+      "loss": 0.602,
+      "step": 1160
+    },
+    {
+      "epoch": 4.285714285714286,
+      "grad_norm": 0.12001339346170425,
+      "learning_rate": 0.00018570474807756452,
+      "loss": 0.6041,
+      "step": 1170
+    },
+    {
+      "epoch": 4.322344322344322,
+      "grad_norm": 0.11746009439229965,
+      "learning_rate": 0.0001853919664319546,
+      "loss": 0.6096,
+      "step": 1180
+    },
+    {
+      "epoch": 4.358974358974359,
+      "grad_norm": 0.10340318828821182,
+      "learning_rate": 0.0001850760706866248,
+      "loss": 0.6023,
+      "step": 1190
+    },
+    {
+      "epoch": 4.395604395604396,
+      "grad_norm": 0.11639545857906342,
+      "learning_rate": 0.00018475707242960742,
+      "loss": 0.6191,
+      "step": 1200
+    },
+    {
+      "epoch": 4.4322344322344325,
+      "grad_norm": 0.15589426457881927,
+      "learning_rate": 0.00018443498336274462,
+      "loss": 0.6054,
+      "step": 1210
+    },
+    {
+      "epoch": 4.468864468864469,
+      "grad_norm": 0.09896785765886307,
+      "learning_rate": 0.00018410981530125875,
+      "loss": 0.6215,
+      "step": 1220
+    },
+    {
+      "epoch": 4.5054945054945055,
+      "grad_norm": 0.12751416862010956,
+      "learning_rate": 0.0001837815801733195,
+      "loss": 0.5995,
+      "step": 1230
+    },
+    {
+      "epoch": 4.542124542124542,
+      "grad_norm": 0.08719929307699203,
+      "learning_rate": 0.0001834502900196058,
+      "loss": 0.6196,
+      "step": 1240
+    },
+    {
+      "epoch": 4.5787545787545785,
+      "grad_norm": 0.1458660364151001,
+      "learning_rate": 0.00018311595699286454,
+      "loss": 0.6178,
+      "step": 1250
+    },
+    {
+      "epoch": 4.615384615384615,
+      "grad_norm": 0.14082370698451996,
+      "learning_rate": 0.0001827785933574645,
+      "loss": 0.6038,
+      "step": 1260
+    },
+    {
+      "epoch": 4.652014652014652,
+      "grad_norm": 0.09783344715833664,
+      "learning_rate": 0.00018243821148894654,
+      "loss": 0.6152,
+      "step": 1270
+    },
+    {
+      "epoch": 4.688644688644689,
+      "grad_norm": 0.13895457983016968,
+      "learning_rate": 0.00018209482387356978,
+      "loss": 0.611,
+      "step": 1280
+    },
+    {
+      "epoch": 4.725274725274725,
+      "grad_norm": 0.12415605038404465,
+      "learning_rate": 0.00018174844310785322,
+      "loss": 0.6189,
+      "step": 1290
+    },
+    {
+      "epoch": 4.761904761904762,
+      "grad_norm": 0.1300516426563263,
+      "learning_rate": 0.000181399081898114,
+      "loss": 0.601,
+      "step": 1300
+    },
+    {
+      "epoch": 4.798534798534798,
+      "grad_norm": 0.10010094940662384,
+      "learning_rate": 0.0001810467530600011,
+      "loss": 0.6111,
+      "step": 1310
+    },
+    {
+      "epoch": 4.835164835164835,
+      "grad_norm": 0.10929550230503082,
+      "learning_rate": 0.00018069146951802542,
+      "loss": 0.6053,
+      "step": 1320
+    },
+    {
+      "epoch": 4.871794871794872,
+      "grad_norm": 0.1457078754901886,
+      "learning_rate": 0.0001803332443050853,
+      "loss": 0.6112,
+      "step": 1330
+    },
+    {
+      "epoch": 4.908424908424909,
+      "grad_norm": 0.1061316654086113,
+      "learning_rate": 0.00017997209056198882,
+      "loss": 0.6199,
+      "step": 1340
+    },
+    {
+      "epoch": 4.945054945054945,
+      "grad_norm": 0.1740475445985794,
+      "learning_rate": 0.0001796080215369716,
+      "loss": 0.6048,
+      "step": 1350
+    },
+    {
+      "epoch": 4.981684981684982,
+      "grad_norm": 0.12337993085384369,
+      "learning_rate": 0.0001792410505852107,
+      "loss": 0.6091,
+      "step": 1360
+    },
+    {
+      "epoch": 5.018315018315018,
+      "grad_norm": 0.11758992075920105,
+      "learning_rate": 0.00017887119116833495,
+      "loss": 0.62,
+      "step": 1370
+    },
+    {
+      "epoch": 5.054945054945055,
+      "grad_norm": 0.11902181804180145,
+      "learning_rate": 0.00017849845685393093,
+      "loss": 0.6145,
+      "step": 1380
+    },
+    {
+      "epoch": 5.091575091575091,
+      "grad_norm": 0.10019928216934204,
+      "learning_rate": 0.0001781228613150454,
+      "loss": 0.6124,
+      "step": 1390
+    },
+    {
+      "epoch": 5.128205128205128,
+      "grad_norm": 0.14300206303596497,
+      "learning_rate": 0.0001777444183296836,
+      "loss": 0.6141,
+      "step": 1400
+    },
+    {
+      "epoch": 5.164835164835165,
+      "grad_norm": 0.15143392980098724,
+      "learning_rate": 0.000177363141780304,
+      "loss": 0.6066,
+      "step": 1410
+    },
+    {
+      "epoch": 5.201465201465202,
+      "grad_norm": 0.11212984472513199,
+      "learning_rate": 0.0001769790456533089,
+      "loss": 0.6041,
+      "step": 1420
+    },
+    {
+      "epoch": 5.238095238095238,
+      "grad_norm": 0.11619605123996735,
+      "learning_rate": 0.00017659214403853145,
+      "loss": 0.614,
+      "step": 1430
+    },
+    {
+      "epoch": 5.274725274725275,
+      "grad_norm": 0.10530748963356018,
+      "learning_rate": 0.00017620245112871867,
+      "loss": 0.6131,
+      "step": 1440
+    },
+    {
+      "epoch": 5.311355311355311,
+      "grad_norm": 0.18820564448833466,
+      "learning_rate": 0.00017580998121901107,
+      "loss": 0.6058,
+      "step": 1450
+    },
+    {
+      "epoch": 5.347985347985348,
+      "grad_norm": 0.10431553423404694,
+      "learning_rate": 0.00017541474870641797,
+      "loss": 0.6058,
+      "step": 1460
+    },
+    {
+      "epoch": 5.384615384615385,
+      "grad_norm": 0.09268366545438766,
+      "learning_rate": 0.0001750167680892896,
+      "loss": 0.5981,
+      "step": 1470
+    },
+    {
+      "epoch": 5.4212454212454215,
+      "grad_norm": 0.1386278122663498,
+      "learning_rate": 0.00017461605396678506,
+      "loss": 0.6086,
+      "step": 1480
+    },
+    {
+      "epoch": 5.457875457875458,
+      "grad_norm": 0.11717536300420761,
+      "learning_rate": 0.000174212621038337,
+      "loss": 0.6032,
+      "step": 1490
+    },
+    {
+      "epoch": 5.4945054945054945,
+      "grad_norm": 0.10783743113279343,
+      "learning_rate": 0.00017380648410311207,
+      "loss": 0.5993,
+      "step": 1500
+    },
+    {
+      "epoch": 5.531135531135531,
+      "grad_norm": 0.13064159452915192,
+      "learning_rate": 0.00017339765805946846,
+      "loss": 0.6107,
+      "step": 1510
+    },
+    {
+      "epoch": 5.5677655677655675,
+      "grad_norm": 0.12426242977380753,
+      "learning_rate": 0.0001729861579044091,
+      "loss": 0.6029,
+      "step": 1520
+    },
+    {
+      "epoch": 5.604395604395604,
+      "grad_norm": 0.12688453495502472,
+      "learning_rate": 0.00017257199873303155,
+      "loss": 0.6154,
+      "step": 1530
+    },
+    {
+      "epoch": 5.641025641025641,
+      "grad_norm": 0.11696295440196991,
+      "learning_rate": 0.0001721551957379743,
+      "loss": 0.61,
+      "step": 1540
+    },
+    {
+      "epoch": 5.677655677655678,
+      "grad_norm": 0.11864680051803589,
+      "learning_rate": 0.00017173576420885956,
+      "loss": 0.6085,
+      "step": 1550
+    },
+    {
+      "epoch": 5.714285714285714,
+      "grad_norm": 0.09387882798910141,
+      "learning_rate": 0.00017131371953173222,
+      "loss": 0.6015,
+      "step": 1560
+    },
+    {
+      "epoch": 5.750915750915751,
+      "grad_norm": 0.09436681121587753,
+      "learning_rate": 0.00017088907718849545,
+      "loss": 0.6032,
+      "step": 1570
+    },
+    {
+      "epoch": 5.787545787545787,
+      "grad_norm": 0.11666887998580933,
+      "learning_rate": 0.00017046185275634303,
+      "loss": 0.6067,
+      "step": 1580
+    },
+    {
+      "epoch": 5.824175824175824,
+      "grad_norm": 0.11687021702528,
+      "learning_rate": 0.00017003206190718758,
+      "loss": 0.6112,
+      "step": 1590
+    },
+    {
+      "epoch": 5.860805860805861,
+      "grad_norm": 0.11235155165195465,
+      "learning_rate": 0.00016959972040708586,
+      "loss": 0.6169,
+      "step": 1600
+    },
+    {
+      "epoch": 5.897435897435898,
+      "grad_norm": 0.10137765854597092,
+      "learning_rate": 0.00016916484411566045,
+      "loss": 0.6028,
+      "step": 1610
+    },
+    {
+      "epoch": 5.934065934065934,
+      "grad_norm": 0.09540582448244095,
+      "learning_rate": 0.00016872744898551787,
+      "loss": 0.6091,
+      "step": 1620
+    },
+    {
+      "epoch": 5.970695970695971,
+      "grad_norm": 0.08847875148057938,
+      "learning_rate": 0.0001682875510616634,
+      "loss": 0.6279,
+      "step": 1630
+    },
+    {
+      "epoch": 6.007326007326007,
+      "grad_norm": 0.13143204152584076,
+      "learning_rate": 0.00016784516648091254,
+      "loss": 0.601,
+      "step": 1640
+    },
+    {
+      "epoch": 6.043956043956044,
+      "grad_norm": 0.11306729912757874,
+      "learning_rate": 0.00016740031147129917,
+      "loss": 0.6093,
+      "step": 1650
+    },
+    {
+      "epoch": 6.08058608058608,
+      "grad_norm": 0.08409835398197174,
+      "learning_rate": 0.00016695300235147994,
+      "loss": 0.6051,
+      "step": 1660
+    },
+    {
+      "epoch": 6.117216117216117,
+      "grad_norm": 0.0728335753083229,
+      "learning_rate": 0.000166503255530136,
+      "loss": 0.5947,
+      "step": 1670
+    },
+    {
+      "epoch": 6.153846153846154,
+      "grad_norm": 0.11714360117912292,
+      "learning_rate": 0.00016605108750537083,
+      "loss": 0.6103,
+      "step": 1680
+    },
+    {
+      "epoch": 6.190476190476191,
+      "grad_norm": 0.10780514776706696,
+      "learning_rate": 0.0001655965148641052,
+      "loss": 0.59,
+      "step": 1690
+    },
+    {
+      "epoch": 6.227106227106227,
+      "grad_norm": 0.10457523167133331,
+      "learning_rate": 0.00016513955428146864,
+      "loss": 0.6014,
+      "step": 1700
+    },
+    {
+      "epoch": 6.263736263736264,
+      "grad_norm": 0.11372072249650955,
+      "learning_rate": 0.00016468022252018774,
+      "loss": 0.6053,
+      "step": 1710
+    },
+    {
+      "epoch": 6.3003663003663,
+      "grad_norm": 0.10791448503732681,
+      "learning_rate": 0.00016421853642997118,
+      "loss": 0.6123,
+      "step": 1720
+    },
+    {
+      "epoch": 6.336996336996337,
+      "grad_norm": 0.10883180052042007,
+      "learning_rate": 0.00016375451294689186,
+      "loss": 0.6127,
+      "step": 1730
+    },
+    {
+      "epoch": 6.373626373626374,
+      "grad_norm": 0.1315256953239441,
+      "learning_rate": 0.00016328816909276525,
+      "loss": 0.6111,
+      "step": 1740
+    },
+    {
+      "epoch": 6.410256410256411,
+      "grad_norm": 0.10748538374900818,
+      "learning_rate": 0.00016281952197452545,
+      "loss": 0.6016,
+      "step": 1750
+    },
+    {
+      "epoch": 6.446886446886447,
+      "grad_norm": 0.10810081660747528,
+      "learning_rate": 0.00016234858878359725,
+      "loss": 0.6128,
+      "step": 1760
+    },
+    {
+      "epoch": 6.483516483516484,
+      "grad_norm": 0.10564674437046051,
+      "learning_rate": 0.0001618753867952657,
+      "loss": 0.6025,
+      "step": 1770
+    },
+    {
+      "epoch": 6.52014652014652,
+      "grad_norm": 0.13300639390945435,
+      "learning_rate": 0.00016139993336804234,
+      "loss": 0.614,
+      "step": 1780
+    },
+    {
+      "epoch": 6.556776556776557,
+      "grad_norm": 0.09764187037944794,
+      "learning_rate": 0.00016092224594302845,
+      "loss": 0.6077,
+      "step": 1790
+    },
+    {
+      "epoch": 6.593406593406593,
+      "grad_norm": 0.08439364284276962,
+      "learning_rate": 0.00016044234204327527,
+      "loss": 0.6098,
+      "step": 1800
+    },
+    {
+      "epoch": 6.63003663003663,
+      "grad_norm": 0.13017664849758148,
+      "learning_rate": 0.00015996023927314113,
+      "loss": 0.6111,
+      "step": 1810
+    },
+    {
+      "epoch": 6.666666666666667,
+      "grad_norm": 0.08291947841644287,
+      "learning_rate": 0.00015947595531764578,
+      "loss": 0.614,
+      "step": 1820
+    },
+    {
+      "epoch": 6.7032967032967035,
+      "grad_norm": 0.14750197529792786,
+      "learning_rate": 0.0001589895079418216,
+      "loss": 0.5958,
+      "step": 1830
+    },
+    {
+      "epoch": 6.73992673992674,
+      "grad_norm": 0.10692450404167175,
+      "learning_rate": 0.00015850091499006188,
+      "loss": 0.599,
+      "step": 1840
+    },
+    {
+      "epoch": 6.7765567765567765,
+      "grad_norm": 0.13206391036510468,
+      "learning_rate": 0.00015801019438546626,
+      "loss": 0.6066,
+      "step": 1850
+    },
+    {
+      "epoch": 6.813186813186813,
+      "grad_norm": 0.08591387420892715,
+      "learning_rate": 0.00015751736412918327,
+      "loss": 0.6093,
+      "step": 1860
+    },
+    {
+      "epoch": 6.8498168498168495,
+      "grad_norm": 0.09769956022500992,
+      "learning_rate": 0.00015702244229975,
+      "loss": 0.6002,
+      "step": 1870
+    },
+    {
+      "epoch": 6.886446886446887,
+      "grad_norm": 0.10705573111772537,
+      "learning_rate": 0.0001565254470524289,
+      "loss": 0.6071,
+      "step": 1880
+    },
+    {
+      "epoch": 6.923076923076923,
+      "grad_norm": 0.11658084392547607,
+      "learning_rate": 0.00015602639661854183,
+      "loss": 0.607,
+      "step": 1890
+    },
+    {
+      "epoch": 6.95970695970696,
+      "grad_norm": 0.10922308266162872,
+      "learning_rate": 0.00015552530930480114,
+      "loss": 0.6082,
+      "step": 1900
+    },
+    {
+      "epoch": 6.996336996336996,
+      "grad_norm": 0.10610403120517731,
+      "learning_rate": 0.00015502220349263837,
+      "loss": 0.6099,
+      "step": 1910
+    },
+    {
+      "epoch": 7.032967032967033,
+      "grad_norm": 0.14902709424495697,
+      "learning_rate": 0.0001545170976375297,
+      "loss": 0.5997,
+      "step": 1920
+    },
+    {
+      "epoch": 7.069597069597069,
+      "grad_norm": 0.1101558580994606,
+      "learning_rate": 0.00015401001026831915,
+      "loss": 0.6117,
+      "step": 1930
+    },
+    {
+      "epoch": 7.106227106227106,
+      "grad_norm": 0.11664875596761703,
+      "learning_rate": 0.00015350095998653874,
+      "loss": 0.6002,
+      "step": 1940
+    },
+    {
+      "epoch": 7.142857142857143,
+      "grad_norm": 0.13412544131278992,
+      "learning_rate": 0.00015298996546572627,
+      "loss": 0.5914,
+      "step": 1950
+    },
+    {
+      "epoch": 7.17948717948718,
+      "grad_norm": 0.15238548815250397,
+      "learning_rate": 0.00015247704545074008,
+      "loss": 0.6113,
+      "step": 1960
+    },
+    {
+      "epoch": 7.216117216117216,
+      "grad_norm": 0.12290948629379272,
+      "learning_rate": 0.00015196221875707184,
+      "loss": 0.6094,
+      "step": 1970
+    },
+    {
+      "epoch": 7.252747252747253,
+      "grad_norm": 0.0918455421924591,
+      "learning_rate": 0.00015144550427015588,
+      "loss": 0.6126,
+      "step": 1980
+    },
+    {
+      "epoch": 7.289377289377289,
+      "grad_norm": 0.12030185759067535,
+      "learning_rate": 0.00015092692094467668,
+      "loss": 0.5912,
+      "step": 1990
+    },
+    {
+      "epoch": 7.326007326007326,
+      "grad_norm": 0.10280390083789825,
+      "learning_rate": 0.0001504064878038735,
+      "loss": 0.6072,
+      "step": 2000
+    },
+    {
+      "epoch": 7.362637362637362,
+      "grad_norm": 0.10017659515142441,
+      "learning_rate": 0.00014988422393884254,
+      "loss": 0.6053,
+      "step": 2010
+    },
+    {
+      "epoch": 7.3992673992674,
+      "grad_norm": 0.08514354377985,
+      "learning_rate": 0.0001493601485078366,
+      "loss": 0.6218,
+      "step": 2020
+    },
+    {
+      "epoch": 7.435897435897436,
+      "grad_norm": 0.10472512245178223,
+      "learning_rate": 0.00014883428073556238,
+      "loss": 0.6064,
+      "step": 2030
+    },
+    {
+      "epoch": 7.472527472527473,
+      "grad_norm": 0.11152771860361099,
+      "learning_rate": 0.0001483066399124752,
+      "loss": 0.5951,
+      "step": 2040
+    },
+    {
+      "epoch": 7.509157509157509,
+      "grad_norm": 0.11035049706697464,
+      "learning_rate": 0.00014777724539407122,
+      "loss": 0.6017,
+      "step": 2050
+    },
+    {
+      "epoch": 7.545787545787546,
+      "grad_norm": 0.11814115196466446,
+      "learning_rate": 0.0001472461166001778,
+      "loss": 0.5947,
+      "step": 2060
+    },
+    {
+      "epoch": 7.582417582417582,
+      "grad_norm": 0.11158854514360428,
+      "learning_rate": 0.00014671327301424067,
+      "loss": 0.6004,
+      "step": 2070
+    },
+    {
+      "epoch": 7.619047619047619,
+      "grad_norm": 0.1224694699048996,
+      "learning_rate": 0.00014617873418260952,
+      "loss": 0.6072,
+      "step": 2080
+    },
+    {
+      "epoch": 7.655677655677656,
+      "grad_norm": 0.1077955961227417,
+      "learning_rate": 0.00014564251971382097,
+      "loss": 0.6007,
+      "step": 2090
+    },
+    {
+      "epoch": 7.6923076923076925,
+      "grad_norm": 0.10596826672554016,
+      "learning_rate": 0.00014510464927787898,
+      "loss": 0.6029,
+      "step": 2100
+    },
+    {
+      "epoch": 7.728937728937729,
+      "grad_norm": 0.13761618733406067,
+      "learning_rate": 0.00014456514260553375,
+      "loss": 0.6025,
+      "step": 2110
+    },
+    {
+      "epoch": 7.7655677655677655,
+      "grad_norm": 0.13236203789710999,
+      "learning_rate": 0.00014402401948755754,
+      "loss": 0.6041,
+      "step": 2120
+    },
+    {
+      "epoch": 7.802197802197802,
+      "grad_norm": 0.09635107964277267,
+      "learning_rate": 0.00014348129977401886,
+      "loss": 0.606,
+      "step": 2130
+    },
+    {
+      "epoch": 7.8388278388278385,
+      "grad_norm": 0.0904504582285881,
+      "learning_rate": 0.00014293700337355427,
+      "loss": 0.6049,
+      "step": 2140
+    },
+    {
+      "epoch": 7.875457875457876,
+      "grad_norm": 0.10969776660203934,
+      "learning_rate": 0.00014239115025263812,
+      "loss": 0.5982,
+      "step": 2150
+    },
+    {
+      "epoch": 7.912087912087912,
+      "grad_norm": 0.1088518351316452,
+      "learning_rate": 0.00014184376043485005,
+      "loss": 0.6021,
+      "step": 2160
+    },
+    {
+      "epoch": 7.948717948717949,
+      "grad_norm": 0.11425957083702087,
+      "learning_rate": 0.0001412948540001405,
+      "loss": 0.6022,
+      "step": 2170
+    },
+    {
+      "epoch": 7.985347985347985,
+      "grad_norm": 0.11807142943143845,
+      "learning_rate": 0.00014074445108409406,
+      "loss": 0.616,
+      "step": 2180
+    },
+    {
+      "epoch": 8.021978021978022,
+      "grad_norm": 0.13567444682121277,
+      "learning_rate": 0.00014019257187719098,
+      "loss": 0.6025,
+      "step": 2190
+    },
+    {
+      "epoch": 8.058608058608058,
+      "grad_norm": 0.12121989578008652,
+      "learning_rate": 0.00013963923662406634,
+      "loss": 0.6046,
+      "step": 2200
+    },
+    {
+      "epoch": 8.095238095238095,
+      "grad_norm": 0.11369021236896515,
+      "learning_rate": 0.0001390844656227675,
+      "loss": 0.6061,
+      "step": 2210
+    },
+    {
+      "epoch": 8.131868131868131,
+      "grad_norm": 0.10813486576080322,
+      "learning_rate": 0.00013852827922400954,
+      "loss": 0.6025,
+      "step": 2220
+    },
+    {
+      "epoch": 8.168498168498168,
+      "grad_norm": 0.11306055635213852,
+      "learning_rate": 0.00013797069783042865,
+      "loss": 0.6105,
+      "step": 2230
+    },
+    {
+      "epoch": 8.205128205128204,
+      "grad_norm": 0.171703040599823,
+      "learning_rate": 0.0001374117418958338,
+      "loss": 0.5866,
+      "step": 2240
+    },
+    {
+      "epoch": 8.241758241758241,
+      "grad_norm": 0.10016179084777832,
+      "learning_rate": 0.00013685143192445633,
+      "loss": 0.6078,
+      "step": 2250
+    },
+    {
+      "epoch": 8.27838827838828,
+      "grad_norm": 0.09870107471942902,
+      "learning_rate": 0.00013628978847019782,
+      "loss": 0.6074,
+      "step": 2260
+    },
+    {
+      "epoch": 8.315018315018316,
+      "grad_norm": 0.11638665944337845,
+      "learning_rate": 0.00013572683213587618,
+      "loss": 0.6018,
+      "step": 2270
+    },
+    {
+      "epoch": 8.351648351648352,
+      "grad_norm": 0.11457157135009766,
+      "learning_rate": 0.00013516258357246976,
+      "loss": 0.5986,
+      "step": 2280
+    },
+    {
+      "epoch": 8.388278388278389,
+      "grad_norm": 0.09893471002578735,
+      "learning_rate": 0.00013459706347835988,
+      "loss": 0.6042,
+      "step": 2290
+    },
+    {
+      "epoch": 8.424908424908425,
+      "grad_norm": 0.09442324936389923,
+      "learning_rate": 0.0001340302925985716,
+      "loss": 0.604,
+      "step": 2300
+    },
+    {
+      "epoch": 8.461538461538462,
+      "grad_norm": 0.11142978072166443,
+      "learning_rate": 0.00013346229172401256,
+      "loss": 0.5966,
+      "step": 2310
+    },
+    {
+      "epoch": 8.498168498168498,
+      "grad_norm": 0.13073523342609406,
+      "learning_rate": 0.00013289308169071051,
+      "loss": 0.5908,
+      "step": 2320
+    },
+    {
+      "epoch": 8.534798534798535,
+      "grad_norm": 0.11161026358604431,
+      "learning_rate": 0.0001323226833790488,
+      "loss": 0.6105,
+      "step": 2330
+    },
+    {
+      "epoch": 8.571428571428571,
+      "grad_norm": 0.09900886565446854,
+      "learning_rate": 0.00013175111771300053,
+      "loss": 0.6013,
+      "step": 2340
+    },
+    {
+      "epoch": 8.608058608058608,
+      "grad_norm": 0.130936399102211,
+      "learning_rate": 0.00013117840565936103,
+      "loss": 0.6022,
+      "step": 2350
+    },
+    {
+      "epoch": 8.644688644688644,
+      "grad_norm": 0.0852714255452156,
+      "learning_rate": 0.00013060456822697848,
+      "loss": 0.606,
+      "step": 2360
+    },
+    {
+      "epoch": 8.68131868131868,
+      "grad_norm": 0.09974522143602371,
+      "learning_rate": 0.00013002962646598373,
+      "loss": 0.5918,
+      "step": 2370
+    },
+    {
+      "epoch": 8.717948717948717,
+      "grad_norm": 0.08324407786130905,
+      "learning_rate": 0.0001294536014670176,
+      "loss": 0.6011,
+      "step": 2380
+    },
+    {
+      "epoch": 8.754578754578755,
+      "grad_norm": 0.09671464562416077,
+      "learning_rate": 0.00012887651436045737,
+      "loss": 0.606,
+      "step": 2390
+    },
+    {
+      "epoch": 8.791208791208792,
+      "grad_norm": 0.10795936733484268,
+      "learning_rate": 0.00012829838631564196,
+      "loss": 0.6008,
+      "step": 2400
+    },
+    {
+      "epoch": 8.827838827838828,
+      "grad_norm": 0.10716196894645691,
+      "learning_rate": 0.00012771923854009495,
+      "loss": 0.5983,
+      "step": 2410
+    },
+    {
+      "epoch": 8.864468864468865,
+      "grad_norm": 0.09245418757200241,
+      "learning_rate": 0.00012713909227874675,
+      "loss": 0.6045,
+      "step": 2420
+    },
+    {
+      "epoch": 8.901098901098901,
+      "grad_norm": 0.11781131476163864,
+      "learning_rate": 0.00012655796881315547,
+      "loss": 0.6028,
+      "step": 2430
+    },
+    {
+      "epoch": 8.937728937728938,
+      "grad_norm": 0.12876753509044647,
+      "learning_rate": 0.00012597588946072595,
+      "loss": 0.6002,
+      "step": 2440
+    },
+    {
+      "epoch": 8.974358974358974,
+      "grad_norm": 0.08043269068002701,
+      "learning_rate": 0.00012539287557392803,
+      "loss": 0.5973,
+      "step": 2450
+    },
+    {
+      "epoch": 9.010989010989011,
+      "grad_norm": 0.11363924294710159,
+      "learning_rate": 0.0001248089485395131,
+      "loss": 0.591,
+      "step": 2460
+    },
+    {
+      "epoch": 9.047619047619047,
+      "grad_norm": 0.09732921421527863,
+      "learning_rate": 0.00012422412977772956,
+      "loss": 0.6034,
+      "step": 2470
+    },
+    {
+      "epoch": 9.084249084249084,
+      "grad_norm": 0.1470343917608261,
+      "learning_rate": 0.0001236384407415373,
+      "loss": 0.5994,
+      "step": 2480
+    },
+    {
+      "epoch": 9.12087912087912,
+      "grad_norm": 0.11025113612413406,
+      "learning_rate": 0.0001230519029158204,
+      "loss": 0.5965,
+      "step": 2490
+    },
+    {
+      "epoch": 9.157509157509157,
+      "grad_norm": 0.1023898497223854,
+      "learning_rate": 0.00012246453781659928,
+      "loss": 0.5966,
+      "step": 2500
+    },
+    {
+      "epoch": 9.194139194139193,
+      "grad_norm": 0.09586931765079498,
+      "learning_rate": 0.00012187636699024118,
+      "loss": 0.6094,
+      "step": 2510
+    },
+    {
+      "epoch": 9.23076923076923,
+      "grad_norm": 0.1178598701953888,
+      "learning_rate": 0.00012128741201267007,
+      "loss": 0.5996,
+      "step": 2520
+    },
+    {
+      "epoch": 9.267399267399268,
+      "grad_norm": 0.10787366330623627,
+      "learning_rate": 0.00012069769448857484,
+      "loss": 0.5964,
+      "step": 2530
+    },
+    {
+      "epoch": 9.304029304029305,
+      "grad_norm": 0.10359363257884979,
+      "learning_rate": 0.00012010723605061705,
+      "loss": 0.5919,
+      "step": 2540
+    },
+    {
+      "epoch": 9.340659340659341,
+      "grad_norm": 0.1379256248474121,
+      "learning_rate": 0.00011951605835863724,
+      "loss": 0.6008,
+      "step": 2550
+    },
+    {
+      "epoch": 9.377289377289378,
+      "grad_norm": 0.12038258463144302,
+      "learning_rate": 0.00011892418309886044,
+      "loss": 0.6027,
+      "step": 2560
+    },
+    {
+      "epoch": 9.413919413919414,
+      "grad_norm": 0.07978381216526031,
+      "learning_rate": 0.00011833163198310055,
+      "loss": 0.5954,
+      "step": 2570
+    },
+    {
+      "epoch": 9.45054945054945,
+      "grad_norm": 0.11855828762054443,
+      "learning_rate": 0.00011773842674796405,
+      "loss": 0.5904,
+      "step": 2580
+    },
+    {
+      "epoch": 9.487179487179487,
+      "grad_norm": 0.10759320110082626,
+      "learning_rate": 0.00011714458915405246,
+      "loss": 0.6129,
+      "step": 2590
+    },
+    {
+      "epoch": 9.523809523809524,
+      "grad_norm": 0.1048365980386734,
+      "learning_rate": 0.00011655014098516422,
+      "loss": 0.5933,
+      "step": 2600
+    },
+    {
+      "epoch": 9.56043956043956,
+      "grad_norm": 0.08056604862213135,
+      "learning_rate": 0.0001159551040474955,
+      "loss": 0.6053,
+      "step": 2610
+    },
+    {
+      "epoch": 9.597069597069597,
+      "grad_norm": 0.16295000910758972,
+      "learning_rate": 0.00011535950016884043,
+      "loss": 0.5994,
+      "step": 2620
+    },
+    {
+      "epoch": 9.633699633699633,
+      "grad_norm": 0.09486577659845352,
+      "learning_rate": 0.00011476335119779015,
+      "loss": 0.6012,
+      "step": 2630
+    },
+    {
+      "epoch": 9.67032967032967,
+      "grad_norm": 0.11261876672506332,
+      "learning_rate": 0.00011416667900293163,
+      "loss": 0.6046,
+      "step": 2640
+    },
+    {
+      "epoch": 9.706959706959706,
+      "grad_norm": 0.14503304660320282,
+      "learning_rate": 0.00011356950547204514,
+      "loss": 0.5988,
+      "step": 2650
+    },
+    {
+      "epoch": 9.743589743589745,
+      "grad_norm": 0.11020506173372269,
+      "learning_rate": 0.00011297185251130165,
+      "loss": 0.6072,
+      "step": 2660
+    },
+    {
+      "epoch": 9.780219780219781,
+      "grad_norm": 0.11951778084039688,
+      "learning_rate": 0.00011237374204445894,
+      "loss": 0.6145,
+      "step": 2670
+    },
+    {
+      "epoch": 9.816849816849818,
+      "grad_norm": 0.13774767518043518,
+      "learning_rate": 0.00011177519601205772,
+      "loss": 0.5822,
+      "step": 2680
+    },
+    {
+      "epoch": 9.853479853479854,
+      "grad_norm": 0.09933178126811981,
+      "learning_rate": 0.00011117623637061649,
+      "loss": 0.6056,
+      "step": 2690
+    },
+    {
+      "epoch": 9.89010989010989,
+      "grad_norm": 0.09804636985063553,
+      "learning_rate": 0.00011057688509182616,
+      "loss": 0.6002,
+      "step": 2700
+    },
+    {
+      "epoch": 9.926739926739927,
+      "grad_norm": 0.14475852251052856,
+      "learning_rate": 0.0001099771641617442,
+      "loss": 0.6005,
+      "step": 2710
+    },
+    {
+      "epoch": 9.963369963369964,
+      "grad_norm": 0.12410984933376312,
+      "learning_rate": 0.00010937709557998797,
+      "loss": 0.597,
+      "step": 2720
+    },
+    {
+      "epoch": 10.0,
+      "grad_norm": 0.12814611196517944,
+      "learning_rate": 0.00010877670135892781,
+      "loss": 0.5909,
+      "step": 2730
+    },
+    {
+      "epoch": 10.036630036630036,
+      "grad_norm": 0.1395956575870514,
+      "learning_rate": 0.00010817600352287946,
+      "loss": 0.6083,
+      "step": 2740
+    },
+    {
+      "epoch": 10.073260073260073,
+      "grad_norm": 0.1210312694311142,
+      "learning_rate": 0.00010757502410729625,
+      "loss": 0.5917,
+      "step": 2750
+    },
+    {
+      "epoch": 10.10989010989011,
+      "grad_norm": 0.10577362775802612,
+      "learning_rate": 0.00010697378515796078,
+      "loss": 0.6027,
+      "step": 2760
+    },
+    {
+      "epoch": 10.146520146520146,
+      "grad_norm": 0.10341820120811462,
+      "learning_rate": 0.00010637230873017592,
+      "loss": 0.5922,
+      "step": 2770
+    },
+    {
+      "epoch": 10.183150183150182,
+      "grad_norm": 0.09854844957590103,
+      "learning_rate": 0.00010577061688795627,
+      "loss": 0.6048,
+      "step": 2780
+    },
+    {
+      "epoch": 10.219780219780219,
+      "grad_norm": 0.13221371173858643,
+      "learning_rate": 0.00010516873170321829,
+      "loss": 0.594,
+      "step": 2790
+    },
+    {
+      "epoch": 10.256410256410255,
+      "grad_norm": 0.11323889344930649,
+      "learning_rate": 0.0001045666752549709,
+      "loss": 0.5854,
+      "step": 2800
+    },
+    {
+      "epoch": 10.293040293040294,
+      "grad_norm": 0.10400225222110748,
+      "learning_rate": 0.00010396446962850557,
+      "loss": 0.5999,
+      "step": 2810
+    },
+    {
+      "epoch": 10.32967032967033,
+      "grad_norm": 0.11270953714847565,
+      "learning_rate": 0.00010336213691458601,
+      "loss": 0.5936,
+      "step": 2820
+    },
+    {
+      "epoch": 10.366300366300367,
+      "grad_norm": 0.11540117859840393,
+      "learning_rate": 0.00010275969920863786,
+      "loss": 0.598,
+      "step": 2830
+    },
+    {
+      "epoch": 10.402930402930403,
+      "grad_norm": 0.09625425189733505,
+      "learning_rate": 0.00010215717860993831,
+      "loss": 0.593,
+      "step": 2840
+    },
+    {
+      "epoch": 10.43956043956044,
+      "grad_norm": 0.10338188707828522,
+      "learning_rate": 0.00010155459722080518,
+      "loss": 0.5903,
+      "step": 2850
+    },
+    {
+      "epoch": 10.476190476190476,
+      "grad_norm": 0.1506458967924118,
+      "learning_rate": 0.00010095197714578635,
+      "loss": 0.5821,
+      "step": 2860
+    },
+    {
+      "epoch": 10.512820512820513,
+      "grad_norm": 0.12155159562826157,
+      "learning_rate": 0.00010034934049084881,
+      "loss": 0.6108,
+      "step": 2870
+    },
+    {
+      "epoch": 10.54945054945055,
+      "grad_norm": 0.10831907391548157,
+      "learning_rate": 9.974670936256771e-05,
+      "loss": 0.5872,
+      "step": 2880
+    },
+    {
+      "epoch": 10.586080586080586,
+      "grad_norm": 0.1336948573589325,
+      "learning_rate": 9.914410586731555e-05,
+      "loss": 0.5929,
+      "step": 2890
+    },
+    {
+      "epoch": 10.622710622710622,
+      "grad_norm": 0.10216037929058075,
+      "learning_rate": 9.854155211045111e-05,
+      "loss": 0.6144,
+      "step": 2900
+    },
+    {
+      "epoch": 10.659340659340659,
+      "grad_norm": 0.11720311641693115,
+      "learning_rate": 9.793907019550857e-05,
+      "loss": 0.598,
+      "step": 2910
+    },
+    {
+      "epoch": 10.695970695970695,
+      "grad_norm": 0.11751140654087067,
+      "learning_rate": 9.733668222338677e-05,
+      "loss": 0.6086,
+      "step": 2920
+    },
+    {
+      "epoch": 10.732600732600732,
+      "grad_norm": 0.09198504686355591,
+      "learning_rate": 9.67344102915385e-05,
+      "loss": 0.604,
+      "step": 2930
+    },
+    {
+      "epoch": 10.76923076923077,
+      "grad_norm": 0.1153954491019249,
+      "learning_rate": 9.613227649315973e-05,
+      "loss": 0.6086,
+      "step": 2940
+    },
+    {
+      "epoch": 10.805860805860807,
+      "grad_norm": 0.09167054295539856,
+      "learning_rate": 9.553030291637929e-05,
+      "loss": 0.5991,
+      "step": 2950
+    },
+    {
+      "epoch": 10.842490842490843,
+      "grad_norm": 0.11931636184453964,
+      "learning_rate": 9.492851164344865e-05,
+      "loss": 0.6154,
+      "step": 2960
+    },
+    {
+      "epoch": 10.87912087912088,
+      "grad_norm": 0.07273609191179276,
+      "learning_rate": 9.432692474993173e-05,
+      "loss": 0.5948,
+      "step": 2970
+    },
+    {
+      "epoch": 10.915750915750916,
+      "grad_norm": 0.12160748243331909,
+      "learning_rate": 9.372556430389523e-05,
+      "loss": 0.602,
+      "step": 2980
+    },
+    {
+      "epoch": 10.952380952380953,
+      "grad_norm": 0.14035232365131378,
+      "learning_rate": 9.312445236509911e-05,
+      "loss": 0.5985,
+      "step": 2990
+    },
+    {
+      "epoch": 10.989010989010989,
+      "grad_norm": 0.12618041038513184,
+      "learning_rate": 9.252361098418715e-05,
+      "loss": 0.6169,
+      "step": 3000
+    },
+    {
+      "epoch": 11.025641025641026,
+      "grad_norm": 0.1227792352437973,
+      "learning_rate": 9.192306220187838e-05,
+      "loss": 0.5856,
+      "step": 3010
+    },
+    {
+      "epoch": 11.062271062271062,
+      "grad_norm": 0.12947812676429749,
+      "learning_rate": 9.132282804815835e-05,
+      "loss": 0.5869,
+      "step": 3020
+    },
+    {
+      "epoch": 11.098901098901099,
+      "grad_norm": 0.14743156731128693,
+      "learning_rate": 9.072293054147112e-05,
+      "loss": 0.6026,
+      "step": 3030
+    },
+    {
+      "epoch": 11.135531135531135,
+      "grad_norm": 0.10158231854438782,
+      "learning_rate": 9.012339168791138e-05,
+      "loss": 0.604,
+      "step": 3040
+    },
+    {
+      "epoch": 11.172161172161172,
+      "grad_norm": 0.10932044684886932,
+      "learning_rate": 8.95242334804174e-05,
+      "loss": 0.6072,
+      "step": 3050
+    },
+    {
+      "epoch": 11.208791208791208,
+      "grad_norm": 0.10173647105693817,
+      "learning_rate": 8.892547789796422e-05,
+      "loss": 0.6042,
+      "step": 3060
+    },
+    {
+      "epoch": 11.245421245421245,
+      "grad_norm": 0.11170593649148941,
+      "learning_rate": 8.832714690475722e-05,
+      "loss": 0.5958,
+      "step": 3070
+    },
+    {
+      "epoch": 11.282051282051283,
+      "grad_norm": 0.1408560425043106,
+      "learning_rate": 8.77292624494266e-05,
+      "loss": 0.5912,
+      "step": 3080
+    },
+    {
+      "epoch": 11.31868131868132,
+      "grad_norm": 0.08870992064476013,
+      "learning_rate": 8.71318464642222e-05,
+      "loss": 0.593,
+      "step": 3090
+    },
+    {
+      "epoch": 11.355311355311356,
+      "grad_norm": 0.11843115091323853,
+      "learning_rate": 8.653492086420894e-05,
+      "loss": 0.6056,
+      "step": 3100
+    },
+    {
+      "epoch": 11.391941391941392,
+      "grad_norm": 0.09680238366127014,
+      "learning_rate": 8.593850754646276e-05,
+      "loss": 0.6032,
+      "step": 3110
+    },
+    {
+      "epoch": 11.428571428571429,
+      "grad_norm": 0.08638516068458557,
+      "learning_rate": 8.534262838926766e-05,
+      "loss": 0.5969,
+      "step": 3120
+    },
+    {
+      "epoch": 11.465201465201465,
+      "grad_norm": 0.10992743074893951,
+      "learning_rate": 8.474730525131291e-05,
+      "loss": 0.6078,
+      "step": 3130
+    },
+    {
+      "epoch": 11.501831501831502,
+      "grad_norm": 0.12074992805719376,
+      "learning_rate": 8.415255997089129e-05,
+      "loss": 0.6033,
+      "step": 3140
+    },
+    {
+      "epoch": 11.538461538461538,
+      "grad_norm": 0.14779578149318695,
+      "learning_rate": 8.355841436509795e-05,
+      "loss": 0.5936,
+      "step": 3150
+    },
+    {
+      "epoch": 11.575091575091575,
+      "grad_norm": 0.1313851773738861,
+      "learning_rate": 8.29648902290302e-05,
+      "loss": 0.5947,
+      "step": 3160
+    },
+    {
+      "epoch": 11.611721611721611,
+      "grad_norm": 0.12997916340827942,
+      "learning_rate": 8.23720093349878e-05,
+      "loss": 0.5981,
+      "step": 3170
+    },
+    {
+      "epoch": 11.648351648351648,
+      "grad_norm": 0.1216355711221695,
+      "learning_rate": 8.177979343167449e-05,
+      "loss": 0.6024,
+      "step": 3180
+    },
+    {
+      "epoch": 11.684981684981684,
+      "grad_norm": 0.12574613094329834,
+      "learning_rate": 8.118826424340013e-05,
+      "loss": 0.5986,
+      "step": 3190
+    },
+    {
+      "epoch": 11.72161172161172,
+      "grad_norm": 0.13364483416080475,
+      "learning_rate": 8.05974434692836e-05,
+      "loss": 0.6075,
+      "step": 3200
+    },
+    {
+      "epoch": 11.758241758241759,
+      "grad_norm": 0.10564299672842026,
+      "learning_rate": 8.000735278245708e-05,
+      "loss": 0.5886,
+      "step": 3210
+    },
+    {
+      "epoch": 11.794871794871796,
+      "grad_norm": 0.12722384929656982,
+      "learning_rate": 7.941801382927094e-05,
+      "loss": 0.5962,
+      "step": 3220
+    },
+    {
+      "epoch": 11.831501831501832,
+      "grad_norm": 0.09655703604221344,
+      "learning_rate": 7.882944822849946e-05,
+      "loss": 0.5918,
+      "step": 3230
+    },
+    {
+      "epoch": 11.868131868131869,
+      "grad_norm": 0.10955348610877991,
+      "learning_rate": 7.82416775705481e-05,
+      "loss": 0.5958,
+      "step": 3240
+    },
+    {
+      "epoch": 11.904761904761905,
+      "grad_norm": 0.08911681175231934,
+      "learning_rate": 7.765472341666133e-05,
+      "loss": 0.5923,
+      "step": 3250
+    },
+    {
+      "epoch": 11.941391941391942,
+      "grad_norm": 0.11370093375444412,
+      "learning_rate": 7.706860729813177e-05,
+      "loss": 0.6032,
+      "step": 3260
+    },
+    {
+      "epoch": 11.978021978021978,
+      "grad_norm": 0.1376587450504303,
+      "learning_rate": 7.648335071551016e-05,
+      "loss": 0.582,
+      "step": 3270
+    },
+    {
+      "epoch": 12.014652014652015,
+      "grad_norm": 0.09854473918676376,
+      "learning_rate": 7.589897513781698e-05,
+      "loss": 0.5944,
+      "step": 3280
+    },
+    {
+      "epoch": 12.051282051282051,
+      "grad_norm": 0.10749837011098862,
+      "learning_rate": 7.531550200175465e-05,
+      "loss": 0.5958,
+      "step": 3290
+    },
+    {
+      "epoch": 12.087912087912088,
+      "grad_norm": 0.11621621996164322,
+      "learning_rate": 7.473295271092125e-05,
+      "loss": 0.601,
+      "step": 3300
+    },
+    {
+      "epoch": 12.124542124542124,
+      "grad_norm": 0.15026544034481049,
+      "learning_rate": 7.415134863502532e-05,
+      "loss": 0.5934,
+      "step": 3310
+    },
+    {
+      "epoch": 12.16117216117216,
+      "grad_norm": 0.09685179591178894,
+      "learning_rate": 7.35707111091021e-05,
+      "loss": 0.6051,
+      "step": 3320
+    },
+    {
+      "epoch": 12.197802197802197,
+      "grad_norm": 0.09188985824584961,
+      "learning_rate": 7.299106143273067e-05,
+      "loss": 0.5978,
+      "step": 3330
+    },
+    {
+      "epoch": 12.234432234432234,
+      "grad_norm": 0.10112758725881577,
+      "learning_rate": 7.24124208692528e-05,
+      "loss": 0.5981,
+      "step": 3340
+    },
+    {
+      "epoch": 12.271062271062272,
+      "grad_norm": 0.10160616040229797,
+      "learning_rate": 7.18348106449929e-05,
+      "loss": 0.5923,
+      "step": 3350
+    },
+    {
+      "epoch": 12.307692307692308,
+      "grad_norm": 0.12250513583421707,
+      "learning_rate": 7.125825194847929e-05,
+      "loss": 0.5819,
+      "step": 3360
+    },
+    {
+      "epoch": 12.344322344322345,
+      "grad_norm": 0.12580272555351257,
+      "learning_rate": 7.068276592966702e-05,
+      "loss": 0.5921,
+      "step": 3370
+    },
+    {
+      "epoch": 12.380952380952381,
+      "grad_norm": 0.09846355020999908,
+      "learning_rate": 7.010837369916204e-05,
+      "loss": 0.6016,
+      "step": 3380
+    },
+    {
+      "epoch": 12.417582417582418,
+      "grad_norm": 0.1316191852092743,
+      "learning_rate": 6.953509632744678e-05,
+      "loss": 0.5908,
+      "step": 3390
+    },
+    {
+      "epoch": 12.454212454212454,
+      "grad_norm": 0.10895272344350815,
+      "learning_rate": 6.896295484410712e-05,
+      "loss": 0.5953,
+      "step": 3400
+    },
+    {
+      "epoch": 12.49084249084249,
+      "grad_norm": 0.1075824722647667,
+      "learning_rate": 6.839197023706109e-05,
+      "loss": 0.5882,
+      "step": 3410
+    },
+    {
+      "epoch": 12.527472527472527,
+      "grad_norm": 0.09522432088851929,
+      "learning_rate": 6.782216345178898e-05,
+      "loss": 0.5873,
+      "step": 3420
+    },
+    {
+      "epoch": 12.564102564102564,
+      "grad_norm": 0.11377083510160446,
+      "learning_rate": 6.725355539056478e-05,
+      "loss": 0.6092,
+      "step": 3430
+    },
+    {
+      "epoch": 12.6007326007326,
+      "grad_norm": 0.10006774961948395,
+      "learning_rate": 6.668616691168976e-05,
+      "loss": 0.5863,
+      "step": 3440
+    },
+    {
+      "epoch": 12.637362637362637,
+      "grad_norm": 0.10102878510951996,
+      "learning_rate": 6.612001882872705e-05,
+      "loss": 0.5885,
+      "step": 3450
+    },
+    {
+      "epoch": 12.673992673992673,
+      "grad_norm": 0.09821674972772598,
+      "learning_rate": 6.555513190973818e-05,
+      "loss": 0.5988,
+      "step": 3460
+    },
+    {
+      "epoch": 12.71062271062271,
+      "grad_norm": 0.09962121397256851,
+      "learning_rate": 6.499152687652136e-05,
+      "loss": 0.5859,
+      "step": 3470
+    },
+    {
+      "epoch": 12.747252747252748,
+      "grad_norm": 0.1359899640083313,
+      "learning_rate": 6.442922440385134e-05,
+      "loss": 0.5915,
+      "step": 3480
+    },
+    {
+      "epoch": 12.783882783882785,
+      "grad_norm": 0.10209459066390991,
+      "learning_rate": 6.386824511872068e-05,
+      "loss": 0.6009,
+      "step": 3490
+    },
+    {
+      "epoch": 12.820512820512821,
+      "grad_norm": 0.10285753011703491,
+      "learning_rate": 6.330860959958354e-05,
+      "loss": 0.5894,
+      "step": 3500
+    },
+    {
+      "epoch": 12.857142857142858,
+      "grad_norm": 0.10430514812469482,
+      "learning_rate": 6.275033837560054e-05,
+      "loss": 0.5952,
+      "step": 3510
+    },
+    {
+      "epoch": 12.893772893772894,
+      "grad_norm": 0.09391837567090988,
+      "learning_rate": 6.219345192588576e-05,
+      "loss": 0.6002,
+      "step": 3520
+    },
+    {
+      "epoch": 12.93040293040293,
+      "grad_norm": 0.11428876221179962,
+      "learning_rate": 6.163797067875537e-05,
+      "loss": 0.6024,
+      "step": 3530
+    },
+    {
+      "epoch": 12.967032967032967,
+      "grad_norm": 0.09616833925247192,
+      "learning_rate": 6.108391501097848e-05,
+      "loss": 0.5969,
+      "step": 3540
+    },
+    {
+      "epoch": 13.003663003663004,
+      "grad_norm": 0.111970454454422,
+      "learning_rate": 6.053130524702956e-05,
+      "loss": 0.592,
+      "step": 3550
+    },
+    {
+      "epoch": 13.04029304029304,
+      "grad_norm": 0.14766483008861542,
+      "learning_rate": 5.99801616583427e-05,
+      "loss": 0.6047,
+      "step": 3560
+    },
+    {
+      "epoch": 13.076923076923077,
+      "grad_norm": 0.09645646065473557,
+      "learning_rate": 5.9430504462568315e-05,
+      "loss": 0.5963,
+      "step": 3570
+    },
+    {
+      "epoch": 13.113553113553113,
+      "grad_norm": 0.10448257625102997,
+      "learning_rate": 5.8882353822831294e-05,
+      "loss": 0.6018,
+      "step": 3580
+    },
+    {
+      "epoch": 13.15018315018315,
+      "grad_norm": 0.09417391568422318,
+      "learning_rate": 5.8335729846991304e-05,
+      "loss": 0.5873,
+      "step": 3590
+    },
+    {
+      "epoch": 13.186813186813186,
+      "grad_norm": 0.12039947509765625,
+      "learning_rate": 5.779065258690537e-05,
+      "loss": 0.5941,
+      "step": 3600
+    },
+    {
+      "epoch": 13.223443223443223,
+      "grad_norm": 0.17030514776706696,
+      "learning_rate": 5.724714203769218e-05,
+      "loss": 0.5967,
+      "step": 3610
+    },
+    {
+      "epoch": 13.260073260073261,
+      "grad_norm": 0.10252691805362701,
+      "learning_rate": 5.670521813699856e-05,
+      "loss": 0.5962,
+      "step": 3620
+    },
+    {
+      "epoch": 13.296703296703297,
+      "grad_norm": 0.10711918771266937,
+      "learning_rate": 5.6164900764268235e-05,
+      "loss": 0.6012,
+      "step": 3630
+    },
+    {
+      "epoch": 13.333333333333334,
+      "grad_norm": 0.11735249310731888,
+      "learning_rate": 5.56262097400125e-05,
+      "loss": 0.5935,
+      "step": 3640
+    },
+    {
+      "epoch": 13.36996336996337,
+      "grad_norm": 0.09693988412618637,
+      "learning_rate": 5.508916482508318e-05,
+      "loss": 0.6063,
+      "step": 3650
+    },
+    {
+      "epoch": 13.406593406593407,
+      "grad_norm": 0.1074470654129982,
+      "learning_rate": 5.455378571994764e-05,
+      "loss": 0.5857,
+      "step": 3660
+    },
+    {
+      "epoch": 13.443223443223443,
+      "grad_norm": 0.15257665514945984,
+      "learning_rate": 5.402009206396627e-05,
+      "loss": 0.5926,
+      "step": 3670
+    },
+    {
+      "epoch": 13.47985347985348,
+      "grad_norm": 0.12574537098407745,
+      "learning_rate": 5.348810343467197e-05,
+      "loss": 0.5909,
+      "step": 3680
+    },
+    {
+      "epoch": 13.516483516483516,
+      "grad_norm": 0.1099851056933403,
+      "learning_rate": 5.295783934705201e-05,
+      "loss": 0.5953,
+      "step": 3690
+    },
+    {
+      "epoch": 13.553113553113553,
+      "grad_norm": 0.09711455553770065,
+      "learning_rate": 5.2429319252832045e-05,
+      "loss": 0.6059,
+      "step": 3700
+    },
+    {
+      "epoch": 13.58974358974359,
+      "grad_norm": 0.10976620763540268,
+      "learning_rate": 5.1902562539762736e-05,
+      "loss": 0.5806,
+      "step": 3710
+    },
+    {
+      "epoch": 13.626373626373626,
+      "grad_norm": 0.1407400518655777,
+      "learning_rate": 5.137758853090848e-05,
+      "loss": 0.589,
+      "step": 3720
+    },
+    {
+      "epoch": 13.663003663003662,
+      "grad_norm": 0.09327317029237747,
+      "learning_rate": 5.085441648393856e-05,
+      "loss": 0.5897,
+      "step": 3730
+    },
+    {
+      "epoch": 13.699633699633699,
+      "grad_norm": 0.1264810413122177,
+      "learning_rate": 5.033306559042075e-05,
+      "loss": 0.5972,
+      "step": 3740
+    },
+    {
+      "epoch": 13.736263736263737,
+      "grad_norm": 0.12380846589803696,
+      "learning_rate": 4.98135549751172e-05,
+      "loss": 0.6,
+      "step": 3750
+    },
+    {
+      "epoch": 13.772893772893774,
+      "grad_norm": 0.12347488850355148,
+      "learning_rate": 4.929590369528302e-05,
+      "loss": 0.5797,
+      "step": 3760
+    },
+    {
+      "epoch": 13.80952380952381,
+      "grad_norm": 0.11884108930826187,
+      "learning_rate": 4.8780130739967216e-05,
+      "loss": 0.5931,
+      "step": 3770
+    },
+    {
+      "epoch": 13.846153846153847,
+      "grad_norm": 0.11143997311592102,
+      "learning_rate": 4.826625502931597e-05,
+      "loss": 0.5803,
+      "step": 3780
+    },
+    {
+      "epoch": 13.882783882783883,
+      "grad_norm": 0.10896267741918564,
+      "learning_rate": 4.775429541387864e-05,
+      "loss": 0.602,
+      "step": 3790
+    },
+    {
+      "epoch": 13.91941391941392,
+      "grad_norm": 0.10565280169248581,
+      "learning_rate": 4.724427067391632e-05,
+      "loss": 0.5949,
+      "step": 3800
+    },
+    {
+      "epoch": 13.956043956043956,
+      "grad_norm": 0.12976156175136566,
+      "learning_rate": 4.673619951871299e-05,
+      "loss": 0.5924,
+      "step": 3810
+    },
+    {
+      "epoch": 13.992673992673993,
+      "grad_norm": 0.1173262670636177,
+      "learning_rate": 4.6230100585888905e-05,
+      "loss": 0.594,
+      "step": 3820
+    },
+    {
+      "epoch": 14.02930402930403,
+      "grad_norm": 0.12908582389354706,
+      "learning_rate": 4.572599244071726e-05,
+      "loss": 0.5992,
+      "step": 3830
+    },
+    {
+      "epoch": 14.065934065934066,
+      "grad_norm": 0.11487755179405212,
+      "learning_rate": 4.522389357544297e-05,
+      "loss": 0.5932,
+      "step": 3840
+    },
+    {
+      "epoch": 14.102564102564102,
+      "grad_norm": 0.1192779615521431,
+      "learning_rate": 4.472382240860431e-05,
+      "loss": 0.5866,
+      "step": 3850
+    },
+    {
+      "epoch": 14.139194139194139,
+      "grad_norm": 0.13515004515647888,
+      "learning_rate": 4.4225797284357354e-05,
+      "loss": 0.5834,
+      "step": 3860
+    },
+    {
+      "epoch": 14.175824175824175,
+      "grad_norm": 0.12698876857757568,
+      "learning_rate": 4.372983647180302e-05,
+      "loss": 0.5984,
+      "step": 3870
+    },
+    {
+      "epoch": 14.212454212454212,
+      "grad_norm": 0.1023423969745636,
+      "learning_rate": 4.323595816431687e-05,
+      "loss": 0.5906,
+      "step": 3880
+    },
+    {
+      "epoch": 14.249084249084248,
+      "grad_norm": 0.1128733679652214,
+      "learning_rate": 4.27441804788817e-05,
+      "loss": 0.5964,
+      "step": 3890
+    },
+    {
+      "epoch": 14.285714285714286,
+      "grad_norm": 0.15929995477199554,
+      "learning_rate": 4.225452145542305e-05,
+      "loss": 0.5722,
+      "step": 3900
+    },
+    {
+      "epoch": 14.322344322344323,
+      "grad_norm": 0.12099599838256836,
+      "learning_rate": 4.1766999056147435e-05,
+      "loss": 0.5987,
+      "step": 3910
+    },
+    {
+      "epoch": 14.35897435897436,
+      "grad_norm": 0.12424413114786148,
+      "learning_rate": 4.1281631164883264e-05,
+      "loss": 0.5856,
+      "step": 3920
+    },
+    {
+      "epoch": 14.395604395604396,
+      "grad_norm": 0.1399509310722351,
+      "learning_rate": 4.079843558642504e-05,
+      "loss": 0.5883,
+      "step": 3930
+    },
+    {
+      "epoch": 14.432234432234432,
+      "grad_norm": 0.11879005283117294,
+      "learning_rate": 4.031743004588009e-05,
+      "loss": 0.595,
+      "step": 3940
+    },
+    {
+      "epoch": 14.468864468864469,
+      "grad_norm": 0.11288290470838547,
+      "learning_rate": 3.9838632188018345e-05,
+      "loss": 0.5859,
+      "step": 3950
+    },
+    {
+      "epoch": 14.505494505494505,
+      "grad_norm": 0.08327735960483551,
+      "learning_rate": 3.936205957662514e-05,
+      "loss": 0.6001,
+      "step": 3960
+    },
+    {
+      "epoch": 14.542124542124542,
+      "grad_norm": 0.11755986511707306,
+      "learning_rate": 3.888772969385695e-05,
+      "loss": 0.5834,
+      "step": 3970
+    },
+    {
+      "epoch": 14.578754578754578,
+      "grad_norm": 0.13514429330825806,
+      "learning_rate": 3.8415659939599886e-05,
+      "loss": 0.6095,
+      "step": 3980
+    },
+    {
+      "epoch": 14.615384615384615,
+      "grad_norm": 0.12506259977817535,
+      "learning_rate": 3.79458676308317e-05,
+      "loss": 0.5889,
+      "step": 3990
+    },
+    {
+      "epoch": 14.652014652014651,
+      "grad_norm": 0.10506050288677216,
+      "learning_rate": 3.747837000098633e-05,
+      "loss": 0.5969,
+      "step": 4000
+    },
+    {
+      "epoch": 14.688644688644688,
+      "grad_norm": 0.12048743665218353,
+      "learning_rate": 3.7013184199321865e-05,
+      "loss": 0.5928,
+      "step": 4010
+    },
+    {
+      "epoch": 14.725274725274724,
+      "grad_norm": 0.10874942690134048,
+      "learning_rate": 3.655032729029127e-05,
+      "loss": 0.5942,
+      "step": 4020
+    },
+    {
+      "epoch": 14.761904761904763,
+      "grad_norm": 0.12899090349674225,
+      "learning_rate": 3.608981625291665e-05,
+      "loss": 0.5906,
+      "step": 4030
+    },
+    {
+      "epoch": 14.7985347985348,
+      "grad_norm": 0.10427646338939667,
+      "learning_rate": 3.563166798016624e-05,
+      "loss": 0.5942,
+      "step": 4040
+    },
+    {
+      "epoch": 14.835164835164836,
+      "grad_norm": 0.08348722010850906,
+      "learning_rate": 3.517589927833476e-05,
+      "loss": 0.6057,
+      "step": 4050
+    },
+    {
+      "epoch": 14.871794871794872,
+      "grad_norm": 0.1019279882311821,
+      "learning_rate": 3.472252686642692e-05,
+      "loss": 0.5883,
+      "step": 4060
+    },
+    {
+      "epoch": 14.908424908424909,
+      "grad_norm": 0.13301706314086914,
+      "learning_rate": 3.427156737554416e-05,
+      "loss": 0.5862,
+      "step": 4070
+    },
+    {
+      "epoch": 14.945054945054945,
+      "grad_norm": 0.07168400287628174,
+      "learning_rate": 3.382303734827438e-05,
+      "loss": 0.5917,
+      "step": 4080
+    },
+    {
+      "epoch": 14.981684981684982,
+      "grad_norm": 0.13180795311927795,
+      "learning_rate": 3.337695323808538e-05,
+      "loss": 0.5866,
+      "step": 4090
+    },
+    {
+      "epoch": 15.018315018315018,
+      "grad_norm": 0.11809457838535309,
+      "learning_rate": 3.293333140872115e-05,
+      "loss": 0.5933,
+      "step": 4100
+    },
+    {
+      "epoch": 15.054945054945055,
+      "grad_norm": 0.15424320101737976,
+      "learning_rate": 3.24921881336015e-05,
+      "loss": 0.5883,
+      "step": 4110
+    },
+    {
+      "epoch": 15.091575091575091,
+      "grad_norm": 0.10090993344783783,
+      "learning_rate": 3.205353959522531e-05,
+      "loss": 0.5962,
+      "step": 4120
+    },
+    {
+      "epoch": 15.128205128205128,
+      "grad_norm": 0.10715416073799133,
+      "learning_rate": 3.161740188457681e-05,
+      "loss": 0.5892,
+      "step": 4130
+    },
+    {
+      "epoch": 15.164835164835164,
+      "grad_norm": 0.11065078526735306,
+      "learning_rate": 3.118379100053528e-05,
+      "loss": 0.5763,
+      "step": 4140
+    },
+    {
+      "epoch": 15.2014652014652,
+      "grad_norm": 0.0856284573674202,
+      "learning_rate": 3.075272284928813e-05,
+      "loss": 0.5919,
+      "step": 4150
+    },
+    {
+      "epoch": 15.238095238095237,
+      "grad_norm": 0.10988280922174454,
+      "learning_rate": 3.0324213243747567e-05,
+      "loss": 0.6042,
+      "step": 4160
+    },
+    {
+      "epoch": 15.274725274725276,
+      "grad_norm": 0.12171415239572525,
+      "learning_rate": 2.989827790297043e-05,
+      "loss": 0.5836,
+      "step": 4170
+    },
+    {
+      "epoch": 15.311355311355312,
+      "grad_norm": 0.16990408301353455,
+      "learning_rate": 2.947493245158146e-05,
+      "loss": 0.6058,
+      "step": 4180
+    },
+    {
+      "epoch": 15.347985347985349,
+      "grad_norm": 0.12295118719339371,
+      "learning_rate": 2.9054192419200365e-05,
+      "loss": 0.6056,
+      "step": 4190
+    },
+    {
+      "epoch": 15.384615384615385,
+      "grad_norm": 0.10608595609664917,
+      "learning_rate": 2.863607323987203e-05,
+      "loss": 0.5909,
+      "step": 4200
+    },
+    {
+      "epoch": 15.421245421245422,
+      "grad_norm": 0.12604480981826782,
+      "learning_rate": 2.8220590251500228e-05,
+      "loss": 0.5878,
+      "step": 4210
+    },
+    {
+      "epoch": 15.457875457875458,
+      "grad_norm": 0.11635804176330566,
+      "learning_rate": 2.780775869528522e-05,
+      "loss": 0.5818,
+      "step": 4220
+    },
+    {
+      "epoch": 15.494505494505495,
+      "grad_norm": 0.11485312134027481,
+      "learning_rate": 2.7397593715164548e-05,
+      "loss": 0.5871,
+      "step": 4230
+    },
+    {
+      "epoch": 15.531135531135531,
+      "grad_norm": 0.13255690038204193,
+      "learning_rate": 2.699011035725738e-05,
+      "loss": 0.5851,
+      "step": 4240
+    },
+    {
+      "epoch": 15.567765567765568,
+      "grad_norm": 0.06791131943464279,
+      "learning_rate": 2.6585323569312786e-05,
+      "loss": 0.5916,
+      "step": 4250
+    },
+    {
+      "epoch": 15.604395604395604,
+      "grad_norm": 0.12282761931419373,
+      "learning_rate": 2.618324820016133e-05,
+      "loss": 0.5909,
+      "step": 4260
+    },
+    {
+      "epoch": 15.64102564102564,
+      "grad_norm": 0.08577387779951096,
+      "learning_rate": 2.5783898999170314e-05,
+      "loss": 0.5876,
+      "step": 4270
+    },
+    {
+      "epoch": 15.677655677655677,
+      "grad_norm": 0.12403818964958191,
+      "learning_rate": 2.538729061570274e-05,
+      "loss": 0.5888,
+      "step": 4280
+    },
+    {
+      "epoch": 15.714285714285714,
+      "grad_norm": 0.11415976285934448,
+      "learning_rate": 2.4993437598579994e-05,
+      "loss": 0.5947,
+      "step": 4290
+    },
+    {
+      "epoch": 15.750915750915752,
+      "grad_norm": 0.12366656213998795,
+      "learning_rate": 2.4602354395548148e-05,
+      "loss": 0.592,
+      "step": 4300
+    },
+    {
+      "epoch": 15.787545787545788,
+      "grad_norm": 0.09358074516057968,
+      "learning_rate": 2.421405535274781e-05,
+      "loss": 0.609,
+      "step": 4310
+    },
+    {
+      "epoch": 15.824175824175825,
+      "grad_norm": 0.08951131254434586,
+      "learning_rate": 2.3828554714188076e-05,
+      "loss": 0.5835,
+      "step": 4320
+    },
+    {
+      "epoch": 15.860805860805861,
+      "grad_norm": 0.1570052057504654,
+      "learning_rate": 2.3445866621223954e-05,
+      "loss": 0.5947,
+      "step": 4330
+    },
+    {
+      "epoch": 15.897435897435898,
+      "grad_norm": 0.09661392867565155,
+      "learning_rate": 2.3066005112037474e-05,
+      "loss": 0.5992,
+      "step": 4340
+    },
+    {
+      "epoch": 15.934065934065934,
+      "grad_norm": 0.11398851126432419,
+      "learning_rate": 2.2688984121122932e-05,
+      "loss": 0.5893,
+      "step": 4350
+    },
+    {
+      "epoch": 15.97069597069597,
+      "grad_norm": 0.15302923321723938,
+      "learning_rate": 2.2314817478775594e-05,
+      "loss": 0.6001,
+      "step": 4360
+    },
+    {
+      "epoch": 16.007326007326007,
+      "grad_norm": 0.09366681426763535,
+      "learning_rate": 2.1943518910584413e-05,
+      "loss": 0.5932,
+      "step": 4370
+    },
+    {
+      "epoch": 16.043956043956044,
+      "grad_norm": 0.08003173023462296,
+      "learning_rate": 2.157510203692851e-05,
+      "loss": 0.6002,
+      "step": 4380
+    },
+    {
+      "epoch": 16.08058608058608,
+      "grad_norm": 0.13794177770614624,
+      "learning_rate": 2.1209580372477568e-05,
+      "loss": 0.5839,
+      "step": 4390
+    },
+    {
+      "epoch": 16.117216117216117,
+      "grad_norm": 0.15384089946746826,
+      "learning_rate": 2.0846967325695974e-05,
+      "loss": 0.5906,
+      "step": 4400
+    },
+    {
+      "epoch": 16.153846153846153,
+      "grad_norm": 0.11059562861919403,
+      "learning_rate": 2.0487276198351102e-05,
+      "loss": 0.5955,
+      "step": 4410
+    },
+    {
+      "epoch": 16.19047619047619,
+      "grad_norm": 0.10853768140077591,
+      "learning_rate": 2.0130520185025272e-05,
+      "loss": 0.5871,
+      "step": 4420
+    },
+    {
+      "epoch": 16.227106227106226,
+      "grad_norm": 0.10565737634897232,
+      "learning_rate": 1.9776712372631758e-05,
+      "loss": 0.5824,
+      "step": 4430
+    },
+    {
+      "epoch": 16.263736263736263,
+      "grad_norm": 0.14627011120319366,
+      "learning_rate": 1.942586573993467e-05,
+      "loss": 0.5904,
+      "step": 4440
+    },
+    {
+      "epoch": 16.3003663003663,
+      "grad_norm": 0.08151823282241821,
+      "learning_rate": 1.907799315707292e-05,
+      "loss": 0.5884,
+      "step": 4450
+    },
+    {
+      "epoch": 16.336996336996336,
+      "grad_norm": 0.11814941465854645,
+      "learning_rate": 1.8733107385088118e-05,
+      "loss": 0.6046,
+      "step": 4460
+    },
+    {
+      "epoch": 16.373626373626372,
+      "grad_norm": 0.1151585578918457,
+      "learning_rate": 1.839122107545633e-05,
+      "loss": 0.5931,
+      "step": 4470
+    },
+    {
+      "epoch": 16.41025641025641,
+      "grad_norm": 0.11447068303823471,
+      "learning_rate": 1.805234676962415e-05,
+      "loss": 0.5864,
+      "step": 4480
+    },
+    {
+      "epoch": 16.446886446886445,
+      "grad_norm": 0.14155898988246918,
+      "learning_rate": 1.7716496898548574e-05,
+      "loss": 0.5807,
+      "step": 4490
+    },
+    {
+      "epoch": 16.483516483516482,
+      "grad_norm": 0.12283307313919067,
+      "learning_rate": 1.7383683782240874e-05,
+      "loss": 0.5875,
+      "step": 4500
+    },
+    {
+      "epoch": 16.520146520146522,
+      "grad_norm": 0.1172984391450882,
+      "learning_rate": 1.7053919629314873e-05,
+      "loss": 0.5904,
+      "step": 4510
+    },
+    {
+      "epoch": 16.55677655677656,
+      "grad_norm": 0.10092102736234665,
+      "learning_rate": 1.6727216536538968e-05,
+      "loss": 0.5808,
+      "step": 4520
+    },
+    {
+      "epoch": 16.593406593406595,
+      "grad_norm": 0.08429203927516937,
+      "learning_rate": 1.640358648839244e-05,
+      "loss": 0.588,
+      "step": 4530
+    },
+    {
+      "epoch": 16.63003663003663,
+      "grad_norm": 0.10441552102565765,
+      "learning_rate": 1.608304135662568e-05,
+      "loss": 0.581,
+      "step": 4540
+    },
+    {
+      "epoch": 16.666666666666668,
+      "grad_norm": 0.12228482216596603,
+      "learning_rate": 1.5765592899824945e-05,
+      "loss": 0.5886,
+      "step": 4550
+    },
+    {
+      "epoch": 16.703296703296704,
+      "grad_norm": 0.12790054082870483,
+      "learning_rate": 1.545125276298082e-05,
+      "loss": 0.6001,
+      "step": 4560
+    },
+    {
+      "epoch": 16.73992673992674,
+      "grad_norm": 0.10372098535299301,
+      "learning_rate": 1.5140032477061087e-05,
+      "loss": 0.5895,
+      "step": 4570
+    },
+    {
+      "epoch": 16.776556776556777,
+      "grad_norm": 0.09895550459623337,
+      "learning_rate": 1.483194345858779e-05,
+      "loss": 0.5913,
+      "step": 4580
+    },
+    {
+      "epoch": 16.813186813186814,
+      "grad_norm": 0.09903260320425034,
+      "learning_rate": 1.4526997009218429e-05,
+      "loss": 0.594,
+      "step": 4590
+    },
+    {
+      "epoch": 16.84981684981685,
+      "grad_norm": 0.10232320427894592,
+      "learning_rate": 1.4225204315331267e-05,
+      "loss": 0.6048,
+      "step": 4600
+    },
+    {
+      "epoch": 16.886446886446887,
+      "grad_norm": 0.12157467007637024,
+      "learning_rate": 1.3926576447615141e-05,
+      "loss": 0.591,
+      "step": 4610
+    },
+    {
+      "epoch": 16.923076923076923,
+      "grad_norm": 0.10378751158714294,
+      "learning_rate": 1.3631124360663295e-05,
+      "loss": 0.5841,
+      "step": 4620
+    },
+    {
+      "epoch": 16.95970695970696,
+      "grad_norm": 0.12422292679548264,
+      "learning_rate": 1.3338858892571438e-05,
+      "loss": 0.5997,
+      "step": 4630
+    },
+    {
+      "epoch": 16.996336996336996,
+      "grad_norm": 0.10485445708036423,
+      "learning_rate": 1.3049790764540315e-05,
+      "loss": 0.5861,
+      "step": 4640
+    },
+    {
+      "epoch": 17.032967032967033,
+      "grad_norm": 0.11898715049028397,
+      "learning_rate": 1.2763930580482347e-05,
+      "loss": 0.5898,
+      "step": 4650
+    },
+    {
+      "epoch": 17.06959706959707,
+      "grad_norm": 0.10091210901737213,
+      "learning_rate": 1.248128882663267e-05,
+      "loss": 0.6032,
+      "step": 4660
+    },
+    {
+      "epoch": 17.106227106227106,
+      "grad_norm": 0.1248786523938179,
+      "learning_rate": 1.220187587116436e-05,
+      "loss": 0.5856,
+      "step": 4670
+    },
+    {
+      "epoch": 17.142857142857142,
+      "grad_norm": 0.09647556394338608,
+      "learning_rate": 1.1925701963808304e-05,
+      "loss": 0.5955,
+      "step": 4680
+    },
+    {
+      "epoch": 17.17948717948718,
+      "grad_norm": 0.10741813480854034,
+      "learning_rate": 1.1652777235477038e-05,
+      "loss": 0.587,
+      "step": 4690
+    },
+    {
+      "epoch": 17.216117216117215,
+      "grad_norm": 0.10575282573699951,
+      "learning_rate": 1.1383111697893177e-05,
+      "loss": 0.5918,
+      "step": 4700
+    },
+    {
+      "epoch": 17.252747252747252,
+      "grad_norm": 0.11025935411453247,
+      "learning_rate": 1.1116715243222157e-05,
+      "loss": 0.5894,
+      "step": 4710
+    },
+    {
+      "epoch": 17.28937728937729,
+      "grad_norm": 0.13426312804222107,
+      "learning_rate": 1.0853597643709306e-05,
+      "loss": 0.5846,
+      "step": 4720
+    },
+    {
+      "epoch": 17.326007326007325,
+      "grad_norm": 0.14736682176589966,
+      "learning_rate": 1.0593768551321449e-05,
+      "loss": 0.5953,
+      "step": 4730
+    },
+    {
+      "epoch": 17.36263736263736,
+      "grad_norm": 0.12300019711256027,
+      "learning_rate": 1.03372374973928e-05,
+      "loss": 0.5892,
+      "step": 4740
+    },
+    {
+      "epoch": 17.399267399267398,
+      "grad_norm": 0.07489597797393799,
+      "learning_rate": 1.008401389227533e-05,
+      "loss": 0.5895,
+      "step": 4750
+    },
+    {
+      "epoch": 17.435897435897434,
+      "grad_norm": 0.09103349596261978,
+      "learning_rate": 9.834107024993564e-06,
+      "loss": 0.5897,
+      "step": 4760
+    },
+    {
+      "epoch": 17.47252747252747,
+      "grad_norm": 0.10881728678941727,
+      "learning_rate": 9.587526062903813e-06,
+      "loss": 0.5944,
+      "step": 4770
+    },
+    {
+      "epoch": 17.50915750915751,
+      "grad_norm": 0.1256643533706665,
+      "learning_rate": 9.34428005135792e-06,
+      "loss": 0.5954,
+      "step": 4780
+    },
+    {
+      "epoch": 17.545787545787547,
+      "grad_norm": 0.09786739200353622,
+      "learning_rate": 9.10437791337146e-06,
+      "loss": 0.586,
+      "step": 4790
+    },
+    {
+      "epoch": 17.582417582417584,
+      "grad_norm": 0.12305913120508194,
+      "learning_rate": 8.86782844929632e-06,
+      "loss": 0.5849,
+      "step": 4800
+    },
+    {
+      "epoch": 17.61904761904762,
+      "grad_norm": 0.094321109354496,
+      "learning_rate": 8.634640336498005e-06,
+      "loss": 0.6047,
+      "step": 4810
+    },
+    {
+      "epoch": 17.655677655677657,
+      "grad_norm": 0.10959235578775406,
+      "learning_rate": 8.404822129037282e-06,
+      "loss": 0.5845,
+      "step": 4820
+    },
+    {
+      "epoch": 17.692307692307693,
+      "grad_norm": 0.12101900577545166,
+      "learning_rate": 8.178382257356299e-06,
+      "loss": 0.5957,
+      "step": 4830
+    },
+    {
+      "epoch": 17.72893772893773,
+      "grad_norm": 0.0871628075838089,
+      "learning_rate": 7.955329027969454e-06,
+      "loss": 0.5982,
+      "step": 4840
+    },
+    {
+      "epoch": 17.765567765567766,
+      "grad_norm": 0.11070037633180618,
+      "learning_rate": 7.735670623158669e-06,
+      "loss": 0.5968,
+      "step": 4850
+    },
+    {
+      "epoch": 17.802197802197803,
+      "grad_norm": 0.12083383649587631,
+      "learning_rate": 7.519415100673134e-06,
+      "loss": 0.5903,
+      "step": 4860
+    },
+    {
+      "epoch": 17.83882783882784,
+      "grad_norm": 0.12383836507797241,
+      "learning_rate": 7.306570393433841e-06,
+      "loss": 0.5816,
+      "step": 4870
+    },
+    {
+      "epoch": 17.875457875457876,
+      "grad_norm": 0.127301886677742,
+      "learning_rate": 7.09714430924253e-06,
+      "loss": 0.5868,
+      "step": 4880
+    },
+    {
+      "epoch": 17.912087912087912,
+      "grad_norm": 0.10558178275823593,
+      "learning_rate": 6.891144530495308e-06,
+      "loss": 0.5934,
+      "step": 4890
+    },
+    {
+      "epoch": 17.94871794871795,
+      "grad_norm": 0.1014784723520279,
+      "learning_rate": 6.688578613900741e-06,
+      "loss": 0.5774,
+      "step": 4900
+    },
+    {
+      "epoch": 17.985347985347985,
+      "grad_norm": 0.09873995184898376,
+      "learning_rate": 6.489453990202765e-06,
+      "loss": 0.5889,
+      "step": 4910
+    },
+    {
+      "epoch": 18.021978021978022,
+      "grad_norm": 0.08788427710533142,
+      "learning_rate": 6.29377796390809e-06,
+      "loss": 0.6059,
+      "step": 4920
+    },
+    {
+      "epoch": 18.05860805860806,
+      "grad_norm": 0.11927574127912521,
+      "learning_rate": 6.101557713018109e-06,
+      "loss": 0.5867,
+      "step": 4930
+    },
+    {
+      "epoch": 18.095238095238095,
+      "grad_norm": 0.11732903867959976,
+      "learning_rate": 5.912800288765798e-06,
+      "loss": 0.5955,
+      "step": 4940
+    },
+    {
+      "epoch": 18.13186813186813,
+      "grad_norm": 0.10442673414945602,
+      "learning_rate": 5.727512615356896e-06,
+      "loss": 0.5861,
+      "step": 4950
+    },
+    {
+      "epoch": 18.168498168498168,
+      "grad_norm": 0.1446387618780136,
+      "learning_rate": 5.54570148971596e-06,
+      "loss": 0.5933,
+      "step": 4960
+    },
+    {
+      "epoch": 18.205128205128204,
+      "grad_norm": 0.12358086556196213,
+      "learning_rate": 5.367373581237032e-06,
+      "loss": 0.5851,
+      "step": 4970
+    },
+    {
+      "epoch": 18.24175824175824,
+      "grad_norm": 0.13660740852355957,
+      "learning_rate": 5.192535431539018e-06,
+      "loss": 0.5857,
+      "step": 4980
+    },
+    {
+      "epoch": 18.278388278388277,
+      "grad_norm": 0.16124746203422546,
+      "learning_rate": 5.021193454225608e-06,
+      "loss": 0.5818,
+      "step": 4990
+    },
+    {
+      "epoch": 18.315018315018314,
+      "grad_norm": 0.11943217366933823,
+      "learning_rate": 4.853353934650142e-06,
+      "loss": 0.5777,
+      "step": 5000
+    },
+    {
+      "epoch": 18.35164835164835,
+      "grad_norm": 0.13698545098304749,
+      "learning_rate": 4.689023029684991e-06,
+      "loss": 0.5885,
+      "step": 5010
+    },
+    {
+      "epoch": 18.388278388278387,
+      "grad_norm": 0.12822367250919342,
+      "learning_rate": 4.528206767495648e-06,
+      "loss": 0.5958,
+      "step": 5020
+    },
+    {
+      "epoch": 18.424908424908423,
+      "grad_norm": 0.09347227960824966,
+      "learning_rate": 4.370911047319707e-06,
+      "loss": 0.5865,
+      "step": 5030
+    },
+    {
+      "epoch": 18.46153846153846,
+      "grad_norm": 0.09904488921165466,
+      "learning_rate": 4.2171416392503395e-06,
+      "loss": 0.5998,
+      "step": 5040
+    },
+    {
+      "epoch": 18.498168498168496,
+      "grad_norm": 0.0935777798295021,
+      "learning_rate": 4.066904184024721e-06,
+      "loss": 0.602,
+      "step": 5050
+    },
+    {
+      "epoch": 18.534798534798536,
+      "grad_norm": 0.12104463577270508,
+      "learning_rate": 3.920204192817085e-06,
+      "loss": 0.5911,
+      "step": 5060
+    },
+    {
+      "epoch": 18.571428571428573,
+      "grad_norm": 0.10362918674945831,
+      "learning_rate": 3.7770470470365394e-06,
+      "loss": 0.5977,
+      "step": 5070
+    },
+    {
+      "epoch": 18.60805860805861,
+      "grad_norm": 0.08118361979722977,
+      "learning_rate": 3.6374379981296724e-06,
+      "loss": 0.6034,
+      "step": 5080
+    },
+    {
+      "epoch": 18.644688644688646,
+      "grad_norm": 0.11816778033971786,
+      "learning_rate": 3.5013821673879024e-06,
+      "loss": 0.5813,
+      "step": 5090
+    },
+    {
+      "epoch": 18.681318681318682,
+      "grad_norm": 0.12376552075147629,
+      "learning_rate": 3.368884545759655e-06,
+      "loss": 0.6031,
+      "step": 5100
+    },
+    {
+      "epoch": 18.71794871794872,
+      "grad_norm": 0.1341128796339035,
+      "learning_rate": 3.2399499936672143e-06,
+      "loss": 0.5848,
+      "step": 5110
+    },
+    {
+      "epoch": 18.754578754578755,
+      "grad_norm": 0.108884297311306,
+      "learning_rate": 3.1145832408284605e-06,
+      "loss": 0.5881,
+      "step": 5120
+    },
+    {
+      "epoch": 18.791208791208792,
+      "grad_norm": 0.12129422277212143,
+      "learning_rate": 2.9927888860833994e-06,
+      "loss": 0.5913,
+      "step": 5130
+    },
+    {
+      "epoch": 18.82783882783883,
+      "grad_norm": 0.12739436328411102,
+      "learning_rate": 2.8745713972254186e-06,
+      "loss": 0.5986,
+      "step": 5140
+    },
+    {
+      "epoch": 18.864468864468865,
+      "grad_norm": 0.11512988805770874,
+      "learning_rate": 2.759935110837394e-06,
+      "loss": 0.5859,
+      "step": 5150
+    },
+    {
+      "epoch": 18.9010989010989,
+      "grad_norm": 0.09803087264299393,
+      "learning_rate": 2.6488842321326404e-06,
+      "loss": 0.5956,
+      "step": 5160
+    },
+    {
+      "epoch": 18.937728937728938,
+      "grad_norm": 0.1529194414615631,
+      "learning_rate": 2.541422834800623e-06,
+      "loss": 0.587,
+      "step": 5170
+    },
+    {
+      "epoch": 18.974358974358974,
+      "grad_norm": 0.13892097771167755,
+      "learning_rate": 2.437554860857573e-06,
+      "loss": 0.593,
+      "step": 5180
+    },
+    {
+      "epoch": 19.01098901098901,
+      "grad_norm": 0.14193354547023773,
+      "learning_rate": 2.337284120501794e-06,
+      "loss": 0.59,
+      "step": 5190
+    },
+    {
+      "epoch": 19.047619047619047,
+      "grad_norm": 0.12481102347373962,
+      "learning_rate": 2.2406142919739795e-06,
+      "loss": 0.6003,
+      "step": 5200
+    },
+    {
+      "epoch": 19.084249084249084,
+      "grad_norm": 0.09259341657161713,
+      "learning_rate": 2.147548921422245e-06,
+      "loss": 0.5781,
+      "step": 5210
+    },
+    {
+      "epoch": 19.12087912087912,
+      "grad_norm": 0.12095209211111069,
+      "learning_rate": 2.058091422772005e-06,
+      "loss": 0.579,
+      "step": 5220
+    },
+    {
+      "epoch": 19.157509157509157,
+      "grad_norm": 0.12661872804164886,
+      "learning_rate": 1.9722450776008397e-06,
+      "loss": 0.5917,
+      "step": 5230
+    },
+    {
+      "epoch": 19.194139194139193,
+      "grad_norm": 0.13092070817947388,
+      "learning_rate": 1.8900130350180266e-06,
+      "loss": 0.5867,
+      "step": 5240
+    },
+    {
+      "epoch": 19.23076923076923,
+      "grad_norm": 0.12919223308563232,
+      "learning_rate": 1.8113983115490504e-06,
+      "loss": 0.6015,
+      "step": 5250
+    },
+    {
+      "epoch": 19.267399267399266,
+      "grad_norm": 0.11888205260038376,
+      "learning_rate": 1.7364037910249573e-06,
+      "loss": 0.5846,
+      "step": 5260
+    },
+    {
+      "epoch": 19.304029304029303,
+      "grad_norm": 0.11549215018749237,
+      "learning_rate": 1.6650322244765616e-06,
+      "loss": 0.5927,
+      "step": 5270
+    },
+    {
+      "epoch": 19.34065934065934,
+      "grad_norm": 0.1435500979423523,
+      "learning_rate": 1.5972862300335327e-06,
+      "loss": 0.5918,
+      "step": 5280
+    },
+    {
+      "epoch": 19.377289377289376,
+      "grad_norm": 0.10421594977378845,
+      "learning_rate": 1.5331682928283338e-06,
+      "loss": 0.6008,
+      "step": 5290
+    },
+    {
+      "epoch": 19.413919413919412,
+      "grad_norm": 0.10365405678749084,
+      "learning_rate": 1.4726807649050746e-06,
+      "loss": 0.5876,
+      "step": 5300
+    },
+    {
+      "epoch": 19.45054945054945,
+      "grad_norm": 0.1124468669295311,
+      "learning_rate": 1.4158258651332599e-06,
+      "loss": 0.603,
+      "step": 5310
+    },
+    {
+      "epoch": 19.487179487179485,
+      "grad_norm": 0.07424788922071457,
+      "learning_rate": 1.3626056791263295e-06,
+      "loss": 0.5959,
+      "step": 5320
+    },
+    {
+      "epoch": 19.523809523809526,
+      "grad_norm": 0.12210869044065475,
+      "learning_rate": 1.3130221591652044e-06,
+      "loss": 0.5884,
+      "step": 5330
+    },
+    {
+      "epoch": 19.560439560439562,
+      "grad_norm": 0.14640656113624573,
+      "learning_rate": 1.267077124126661e-06,
+      "loss": 0.5899,
+      "step": 5340
+    },
+    {
+      "epoch": 19.5970695970696,
+      "grad_norm": 0.11317311227321625,
+      "learning_rate": 1.2247722594165962e-06,
+      "loss": 0.5853,
+      "step": 5350
+    },
+    {
+      "epoch": 19.633699633699635,
+      "grad_norm": 0.12382689118385315,
+      "learning_rate": 1.1861091169081995e-06,
+      "loss": 0.6015,
+      "step": 5360
+    },
+    {
+      "epoch": 19.67032967032967,
+      "grad_norm": 0.12879708409309387,
+      "learning_rate": 1.1510891148850406e-06,
+      "loss": 0.5851,
+      "step": 5370
+    },
+    {
+      "epoch": 19.706959706959708,
+      "grad_norm": 0.18461808562278748,
+      "learning_rate": 1.1197135379890394e-06,
+      "loss": 0.5805,
+      "step": 5380
+    },
+    {
+      "epoch": 19.743589743589745,
+      "grad_norm": 0.09487687051296234,
+      "learning_rate": 1.0919835371733294e-06,
+      "loss": 0.5875,
+      "step": 5390
+    },
+    {
+      "epoch": 19.78021978021978,
+      "grad_norm": 0.1067289263010025,
+      "learning_rate": 1.0679001296600698e-06,
+      "loss": 0.5817,
+      "step": 5400
+    },
+    {
+      "epoch": 19.816849816849818,
+      "grad_norm": 0.12654584646224976,
+      "learning_rate": 1.0474641989030647e-06,
+      "loss": 0.5846,
+      "step": 5410
+    },
+    {
+      "epoch": 19.853479853479854,
+      "grad_norm": 0.09692192822694778,
+      "learning_rate": 1.0306764945554389e-06,
+      "loss": 0.5921,
+      "step": 5420
+    },
+    {
+      "epoch": 19.89010989010989,
+      "grad_norm": 0.10969322174787521,
+      "learning_rate": 1.017537632442055e-06,
+      "loss": 0.5885,
+      "step": 5430
+    },
+    {
+      "epoch": 19.926739926739927,
+      "grad_norm": 0.089121513068676,
+      "learning_rate": 1.0080480945370008e-06,
+      "loss": 0.601,
+      "step": 5440
+    },
+    {
+      "epoch": 19.963369963369964,
+      "grad_norm": 0.08970560878515244,
+      "learning_rate": 1.002208228945858e-06,
+      "loss": 0.5843,
+      "step": 5450
+    },
+    {
+      "epoch": 20.0,
+      "grad_norm": 0.13014668226242065,
+      "learning_rate": 1.0000182498929442e-06,
+      "loss": 0.5894,
+      "step": 5460
+    },
+    {
+      "epoch": 20.0,
+      "step": 5460,
+      "total_flos": 6.021259110266175e+18,
+      "train_loss": 0.6025372538374457,
+      "train_runtime": 24164.0932,
+      "train_samples_per_second": 86.611,
+      "train_steps_per_second": 0.226
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 5460,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 20,
+  "save_steps": 5000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 6.021259110266175e+18,
+  "train_batch_size": 24,
+  "trial_name": null,
+  "trial_params": null
+}