diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,5 +1,5 @@
 {
-  "best_metric": 0.35884907841682434,
+  "best_metric": 0.3794967830181122,
   "best_model_checkpoint": "videomae-large-finetuned-deepfake-subset/checkpoint-2235",
   "epoch": 9.1,
   "eval_steps": 500,
@@ -10,3236 +10,3236 @@
   "log_history": [
     {
       "epoch": 0.0022371364653243847,
-      "grad_norm": 8.820916175842285,
+      "grad_norm": 8.840668678283691,
       "learning_rate": 1.1185682326621925e-06,
-      "loss": 0.731,
+      "loss": 0.7314,
       "step": 10
     },
     {
       "epoch": 0.0044742729306487695,
-      "grad_norm": 12.290339469909668,
+      "grad_norm": 14.85615348815918,
       "learning_rate": 2.237136465324385e-06,
-      "loss": 0.705,
+      "loss": 0.7059,
       "step": 20
     },
     {
       "epoch": 0.006711409395973154,
-      "grad_norm": 10.981163024902344,
+      "grad_norm": 13.392386436462402,
       "learning_rate": 3.3557046979865773e-06,
-      "loss": 0.7033,
+      "loss": 0.677,
       "step": 30
     },
     {
       "epoch": 0.008948545861297539,
-      "grad_norm": 9.063810348510742,
+      "grad_norm": 9.807293891906738,
       "learning_rate": 4.47427293064877e-06,
-      "loss": 0.6902,
+      "loss": 0.7136,
       "step": 40
     },
     {
       "epoch": 0.011185682326621925,
-      "grad_norm": 14.12073802947998,
+      "grad_norm": 7.858952045440674,
       "learning_rate": 5.592841163310962e-06,
-      "loss": 0.651,
+      "loss": 0.679,
       "step": 50
     },
     {
       "epoch": 0.013422818791946308,
-      "grad_norm": 11.928865432739258,
+      "grad_norm": 11.944358825683594,
       "learning_rate": 6.7114093959731546e-06,
-      "loss": 0.6963,
+      "loss": 0.6539,
       "step": 60
     },
     {
       "epoch": 0.015659955257270694,
-      "grad_norm": 11.369632720947266,
+      "grad_norm": 11.665265083312988,
       "learning_rate": 7.829977628635348e-06,
-      "loss": 0.6992,
+      "loss": 0.7147,
       "step": 70
     },
     {
       "epoch": 0.017897091722595078,
-      "grad_norm": 13.989575386047363,
+      "grad_norm": 17.402263641357422,
       "learning_rate": 8.94854586129754e-06,
-      "loss": 0.6777,
+      "loss": 0.6939,
       "step": 80
     },
     {
       "epoch": 0.020134228187919462,
-      "grad_norm": 14.570990562438965,
+      "grad_norm": 13.367256164550781,
       "learning_rate": 1.006711409395973e-05,
-      "loss": 0.6896,
+      "loss": 0.6801,
       "step": 90
     },
     {
       "epoch": 0.02237136465324385,
-      "grad_norm": 10.531535148620605,
+      "grad_norm": 11.276527404785156,
       "learning_rate": 1.1185682326621925e-05,
-      "loss": 0.6827,
+      "loss": 0.6904,
       "step": 100
     },
     {
       "epoch": 0.024608501118568233,
-      "grad_norm": 11.762115478515625,
+      "grad_norm": 8.762359619140625,
       "learning_rate": 1.2304250559284117e-05,
-      "loss": 0.6952,
+      "loss": 0.6742,
       "step": 110
     },
     {
       "epoch": 0.026845637583892617,
-      "grad_norm": 10.128353118896484,
+      "grad_norm": 10.632108688354492,
       "learning_rate": 1.3422818791946309e-05,
-      "loss": 0.7187,
+      "loss": 0.6967,
       "step": 120
     },
     {
       "epoch": 0.029082774049217,
-      "grad_norm": 5.641812324523926,
+      "grad_norm": 8.825544357299805,
       "learning_rate": 1.4541387024608501e-05,
-      "loss": 0.6474,
+      "loss": 0.631,
       "step": 130
     },
     {
       "epoch": 0.03131991051454139,
-      "grad_norm": 9.743863105773926,
+      "grad_norm": 10.079858779907227,
       "learning_rate": 1.5659955257270695e-05,
-      "loss": 0.6943,
+      "loss": 0.698,
       "step": 140
     },
     {
       "epoch": 0.03355704697986577,
-      "grad_norm": 7.398090362548828,
+      "grad_norm": 12.647045135498047,
       "learning_rate": 1.6778523489932888e-05,
-      "loss": 0.6404,
+      "loss": 0.6871,
       "step": 150
     },
     {
       "epoch": 0.035794183445190156,
-      "grad_norm": 6.879280090332031,
+      "grad_norm": 7.6588311195373535,
       "learning_rate": 1.789709172259508e-05,
-      "loss": 0.7416,
+      "loss": 0.7482,
       "step": 160
     },
     {
       "epoch": 0.03803131991051454,
-      "grad_norm": 5.413578510284424,
+      "grad_norm": 8.198958396911621,
       "learning_rate": 1.9015659955257272e-05,
-      "loss": 0.6926,
+      "loss": 0.641,
       "step": 170
     },
     {
       "epoch": 0.040268456375838924,
-      "grad_norm": 6.2413411140441895,
+      "grad_norm": 8.8464937210083,
       "learning_rate": 2.013422818791946e-05,
-      "loss": 0.8072,
+      "loss": 0.7909,
       "step": 180
     },
     {
       "epoch": 0.042505592841163314,
-      "grad_norm": 3.8624861240386963,
+      "grad_norm": 4.950939178466797,
       "learning_rate": 2.1252796420581657e-05,
-      "loss": 0.685,
+      "loss": 0.7335,
       "step": 190
     },
     {
       "epoch": 0.0447427293064877,
-      "grad_norm": 2.6995649337768555,
+      "grad_norm": 5.206540584564209,
       "learning_rate": 2.237136465324385e-05,
-      "loss": 0.7004,
+      "loss": 0.744,
       "step": 200
     },
     {
       "epoch": 0.04697986577181208,
-      "grad_norm": 6.890233993530273,
+      "grad_norm": 6.270852088928223,
       "learning_rate": 2.348993288590604e-05,
-      "loss": 0.7119,
+      "loss": 0.691,
       "step": 210
     },
     {
       "epoch": 0.049217002237136466,
-      "grad_norm": 4.524689197540283,
+      "grad_norm": 6.907114028930664,
       "learning_rate": 2.4608501118568234e-05,
-      "loss": 0.6801,
+      "loss": 0.6868,
       "step": 220
     },
     {
       "epoch": 0.05145413870246085,
-      "grad_norm": 4.578395843505859,
+      "grad_norm": 6.208651065826416,
       "learning_rate": 2.5727069351230426e-05,
-      "loss": 0.6926,
+      "loss": 0.7205,
       "step": 230
     },
     {
       "epoch": 0.053691275167785234,
-      "grad_norm": 4.124147891998291,
+      "grad_norm": 4.508482456207275,
       "learning_rate": 2.6845637583892618e-05,
-      "loss": 0.6538,
+      "loss": 0.6688,
       "step": 240
     },
     {
       "epoch": 0.05592841163310962,
-      "grad_norm": 6.061787128448486,
+      "grad_norm": 8.969482421875,
       "learning_rate": 2.796420581655481e-05,
-      "loss": 0.6506,
+      "loss": 0.6811,
       "step": 250
     },
     {
       "epoch": 0.058165548098434,
-      "grad_norm": 12.14808177947998,
+      "grad_norm": 6.105631351470947,
       "learning_rate": 2.9082774049217003e-05,
-      "loss": 0.7799,
+      "loss": 0.7777,
       "step": 260
     },
     {
       "epoch": 0.06040268456375839,
-      "grad_norm": 3.3967928886413574,
+      "grad_norm": 4.278919696807861,
       "learning_rate": 3.02013422818792e-05,
-      "loss": 0.687,
+      "loss": 0.7046,
       "step": 270
     },
     {
       "epoch": 0.06263982102908278,
-      "grad_norm": 3.4164726734161377,
+      "grad_norm": 5.356738567352295,
       "learning_rate": 3.131991051454139e-05,
-      "loss": 0.6865,
+      "loss": 0.6642,
       "step": 280
     },
     {
       "epoch": 0.06487695749440715,
-      "grad_norm": 5.354477405548096,
+      "grad_norm": 5.4614691734313965,
       "learning_rate": 3.243847874720358e-05,
-      "loss": 0.6474,
+      "loss": 0.6817,
       "step": 290
     },
     {
       "epoch": 0.06711409395973154,
-      "grad_norm": 6.427885055541992,
+      "grad_norm": 5.7568278312683105,
       "learning_rate": 3.3557046979865775e-05,
-      "loss": 0.6265,
+      "loss": 0.6382,
       "step": 300
     },
     {
       "epoch": 0.06935123042505593,
-      "grad_norm": 5.238284111022949,
+      "grad_norm": 3.24556565284729,
       "learning_rate": 3.4675615212527964e-05,
-      "loss": 0.6951,
+      "loss": 0.679,
       "step": 310
     },
     {
       "epoch": 0.07158836689038031,
-      "grad_norm": 3.8965413570404053,
+      "grad_norm": 3.6615562438964844,
       "learning_rate": 3.579418344519016e-05,
-      "loss": 0.702,
+      "loss": 0.6796,
       "step": 320
     },
     {
       "epoch": 0.0738255033557047,
-      "grad_norm": 5.09553861618042,
+      "grad_norm": 7.1600494384765625,
       "learning_rate": 3.6912751677852356e-05,
-      "loss": 0.6297,
+      "loss": 0.675,
       "step": 330
     },
     {
       "epoch": 0.07606263982102908,
-      "grad_norm": 10.964111328125,
+      "grad_norm": 3.104412317276001,
       "learning_rate": 3.8031319910514545e-05,
-      "loss": 0.5219,
+      "loss": 0.6024,
       "step": 340
     },
     {
       "epoch": 0.07829977628635347,
-      "grad_norm": 2.5747010707855225,
+      "grad_norm": 4.062417030334473,
       "learning_rate": 3.914988814317674e-05,
-      "loss": 0.7686,
+      "loss": 0.7211,
       "step": 350
     },
     {
       "epoch": 0.08053691275167785,
-      "grad_norm": 8.747536659240723,
+      "grad_norm": 16.75772476196289,
       "learning_rate": 4.026845637583892e-05,
-      "loss": 0.6403,
+      "loss": 0.6736,
       "step": 360
     },
     {
       "epoch": 0.08277404921700224,
-      "grad_norm": 6.88169527053833,
+      "grad_norm": 6.7340168952941895,
       "learning_rate": 4.138702460850112e-05,
-      "loss": 0.7298,
+      "loss": 0.6964,
       "step": 370
     },
     {
       "epoch": 0.08501118568232663,
-      "grad_norm": 3.903372049331665,
+      "grad_norm": 4.515945911407471,
       "learning_rate": 4.2505592841163314e-05,
-      "loss": 0.725,
+      "loss": 0.7276,
       "step": 380
     },
     {
       "epoch": 0.087248322147651,
-      "grad_norm": 3.645214319229126,
+      "grad_norm": 2.2911009788513184,
       "learning_rate": 4.36241610738255e-05,
-      "loss": 0.6768,
+      "loss": 0.5794,
       "step": 390
     },
     {
       "epoch": 0.0894854586129754,
-      "grad_norm": 4.003920555114746,
+      "grad_norm": 5.8774027824401855,
       "learning_rate": 4.47427293064877e-05,
-      "loss": 0.7701,
+      "loss": 0.5691,
       "step": 400
     },
     {
       "epoch": 0.09172259507829977,
-      "grad_norm": 8.795763969421387,
+      "grad_norm": 4.3128743171691895,
       "learning_rate": 4.586129753914989e-05,
-      "loss": 0.7623,
+      "loss": 0.7606,
       "step": 410
     },
     {
       "epoch": 0.09395973154362416,
-      "grad_norm": 5.5298357009887695,
+      "grad_norm": 9.608476638793945,
       "learning_rate": 4.697986577181208e-05,
-      "loss": 0.7405,
+      "loss": 0.775,
       "step": 420
     },
     {
       "epoch": 0.09619686800894854,
-      "grad_norm": 5.45560884475708,
+      "grad_norm": 7.786107540130615,
       "learning_rate": 4.809843400447427e-05,
-      "loss": 0.6677,
+      "loss": 0.6394,
       "step": 430
     },
     {
       "epoch": 0.09843400447427293,
-      "grad_norm": 5.385969161987305,
+      "grad_norm": 8.117825508117676,
       "learning_rate": 4.921700223713647e-05,
-      "loss": 0.587,
+      "loss": 0.6361,
       "step": 440
     },
     {
       "epoch": 0.1,
-      "eval_loss": 0.6548025012016296,
-      "eval_runtime": 935.5761,
-      "eval_samples_per_second": 5.284,
-      "eval_steps_per_second": 0.661,
+      "eval_loss": 0.6478366255760193,
+      "eval_runtime": 887.1172,
+      "eval_samples_per_second": 8.36,
+      "eval_steps_per_second": 1.045,
       "step": 447
     },
     {
       "epoch": 1.0006711409395974,
-      "grad_norm": 3.5708768367767334,
+      "grad_norm": 4.550715446472168,
       "learning_rate": 4.99627143922446e-05,
-      "loss": 0.6714,
+      "loss": 0.6042,
       "step": 450
     },
     {
       "epoch": 1.0029082774049216,
-      "grad_norm": 3.380483865737915,
+      "grad_norm": 6.292792797088623,
       "learning_rate": 4.9838429033059906e-05,
-      "loss": 0.6215,
+      "loss": 0.6412,
       "step": 460
     },
     {
       "epoch": 1.005145413870246,
-      "grad_norm": 2.324326276779175,
+      "grad_norm": 4.233644485473633,
       "learning_rate": 4.971414367387522e-05,
-      "loss": 0.6885,
+      "loss": 0.6035,
       "step": 470
     },
     {
       "epoch": 1.0073825503355704,
-      "grad_norm": 4.357543468475342,
+      "grad_norm": 5.84645938873291,
       "learning_rate": 4.958985831469053e-05,
-      "loss": 0.678,
+      "loss": 0.5547,
       "step": 480
     },
     {
       "epoch": 1.0096196868008949,
-      "grad_norm": 3.158769369125366,
+      "grad_norm": 5.941657066345215,
       "learning_rate": 4.946557295550584e-05,
-      "loss": 0.7105,
+      "loss": 0.6611,
       "step": 490
     },
     {
       "epoch": 1.0118568232662193,
-      "grad_norm": 2.6318163871765137,
+      "grad_norm": 2.766065835952759,
       "learning_rate": 4.9341287596321155e-05,
-      "loss": 0.6218,
+      "loss": 0.7369,
       "step": 500
     },
     {
       "epoch": 1.0140939597315437,
-      "grad_norm": 3.114478349685669,
+      "grad_norm": 7.194911003112793,
       "learning_rate": 4.921700223713647e-05,
-      "loss": 0.6693,
+      "loss": 0.7586,
       "step": 510
     },
     {
       "epoch": 1.016331096196868,
-      "grad_norm": 5.514845371246338,
+      "grad_norm": 5.610304832458496,
       "learning_rate": 4.909271687795178e-05,
-      "loss": 0.6839,
+      "loss": 0.6815,
       "step": 520
     },
     {
       "epoch": 1.0185682326621923,
-      "grad_norm": 4.107670307159424,
+      "grad_norm": 5.830178260803223,
       "learning_rate": 4.896843151876709e-05,
-      "loss": 0.6526,
+      "loss": 0.5949,
       "step": 530
     },
     {
       "epoch": 1.0208053691275167,
-      "grad_norm": 9.080002784729004,
+      "grad_norm": 7.435003280639648,
       "learning_rate": 4.8844146159582404e-05,
-      "loss": 0.5914,
+      "loss": 0.6054,
       "step": 540
     },
     {
       "epoch": 1.0230425055928412,
-      "grad_norm": 2.4229257106781006,
+      "grad_norm": 3.300534248352051,
       "learning_rate": 4.871986080039772e-05,
-      "loss": 0.6653,
+      "loss": 0.644,
       "step": 550
     },
     {
       "epoch": 1.0252796420581656,
-      "grad_norm": 7.172940731048584,
+      "grad_norm": 5.10628080368042,
       "learning_rate": 4.859557544121303e-05,
-      "loss": 0.6491,
+      "loss": 0.7789,
       "step": 560
     },
     {
       "epoch": 1.02751677852349,
-      "grad_norm": 3.7052698135375977,
+      "grad_norm": 3.5758488178253174,
       "learning_rate": 4.8471290082028335e-05,
-      "loss": 0.6932,
+      "loss": 0.6373,
       "step": 570
     },
     {
       "epoch": 1.0297539149888144,
-      "grad_norm": 3.1077563762664795,
+      "grad_norm": 2.3225715160369873,
       "learning_rate": 4.8347004722843654e-05,
-      "loss": 0.6827,
+      "loss": 0.6763,
       "step": 580
     },
     {
       "epoch": 1.0319910514541386,
-      "grad_norm": 3.718585968017578,
+      "grad_norm": 2.920625686645508,
       "learning_rate": 4.8222719363658966e-05,
-      "loss": 0.7414,
+      "loss": 0.6733,
       "step": 590
     },
     {
       "epoch": 1.034228187919463,
-      "grad_norm": 2.7453243732452393,
+      "grad_norm": 5.328765869140625,
       "learning_rate": 4.809843400447427e-05,
-      "loss": 0.6593,
+      "loss": 0.6051,
       "step": 600
     },
     {
       "epoch": 1.0364653243847874,
-      "grad_norm": 2.820056438446045,
+      "grad_norm": 5.5468878746032715,
       "learning_rate": 4.7974148645289584e-05,
-      "loss": 0.674,
+      "loss": 0.6439,
       "step": 610
     },
     {
       "epoch": 1.0387024608501119,
-      "grad_norm": 5.4566192626953125,
+      "grad_norm": 3.9731194972991943,
       "learning_rate": 4.78498632861049e-05,
-      "loss": 0.6664,
+      "loss": 0.6416,
       "step": 620
     },
     {
       "epoch": 1.0409395973154363,
-      "grad_norm": 4.904478549957275,
+      "grad_norm": 5.598540782928467,
       "learning_rate": 4.772557792692021e-05,
-      "loss": 0.6305,
+      "loss": 0.6182,
       "step": 630
     },
     {
       "epoch": 1.0431767337807607,
-      "grad_norm": 4.792154312133789,
+      "grad_norm": 3.5205001831054688,
       "learning_rate": 4.760129256773552e-05,
-      "loss": 0.6781,
+      "loss": 0.6307,
       "step": 640
     },
     {
       "epoch": 1.045413870246085,
-      "grad_norm": 8.574151039123535,
+      "grad_norm": 8.040474891662598,
       "learning_rate": 4.7477007208550834e-05,
-      "loss": 0.6938,
+      "loss": 0.6298,
       "step": 650
     },
     {
       "epoch": 1.0476510067114093,
-      "grad_norm": 3.2592074871063232,
+      "grad_norm": 7.812734127044678,
       "learning_rate": 4.735272184936615e-05,
-      "loss": 0.7068,
+      "loss": 0.5933,
       "step": 660
     },
     {
       "epoch": 1.0498881431767337,
-      "grad_norm": 3.684659957885742,
+      "grad_norm": 8.276246070861816,
       "learning_rate": 4.722843649018146e-05,
-      "loss": 0.6925,
+      "loss": 0.68,
       "step": 670
     },
     {
       "epoch": 1.0521252796420582,
-      "grad_norm": 3.7525546550750732,
+      "grad_norm": 9.207062721252441,
       "learning_rate": 4.710415113099677e-05,
-      "loss": 0.7036,
+      "loss": 0.6591,
       "step": 680
     },
     {
       "epoch": 1.0543624161073826,
-      "grad_norm": 2.746760129928589,
+      "grad_norm": 2.244326591491699,
       "learning_rate": 4.697986577181208e-05,
-      "loss": 0.6284,
+      "loss": 0.5858,
       "step": 690
     },
     {
       "epoch": 1.056599552572707,
-      "grad_norm": 2.086256504058838,
+      "grad_norm": 4.339087963104248,
       "learning_rate": 4.6855580412627395e-05,
-      "loss": 0.6403,
+      "loss": 0.6173,
       "step": 700
     },
     {
       "epoch": 1.0588366890380314,
-      "grad_norm": 3.4739294052124023,
+      "grad_norm": 4.587048053741455,
       "learning_rate": 4.673129505344271e-05,
-      "loss": 0.6529,
+      "loss": 0.5405,
       "step": 710
     },
     {
       "epoch": 1.0610738255033556,
-      "grad_norm": 6.01228141784668,
+      "grad_norm": 6.943038463592529,
       "learning_rate": 4.660700969425802e-05,
-      "loss": 0.6661,
+      "loss": 0.5336,
       "step": 720
     },
     {
       "epoch": 1.06331096196868,
-      "grad_norm": 3.2344679832458496,
+      "grad_norm": 3.565220832824707,
       "learning_rate": 4.648272433507333e-05,
-      "loss": 0.6968,
+      "loss": 0.6332,
       "step": 730
     },
     {
       "epoch": 1.0655480984340044,
-      "grad_norm": 2.684730052947998,
+      "grad_norm": 4.71682596206665,
       "learning_rate": 4.635843897588864e-05,
-      "loss": 0.618,
+      "loss": 0.6398,
       "step": 740
     },
     {
       "epoch": 1.0677852348993289,
-      "grad_norm": 2.324744701385498,
+      "grad_norm": 8.337652206420898,
       "learning_rate": 4.623415361670396e-05,
-      "loss": 0.6689,
+      "loss": 0.6597,
       "step": 750
     },
     {
       "epoch": 1.0700223713646533,
-      "grad_norm": 4.5876569747924805,
+      "grad_norm": 6.056807994842529,
       "learning_rate": 4.610986825751927e-05,
-      "loss": 0.6356,
+      "loss": 0.68,
       "step": 760
     },
     {
       "epoch": 1.0722595078299777,
-      "grad_norm": 4.345230579376221,
+      "grad_norm": 3.2448365688323975,
       "learning_rate": 4.5985582898334575e-05,
-      "loss": 0.6123,
+      "loss": 0.6275,
       "step": 770
     },
     {
       "epoch": 1.0744966442953021,
-      "grad_norm": 5.078761577606201,
+      "grad_norm": 7.669735431671143,
       "learning_rate": 4.586129753914989e-05,
-      "loss": 0.7232,
+      "loss": 0.6847,
       "step": 780
     },
     {
       "epoch": 1.0767337807606263,
-      "grad_norm": 5.6790266036987305,
+      "grad_norm": 3.858572483062744,
       "learning_rate": 4.57370121799652e-05,
-      "loss": 0.6505,
+      "loss": 0.5387,
       "step": 790
     },
     {
       "epoch": 1.0789709172259507,
-      "grad_norm": 4.5163445472717285,
+      "grad_norm": 3.0774307250976562,
       "learning_rate": 4.561272682078052e-05,
-      "loss": 0.6364,
+      "loss": 0.6463,
       "step": 800
     },
     {
       "epoch": 1.0812080536912752,
-      "grad_norm": 5.686500549316406,
+      "grad_norm": 5.94879674911499,
       "learning_rate": 4.5488441461595824e-05,
-      "loss": 0.5556,
+      "loss": 0.5771,
       "step": 810
     },
     {
       "epoch": 1.0834451901565996,
-      "grad_norm": 2.98331356048584,
+      "grad_norm": 7.283985137939453,
       "learning_rate": 4.5364156102411137e-05,
-      "loss": 0.6267,
+      "loss": 0.4675,
       "step": 820
     },
     {
       "epoch": 1.085682326621924,
-      "grad_norm": 6.326406478881836,
+      "grad_norm": 5.343263149261475,
       "learning_rate": 4.523987074322645e-05,
-      "loss": 0.5947,
+      "loss": 0.619,
       "step": 830
     },
     {
       "epoch": 1.0879194630872484,
-      "grad_norm": 3.6593778133392334,
+      "grad_norm": 2.982252597808838,
       "learning_rate": 4.511558538404176e-05,
-      "loss": 0.6253,
+      "loss": 0.6595,
       "step": 840
     },
     {
       "epoch": 1.0901565995525728,
-      "grad_norm": 2.3336687088012695,
+      "grad_norm": 3.7833807468414307,
       "learning_rate": 4.4991300024857074e-05,
-      "loss": 0.5903,
+      "loss": 0.4992,
       "step": 850
     },
     {
       "epoch": 1.092393736017897,
-      "grad_norm": 3.4927194118499756,
+      "grad_norm": 4.744262218475342,
       "learning_rate": 4.4867014665672386e-05,
-      "loss": 0.5819,
+      "loss": 0.5481,
       "step": 860
     },
     {
       "epoch": 1.0946308724832214,
-      "grad_norm": 3.6458919048309326,
+      "grad_norm": 4.21218204498291,
       "learning_rate": 4.47427293064877e-05,
-      "loss": 0.5121,
+      "loss": 0.4745,
       "step": 870
     },
     {
       "epoch": 1.0968680089485459,
-      "grad_norm": 5.987964630126953,
+      "grad_norm": 6.225035667419434,
       "learning_rate": 4.461844394730301e-05,
-      "loss": 0.6405,
+      "loss": 0.6044,
       "step": 880
     },
     {
       "epoch": 1.0991051454138703,
-      "grad_norm": 5.051669597625732,
+      "grad_norm": 6.521196365356445,
       "learning_rate": 4.449415858811832e-05,
-      "loss": 0.6739,
+      "loss": 0.6774,
       "step": 890
     },
     {
       "epoch": 1.1,
-      "eval_loss": 0.5807220339775085,
-      "eval_runtime": 928.2211,
-      "eval_samples_per_second": 5.326,
-      "eval_steps_per_second": 0.666,
+      "eval_loss": 0.6047455072402954,
+      "eval_runtime": 892.7591,
+      "eval_samples_per_second": 8.307,
+      "eval_steps_per_second": 1.038,
       "step": 894
     },
     {
       "epoch": 2.001342281879195,
-      "grad_norm": 2.9489381313323975,
+      "grad_norm": 4.137167930603027,
       "learning_rate": 4.4369873228933635e-05,
-      "loss": 0.5788,
+      "loss": 0.6044,
       "step": 900
     },
     {
       "epoch": 2.003579418344519,
-      "grad_norm": 4.1599273681640625,
+      "grad_norm": 5.822400093078613,
       "learning_rate": 4.424558786974895e-05,
-      "loss": 0.4658,
+      "loss": 0.4379,
       "step": 910
     },
     {
       "epoch": 2.005816554809843,
-      "grad_norm": 3.328503131866455,
+      "grad_norm": 5.85886812210083,
       "learning_rate": 4.412130251056425e-05,
-      "loss": 0.6324,
+      "loss": 0.6157,
       "step": 920
     },
     {
       "epoch": 2.0080536912751676,
-      "grad_norm": 4.6580681800842285,
+      "grad_norm": 3.880828857421875,
       "learning_rate": 4.399701715137957e-05,
-      "loss": 0.5451,
+      "loss": 0.457,
       "step": 930
     },
     {
       "epoch": 2.010290827740492,
-      "grad_norm": 9.446999549865723,
+      "grad_norm": 6.341732025146484,
       "learning_rate": 4.3872731792194885e-05,
-      "loss": 0.5531,
+      "loss": 0.522,
       "step": 940
     },
     {
       "epoch": 2.0125279642058165,
-      "grad_norm": 2.4541642665863037,
+      "grad_norm": 10.094740867614746,
       "learning_rate": 4.374844643301019e-05,
-      "loss": 0.6491,
+      "loss": 0.6731,
       "step": 950
     },
     {
       "epoch": 2.014765100671141,
-      "grad_norm": 3.4880592823028564,
+      "grad_norm": 5.733267784118652,
       "learning_rate": 4.36241610738255e-05,
-      "loss": 0.6484,
+      "loss": 0.4796,
       "step": 960
     },
     {
       "epoch": 2.0170022371364653,
-      "grad_norm": 1.96310293674469,
+      "grad_norm": 3.963778257369995,
       "learning_rate": 4.349987571464082e-05,
-      "loss": 0.557,
+      "loss": 0.5343,
       "step": 970
     },
     {
       "epoch": 2.0192393736017897,
-      "grad_norm": 5.5436577796936035,
+      "grad_norm": 4.41682243347168,
       "learning_rate": 4.337559035545613e-05,
-      "loss": 0.5996,
+      "loss": 0.5492,
       "step": 980
     },
     {
       "epoch": 2.021476510067114,
-      "grad_norm": 10.744361877441406,
+      "grad_norm": 1.9658536911010742,
       "learning_rate": 4.325130499627144e-05,
-      "loss": 0.4878,
+      "loss": 0.4044,
       "step": 990
     },
     {
       "epoch": 2.0237136465324386,
-      "grad_norm": 5.160739421844482,
+      "grad_norm": 8.907441139221191,
       "learning_rate": 4.312701963708675e-05,
-      "loss": 0.6356,
+      "loss": 0.486,
       "step": 1000
     },
     {
       "epoch": 2.025950782997763,
-      "grad_norm": 4.501141548156738,
+      "grad_norm": 4.796280860900879,
       "learning_rate": 4.3002734277902064e-05,
-      "loss": 0.6026,
+      "loss": 0.5449,
       "step": 1010
     },
     {
       "epoch": 2.0281879194630874,
-      "grad_norm": 10.099557876586914,
+      "grad_norm": 13.880756378173828,
       "learning_rate": 4.287844891871738e-05,
-      "loss": 0.5309,
+      "loss": 0.5926,
       "step": 1020
     },
     {
       "epoch": 2.030425055928412,
-      "grad_norm": 5.0861029624938965,
+      "grad_norm": 5.1638312339782715,
       "learning_rate": 4.275416355953269e-05,
-      "loss": 0.7049,
+      "loss": 0.7223,
       "step": 1030
     },
     {
       "epoch": 2.032662192393736,
-      "grad_norm": 2.878390073776245,
+      "grad_norm": 2.8141660690307617,
       "learning_rate": 4.2629878200348e-05,
-      "loss": 0.6,
+      "loss": 0.5354,
       "step": 1040
     },
     {
       "epoch": 2.03489932885906,
-      "grad_norm": 4.252883434295654,
+      "grad_norm": 5.402659893035889,
       "learning_rate": 4.2505592841163314e-05,
-      "loss": 0.6321,
+      "loss": 0.5388,
       "step": 1050
     },
     {
       "epoch": 2.0371364653243846,
-      "grad_norm": 4.990085124969482,
+      "grad_norm": 9.897317886352539,
       "learning_rate": 4.2381307481978626e-05,
-      "loss": 0.6085,
+      "loss": 0.5581,
       "step": 1060
     },
     {
       "epoch": 2.039373601789709,
-      "grad_norm": 3.3724958896636963,
+      "grad_norm": 5.52318811416626,
       "learning_rate": 4.225702212279394e-05,
-      "loss": 0.4212,
+      "loss": 0.4331,
       "step": 1070
     },
     {
       "epoch": 2.0416107382550335,
-      "grad_norm": 7.132399559020996,
+      "grad_norm": 6.323409080505371,
       "learning_rate": 4.213273676360925e-05,
-      "loss": 0.6055,
+      "loss": 0.6491,
       "step": 1080
     },
     {
       "epoch": 2.043847874720358,
-      "grad_norm": 6.623457908630371,
+      "grad_norm": 5.7435503005981445,
       "learning_rate": 4.2008451404424556e-05,
-      "loss": 0.5264,
+      "loss": 0.494,
       "step": 1090
     },
     {
       "epoch": 2.0460850111856823,
-      "grad_norm": 3.902120351791382,
+      "grad_norm": 2.975238084793091,
       "learning_rate": 4.1884166045239875e-05,
-      "loss": 0.5448,
+      "loss": 0.5629,
       "step": 1100
     },
     {
       "epoch": 2.0483221476510067,
-      "grad_norm": 7.523580551147461,
+      "grad_norm": 7.936377048492432,
       "learning_rate": 4.175988068605519e-05,
-      "loss": 0.5755,
+      "loss": 0.5487,
       "step": 1110
     },
     {
       "epoch": 2.050559284116331,
-      "grad_norm": 6.084871768951416,
+      "grad_norm": 9.979763984680176,
       "learning_rate": 4.16355953268705e-05,
-      "loss": 0.6067,
+      "loss": 0.6606,
       "step": 1120
     },
     {
       "epoch": 2.0527964205816556,
-      "grad_norm": 10.225180625915527,
+      "grad_norm": 8.900005340576172,
       "learning_rate": 4.1511309967685806e-05,
-      "loss": 0.5577,
+      "loss": 0.537,
       "step": 1130
     },
     {
       "epoch": 2.05503355704698,
-      "grad_norm": 6.4241790771484375,
+      "grad_norm": 5.619589328765869,
       "learning_rate": 4.138702460850112e-05,
-      "loss": 0.6698,
+      "loss": 0.5785,
       "step": 1140
     },
     {
       "epoch": 2.0572706935123044,
-      "grad_norm": 2.930966377258301,
+      "grad_norm": 4.075002193450928,
       "learning_rate": 4.126273924931644e-05,
-      "loss": 0.5346,
+      "loss": 0.3999,
       "step": 1150
     },
     {
       "epoch": 2.059507829977629,
-      "grad_norm": 7.139683723449707,
+      "grad_norm": 5.566099166870117,
       "learning_rate": 4.113845389013174e-05,
-      "loss": 0.6482,
+      "loss": 0.4934,
       "step": 1160
     },
     {
       "epoch": 2.0617449664429532,
-      "grad_norm": 8.462257385253906,
+      "grad_norm": 7.996336936950684,
       "learning_rate": 4.1014168530947055e-05,
-      "loss": 0.5197,
+      "loss": 0.3563,
       "step": 1170
     },
     {
       "epoch": 2.063982102908277,
-      "grad_norm": 6.358745574951172,
+      "grad_norm": 8.547967910766602,
       "learning_rate": 4.088988317176237e-05,
-      "loss": 0.5159,
+      "loss": 0.6231,
       "step": 1180
     },
     {
       "epoch": 2.0662192393736016,
-      "grad_norm": 4.412177562713623,
+      "grad_norm": 4.0213236808776855,
       "learning_rate": 4.076559781257768e-05,
-      "loss": 0.7253,
+      "loss": 0.7292,
       "step": 1190
     },
     {
       "epoch": 2.068456375838926,
-      "grad_norm": 2.295029640197754,
+      "grad_norm": 4.820833206176758,
       "learning_rate": 4.064131245339299e-05,
-      "loss": 0.5456,
+      "loss": 0.4553,
       "step": 1200
     },
     {
       "epoch": 2.0706935123042505,
-      "grad_norm": 6.339975357055664,
+      "grad_norm": 9.791057586669922,
       "learning_rate": 4.0517027094208304e-05,
-      "loss": 0.5454,
+      "loss": 0.5081,
       "step": 1210
     },
     {
       "epoch": 2.072930648769575,
-      "grad_norm": 3.029376745223999,
+      "grad_norm": 2.710472345352173,
       "learning_rate": 4.039274173502362e-05,
-      "loss": 0.5875,
+      "loss": 0.4875,
       "step": 1220
     },
     {
       "epoch": 2.0751677852348993,
-      "grad_norm": 5.36366081237793,
+      "grad_norm": 4.432290554046631,
       "learning_rate": 4.026845637583892e-05,
-      "loss": 0.5715,
+      "loss": 0.4762,
       "step": 1230
     },
     {
       "epoch": 2.0774049217002237,
-      "grad_norm": 8.790428161621094,
+      "grad_norm": 12.281310081481934,
       "learning_rate": 4.014417101665424e-05,
-      "loss": 0.6075,
+      "loss": 0.5381,
       "step": 1240
     },
     {
       "epoch": 2.079642058165548,
-      "grad_norm": 12.27261734008789,
+      "grad_norm": 12.222536087036133,
       "learning_rate": 4.0019885657469554e-05,
-      "loss": 0.4774,
+      "loss": 0.4633,
       "step": 1250
     },
     {
       "epoch": 2.0818791946308726,
-      "grad_norm": 9.575027465820312,
+      "grad_norm": 10.08840274810791,
       "learning_rate": 3.9895600298284866e-05,
-      "loss": 0.5224,
+      "loss": 0.4012,
       "step": 1260
     },
     {
       "epoch": 2.084116331096197,
-      "grad_norm": 4.080117702484131,
+      "grad_norm": 12.933877944946289,
       "learning_rate": 3.977131493910017e-05,
-      "loss": 0.5142,
+      "loss": 0.4828,
       "step": 1270
     },
     {
       "epoch": 2.0863534675615214,
-      "grad_norm": 9.554539680480957,
+      "grad_norm": 8.704557418823242,
       "learning_rate": 3.964702957991549e-05,
-      "loss": 0.5001,
+      "loss": 0.4824,
       "step": 1280
     },
     {
       "epoch": 2.088590604026846,
-      "grad_norm": 5.12808084487915,
+      "grad_norm": 5.015852451324463,
       "learning_rate": 3.95227442207308e-05,
-      "loss": 0.5425,
+      "loss": 0.5165,
       "step": 1290
     },
     {
       "epoch": 2.09082774049217,
-      "grad_norm": 5.651499271392822,
+      "grad_norm": 11.11552906036377,
       "learning_rate": 3.939845886154611e-05,
-      "loss": 0.6167,
+      "loss": 0.5092,
       "step": 1300
     },
     {
       "epoch": 2.093064876957494,
-      "grad_norm": 10.270825386047363,
+      "grad_norm": 10.659283638000488,
       "learning_rate": 3.927417350236142e-05,
-      "loss": 0.553,
+      "loss": 0.5005,
       "step": 1310
     },
     {
       "epoch": 2.0953020134228186,
-      "grad_norm": 3.7876462936401367,
+      "grad_norm": 4.979311466217041,
       "learning_rate": 3.914988814317674e-05,
-      "loss": 0.7124,
+      "loss": 0.7376,
       "step": 1320
     },
     {
       "epoch": 2.097539149888143,
-      "grad_norm": 2.0910379886627197,
+      "grad_norm": 2.4167163372039795,
       "learning_rate": 3.9025602783992046e-05,
-      "loss": 0.5093,
+      "loss": 0.4397,
       "step": 1330
     },
     {
       "epoch": 2.0997762863534675,
-      "grad_norm": 6.5479207038879395,
+      "grad_norm": 7.5874857902526855,
       "learning_rate": 3.890131742480736e-05,
-      "loss": 0.444,
+      "loss": 0.4168,
       "step": 1340
     },
     {
       "epoch": 2.1,
-      "eval_loss": 0.5115740895271301,
-      "eval_runtime": 933.3886,
-      "eval_samples_per_second": 5.297,
-      "eval_steps_per_second": 0.662,
+      "eval_loss": 0.48516377806663513,
+      "eval_runtime": 891.9205,
+      "eval_samples_per_second": 8.315,
+      "eval_steps_per_second": 1.039,
       "step": 1341
     },
     {
       "epoch": 3.002013422818792,
-      "grad_norm": 11.015625,
+      "grad_norm": 9.898842811584473,
       "learning_rate": 3.877703206562267e-05,
-      "loss": 0.4018,
+      "loss": 0.4532,
       "step": 1350
     },
     {
       "epoch": 3.004250559284116,
-      "grad_norm": 8.430967330932617,
+      "grad_norm": 10.05817699432373,
       "learning_rate": 3.865274670643798e-05,
-      "loss": 0.767,
+      "loss": 0.7569,
       "step": 1360
     },
     {
       "epoch": 3.0064876957494406,
-      "grad_norm": 1.8640131950378418,
+      "grad_norm": 1.2511606216430664,
       "learning_rate": 3.8528461347253295e-05,
-      "loss": 0.4937,
+      "loss": 0.4412,
       "step": 1370
     },
     {
       "epoch": 3.008724832214765,
-      "grad_norm": 6.5426177978515625,
+      "grad_norm": 11.480599403381348,
       "learning_rate": 3.840417598806861e-05,
-      "loss": 0.4556,
+      "loss": 0.3935,
       "step": 1380
     },
     {
       "epoch": 3.0109619686800895,
-      "grad_norm": 8.181113243103027,
+      "grad_norm": 11.63521671295166,
       "learning_rate": 3.827989062888392e-05,
-      "loss": 0.4801,
+      "loss": 0.5626,
       "step": 1390
     },
     {
       "epoch": 3.013199105145414,
-      "grad_norm": 5.536498069763184,
+      "grad_norm": 9.114398956298828,
       "learning_rate": 3.815560526969923e-05,
-      "loss": 0.5512,
+      "loss": 0.6732,
       "step": 1400
     },
     {
       "epoch": 3.0154362416107383,
-      "grad_norm": 8.874992370605469,
+      "grad_norm": 7.543931484222412,
       "learning_rate": 3.8031319910514545e-05,
-      "loss": 0.6437,
+      "loss": 0.4275,
       "step": 1410
     },
     {
       "epoch": 3.0176733780760627,
-      "grad_norm": 8.250784873962402,
+      "grad_norm": 10.644927024841309,
       "learning_rate": 3.790703455132986e-05,
-      "loss": 0.5753,
+      "loss": 0.4325,
       "step": 1420
     },
     {
       "epoch": 3.019910514541387,
-      "grad_norm": 6.462278366088867,
+      "grad_norm": 3.5276317596435547,
       "learning_rate": 3.778274919214517e-05,
-      "loss": 0.4336,
+      "loss": 0.3605,
       "step": 1430
     },
     {
       "epoch": 3.0221476510067116,
-      "grad_norm": 2.823152780532837,
+      "grad_norm": 5.080909729003906,
       "learning_rate": 3.7658463832960475e-05,
-      "loss": 0.6393,
+      "loss": 0.5921,
       "step": 1440
     },
     {
       "epoch": 3.024384787472036,
-      "grad_norm": 1.859449028968811,
+      "grad_norm": 1.2971785068511963,
       "learning_rate": 3.753417847377579e-05,
-      "loss": 0.3849,
+      "loss": 0.3565,
       "step": 1450
     },
     {
       "epoch": 3.02662192393736,
-      "grad_norm": 7.397490501403809,
+      "grad_norm": 3.7884743213653564,
       "learning_rate": 3.7409893114591106e-05,
-      "loss": 0.5019,
+      "loss": 0.287,
       "step": 1460
     },
     {
       "epoch": 3.0288590604026844,
-      "grad_norm": 3.8984179496765137,
+      "grad_norm": 5.833460807800293,
       "learning_rate": 3.728560775540642e-05,
-      "loss": 0.5481,
+      "loss": 0.4596,
       "step": 1470
     },
     {
       "epoch": 3.031096196868009,
-      "grad_norm": 9.145849227905273,
+      "grad_norm": 10.393218994140625,
       "learning_rate": 3.7161322396221724e-05,
-      "loss": 0.5017,
+      "loss": 0.3767,
       "step": 1480
     },
     {
       "epoch": 3.033333333333333,
-      "grad_norm": 9.390955924987793,
+      "grad_norm": 12.434408187866211,
       "learning_rate": 3.7037037037037037e-05,
-      "loss": 0.4875,
+      "loss": 0.5202,
       "step": 1490
     },
     {
       "epoch": 3.0355704697986576,
-      "grad_norm": 7.330883979797363,
+      "grad_norm": 7.507827281951904,
       "learning_rate": 3.6912751677852356e-05,
-      "loss": 0.4656,
+      "loss": 0.4699,
       "step": 1500
     },
     {
       "epoch": 3.037807606263982,
-      "grad_norm": 4.043966293334961,
+      "grad_norm": 4.108563423156738,
       "learning_rate": 3.678846631866766e-05,
-      "loss": 0.4502,
+      "loss": 0.4836,
       "step": 1510
     },
     {
       "epoch": 3.0400447427293065,
-      "grad_norm": 4.5185089111328125,
+      "grad_norm": 6.502699851989746,
       "learning_rate": 3.6664180959482974e-05,
-      "loss": 0.4251,
+      "loss": 0.2626,
       "step": 1520
     },
     {
       "epoch": 3.042281879194631,
-      "grad_norm": 5.390605926513672,
+      "grad_norm": 11.717517852783203,
       "learning_rate": 3.6539895600298286e-05,
-      "loss": 0.5748,
+      "loss": 0.6785,
       "step": 1530
     },
     {
       "epoch": 3.0445190156599553,
-      "grad_norm": 2.755403995513916,
+      "grad_norm": 9.127665519714355,
       "learning_rate": 3.64156102411136e-05,
-      "loss": 0.4585,
+      "loss": 0.5052,
       "step": 1540
     },
     {
       "epoch": 3.0467561521252797,
-      "grad_norm": 4.787998676300049,
+      "grad_norm": 6.234489917755127,
       "learning_rate": 3.629132488192891e-05,
-      "loss": 0.6129,
+      "loss": 0.4078,
       "step": 1550
     },
     {
       "epoch": 3.048993288590604,
-      "grad_norm": 3.5947635173797607,
+      "grad_norm": 13.028934478759766,
       "learning_rate": 3.616703952274422e-05,
-      "loss": 0.4497,
+      "loss": 0.5171,
       "step": 1560
     },
     {
       "epoch": 3.0512304250559286,
-      "grad_norm": 9.294822692871094,
+      "grad_norm": 9.663383483886719,
       "learning_rate": 3.6042754163559535e-05,
-      "loss": 0.5038,
+      "loss": 0.6043,
       "step": 1570
     },
     {
       "epoch": 3.053467561521253,
-      "grad_norm": 4.422880172729492,
+      "grad_norm": 3.3989367485046387,
       "learning_rate": 3.591846880437484e-05,
-      "loss": 0.4785,
+      "loss": 0.4231,
       "step": 1580
     },
     {
       "epoch": 3.0557046979865774,
-      "grad_norm": 8.102038383483887,
+      "grad_norm": 6.579158782958984,
       "learning_rate": 3.579418344519016e-05,
-      "loss": 0.6245,
+      "loss": 0.515,
       "step": 1590
     },
     {
       "epoch": 3.0579418344519014,
-      "grad_norm": 4.504268169403076,
+      "grad_norm": 5.151082515716553,
       "learning_rate": 3.566989808600547e-05,
-      "loss": 0.4765,
+      "loss": 0.3919,
       "step": 1600
     },
     {
       "epoch": 3.060178970917226,
-      "grad_norm": 9.747063636779785,
+      "grad_norm": 2.145969867706299,
       "learning_rate": 3.5545612726820785e-05,
-      "loss": 0.6108,
+      "loss": 0.4733,
       "step": 1610
     },
     {
       "epoch": 3.06241610738255,
-      "grad_norm": 2.142608165740967,
+      "grad_norm": 5.741364002227783,
       "learning_rate": 3.542132736763609e-05,
-      "loss": 0.5031,
+      "loss": 0.4354,
       "step": 1620
     },
     {
       "epoch": 3.0646532438478746,
-      "grad_norm": 4.941412448883057,
+      "grad_norm": 3.9511780738830566,
       "learning_rate": 3.529704200845141e-05,
-      "loss": 0.4812,
+      "loss": 0.443,
       "step": 1630
     },
     {
       "epoch": 3.066890380313199,
-      "grad_norm": 9.923797607421875,
+      "grad_norm": 6.973093509674072,
       "learning_rate": 3.517275664926672e-05,
-      "loss": 0.4473,
+      "loss": 0.4201,
       "step": 1640
     },
     {
       "epoch": 3.0691275167785235,
-      "grad_norm": 4.434045791625977,
+      "grad_norm": 1.0698981285095215,
       "learning_rate": 3.504847129008203e-05,
-      "loss": 0.4739,
+      "loss": 0.485,
       "step": 1650
     },
     {
       "epoch": 3.071364653243848,
-      "grad_norm": 4.894314765930176,
+      "grad_norm": 6.2486701011657715,
       "learning_rate": 3.492418593089734e-05,
-      "loss": 0.5424,
+      "loss": 0.4176,
       "step": 1660
     },
     {
       "epoch": 3.0736017897091723,
-      "grad_norm": 3.4307405948638916,
+      "grad_norm": 2.134953022003174,
       "learning_rate": 3.479990057171265e-05,
-      "loss": 0.5213,
+      "loss": 0.3947,
       "step": 1670
     },
     {
       "epoch": 3.0758389261744967,
-      "grad_norm": 2.12326979637146,
+      "grad_norm": 1.0479804277420044,
       "learning_rate": 3.4675615212527964e-05,
-      "loss": 0.4881,
+      "loss": 0.3886,
       "step": 1680
     },
     {
       "epoch": 3.078076062639821,
-      "grad_norm": 3.1597368717193604,
+      "grad_norm": 1.2134567499160767,
       "learning_rate": 3.455132985334328e-05,
-      "loss": 0.4149,
+      "loss": 0.3022,
       "step": 1690
     },
     {
       "epoch": 3.0803131991051456,
-      "grad_norm": 8.707585334777832,
+      "grad_norm": 10.898287773132324,
       "learning_rate": 3.442704449415859e-05,
-      "loss": 0.4081,
+      "loss": 0.3078,
       "step": 1700
     },
     {
       "epoch": 3.08255033557047,
-      "grad_norm": 14.810624122619629,
+      "grad_norm": 18.389766693115234,
       "learning_rate": 3.43027591349739e-05,
-      "loss": 0.569,
+      "loss": 0.663,
       "step": 1710
     },
     {
       "epoch": 3.0847874720357944,
-      "grad_norm": 2.331845998764038,
+      "grad_norm": 2.9712672233581543,
       "learning_rate": 3.4178473775789214e-05,
-      "loss": 0.4592,
+      "loss": 0.4882,
       "step": 1720
     },
     {
       "epoch": 3.0870246085011184,
-      "grad_norm": 0.9509165287017822,
+      "grad_norm": 4.190480709075928,
       "learning_rate": 3.4054188416604526e-05,
-      "loss": 0.3231,
+      "loss": 0.4099,
       "step": 1730
     },
     {
       "epoch": 3.089261744966443,
-      "grad_norm": 7.728116989135742,
+      "grad_norm": 5.036893367767334,
       "learning_rate": 3.392990305741984e-05,
-      "loss": 0.4468,
+      "loss": 0.3227,
       "step": 1740
     },
     {
       "epoch": 3.091498881431767,
-      "grad_norm": 5.856037616729736,
+      "grad_norm": 3.94989013671875,
       "learning_rate": 3.380561769823515e-05,
-      "loss": 0.4978,
+      "loss": 0.3345,
       "step": 1750
     },
     {
       "epoch": 3.0937360178970916,
-      "grad_norm": 5.866270065307617,
+      "grad_norm": 10.000751495361328,
       "learning_rate": 3.3681332339050456e-05,
-      "loss": 0.3826,
+      "loss": 0.3655,
       "step": 1760
     },
     {
       "epoch": 3.095973154362416,
-      "grad_norm": 4.168057918548584,
+      "grad_norm": 6.677926063537598,
       "learning_rate": 3.3557046979865775e-05,
-      "loss": 0.4231,
+      "loss": 0.2954,
       "step": 1770
     },
     {
       "epoch": 3.0982102908277405,
-      "grad_norm": 11.238313674926758,
+      "grad_norm": 10.632355690002441,
       "learning_rate": 3.343276162068109e-05,
-      "loss": 0.5038,
+      "loss": 0.4427,
       "step": 1780
     },
     {
       "epoch": 3.1,
-      "eval_loss": 0.7296048998832703,
-      "eval_runtime": 937.8481,
-      "eval_samples_per_second": 5.272,
-      "eval_steps_per_second": 0.659,
+      "eval_loss": 0.8546826243400574,
+      "eval_runtime": 891.5992,
+      "eval_samples_per_second": 8.318,
+      "eval_steps_per_second": 1.04,
       "step": 1788
     },
     {
       "epoch": 4.000447427293065,
-      "grad_norm": 8.892515182495117,
+      "grad_norm": 16.46874237060547,
       "learning_rate": 3.330847626149639e-05,
-      "loss": 0.478,
+      "loss": 0.6448,
       "step": 1790
     },
     {
       "epoch": 4.00268456375839,
-      "grad_norm": 5.279501914978027,
+      "grad_norm": 7.57173490524292,
       "learning_rate": 3.3184190902311706e-05,
-      "loss": 0.4808,
+      "loss": 0.4691,
       "step": 1800
     },
     {
       "epoch": 4.004921700223714,
-      "grad_norm": 1.478691577911377,
+      "grad_norm": 6.603734493255615,
       "learning_rate": 3.3059905543127025e-05,
-      "loss": 0.4722,
+      "loss": 0.4373,
       "step": 1810
     },
     {
       "epoch": 4.007158836689038,
-      "grad_norm": 9.76483154296875,
+      "grad_norm": 5.733815670013428,
       "learning_rate": 3.293562018394234e-05,
-      "loss": 0.4112,
+      "loss": 0.4219,
       "step": 1820
     },
     {
       "epoch": 4.009395973154362,
-      "grad_norm": 3.3786661624908447,
+      "grad_norm": 0.5362582206726074,
       "learning_rate": 3.281133482475764e-05,
-      "loss": 0.4243,
+      "loss": 0.2901,
       "step": 1830
     },
     {
       "epoch": 4.011633109619686,
-      "grad_norm": 4.852506637573242,
+      "grad_norm": 9.654644012451172,
       "learning_rate": 3.2687049465572955e-05,
-      "loss": 0.3942,
+      "loss": 0.4001,
       "step": 1840
     },
     {
       "epoch": 4.013870246085011,
-      "grad_norm": 7.3863301277160645,
+      "grad_norm": 5.657355785369873,
       "learning_rate": 3.2562764106388274e-05,
-      "loss": 0.5132,
+      "loss": 0.3882,
       "step": 1850
     },
     {
       "epoch": 4.016107382550335,
-      "grad_norm": 4.083529472351074,
+      "grad_norm": 4.895392417907715,
       "learning_rate": 3.243847874720358e-05,
-      "loss": 0.355,
+      "loss": 0.402,
       "step": 1860
     },
     {
       "epoch": 4.01834451901566,
-      "grad_norm": 6.427587032318115,
+      "grad_norm": 7.476536750793457,
       "learning_rate": 3.231419338801889e-05,
-      "loss": 0.4309,
+      "loss": 0.2948,
       "step": 1870
     },
     {
       "epoch": 4.020581655480984,
-      "grad_norm": 11.749137878417969,
+      "grad_norm": 15.446544647216797,
       "learning_rate": 3.2189908028834204e-05,
-      "loss": 0.5108,
+      "loss": 0.3745,
       "step": 1880
     },
     {
       "epoch": 4.0228187919463085,
-      "grad_norm": 8.5358304977417,
+      "grad_norm": 9.441873550415039,
       "learning_rate": 3.206562266964952e-05,
-      "loss": 0.3922,
+      "loss": 0.4731,
       "step": 1890
     },
     {
       "epoch": 4.025055928411633,
-      "grad_norm": 5.002346038818359,
+      "grad_norm": 2.744432210922241,
       "learning_rate": 3.194133731046483e-05,
-      "loss": 0.3854,
+      "loss": 0.5575,
       "step": 1900
     },
     {
       "epoch": 4.027293064876957,
-      "grad_norm": 4.695934295654297,
+      "grad_norm": 7.594290733337402,
       "learning_rate": 3.181705195128014e-05,
-      "loss": 0.3627,
+      "loss": 0.463,
       "step": 1910
     },
     {
       "epoch": 4.029530201342282,
-      "grad_norm": 9.114057540893555,
+      "grad_norm": 9.001227378845215,
       "learning_rate": 3.1692766592095454e-05,
-      "loss": 0.3789,
+      "loss": 0.3622,
       "step": 1920
     },
     {
       "epoch": 4.031767337807606,
-      "grad_norm": 13.6981782913208,
+      "grad_norm": 12.734862327575684,
       "learning_rate": 3.156848123291076e-05,
-      "loss": 0.3505,
+      "loss": 0.3435,
       "step": 1930
     },
     {
       "epoch": 4.034004474272931,
-      "grad_norm": 1.9973479509353638,
+      "grad_norm": 1.6699249744415283,
       "learning_rate": 3.144419587372608e-05,
-      "loss": 0.3911,
+      "loss": 0.2984,
       "step": 1940
     },
     {
       "epoch": 4.036241610738255,
-      "grad_norm": 12.077263832092285,
+      "grad_norm": 14.737456321716309,
       "learning_rate": 3.131991051454139e-05,
-      "loss": 0.4104,
+      "loss": 0.4041,
       "step": 1950
     },
     {
       "epoch": 4.0384787472035795,
-      "grad_norm": 9.706670761108398,
+      "grad_norm": 5.71207857131958,
       "learning_rate": 3.11956251553567e-05,
-      "loss": 0.4253,
+      "loss": 0.3152,
       "step": 1960
     },
     {
       "epoch": 4.040715883668904,
-      "grad_norm": 3.5470070838928223,
+      "grad_norm": 12.913744926452637,
       "learning_rate": 3.107133979617201e-05,
-      "loss": 0.4829,
+      "loss": 0.4892,
       "step": 1970
     },
     {
       "epoch": 4.042953020134228,
-      "grad_norm": 9.14087200164795,
+      "grad_norm": 12.614986419677734,
       "learning_rate": 3.094705443698732e-05,
-      "loss": 0.4324,
+      "loss": 0.4399,
       "step": 1980
     },
     {
       "epoch": 4.045190156599553,
-      "grad_norm": 3.2850630283355713,
+      "grad_norm": 10.273175239562988,
       "learning_rate": 3.082276907780264e-05,
-      "loss": 0.4274,
+      "loss": 0.5362,
       "step": 1990
     },
     {
       "epoch": 4.047427293064877,
-      "grad_norm": 6.201521873474121,
+      "grad_norm": 1.00068199634552,
       "learning_rate": 3.0698483718617946e-05,
-      "loss": 0.373,
+      "loss": 0.4019,
       "step": 2000
     },
     {
       "epoch": 4.049664429530202,
-      "grad_norm": 2.3075242042541504,
+      "grad_norm": 3.2468674182891846,
       "learning_rate": 3.057419835943326e-05,
-      "loss": 0.2872,
+      "loss": 0.2908,
       "step": 2010
     },
     {
       "epoch": 4.051901565995526,
-      "grad_norm": 10.908852577209473,
+      "grad_norm": 13.30661392211914,
       "learning_rate": 3.044991300024857e-05,
-      "loss": 0.3557,
+      "loss": 0.4924,
       "step": 2020
     },
     {
       "epoch": 4.05413870246085,
-      "grad_norm": 10.656627655029297,
+      "grad_norm": 7.9545063972473145,
       "learning_rate": 3.0325627641063886e-05,
-      "loss": 0.4079,
+      "loss": 0.3895,
       "step": 2030
     },
     {
       "epoch": 4.056375838926175,
-      "grad_norm": 8.033869743347168,
+      "grad_norm": 9.699666023254395,
       "learning_rate": 3.02013422818792e-05,
-      "loss": 0.2665,
+      "loss": 0.2904,
       "step": 2040
     },
     {
       "epoch": 4.058612975391499,
-      "grad_norm": 3.8938560485839844,
+      "grad_norm": 6.4541707038879395,
       "learning_rate": 3.0077056922694508e-05,
-      "loss": 0.4266,
+      "loss": 0.4361,
       "step": 2050
     },
     {
       "epoch": 4.060850111856824,
-      "grad_norm": 11.372118949890137,
+      "grad_norm": 22.470691680908203,
       "learning_rate": 2.995277156350982e-05,
-      "loss": 0.4161,
+      "loss": 0.3353,
       "step": 2060
     },
     {
       "epoch": 4.063087248322148,
-      "grad_norm": 7.895952224731445,
+      "grad_norm": 8.081043243408203,
       "learning_rate": 2.9828486204325136e-05,
-      "loss": 0.5103,
+      "loss": 0.492,
       "step": 2070
     },
     {
       "epoch": 4.065324384787472,
-      "grad_norm": 5.879835605621338,
+      "grad_norm": 1.5061018466949463,
       "learning_rate": 2.9704200845140445e-05,
-      "loss": 0.424,
+      "loss": 0.3086,
       "step": 2080
     },
     {
       "epoch": 4.067561521252796,
-      "grad_norm": 1.7476272583007812,
+      "grad_norm": 0.6852089762687683,
       "learning_rate": 2.9579915485955757e-05,
-      "loss": 0.403,
+      "loss": 0.389,
       "step": 2090
     },
     {
       "epoch": 4.06979865771812,
-      "grad_norm": 11.164639472961426,
+      "grad_norm": 12.416427612304688,
       "learning_rate": 2.9455630126771066e-05,
-      "loss": 0.3045,
+      "loss": 0.3055,
       "step": 2100
     },
     {
       "epoch": 4.072035794183445,
-      "grad_norm": 4.138221740722656,
+      "grad_norm": 7.986660957336426,
       "learning_rate": 2.9331344767586378e-05,
-      "loss": 0.3905,
+      "loss": 0.5869,
       "step": 2110
     },
     {
       "epoch": 4.074272930648769,
-      "grad_norm": 11.378765106201172,
+      "grad_norm": 14.885111808776855,
       "learning_rate": 2.9207059408401694e-05,
-      "loss": 0.6302,
+      "loss": 0.3224,
       "step": 2120
     },
     {
       "epoch": 4.076510067114094,
-      "grad_norm": 4.001158714294434,
+      "grad_norm": 10.167362213134766,
       "learning_rate": 2.9082774049217003e-05,
-      "loss": 0.3419,
+      "loss": 0.3044,
       "step": 2130
     },
     {
       "epoch": 4.078747203579418,
-      "grad_norm": 5.916733264923096,
+      "grad_norm": 6.552039623260498,
       "learning_rate": 2.8958488690032315e-05,
-      "loss": 0.3233,
+      "loss": 0.2917,
       "step": 2140
     },
     {
       "epoch": 4.0809843400447425,
-      "grad_norm": 9.534823417663574,
+      "grad_norm": 7.694530487060547,
       "learning_rate": 2.8834203330847624e-05,
-      "loss": 0.4643,
+      "loss": 0.4193,
       "step": 2150
     },
     {
       "epoch": 4.083221476510067,
-      "grad_norm": 8.507469177246094,
+      "grad_norm": 7.818331241607666,
       "learning_rate": 2.8709917971662943e-05,
-      "loss": 0.391,
+      "loss": 0.3355,
       "step": 2160
     },
     {
       "epoch": 4.085458612975391,
-      "grad_norm": 12.570788383483887,
+      "grad_norm": 18.491985321044922,
       "learning_rate": 2.8585632612478252e-05,
-      "loss": 0.3059,
+      "loss": 0.2317,
       "step": 2170
     },
     {
       "epoch": 4.087695749440716,
-      "grad_norm": 6.391950607299805,
+      "grad_norm": 6.1848602294921875,
       "learning_rate": 2.8461347253293565e-05,
-      "loss": 0.3738,
+      "loss": 0.2758,
       "step": 2180
     },
     {
       "epoch": 4.08993288590604,
-      "grad_norm": 14.067455291748047,
+      "grad_norm": 13.520557403564453,
       "learning_rate": 2.8337061894108874e-05,
-      "loss": 0.4277,
+      "loss": 0.439,
       "step": 2190
     },
     {
       "epoch": 4.092170022371365,
-      "grad_norm": 6.279323101043701,
+      "grad_norm": 9.846938133239746,
       "learning_rate": 2.8212776534924186e-05,
-      "loss": 0.4808,
+      "loss": 0.4711,
       "step": 2200
     },
     {
       "epoch": 4.094407158836689,
-      "grad_norm": 12.135629653930664,
+      "grad_norm": 5.90399694442749,
       "learning_rate": 2.80884911757395e-05,
-      "loss": 0.2029,
+      "loss": 0.085,
       "step": 2210
     },
     {
       "epoch": 4.0966442953020135,
-      "grad_norm": 11.846600532531738,
+      "grad_norm": 16.939096450805664,
       "learning_rate": 2.796420581655481e-05,
-      "loss": 0.4229,
+      "loss": 0.4099,
       "step": 2220
     },
     {
       "epoch": 4.098881431767338,
-      "grad_norm": 3.3600635528564453,
+      "grad_norm": 1.1104991436004639,
       "learning_rate": 2.7839920457370123e-05,
-      "loss": 0.3927,
+      "loss": 0.4496,
       "step": 2230
     },
     {
       "epoch": 4.1,
-      "eval_loss": 0.35884907841682434,
-      "eval_runtime": 935.0139,
-      "eval_samples_per_second": 5.288,
-      "eval_steps_per_second": 0.661,
+      "eval_loss": 0.3794967830181122,
+      "eval_runtime": 891.9924,
+      "eval_samples_per_second": 8.314,
+      "eval_steps_per_second": 1.039,
       "step": 2235
     },
     {
       "epoch": 5.001118568232662,
-      "grad_norm": 4.892012119293213,
+      "grad_norm": 4.833787441253662,
       "learning_rate": 2.7715635098185432e-05,
-      "loss": 0.308,
+      "loss": 0.2992,
       "step": 2240
     },
     {
       "epoch": 5.003355704697986,
-      "grad_norm": 3.3251285552978516,
+      "grad_norm": 5.847362518310547,
       "learning_rate": 2.7591349739000748e-05,
-      "loss": 0.2273,
+      "loss": 0.2596,
       "step": 2250
     },
     {
       "epoch": 5.005592841163311,
-      "grad_norm": 22.025094985961914,
+      "grad_norm": 18.474834442138672,
       "learning_rate": 2.746706437981606e-05,
-      "loss": 0.5861,
+      "loss": 0.4849,
       "step": 2260
     },
     {
       "epoch": 5.007829977628635,
-      "grad_norm": 16.336713790893555,
+      "grad_norm": 12.474526405334473,
       "learning_rate": 2.734277902063137e-05,
-      "loss": 0.4155,
+      "loss": 0.3275,
       "step": 2270
     },
     {
       "epoch": 5.010067114093959,
-      "grad_norm": 1.52274489402771,
+      "grad_norm": 1.1569851636886597,
       "learning_rate": 2.721849366144668e-05,
-      "loss": 0.3442,
+      "loss": 0.3104,
       "step": 2280
     },
     {
       "epoch": 5.012304250559284,
-      "grad_norm": 7.693943977355957,
+      "grad_norm": 10.059696197509766,
       "learning_rate": 2.7094208302261997e-05,
-      "loss": 0.1632,
+      "loss": 0.1573,
       "step": 2290
     },
     {
       "epoch": 5.014541387024608,
-      "grad_norm": 10.348392486572266,
+      "grad_norm": 4.665347099304199,
       "learning_rate": 2.696992294307731e-05,
-      "loss": 0.3964,
+      "loss": 0.4128,
       "step": 2300
     },
     {
       "epoch": 5.016778523489933,
-      "grad_norm": 11.288122177124023,
+      "grad_norm": 13.133956909179688,
       "learning_rate": 2.6845637583892618e-05,
-      "loss": 0.5684,
+      "loss": 0.5156,
       "step": 2310
     },
     {
       "epoch": 5.019015659955257,
-      "grad_norm": 1.0990941524505615,
+      "grad_norm": 1.0956755876541138,
       "learning_rate": 2.672135222470793e-05,
-      "loss": 0.3866,
+      "loss": 0.3676,
       "step": 2320
     },
     {
       "epoch": 5.0212527964205815,
-      "grad_norm": 7.708190441131592,
+      "grad_norm": 8.179460525512695,
       "learning_rate": 2.659706686552324e-05,
-      "loss": 0.3275,
+      "loss": 0.3082,
       "step": 2330
     },
     {
       "epoch": 5.023489932885906,
-      "grad_norm": 0.9472073912620544,
+      "grad_norm": 1.9563068151474,
       "learning_rate": 2.6472781506338555e-05,
-      "loss": 0.1861,
+      "loss": 0.2365,
       "step": 2340
     },
     {
       "epoch": 5.02572706935123,
-      "grad_norm": 13.988951683044434,
+      "grad_norm": 9.728827476501465,
       "learning_rate": 2.6348496147153868e-05,
-      "loss": 0.4737,
+      "loss": 0.6143,
       "step": 2350
     },
     {
       "epoch": 5.027964205816555,
-      "grad_norm": 18.836071014404297,
+      "grad_norm": 0.4499684274196625,
       "learning_rate": 2.6224210787969177e-05,
-      "loss": 0.3949,
+      "loss": 0.1527,
       "step": 2360
     },
     {
       "epoch": 5.030201342281879,
-      "grad_norm": 11.103631973266602,
+      "grad_norm": 4.692073345184326,
       "learning_rate": 2.609992542878449e-05,
-      "loss": 0.3836,
+      "loss": 0.4215,
       "step": 2370
     },
     {
       "epoch": 5.032438478747204,
-      "grad_norm": 13.336675643920898,
+      "grad_norm": 0.4127592146396637,
       "learning_rate": 2.5975640069599805e-05,
-      "loss": 0.349,
+      "loss": 0.3132,
       "step": 2380
     },
     {
       "epoch": 5.034675615212528,
-      "grad_norm": 10.622421264648438,
+      "grad_norm": 5.006928443908691,
       "learning_rate": 2.5851354710415117e-05,
-      "loss": 0.3322,
+      "loss": 0.243,
       "step": 2390
     },
     {
       "epoch": 5.0369127516778525,
-      "grad_norm": 4.1144843101501465,
+      "grad_norm": 9.491246223449707,
       "learning_rate": 2.5727069351230426e-05,
-      "loss": 0.2823,
+      "loss": 0.4203,
       "step": 2400
     },
     {
       "epoch": 5.039149888143177,
-      "grad_norm": 4.660728454589844,
+      "grad_norm": 1.1350034475326538,
       "learning_rate": 2.560278399204574e-05,
-      "loss": 0.2464,
+      "loss": 0.3657,
       "step": 2410
     },
     {
       "epoch": 5.041387024608501,
-      "grad_norm": 7.401895523071289,
+      "grad_norm": 17.347612380981445,
       "learning_rate": 2.5478498632861047e-05,
-      "loss": 0.3123,
+      "loss": 0.4972,
       "step": 2420
     },
     {
       "epoch": 5.043624161073826,
-      "grad_norm": 2.969099283218384,
+      "grad_norm": 2.2265026569366455,
       "learning_rate": 2.5354213273676363e-05,
-      "loss": 0.2394,
+      "loss": 0.3345,
       "step": 2430
     },
     {
       "epoch": 5.04586129753915,
-      "grad_norm": 9.645813941955566,
+      "grad_norm": 4.642486572265625,
       "learning_rate": 2.5229927914491675e-05,
-      "loss": 0.2217,
+      "loss": 0.2684,
       "step": 2440
     },
     {
       "epoch": 5.0480984340044746,
-      "grad_norm": 6.4596734046936035,
+      "grad_norm": 12.171128273010254,
       "learning_rate": 2.5105642555306984e-05,
-      "loss": 0.5332,
+      "loss": 0.4424,
       "step": 2450
     },
     {
       "epoch": 5.050335570469799,
-      "grad_norm": 5.598334312438965,
+      "grad_norm": 11.319775581359863,
       "learning_rate": 2.49813571961223e-05,
-      "loss": 0.3076,
+      "loss": 0.2878,
       "step": 2460
     },
     {
       "epoch": 5.052572706935123,
-      "grad_norm": 8.19486141204834,
+      "grad_norm": 7.691583156585693,
       "learning_rate": 2.485707183693761e-05,
-      "loss": 0.5466,
+      "loss": 0.3732,
       "step": 2470
     },
     {
       "epoch": 5.054809843400448,
-      "grad_norm": 9.856636047363281,
+      "grad_norm": 11.818854331970215,
       "learning_rate": 2.473278647775292e-05,
-      "loss": 0.3923,
+      "loss": 0.3303,
       "step": 2480
     },
     {
       "epoch": 5.057046979865772,
-      "grad_norm": 9.83061695098877,
+      "grad_norm": 10.7247314453125,
       "learning_rate": 2.4608501118568234e-05,
-      "loss": 0.2615,
+      "loss": 0.2182,
       "step": 2490
     },
     {
       "epoch": 5.059284116331096,
-      "grad_norm": 5.30068302154541,
+      "grad_norm": 11.010503768920898,
       "learning_rate": 2.4484215759383546e-05,
-      "loss": 0.4623,
+      "loss": 0.3295,
       "step": 2500
     },
     {
       "epoch": 5.06152125279642,
-      "grad_norm": 11.414236068725586,
+      "grad_norm": 8.999567985534668,
       "learning_rate": 2.435993040019886e-05,
-      "loss": 0.3316,
+      "loss": 0.2326,
       "step": 2510
     },
     {
       "epoch": 5.063758389261745,
-      "grad_norm": 3.8756637573242188,
+      "grad_norm": 7.073877334594727,
       "learning_rate": 2.4235645041014167e-05,
-      "loss": 0.2236,
+      "loss": 0.2705,
       "step": 2520
     },
     {
       "epoch": 5.065995525727069,
-      "grad_norm": 14.053914070129395,
+      "grad_norm": 11.623409271240234,
       "learning_rate": 2.4111359681829483e-05,
-      "loss": 0.3271,
+      "loss": 0.3234,
       "step": 2530
     },
     {
       "epoch": 5.068232662192393,
-      "grad_norm": 10.499263763427734,
+      "grad_norm": 4.587973594665527,
       "learning_rate": 2.3987074322644792e-05,
-      "loss": 0.5217,
+      "loss": 0.2505,
       "step": 2540
     },
     {
       "epoch": 5.070469798657718,
-      "grad_norm": 6.821963787078857,
+      "grad_norm": 11.883222579956055,
       "learning_rate": 2.3862788963460104e-05,
-      "loss": 0.3827,
+      "loss": 0.456,
       "step": 2550
     },
     {
       "epoch": 5.072706935123042,
-      "grad_norm": 7.142666339874268,
+      "grad_norm": 21.06523323059082,
       "learning_rate": 2.3738503604275417e-05,
-      "loss": 0.3265,
+      "loss": 0.2691,
       "step": 2560
     },
     {
       "epoch": 5.074944071588367,
-      "grad_norm": 3.25323486328125,
+      "grad_norm": 12.439352989196777,
       "learning_rate": 2.361421824509073e-05,
-      "loss": 0.2221,
+      "loss": 0.2216,
       "step": 2570
     },
     {
       "epoch": 5.077181208053691,
-      "grad_norm": 2.06369686126709,
+      "grad_norm": 0.9154367446899414,
       "learning_rate": 2.348993288590604e-05,
-      "loss": 0.2334,
+      "loss": 0.3547,
       "step": 2580
     },
     {
       "epoch": 5.0794183445190155,
-      "grad_norm": 4.083550930023193,
+      "grad_norm": 1.807629942893982,
       "learning_rate": 2.3365647526721354e-05,
-      "loss": 0.2098,
+      "loss": 0.2784,
       "step": 2590
     },
     {
       "epoch": 5.08165548098434,
-      "grad_norm": 22.284122467041016,
+      "grad_norm": 14.866148948669434,
       "learning_rate": 2.3241362167536666e-05,
-      "loss": 0.4255,
+      "loss": 0.4041,
       "step": 2600
     },
     {
       "epoch": 5.083892617449664,
-      "grad_norm": 7.050011157989502,
+      "grad_norm": 9.991622924804688,
       "learning_rate": 2.311707680835198e-05,
-      "loss": 0.4338,
+      "loss": 0.3884,
       "step": 2610
     },
     {
       "epoch": 5.086129753914989,
-      "grad_norm": 12.864763259887695,
+      "grad_norm": 11.396065711975098,
       "learning_rate": 2.2992791449167287e-05,
-      "loss": 0.217,
+      "loss": 0.1861,
       "step": 2620
     },
     {
       "epoch": 5.088366890380313,
-      "grad_norm": 8.292052268981934,
+      "grad_norm": 5.028550624847412,
       "learning_rate": 2.28685060899826e-05,
-      "loss": 0.2749,
+      "loss": 0.1773,
       "step": 2630
     },
     {
       "epoch": 5.090604026845638,
-      "grad_norm": 8.610766410827637,
+      "grad_norm": 7.7053022384643555,
       "learning_rate": 2.2744220730797912e-05,
-      "loss": 0.475,
+      "loss": 0.34,
       "step": 2640
     },
     {
       "epoch": 5.092841163310962,
-      "grad_norm": 6.8138933181762695,
+      "grad_norm": 11.335506439208984,
       "learning_rate": 2.2619935371613224e-05,
-      "loss": 0.2715,
+      "loss": 0.2359,
       "step": 2650
     },
     {
       "epoch": 5.0950782997762865,
-      "grad_norm": 13.458662986755371,
+      "grad_norm": 9.261770248413086,
       "learning_rate": 2.2495650012428537e-05,
-      "loss": 0.558,
+      "loss": 0.3856,
       "step": 2660
     },
     {
       "epoch": 5.097315436241611,
-      "grad_norm": 7.119832515716553,
+      "grad_norm": 8.961913108825684,
       "learning_rate": 2.237136465324385e-05,
-      "loss": 0.3312,
+      "loss": 0.4009,
       "step": 2670
     },
     {
       "epoch": 5.099552572706935,
-      "grad_norm": 10.518315315246582,
+      "grad_norm": 3.3359217643737793,
       "learning_rate": 2.224707929405916e-05,
-      "loss": 0.3167,
+      "loss": 0.3433,
       "step": 2680
     },
     {
       "epoch": 5.1,
-      "eval_loss": 0.4430215656757355,
-      "eval_runtime": 921.6901,
-      "eval_samples_per_second": 5.364,
-      "eval_steps_per_second": 0.671,
+      "eval_loss": 0.4118688702583313,
+      "eval_runtime": 892.4454,
+      "eval_samples_per_second": 8.31,
+      "eval_steps_per_second": 1.039,
       "step": 2682
     },
     {
       "epoch": 6.001789709172259,
-      "grad_norm": 8.830635070800781,
+      "grad_norm": 8.580872535705566,
       "learning_rate": 2.2122793934874474e-05,
-      "loss": 0.1378,
+      "loss": 0.1915,
       "step": 2690
     },
     {
       "epoch": 6.004026845637584,
-      "grad_norm": 8.985321044921875,
+      "grad_norm": 0.8801985383033752,
       "learning_rate": 2.1998508575689786e-05,
-      "loss": 0.2403,
+      "loss": 0.182,
       "step": 2700
     },
     {
       "epoch": 6.006263982102908,
-      "grad_norm": 8.969911575317383,
+      "grad_norm": 15.736348152160645,
       "learning_rate": 2.1874223216505095e-05,
-      "loss": 0.3703,
+      "loss": 0.4147,
       "step": 2710
     },
     {
       "epoch": 6.008501118568232,
-      "grad_norm": 13.08025074005127,
+      "grad_norm": 1.222266674041748,
       "learning_rate": 2.174993785732041e-05,
-      "loss": 0.236,
+      "loss": 0.2243,
       "step": 2720
     },
     {
       "epoch": 6.010738255033557,
-      "grad_norm": 10.97948169708252,
+      "grad_norm": 9.234151840209961,
       "learning_rate": 2.162565249813572e-05,
-      "loss": 0.2386,
+      "loss": 0.2065,
       "step": 2730
     },
     {
       "epoch": 6.012975391498881,
-      "grad_norm": 20.18202781677246,
+      "grad_norm": 17.84362030029297,
       "learning_rate": 2.1501367138951032e-05,
-      "loss": 0.3661,
+      "loss": 0.2676,
       "step": 2740
     },
     {
       "epoch": 6.015212527964206,
-      "grad_norm": 9.394488334655762,
+      "grad_norm": 7.234649658203125,
       "learning_rate": 2.1377081779766345e-05,
-      "loss": 0.3235,
+      "loss": 0.3512,
       "step": 2750
     },
     {
       "epoch": 6.01744966442953,
-      "grad_norm": 8.908599853515625,
+      "grad_norm": 5.13038969039917,
       "learning_rate": 2.1252796420581657e-05,
-      "loss": 0.4476,
+      "loss": 0.3029,
       "step": 2760
     },
     {
       "epoch": 6.0196868008948545,
-      "grad_norm": 10.415188789367676,
+      "grad_norm": 12.59158706665039,
       "learning_rate": 2.112851106139697e-05,
-      "loss": 0.3331,
+      "loss": 0.2794,
       "step": 2770
     },
     {
       "epoch": 6.021923937360179,
-      "grad_norm": 10.19694709777832,
+      "grad_norm": 14.324825286865234,
       "learning_rate": 2.1004225702212278e-05,
-      "loss": 0.2918,
+      "loss": 0.2144,
       "step": 2780
     },
     {
       "epoch": 6.024161073825503,
-      "grad_norm": 5.295567512512207,
+      "grad_norm": 6.485671043395996,
       "learning_rate": 2.0879940343027594e-05,
-      "loss": 0.2004,
+      "loss": 0.249,
       "step": 2790
     },
     {
       "epoch": 6.026398210290828,
-      "grad_norm": 3.8771309852600098,
+      "grad_norm": 1.8053562641143799,
       "learning_rate": 2.0755654983842903e-05,
-      "loss": 0.1936,
+      "loss": 0.2008,
       "step": 2800
     },
     {
       "epoch": 6.028635346756152,
-      "grad_norm": 10.931071281433105,
+      "grad_norm": 5.798049449920654,
       "learning_rate": 2.063136962465822e-05,
-      "loss": 0.2634,
+      "loss": 0.1985,
       "step": 2810
     },
     {
       "epoch": 6.030872483221477,
-      "grad_norm": 13.032493591308594,
+      "grad_norm": 11.126876831054688,
       "learning_rate": 2.0507084265473528e-05,
-      "loss": 0.4461,
+      "loss": 0.3294,
       "step": 2820
     },
     {
       "epoch": 6.033109619686801,
-      "grad_norm": 15.591374397277832,
+      "grad_norm": 20.707542419433594,
       "learning_rate": 2.038279890628884e-05,
-      "loss": 0.4873,
+      "loss": 0.4472,
       "step": 2830
     },
     {
       "epoch": 6.0353467561521255,
-      "grad_norm": 8.528480529785156,
+      "grad_norm": 7.778163909912109,
       "learning_rate": 2.0258513547104152e-05,
-      "loss": 0.2138,
+      "loss": 0.2675,
       "step": 2840
     },
     {
       "epoch": 6.03758389261745,
-      "grad_norm": 12.467988014221191,
+      "grad_norm": 8.803659439086914,
       "learning_rate": 2.013422818791946e-05,
-      "loss": 0.4351,
+      "loss": 0.244,
       "step": 2850
     },
     {
       "epoch": 6.039821029082774,
-      "grad_norm": 6.8514018058776855,
+      "grad_norm": 0.29662925004959106,
       "learning_rate": 2.0009942828734777e-05,
-      "loss": 0.2339,
+      "loss": 0.1816,
       "step": 2860
     },
     {
       "epoch": 6.042058165548099,
-      "grad_norm": 1.4219473600387573,
+      "grad_norm": 0.928829550743103,
       "learning_rate": 1.9885657469550086e-05,
-      "loss": 0.2019,
+      "loss": 0.3033,
       "step": 2870
     },
     {
       "epoch": 6.044295302013423,
-      "grad_norm": 1.5489323139190674,
+      "grad_norm": 0.4924483299255371,
       "learning_rate": 1.97613721103654e-05,
-      "loss": 0.3255,
+      "loss": 0.3565,
       "step": 2880
     },
     {
       "epoch": 6.0465324384787476,
-      "grad_norm": 9.738638877868652,
+      "grad_norm": 10.867852210998535,
       "learning_rate": 1.963708675118071e-05,
-      "loss": 0.2453,
+      "loss": 0.1718,
       "step": 2890
     },
     {
       "epoch": 6.048769574944072,
-      "grad_norm": 10.866888999938965,
+      "grad_norm": 0.17046819627285004,
       "learning_rate": 1.9512801391996023e-05,
-      "loss": 0.289,
+      "loss": 0.3193,
       "step": 2900
     },
     {
       "epoch": 6.051006711409396,
-      "grad_norm": 29.56283187866211,
+      "grad_norm": 21.9335994720459,
       "learning_rate": 1.9388516032811335e-05,
-      "loss": 0.3176,
+      "loss": 0.2725,
       "step": 2910
     },
     {
       "epoch": 6.05324384787472,
-      "grad_norm": 9.453397750854492,
+      "grad_norm": 7.8351030349731445,
       "learning_rate": 1.9264230673626648e-05,
-      "loss": 0.4457,
+      "loss": 0.2877,
       "step": 2920
     },
     {
       "epoch": 6.055480984340044,
-      "grad_norm": 1.6036638021469116,
+      "grad_norm": 1.0813746452331543,
       "learning_rate": 1.913994531444196e-05,
-      "loss": 0.2702,
+      "loss": 0.2619,
       "step": 2930
     },
     {
       "epoch": 6.057718120805369,
-      "grad_norm": 11.915915489196777,
+      "grad_norm": 4.7920241355896,
       "learning_rate": 1.9015659955257272e-05,
-      "loss": 0.3377,
+      "loss": 0.182,
       "step": 2940
     },
     {
       "epoch": 6.059955257270693,
-      "grad_norm": 8.017142295837402,
+      "grad_norm": 18.19488525390625,
       "learning_rate": 1.8891374596072585e-05,
-      "loss": 0.4199,
+      "loss": 0.2848,
       "step": 2950
     },
     {
       "epoch": 6.062192393736018,
-      "grad_norm": 3.0255606174468994,
+      "grad_norm": 2.5479190349578857,
       "learning_rate": 1.8767089236887894e-05,
-      "loss": 0.2819,
+      "loss": 0.3099,
       "step": 2960
     },
     {
       "epoch": 6.064429530201342,
-      "grad_norm": 12.458708763122559,
+      "grad_norm": 21.13016128540039,
       "learning_rate": 1.864280387770321e-05,
-      "loss": 0.3564,
+      "loss": 0.4713,
       "step": 2970
     },
     {
       "epoch": 6.066666666666666,
-      "grad_norm": 0.24730181694030762,
+      "grad_norm": 0.09484589099884033,
       "learning_rate": 1.8518518518518518e-05,
-      "loss": 0.2035,
+      "loss": 0.2758,
       "step": 2980
     },
     {
       "epoch": 6.068903803131991,
-      "grad_norm": 15.799874305725098,
+      "grad_norm": 11.212635040283203,
       "learning_rate": 1.839423315933383e-05,
-      "loss": 0.2655,
+      "loss": 0.2305,
       "step": 2990
     },
     {
       "epoch": 6.071140939597315,
-      "grad_norm": 15.106266021728516,
+      "grad_norm": 8.96382999420166,
       "learning_rate": 1.8269947800149143e-05,
-      "loss": 0.2328,
+      "loss": 0.2464,
       "step": 3000
     },
     {
       "epoch": 6.07337807606264,
-      "grad_norm": 9.420243263244629,
+      "grad_norm": 8.752419471740723,
       "learning_rate": 1.8145662440964455e-05,
-      "loss": 0.2431,
+      "loss": 0.1472,
       "step": 3010
     },
     {
       "epoch": 6.075615212527964,
-      "grad_norm": 6.75723123550415,
+      "grad_norm": 9.717801094055176,
       "learning_rate": 1.8021377081779768e-05,
-      "loss": 0.4296,
+      "loss": 0.3221,
       "step": 3020
     },
     {
       "epoch": 6.0778523489932885,
-      "grad_norm": 8.274847984313965,
+      "grad_norm": 5.287158966064453,
       "learning_rate": 1.789709172259508e-05,
-      "loss": 0.4351,
+      "loss": 0.2747,
       "step": 3030
     },
     {
       "epoch": 6.080089485458613,
-      "grad_norm": 4.256190299987793,
+      "grad_norm": 1.6286725997924805,
       "learning_rate": 1.7772806363410392e-05,
-      "loss": 0.0965,
+      "loss": 0.1127,
       "step": 3040
     },
     {
       "epoch": 6.082326621923937,
-      "grad_norm": 1.681492567062378,
+      "grad_norm": 12.31570053100586,
       "learning_rate": 1.7648521004225705e-05,
-      "loss": 0.4374,
+      "loss": 0.2879,
       "step": 3050
     },
     {
       "epoch": 6.084563758389262,
-      "grad_norm": 0.5820801854133606,
+      "grad_norm": 3.0768423080444336,
       "learning_rate": 1.7524235645041014e-05,
-      "loss": 0.3684,
+      "loss": 0.3264,
       "step": 3060
     },
     {
       "epoch": 6.086800894854586,
-      "grad_norm": 5.919391632080078,
+      "grad_norm": 6.542660713195801,
       "learning_rate": 1.7399950285856326e-05,
-      "loss": 0.2455,
+      "loss": 0.2626,
       "step": 3070
     },
     {
       "epoch": 6.089038031319911,
-      "grad_norm": 7.470072269439697,
+      "grad_norm": 22.778274536132812,
       "learning_rate": 1.727566492667164e-05,
-      "loss": 0.2138,
+      "loss": 0.2602,
       "step": 3080
     },
     {
       "epoch": 6.091275167785235,
-      "grad_norm": 5.051515102386475,
+      "grad_norm": 14.418547630310059,
       "learning_rate": 1.715137956748695e-05,
-      "loss": 0.2726,
+      "loss": 0.2574,
       "step": 3090
     },
     {
       "epoch": 6.0935123042505595,
-      "grad_norm": 0.34526365995407104,
+      "grad_norm": 1.627580165863037,
       "learning_rate": 1.7027094208302263e-05,
-      "loss": 0.178,
+      "loss": 0.2347,
       "step": 3100
     },
     {
       "epoch": 6.095749440715884,
-      "grad_norm": 17.414888381958008,
+      "grad_norm": 2.667323350906372,
       "learning_rate": 1.6902808849117575e-05,
-      "loss": 0.1793,
+      "loss": 0.4904,
       "step": 3110
     },
     {
       "epoch": 6.097986577181208,
-      "grad_norm": 23.270225524902344,
+      "grad_norm": 26.79123878479004,
       "learning_rate": 1.6778523489932888e-05,
-      "loss": 0.2943,
+      "loss": 0.2287,
       "step": 3120
     },
     {
       "epoch": 6.1,
-      "eval_loss": 0.4818915128707886,
-      "eval_runtime": 922.9566,
-      "eval_samples_per_second": 5.357,
-      "eval_steps_per_second": 0.67,
+      "eval_loss": 0.4823199510574341,
+      "eval_runtime": 898.1668,
+      "eval_samples_per_second": 8.257,
+      "eval_steps_per_second": 1.032,
       "step": 3129
     },
     {
       "epoch": 7.000223713646532,
-      "grad_norm": 2.266531229019165,
+      "grad_norm": 10.718249320983887,
       "learning_rate": 1.6654238130748197e-05,
-      "loss": 0.2625,
+      "loss": 0.2243,
       "step": 3130
     },
     {
       "epoch": 7.002460850111857,
-      "grad_norm": 24.349164962768555,
+      "grad_norm": 3.2877581119537354,
       "learning_rate": 1.6529952771563512e-05,
-      "loss": 0.28,
+      "loss": 0.3877,
       "step": 3140
     },
     {
       "epoch": 7.004697986577181,
-      "grad_norm": 0.4873502254486084,
+      "grad_norm": 2.2482826709747314,
       "learning_rate": 1.640566741237882e-05,
-      "loss": 0.1496,
+      "loss": 0.1283,
       "step": 3150
     },
     {
       "epoch": 7.006935123042505,
-      "grad_norm": 3.7403814792633057,
+      "grad_norm": 11.641459465026855,
       "learning_rate": 1.6281382053194137e-05,
-      "loss": 0.2493,
+      "loss": 0.2999,
       "step": 3160
     },
     {
       "epoch": 7.00917225950783,
-      "grad_norm": 0.49554741382598877,
+      "grad_norm": 6.360275745391846,
       "learning_rate": 1.6157096694009446e-05,
-      "loss": 0.3466,
+      "loss": 0.2999,
       "step": 3170
     },
     {
       "epoch": 7.011409395973154,
-      "grad_norm": 0.8320086598396301,
+      "grad_norm": 0.33911630511283875,
       "learning_rate": 1.603281133482476e-05,
-      "loss": 0.3601,
+      "loss": 0.3532,
       "step": 3180
     },
     {
       "epoch": 7.013646532438479,
-      "grad_norm": 2.373836040496826,
+      "grad_norm": 6.708487033843994,
       "learning_rate": 1.590852597564007e-05,
-      "loss": 0.4232,
+      "loss": 0.3293,
       "step": 3190
     },
     {
       "epoch": 7.015883668903803,
-      "grad_norm": 13.097466468811035,
+      "grad_norm": 7.4095025062561035,
       "learning_rate": 1.578424061645538e-05,
-      "loss": 0.2663,
+      "loss": 0.1617,
       "step": 3200
     },
     {
       "epoch": 7.0181208053691275,
-      "grad_norm": 2.8690104484558105,
+      "grad_norm": 23.493078231811523,
       "learning_rate": 1.5659955257270695e-05,
-      "loss": 0.1244,
+      "loss": 0.1497,
       "step": 3210
     },
     {
       "epoch": 7.020357941834452,
-      "grad_norm": 10.713217735290527,
+      "grad_norm": 7.098881244659424,
       "learning_rate": 1.5535669898086004e-05,
-      "loss": 0.2673,
+      "loss": 0.2247,
       "step": 3220
     },
     {
       "epoch": 7.022595078299776,
-      "grad_norm": 16.93208885192871,
+      "grad_norm": 8.431655883789062,
       "learning_rate": 1.541138453890132e-05,
-      "loss": 0.3606,
+      "loss": 0.2487,
       "step": 3230
     },
     {
       "epoch": 7.024832214765101,
-      "grad_norm": 10.654176712036133,
+      "grad_norm": 15.323188781738281,
       "learning_rate": 1.528709917971663e-05,
-      "loss": 0.1426,
+      "loss": 0.1421,
       "step": 3240
     },
     {
       "epoch": 7.027069351230425,
-      "grad_norm": 10.325299263000488,
+      "grad_norm": 7.675280570983887,
       "learning_rate": 1.5162813820531943e-05,
-      "loss": 0.4853,
+      "loss": 0.4883,
       "step": 3250
     },
     {
       "epoch": 7.02930648769575,
-      "grad_norm": 0.8497664332389832,
+      "grad_norm": 3.4720630645751953,
       "learning_rate": 1.5038528461347254e-05,
-      "loss": 0.3557,
+      "loss": 0.2715,
       "step": 3260
     },
     {
       "epoch": 7.031543624161074,
-      "grad_norm": 7.555183410644531,
+      "grad_norm": 1.7522681951522827,
       "learning_rate": 1.4914243102162568e-05,
-      "loss": 0.2883,
+      "loss": 0.1637,
       "step": 3270
     },
     {
       "epoch": 7.0337807606263985,
-      "grad_norm": 0.6843427419662476,
+      "grad_norm": 0.16172055900096893,
       "learning_rate": 1.4789957742977878e-05,
-      "loss": 0.1345,
+      "loss": 0.1717,
       "step": 3280
     },
     {
       "epoch": 7.036017897091723,
-      "grad_norm": 0.9340455532073975,
+      "grad_norm": 12.743756294250488,
       "learning_rate": 1.4665672383793189e-05,
-      "loss": 0.1728,
+      "loss": 0.3252,
       "step": 3290
     },
     {
       "epoch": 7.038255033557047,
-      "grad_norm": 12.420462608337402,
+      "grad_norm": 10.548824310302734,
       "learning_rate": 1.4541387024608501e-05,
-      "loss": 0.1637,
+      "loss": 0.0774,
       "step": 3300
     },
     {
       "epoch": 7.040492170022372,
-      "grad_norm": 18.193172454833984,
+      "grad_norm": 11.808302879333496,
       "learning_rate": 1.4417101665423812e-05,
-      "loss": 0.1866,
+      "loss": 0.3318,
       "step": 3310
     },
     {
       "epoch": 7.042729306487696,
-      "grad_norm": 7.021984100341797,
+      "grad_norm": 11.149969100952148,
       "learning_rate": 1.4292816306239126e-05,
-      "loss": 0.3043,
+      "loss": 0.3435,
       "step": 3320
     },
     {
       "epoch": 7.0449664429530205,
-      "grad_norm": 19.211868286132812,
+      "grad_norm": 16.549835205078125,
       "learning_rate": 1.4168530947054437e-05,
-      "loss": 0.1779,
+      "loss": 0.245,
       "step": 3330
     },
     {
       "epoch": 7.047203579418344,
-      "grad_norm": 1.4456816911697388,
+      "grad_norm": 0.44824355840682983,
       "learning_rate": 1.404424558786975e-05,
-      "loss": 0.2717,
+      "loss": 0.2654,
       "step": 3340
     },
     {
       "epoch": 7.0494407158836685,
-      "grad_norm": 5.062099456787109,
+      "grad_norm": 10.627558708190918,
       "learning_rate": 1.3919960228685061e-05,
-      "loss": 0.3043,
+      "loss": 0.2706,
       "step": 3350
     },
     {
       "epoch": 7.051677852348993,
-      "grad_norm": 1.7384883165359497,
+      "grad_norm": 4.539977073669434,
       "learning_rate": 1.3795674869500374e-05,
-      "loss": 0.2957,
+      "loss": 0.2529,
       "step": 3360
     },
     {
       "epoch": 7.053914988814317,
-      "grad_norm": 10.810503959655762,
+      "grad_norm": 25.88998031616211,
       "learning_rate": 1.3671389510315684e-05,
-      "loss": 0.1791,
+      "loss": 0.1102,
       "step": 3370
     },
     {
       "epoch": 7.056152125279642,
-      "grad_norm": 0.12086842954158783,
+      "grad_norm": 0.056022170931100845,
       "learning_rate": 1.3547104151130999e-05,
-      "loss": 0.0836,
+      "loss": 0.0937,
       "step": 3380
     },
     {
       "epoch": 7.058389261744966,
-      "grad_norm": 14.477333068847656,
+      "grad_norm": 27.33989715576172,
       "learning_rate": 1.3422818791946309e-05,
-      "loss": 0.4779,
+      "loss": 0.3445,
       "step": 3390
     },
     {
       "epoch": 7.060626398210291,
-      "grad_norm": 10.850499153137207,
+      "grad_norm": 27.60544204711914,
       "learning_rate": 1.329853343276162e-05,
-      "loss": 0.2134,
+      "loss": 0.3671,
       "step": 3400
     },
     {
       "epoch": 7.062863534675615,
-      "grad_norm": 0.5340798497200012,
+      "grad_norm": 0.10744742304086685,
       "learning_rate": 1.3174248073576934e-05,
-      "loss": 0.0879,
+      "loss": 0.2315,
       "step": 3410
     },
     {
       "epoch": 7.065100671140939,
-      "grad_norm": 20.357879638671875,
+      "grad_norm": 19.817075729370117,
       "learning_rate": 1.3049962714392244e-05,
-      "loss": 0.364,
+      "loss": 0.2397,
       "step": 3420
     },
     {
       "epoch": 7.067337807606264,
-      "grad_norm": 12.8193998336792,
+      "grad_norm": 20.652551651000977,
       "learning_rate": 1.2925677355207559e-05,
-      "loss": 0.1638,
+      "loss": 0.1723,
       "step": 3430
     },
     {
       "epoch": 7.069574944071588,
-      "grad_norm": 32.92828369140625,
+      "grad_norm": 3.9074671268463135,
       "learning_rate": 1.280139199602287e-05,
-      "loss": 0.438,
+      "loss": 0.2729,
       "step": 3440
     },
     {
       "epoch": 7.071812080536913,
-      "grad_norm": 1.9262471199035645,
+      "grad_norm": 0.27141252160072327,
       "learning_rate": 1.2677106636838182e-05,
-      "loss": 0.2231,
+      "loss": 0.3176,
       "step": 3450
     },
     {
       "epoch": 7.074049217002237,
-      "grad_norm": 13.616302490234375,
+      "grad_norm": 11.46578598022461,
       "learning_rate": 1.2552821277653492e-05,
-      "loss": 0.1296,
+      "loss": 0.2252,
       "step": 3460
     },
     {
       "epoch": 7.0762863534675615,
-      "grad_norm": 17.851036071777344,
+      "grad_norm": 16.85219383239746,
       "learning_rate": 1.2428535918468805e-05,
-      "loss": 0.119,
+      "loss": 0.1388,
       "step": 3470
     },
     {
       "epoch": 7.078523489932886,
-      "grad_norm": 9.76507568359375,
+      "grad_norm": 10.73118782043457,
       "learning_rate": 1.2304250559284117e-05,
-      "loss": 0.3196,
+      "loss": 0.2334,
       "step": 3480
     },
     {
       "epoch": 7.08076062639821,
-      "grad_norm": 3.3353357315063477,
+      "grad_norm": 0.460288941860199,
       "learning_rate": 1.217996520009943e-05,
-      "loss": 0.3447,
+      "loss": 0.0608,
       "step": 3490
     },
     {
       "epoch": 7.082997762863535,
-      "grad_norm": 19.23137664794922,
+      "grad_norm": 1.64030921459198,
       "learning_rate": 1.2055679840914742e-05,
-      "loss": 0.2249,
+      "loss": 0.1595,
       "step": 3500
     },
     {
       "epoch": 7.085234899328859,
-      "grad_norm": 19.729598999023438,
+      "grad_norm": 14.426569938659668,
       "learning_rate": 1.1931394481730052e-05,
-      "loss": 0.151,
+      "loss": 0.2432,
       "step": 3510
     },
     {
       "epoch": 7.087472035794184,
-      "grad_norm": 9.253190994262695,
+      "grad_norm": 13.951542854309082,
       "learning_rate": 1.1807109122545365e-05,
-      "loss": 0.2224,
+      "loss": 0.2007,
       "step": 3520
     },
     {
       "epoch": 7.089709172259508,
-      "grad_norm": 0.1582317054271698,
+      "grad_norm": 25.381717681884766,
       "learning_rate": 1.1682823763360677e-05,
-      "loss": 0.2683,
+      "loss": 0.2316,
       "step": 3530
     },
     {
       "epoch": 7.0919463087248324,
-      "grad_norm": 0.43652433156967163,
+      "grad_norm": 0.6340412497520447,
       "learning_rate": 1.155853840417599e-05,
-      "loss": 0.275,
+      "loss": 0.1972,
       "step": 3540
     },
     {
       "epoch": 7.094183445190157,
-      "grad_norm": 0.7390974760055542,
+      "grad_norm": 11.732870101928711,
       "learning_rate": 1.14342530449913e-05,
-      "loss": 0.1894,
+      "loss": 0.1388,
       "step": 3550
     },
     {
       "epoch": 7.096420581655481,
-      "grad_norm": 1.3036680221557617,
+      "grad_norm": 0.08352160453796387,
       "learning_rate": 1.1309967685806612e-05,
-      "loss": 0.2943,
+      "loss": 0.2276,
       "step": 3560
     },
     {
       "epoch": 7.098657718120806,
-      "grad_norm": 9.086507797241211,
+      "grad_norm": 0.6394329071044922,
       "learning_rate": 1.1185682326621925e-05,
-      "loss": 0.2089,
+      "loss": 0.1297,
       "step": 3570
     },
     {
       "epoch": 7.1,
-      "eval_loss": 0.3785615861415863,
-      "eval_runtime": 932.236,
-      "eval_samples_per_second": 5.303,
-      "eval_steps_per_second": 0.663,
+      "eval_loss": 0.4294538199901581,
+      "eval_runtime": 892.5544,
+      "eval_samples_per_second": 8.309,
+      "eval_steps_per_second": 1.039,
       "step": 3576
     },
     {
       "epoch": 8.00089485458613,
-      "grad_norm": 1.078548550605774,
+      "grad_norm": 12.145805358886719,
       "learning_rate": 1.1061396967437237e-05,
-      "loss": 0.2587,
+      "loss": 0.4217,
       "step": 3580
     },
     {
       "epoch": 8.003131991051454,
-      "grad_norm": 1.228070616722107,
+      "grad_norm": 16.92803382873535,
       "learning_rate": 1.0937111608252548e-05,
-      "loss": 0.1517,
+      "loss": 0.1421,
       "step": 3590
     },
     {
       "epoch": 8.00536912751678,
-      "grad_norm": 6.344273090362549,
+      "grad_norm": 6.773504257202148,
       "learning_rate": 1.081282624906786e-05,
-      "loss": 0.2672,
+      "loss": 0.1891,
       "step": 3600
     },
     {
       "epoch": 8.007606263982103,
-      "grad_norm": 10.512372970581055,
+      "grad_norm": 22.504379272460938,
       "learning_rate": 1.0688540889883172e-05,
-      "loss": 0.1911,
+      "loss": 0.1844,
       "step": 3610
     },
     {
       "epoch": 8.009843400447428,
-      "grad_norm": 9.218599319458008,
+      "grad_norm": 19.881702423095703,
       "learning_rate": 1.0564255530698485e-05,
-      "loss": 0.2245,
+      "loss": 0.2223,
       "step": 3620
     },
     {
       "epoch": 8.012080536912752,
-      "grad_norm": 4.7348952293396,
+      "grad_norm": 1.9907094240188599,
       "learning_rate": 1.0439970171513797e-05,
-      "loss": 0.1359,
+      "loss": 0.2041,
       "step": 3630
     },
     {
       "epoch": 8.014317673378075,
-      "grad_norm": 6.268092632293701,
+      "grad_norm": 0.3308212161064148,
       "learning_rate": 1.031568481232911e-05,
-      "loss": 0.1942,
+      "loss": 0.0968,
       "step": 3640
     },
     {
       "epoch": 8.0165548098434,
-      "grad_norm": 9.11316204071045,
+      "grad_norm": 1.7726011276245117,
       "learning_rate": 1.019139945314442e-05,
-      "loss": 0.0928,
+      "loss": 0.0571,
       "step": 3650
     },
     {
       "epoch": 8.018791946308724,
-      "grad_norm": 40.897552490234375,
+      "grad_norm": 29.911670684814453,
       "learning_rate": 1.006711409395973e-05,
-      "loss": 0.2588,
+      "loss": 0.1089,
       "step": 3660
     },
     {
       "epoch": 8.02102908277405,
-      "grad_norm": 0.3429583013057709,
+      "grad_norm": 5.603979587554932,
       "learning_rate": 9.942828734775043e-06,
-      "loss": 0.3107,
+      "loss": 0.0697,
       "step": 3670
     },
     {
       "epoch": 8.023266219239373,
-      "grad_norm": 0.48826009035110474,
+      "grad_norm": 0.07145686447620392,
       "learning_rate": 9.818543375590355e-06,
-      "loss": 0.2078,
+      "loss": 0.0768,
       "step": 3680
     },
     {
       "epoch": 8.025503355704698,
-      "grad_norm": 7.965569019317627,
+      "grad_norm": 1.745406985282898,
       "learning_rate": 9.694258016405668e-06,
-      "loss": 0.1233,
+      "loss": 0.0412,
       "step": 3690
     },
     {
       "epoch": 8.027740492170022,
-      "grad_norm": 0.34283313155174255,
+      "grad_norm": 0.02010565809905529,
       "learning_rate": 9.56997265722098e-06,
-      "loss": 0.1816,
+      "loss": 0.2096,
       "step": 3700
     },
     {
       "epoch": 8.029977628635347,
-      "grad_norm": 5.221133708953857,
+      "grad_norm": 8.72205638885498,
       "learning_rate": 9.445687298036292e-06,
-      "loss": 0.2169,
+      "loss": 0.1555,
       "step": 3710
     },
     {
       "epoch": 8.03221476510067,
-      "grad_norm": 0.05195072665810585,
+      "grad_norm": 0.01665619947016239,
       "learning_rate": 9.321401938851605e-06,
-      "loss": 0.237,
+      "loss": 0.2793,
       "step": 3720
     },
     {
       "epoch": 8.034451901565996,
-      "grad_norm": 0.1627640724182129,
+      "grad_norm": 0.004611628130078316,
       "learning_rate": 9.197116579666915e-06,
-      "loss": 0.2308,
+      "loss": 0.2735,
       "step": 3730
     },
     {
       "epoch": 8.03668903803132,
-      "grad_norm": 0.33864226937294006,
+      "grad_norm": 1.931004524230957,
       "learning_rate": 9.072831220482228e-06,
-      "loss": 0.1982,
+      "loss": 0.2636,
       "step": 3740
     },
     {
       "epoch": 8.038926174496645,
-      "grad_norm": 0.016831032931804657,
+      "grad_norm": 0.014065139926970005,
       "learning_rate": 8.94854586129754e-06,
-      "loss": 0.2632,
+      "loss": 0.2653,
       "step": 3750
     },
     {
       "epoch": 8.041163310961968,
-      "grad_norm": 1.7836061716079712,
+      "grad_norm": 0.5851568579673767,
       "learning_rate": 8.824260502112852e-06,
-      "loss": 0.0641,
+      "loss": 0.092,
       "step": 3760
     },
     {
       "epoch": 8.043400447427294,
-      "grad_norm": 20.223674774169922,
+      "grad_norm": 0.051146071404218674,
       "learning_rate": 8.699975142928163e-06,
-      "loss": 0.1735,
+      "loss": 0.2241,
       "step": 3770
     },
     {
       "epoch": 8.045637583892617,
-      "grad_norm": 16.769073486328125,
+      "grad_norm": 0.1566874086856842,
       "learning_rate": 8.575689783743475e-06,
-      "loss": 0.2878,
+      "loss": 0.1845,
       "step": 3780
     },
     {
       "epoch": 8.047874720357942,
-      "grad_norm": 26.629791259765625,
+      "grad_norm": 35.4766960144043,
       "learning_rate": 8.451404424558788e-06,
-      "loss": 0.1811,
+      "loss": 0.1556,
       "step": 3790
     },
     {
       "epoch": 8.050111856823266,
-      "grad_norm": 35.164878845214844,
+      "grad_norm": 44.98518753051758,
       "learning_rate": 8.327119065374098e-06,
-      "loss": 0.3123,
+      "loss": 0.2961,
       "step": 3800
     },
     {
       "epoch": 8.052348993288591,
-      "grad_norm": 1.7551312446594238,
+      "grad_norm": 1.2694602012634277,
       "learning_rate": 8.20283370618941e-06,
-      "loss": 0.1989,
+      "loss": 0.0785,
       "step": 3810
     },
     {
       "epoch": 8.054586129753915,
-      "grad_norm": 0.1438596099615097,
+      "grad_norm": 0.6759599447250366,
       "learning_rate": 8.078548347004723e-06,
-      "loss": 0.1427,
+      "loss": 0.31,
       "step": 3820
     },
     {
       "epoch": 8.05682326621924,
-      "grad_norm": 4.8719482421875,
+      "grad_norm": 5.972934246063232,
       "learning_rate": 7.954262987820035e-06,
-      "loss": 0.0963,
+      "loss": 0.0335,
       "step": 3830
     },
     {
       "epoch": 8.059060402684564,
-      "grad_norm": 2.296576499938965,
+      "grad_norm": 0.8728901743888855,
       "learning_rate": 7.829977628635348e-06,
-      "loss": 0.1478,
+      "loss": 0.2005,
       "step": 3840
     },
     {
       "epoch": 8.061297539149889,
-      "grad_norm": 7.730291366577148,
+      "grad_norm": 10.340746879577637,
       "learning_rate": 7.70569226945066e-06,
-      "loss": 0.1619,
+      "loss": 0.1724,
       "step": 3850
     },
     {
       "epoch": 8.063534675615212,
-      "grad_norm": 0.0443468801677227,
+      "grad_norm": 0.060672469437122345,
       "learning_rate": 7.5814069102659716e-06,
-      "loss": 0.1471,
+      "loss": 0.174,
       "step": 3860
     },
     {
       "epoch": 8.065771812080538,
-      "grad_norm": 1.623080849647522,
+      "grad_norm": 4.864527702331543,
       "learning_rate": 7.457121551081284e-06,
-      "loss": 0.2739,
+      "loss": 0.3543,
       "step": 3870
     },
     {
       "epoch": 8.068008948545861,
-      "grad_norm": 21.143207550048828,
+      "grad_norm": 4.48012638092041,
       "learning_rate": 7.3328361918965945e-06,
-      "loss": 0.2317,
+      "loss": 0.122,
       "step": 3880
     },
     {
       "epoch": 8.070246085011185,
-      "grad_norm": 0.43577510118484497,
+      "grad_norm": 0.07548043876886368,
       "learning_rate": 7.208550832711906e-06,
-      "loss": 0.3871,
+      "loss": 0.2838,
       "step": 3890
     },
     {
       "epoch": 8.07248322147651,
-      "grad_norm": 27.5056095123291,
+      "grad_norm": 18.938758850097656,
       "learning_rate": 7.084265473527218e-06,
-      "loss": 0.2158,
+      "loss": 0.1681,
       "step": 3900
     },
     {
       "epoch": 8.074720357941834,
-      "grad_norm": 0.9177855253219604,
+      "grad_norm": 1.3047031164169312,
       "learning_rate": 6.959980114342531e-06,
-      "loss": 0.1795,
+      "loss": 0.2001,
       "step": 3910
     },
     {
       "epoch": 8.076957494407159,
-      "grad_norm": 17.42781639099121,
+      "grad_norm": 12.20274829864502,
       "learning_rate": 6.835694755157842e-06,
-      "loss": 0.2471,
+      "loss": 0.2776,
       "step": 3920
     },
     {
       "epoch": 8.079194630872482,
-      "grad_norm": 10.758467674255371,
+      "grad_norm": 0.47563186287879944,
       "learning_rate": 6.7114093959731546e-06,
-      "loss": 0.0654,
+      "loss": 0.2005,
       "step": 3930
     },
     {
       "epoch": 8.081431767337808,
-      "grad_norm": 8.318502426147461,
+      "grad_norm": 5.735065460205078,
       "learning_rate": 6.587124036788467e-06,
-      "loss": 0.3004,
+      "loss": 0.1499,
       "step": 3940
     },
     {
       "epoch": 8.083668903803131,
-      "grad_norm": 11.846883773803711,
+      "grad_norm": 20.469074249267578,
       "learning_rate": 6.462838677603779e-06,
-      "loss": 0.2625,
+      "loss": 0.2258,
       "step": 3950
     },
     {
       "epoch": 8.085906040268457,
-      "grad_norm": 4.139098644256592,
+      "grad_norm": 3.2545881271362305,
       "learning_rate": 6.338553318419091e-06,
-      "loss": 0.1639,
+      "loss": 0.2216,
       "step": 3960
     },
     {
       "epoch": 8.08814317673378,
-      "grad_norm": 0.2406436949968338,
+      "grad_norm": 1.8031901121139526,
       "learning_rate": 6.214267959234402e-06,
-      "loss": 0.1516,
+      "loss": 0.1462,
       "step": 3970
     },
     {
       "epoch": 8.090380313199105,
-      "grad_norm": 5.129803657531738,
+      "grad_norm": 11.510761260986328,
       "learning_rate": 6.089982600049715e-06,
-      "loss": 0.2214,
+      "loss": 0.2225,
       "step": 3980
     },
     {
       "epoch": 8.092617449664429,
-      "grad_norm": 0.08740751445293427,
+      "grad_norm": 1.421615481376648,
       "learning_rate": 5.965697240865026e-06,
-      "loss": 0.2038,
+      "loss": 0.1203,
       "step": 3990
     },
     {
       "epoch": 8.094854586129754,
-      "grad_norm": 0.03226242959499359,
+      "grad_norm": 0.10247212648391724,
       "learning_rate": 5.8414118816803384e-06,
-      "loss": 0.1007,
+      "loss": 0.0633,
       "step": 4000
     },
     {
       "epoch": 8.097091722595078,
-      "grad_norm": 0.23406432569026947,
+      "grad_norm": 0.038117095828056335,
       "learning_rate": 5.71712652249565e-06,
-      "loss": 0.1954,
+      "loss": 0.1246,
       "step": 4010
     },
     {
       "epoch": 8.099328859060403,
-      "grad_norm": 14.286626815795898,
+      "grad_norm": 14.49740982055664,
       "learning_rate": 5.592841163310962e-06,
-      "loss": 0.2659,
+      "loss": 0.3104,
       "step": 4020
     },
     {
       "epoch": 8.1,
-      "eval_loss": 0.37856465578079224,
-      "eval_runtime": 924.4583,
-      "eval_samples_per_second": 5.348,
-      "eval_steps_per_second": 0.668,
+      "eval_loss": 0.40958455204963684,
+      "eval_runtime": 894.7196,
+      "eval_samples_per_second": 8.289,
+      "eval_steps_per_second": 1.036,
       "step": 4023
     },
     {
       "epoch": 9.001565995525727,
-      "grad_norm": 2.8623125553131104,
+      "grad_norm": 0.1057172641158104,
       "learning_rate": 5.468555804126274e-06,
-      "loss": 0.027,
+      "loss": 0.0831,
       "step": 4030
     },
     {
       "epoch": 9.003803131991052,
-      "grad_norm": 0.19636014103889465,
+      "grad_norm": 1.2070332765579224,
       "learning_rate": 5.344270444941586e-06,
-      "loss": 0.1193,
+      "loss": 0.0479,
       "step": 4040
     },
     {
       "epoch": 9.006040268456376,
-      "grad_norm": 0.22568732500076294,
+      "grad_norm": 0.6095995903015137,
       "learning_rate": 5.2199850857568985e-06,
-      "loss": 0.0185,
+      "loss": 0.2161,
       "step": 4050
     },
     {
       "epoch": 9.0082774049217,
-      "grad_norm": 16.95732307434082,
+      "grad_norm": 6.29830265045166,
       "learning_rate": 5.09569972657221e-06,
-      "loss": 0.1922,
+      "loss": 0.1455,
       "step": 4060
     },
     {
       "epoch": 9.010514541387025,
-      "grad_norm": 0.051336098462343216,
+      "grad_norm": 0.01811257004737854,
       "learning_rate": 4.9714143673875215e-06,
-      "loss": 0.1818,
+      "loss": 0.0465,
       "step": 4070
     },
     {
       "epoch": 9.012751677852348,
-      "grad_norm": 15.473747253417969,
+      "grad_norm": 17.725488662719727,
       "learning_rate": 4.847129008202834e-06,
-      "loss": 0.2227,
+      "loss": 0.3159,
       "step": 4080
     },
     {
       "epoch": 9.014988814317674,
-      "grad_norm": 1.2553575038909912,
+      "grad_norm": 27.672937393188477,
       "learning_rate": 4.722843649018146e-06,
-      "loss": 0.0386,
+      "loss": 0.0538,
       "step": 4090
     },
     {
       "epoch": 9.017225950782997,
-      "grad_norm": 15.090453147888184,
+      "grad_norm": 0.12884105741977692,
       "learning_rate": 4.598558289833458e-06,
-      "loss": 0.1785,
+      "loss": 0.2565,
       "step": 4100
     },
     {
       "epoch": 9.019463087248322,
-      "grad_norm": 3.2336063385009766,
+      "grad_norm": 20.479238510131836,
       "learning_rate": 4.47427293064877e-06,
-      "loss": 0.0169,
+      "loss": 0.1162,
       "step": 4110
     },
     {
       "epoch": 9.021700223713646,
-      "grad_norm": 0.01566535234451294,
+      "grad_norm": 23.51799774169922,
       "learning_rate": 4.3499875714640815e-06,
-      "loss": 0.1087,
+      "loss": 0.181,
       "step": 4120
     },
     {
       "epoch": 9.023937360178971,
-      "grad_norm": 40.838531494140625,
+      "grad_norm": 14.024354934692383,
       "learning_rate": 4.225702212279394e-06,
-      "loss": 0.1926,
+      "loss": 0.2026,
       "step": 4130
     },
     {
       "epoch": 9.026174496644295,
-      "grad_norm": 0.03625281900167465,
+      "grad_norm": 0.20324963331222534,
       "learning_rate": 4.101416853094705e-06,
-      "loss": 0.1113,
+      "loss": 0.0238,
       "step": 4140
     },
     {
       "epoch": 9.02841163310962,
-      "grad_norm": 0.037700217217206955,
+      "grad_norm": 0.17176759243011475,
       "learning_rate": 3.977131493910018e-06,
-      "loss": 0.0486,
+      "loss": 0.036,
       "step": 4150
     },
     {
       "epoch": 9.030648769574944,
-      "grad_norm": 28.63728141784668,
+      "grad_norm": 13.05589485168457,
       "learning_rate": 3.85284613472533e-06,
-      "loss": 0.2895,
+      "loss": 0.3049,
       "step": 4160
     },
     {
       "epoch": 9.032885906040269,
-      "grad_norm": 0.01208668015897274,
+      "grad_norm": 0.03032175451517105,
       "learning_rate": 3.728560775540642e-06,
-      "loss": 0.0776,
+      "loss": 0.043,
       "step": 4170
     },
     {
       "epoch": 9.035123042505592,
-      "grad_norm": 0.07349768280982971,
+      "grad_norm": 2.0046262741088867,
       "learning_rate": 3.604275416355953e-06,
-      "loss": 0.2092,
+      "loss": 0.2365,
       "step": 4180
     },
     {
       "epoch": 9.037360178970918,
-      "grad_norm": 0.4977450966835022,
+      "grad_norm": 0.10407610237598419,
       "learning_rate": 3.4799900571712654e-06,
-      "loss": 0.1928,
+      "loss": 0.0472,
       "step": 4190
     },
     {
       "epoch": 9.039597315436241,
-      "grad_norm": 0.0466020293533802,
+      "grad_norm": 0.7708104252815247,
       "learning_rate": 3.3557046979865773e-06,
-      "loss": 0.1642,
+      "loss": 0.2119,
       "step": 4200
     },
     {
       "epoch": 9.041834451901567,
-      "grad_norm": 0.30247071385383606,
+      "grad_norm": 0.13973136246204376,
       "learning_rate": 3.2314193388018896e-06,
-      "loss": 0.2029,
+      "loss": 0.1322,
       "step": 4210
     },
     {
       "epoch": 9.04407158836689,
-      "grad_norm": 13.724775314331055,
+      "grad_norm": 4.554214000701904,
       "learning_rate": 3.107133979617201e-06,
-      "loss": 0.1069,
+      "loss": 0.0044,
       "step": 4220
     },
     {
       "epoch": 9.046308724832215,
-      "grad_norm": 0.40144094824790955,
+      "grad_norm": 1.251420259475708,
       "learning_rate": 2.982848620432513e-06,
-      "loss": 0.1657,
+      "loss": 0.1559,
       "step": 4230
     },
     {
       "epoch": 9.048545861297539,
-      "grad_norm": 0.06872935593128204,
+      "grad_norm": 0.027974896132946014,
       "learning_rate": 2.858563261247825e-06,
-      "loss": 0.1699,
+      "loss": 0.1958,
       "step": 4240
     },
     {
       "epoch": 9.050782997762864,
-      "grad_norm": 10.61763858795166,
+      "grad_norm": 17.233055114746094,
       "learning_rate": 2.734277902063137e-06,
-      "loss": 0.1787,
+      "loss": 0.1606,
       "step": 4250
     },
     {
       "epoch": 9.053020134228188,
-      "grad_norm": 34.91901779174805,
+      "grad_norm": 0.4418050944805145,
       "learning_rate": 2.6099925428784492e-06,
-      "loss": 0.2871,
+      "loss": 0.19,
       "step": 4260
     },
     {
       "epoch": 9.055257270693513,
-      "grad_norm": 14.127365112304688,
+      "grad_norm": 7.108306884765625,
       "learning_rate": 2.4857071836937607e-06,
-      "loss": 0.2077,
+      "loss": 0.1778,
       "step": 4270
     },
     {
       "epoch": 9.057494407158837,
-      "grad_norm": 0.02785838022828102,
+      "grad_norm": 0.11051614582538605,
       "learning_rate": 2.361421824509073e-06,
-      "loss": 0.1807,
+      "loss": 0.1074,
       "step": 4280
     },
     {
       "epoch": 9.059731543624162,
-      "grad_norm": 0.04847564175724983,
+      "grad_norm": 0.011775967665016651,
       "learning_rate": 2.237136465324385e-06,
-      "loss": 0.0376,
+      "loss": 0.1294,
       "step": 4290
     },
     {
       "epoch": 9.061968680089485,
-      "grad_norm": 0.029158098623156548,
+      "grad_norm": 0.011663487181067467,
       "learning_rate": 2.112851106139697e-06,
-      "loss": 0.1871,
+      "loss": 0.0744,
       "step": 4300
     },
     {
       "epoch": 9.06420581655481,
-      "grad_norm": 23.394874572753906,
+      "grad_norm": 23.934040069580078,
       "learning_rate": 1.988565746955009e-06,
-      "loss": 0.1058,
+      "loss": 0.1064,
       "step": 4310
     },
     {
       "epoch": 9.066442953020134,
-      "grad_norm": 0.8079015016555786,
+      "grad_norm": 0.42981475591659546,
       "learning_rate": 1.864280387770321e-06,
-      "loss": 0.1675,
+      "loss": 0.0546,
       "step": 4320
     },
     {
       "epoch": 9.068680089485458,
-      "grad_norm": 24.60673713684082,
+      "grad_norm": 19.337921142578125,
       "learning_rate": 1.7399950285856327e-06,
-      "loss": 0.2744,
+      "loss": 0.2816,
       "step": 4330
     },
     {
       "epoch": 9.070917225950783,
-      "grad_norm": 0.17073792219161987,
+      "grad_norm": 0.4612930119037628,
       "learning_rate": 1.6157096694009448e-06,
-      "loss": 0.2612,
+      "loss": 0.136,
       "step": 4340
     },
     {
       "epoch": 9.073154362416107,
-      "grad_norm": 0.29150497913360596,
+      "grad_norm": 0.09723786264657974,
       "learning_rate": 1.4914243102162565e-06,
-      "loss": 0.2493,
+      "loss": 0.1669,
       "step": 4350
     },
     {
       "epoch": 9.075391498881432,
-      "grad_norm": 0.5105171203613281,
+      "grad_norm": 0.028177455067634583,
       "learning_rate": 1.3671389510315684e-06,
-      "loss": 0.2066,
+      "loss": 0.0808,
       "step": 4360
     },
     {
       "epoch": 9.077628635346755,
-      "grad_norm": 15.902856826782227,
+      "grad_norm": 13.670249938964844,
       "learning_rate": 1.2428535918468804e-06,
-      "loss": 0.0912,
+      "loss": 0.1512,
       "step": 4370
     },
     {
       "epoch": 9.07986577181208,
-      "grad_norm": 0.061424434185028076,
+      "grad_norm": 0.34152135252952576,
       "learning_rate": 1.1185682326621925e-06,
-      "loss": 0.1446,
+      "loss": 0.1886,
       "step": 4380
     },
     {
       "epoch": 9.082102908277404,
-      "grad_norm": 0.5502781271934509,
+      "grad_norm": 0.24342969059944153,
       "learning_rate": 9.942828734775044e-07,
-      "loss": 0.1979,
+      "loss": 0.2693,
       "step": 4390
     },
     {
       "epoch": 9.08434004474273,
-      "grad_norm": 0.8110822439193726,
+      "grad_norm": 0.17166166007518768,
       "learning_rate": 8.699975142928163e-07,
-      "loss": 0.0716,
+      "loss": 0.1354,
       "step": 4400
     },
     {
       "epoch": 9.086577181208053,
-      "grad_norm": 0.02159685641527176,
+      "grad_norm": 0.027880514040589333,
       "learning_rate": 7.457121551081283e-07,
-      "loss": 0.1669,
+      "loss": 0.1474,
       "step": 4410
     },
     {
       "epoch": 9.088814317673378,
-      "grad_norm": 0.6268171668052673,
+      "grad_norm": 0.05772515758872032,
       "learning_rate": 6.214267959234402e-07,
-      "loss": 0.0994,
+      "loss": 0.0538,
       "step": 4420
     },
     {
       "epoch": 9.091051454138702,
-      "grad_norm": 0.645751416683197,
+      "grad_norm": 0.07866553962230682,
       "learning_rate": 4.971414367387522e-07,
-      "loss": 0.1157,
+      "loss": 0.1434,
       "step": 4430
     },
     {
       "epoch": 9.093288590604027,
-      "grad_norm": 22.347421646118164,
+      "grad_norm": 16.45406723022461,
       "learning_rate": 3.7285607755406413e-07,
-      "loss": 0.1085,
+      "loss": 0.1083,
       "step": 4440
     },
     {
       "epoch": 9.09552572706935,
-      "grad_norm": 0.024207374081015587,
+      "grad_norm": 0.050580114126205444,
       "learning_rate": 2.485707183693761e-07,
-      "loss": 0.2458,
+      "loss": 0.3213,
       "step": 4450
     },
     {
       "epoch": 9.097762863534676,
-      "grad_norm": 28.59840965270996,
+      "grad_norm": 0.4725286364555359,
       "learning_rate": 1.2428535918468805e-07,
-      "loss": 0.1111,
+      "loss": 0.1468,
       "step": 4460
     },
     {
       "epoch": 9.1,
-      "grad_norm": 0.028079470619559288,
+      "grad_norm": 0.048940546810626984,
       "learning_rate": 0.0,
-      "loss": 0.0856,
+      "loss": 0.0525,
       "step": 4470
     },
     {
       "epoch": 9.1,
-      "eval_loss": 0.3906996250152588,
-      "eval_runtime": 934.7366,
-      "eval_samples_per_second": 5.289,
-      "eval_steps_per_second": 0.661,
+      "eval_loss": 0.41542962193489075,
+      "eval_runtime": 893.9438,
+      "eval_samples_per_second": 8.296,
+      "eval_steps_per_second": 1.037,
       "step": 4470
     },
     {
       "epoch": 9.1,
       "step": 4470,
-      "total_flos": 2.3553465155251864e+20,
-      "train_loss": 0.4063819212768168,
-      "train_runtime": 34011.8222,
-      "train_samples_per_second": 1.051,
-      "train_steps_per_second": 0.131
+      "total_flos": 1.5702310103501242e+20,
+      "train_loss": 0.38359242084309025,
+      "train_runtime": 28356.0667,
+      "train_samples_per_second": 1.261,
+      "train_steps_per_second": 0.158
     },
     {
       "epoch": 9.1,
-      "eval_loss": 0.39387446641921997,
-      "eval_runtime": 485.4183,
-      "eval_samples_per_second": 2.266,
-      "eval_steps_per_second": 0.284,
+      "eval_loss": 0.39315128326416016,
+      "eval_runtime": 423.7958,
+      "eval_samples_per_second": 2.596,
+      "eval_steps_per_second": 0.326,
       "step": 4470
     },
     {
       "epoch": 9.1,
-      "eval_loss": 0.3938744366168976,
-      "eval_runtime": 485.0329,
-      "eval_samples_per_second": 2.268,
-      "eval_steps_per_second": 0.285,
+      "eval_loss": 0.39315125346183777,
+      "eval_runtime": 426.8234,
+      "eval_samples_per_second": 2.577,
+      "eval_steps_per_second": 0.323,
       "step": 4470
     }
   ],
@@ -3260,7 +3260,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 2.3553465155251864e+20,
+  "total_flos": 1.5702310103501242e+20,
   "train_batch_size": 8,
   "trial_name": null,
   "trial_params": null