diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,5009 +1,66 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 0.5000436033836225,
+  "epoch": 0.005057992500218017,
   "eval_steps": 100,
-  "global_step": 5734,
+  "global_step": 58,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
       "epoch": 0.0008720676724513823,
-      "grad_norm": 11.64645767211914,
-      "learning_rate": 3.4843205574912896e-07,
-      "loss": 3.0969,
-      "mean_token_accuracy": 0.4389921709895134,
+      "grad_norm": 8.504547119140625,
+      "learning_rate": 1.9709418174260523e-05,
+      "loss": 3.0672,
+      "mean_token_accuracy": 0.4417563572525978,
       "step": 10
     },
     {
       "epoch": 0.0017441353449027645,
-      "grad_norm": 10.649463653564453,
-      "learning_rate": 6.968641114982579e-07,
-      "loss": 3.1172,
-      "mean_token_accuracy": 0.42953766882419586,
+      "grad_norm": 7.896213531494141,
+      "learning_rate": 1.6631226582407954e-05,
+      "loss": 2.9109,
+      "mean_token_accuracy": 0.44279598593711855,
       "step": 20
     },
     {
       "epoch": 0.002616203017354147,
-      "grad_norm": 22.668081283569336,
-      "learning_rate": 1.045296167247387e-06,
-      "loss": 3.1367,
-      "mean_token_accuracy": 0.4332436382770538,
+      "grad_norm": 5.716647148132324,
+      "learning_rate": 1.1205366802553231e-05,
+      "loss": 2.8492,
+      "mean_token_accuracy": 0.4517245590686798,
       "step": 30
     },
     {
       "epoch": 0.003488270689805529,
-      "grad_norm": 9.847553253173828,
-      "learning_rate": 1.3937282229965158e-06,
-      "loss": 3.0719,
-      "mean_token_accuracy": 0.43627690672874453,
+      "grad_norm": 11.646272659301758,
+      "learning_rate": 5.352768279562315e-06,
+      "loss": 2.7641,
+      "mean_token_accuracy": 0.45660469830036166,
       "step": 40
     },
     {
       "epoch": 0.004360338362256911,
-      "grad_norm": 12.409385681152344,
-      "learning_rate": 1.742160278745645e-06,
-      "loss": 3.0055,
-      "mean_token_accuracy": 0.44554794728755953,
+      "grad_norm": 85.00694274902344,
+      "learning_rate": 1.1454397434679022e-06,
+      "loss": 2.6703,
+      "mean_token_accuracy": 0.46954501122236253,
       "step": 50
     },
     {
-      "epoch": 0.005232406034708294,
-      "grad_norm": 10.909235954284668,
-      "learning_rate": 2.090592334494774e-06,
-      "loss": 3.0664,
-      "mean_token_accuracy": 0.4381849318742752,
-      "step": 60
-    },
-    {
-      "epoch": 0.0061044737071596755,
-      "grad_norm": 8.22943115234375,
-      "learning_rate": 2.4390243902439027e-06,
-      "loss": 3.0672,
-      "mean_token_accuracy": 0.4359099805355072,
-      "step": 70
-    },
-    {
-      "epoch": 0.006976541379611058,
-      "grad_norm": 7.6157546043396,
-      "learning_rate": 2.7874564459930316e-06,
-      "loss": 3.0586,
-      "mean_token_accuracy": 0.43613013476133344,
-      "step": 80
-    },
-    {
-      "epoch": 0.00784860905206244,
-      "grad_norm": 23.02543830871582,
-      "learning_rate": 3.13588850174216e-06,
-      "loss": 3.0211,
-      "mean_token_accuracy": 0.4358121335506439,
-      "step": 90
-    },
-    {
-      "epoch": 0.008720676724513822,
-      "grad_norm": 9.992512702941895,
-      "learning_rate": 3.48432055749129e-06,
-      "loss": 2.9414,
-      "mean_token_accuracy": 0.4512353241443634,
-      "step": 100
-    },
-    {
-      "epoch": 0.008720676724513822,
-      "eval_runtime": 4.1423,
-      "eval_samples_per_second": 264.827,
-      "eval_steps_per_second": 66.388,
-      "step": 100
-    },
-    {
-      "epoch": 0.009592744396965204,
-      "grad_norm": 15.101585388183594,
-      "learning_rate": 3.832752613240418e-06,
-      "loss": 2.8867,
-      "mean_token_accuracy": 0.4551247552037239,
-      "step": 110
-    },
-    {
-      "epoch": 0.010464812069416587,
-      "grad_norm": 10.346649169921875,
-      "learning_rate": 4.181184668989548e-06,
-      "loss": 2.8953,
-      "mean_token_accuracy": 0.45013454258441926,
-      "step": 120
-    },
-    {
-      "epoch": 0.01133687974186797,
-      "grad_norm": 10.099631309509277,
-      "learning_rate": 4.529616724738676e-06,
-      "loss": 2.8516,
-      "mean_token_accuracy": 0.4521404102444649,
-      "step": 130
-    },
-    {
-      "epoch": 0.012208947414319351,
-      "grad_norm": 5.679756164550781,
-      "learning_rate": 4.8780487804878055e-06,
-      "loss": 2.7828,
-      "mean_token_accuracy": 0.46083659380674363,
-      "step": 140
-    },
-    {
-      "epoch": 0.013081015086770733,
-      "grad_norm": 5.871412754058838,
-      "learning_rate": 5.226480836236935e-06,
-      "loss": 2.7367,
-      "mean_token_accuracy": 0.4641144797205925,
-      "step": 150
-    },
-    {
-      "epoch": 0.013953082759222116,
-      "grad_norm": 6.499264240264893,
-      "learning_rate": 5.574912891986063e-06,
-      "loss": 2.7617,
-      "mean_token_accuracy": 0.46029843389987946,
-      "step": 160
-    },
-    {
-      "epoch": 0.014825150431673498,
-      "grad_norm": 5.921131610870361,
-      "learning_rate": 5.923344947735193e-06,
-      "loss": 2.768,
-      "mean_token_accuracy": 0.46050635874271395,
-      "step": 170
-    },
-    {
-      "epoch": 0.01569721810412488,
-      "grad_norm": 6.9524078369140625,
-      "learning_rate": 6.27177700348432e-06,
-      "loss": 2.7039,
-      "mean_token_accuracy": 0.46823630183935167,
-      "step": 180
-    },
-    {
-      "epoch": 0.01656928577657626,
-      "grad_norm": 5.634174823760986,
-      "learning_rate": 6.62020905923345e-06,
-      "loss": 2.7242,
-      "mean_token_accuracy": 0.4601394355297089,
-      "step": 190
-    },
-    {
-      "epoch": 0.017441353449027645,
-      "grad_norm": 6.105836868286133,
-      "learning_rate": 6.96864111498258e-06,
-      "loss": 2.6477,
-      "mean_token_accuracy": 0.47765411138534547,
-      "step": 200
-    },
-    {
-      "epoch": 0.017441353449027645,
-      "eval_runtime": 4.1235,
-      "eval_samples_per_second": 266.035,
-      "eval_steps_per_second": 66.691,
-      "step": 200
-    },
-    {
-      "epoch": 0.018313421121479028,
-      "grad_norm": 31.381994247436523,
-      "learning_rate": 7.317073170731707e-06,
-      "loss": 2.6984,
-      "mean_token_accuracy": 0.464762718975544,
-      "step": 210
-    },
-    {
-      "epoch": 0.019185488793930408,
-      "grad_norm": 4.033312797546387,
-      "learning_rate": 7.665505226480837e-06,
-      "loss": 2.6125,
-      "mean_token_accuracy": 0.48017367869615557,
-      "step": 220
-    },
-    {
-      "epoch": 0.02005755646638179,
-      "grad_norm": 6.638462066650391,
-      "learning_rate": 8.013937282229966e-06,
-      "loss": 2.6297,
-      "mean_token_accuracy": 0.4705357149243355,
-      "step": 230
-    },
-    {
-      "epoch": 0.020929624138833175,
-      "grad_norm": 9.26970386505127,
-      "learning_rate": 8.362369337979095e-06,
-      "loss": 2.6555,
-      "mean_token_accuracy": 0.4658512696623802,
-      "step": 240
-    },
-    {
-      "epoch": 0.021801691811284555,
-      "grad_norm": 4.880399703979492,
-      "learning_rate": 8.710801393728223e-06,
-      "loss": 2.5578,
-      "mean_token_accuracy": 0.4821795463562012,
-      "step": 250
-    },
-    {
-      "epoch": 0.02267375948373594,
-      "grad_norm": 9.49838924407959,
-      "learning_rate": 9.059233449477352e-06,
-      "loss": 2.607,
-      "mean_token_accuracy": 0.47185665667057036,
-      "step": 260
-    },
-    {
-      "epoch": 0.023545827156187322,
-      "grad_norm": 5.966403484344482,
-      "learning_rate": 9.407665505226482e-06,
-      "loss": 2.525,
-      "mean_token_accuracy": 0.48724314719438555,
-      "step": 270
-    },
-    {
-      "epoch": 0.024417894828638702,
-      "grad_norm": 3.227234363555908,
-      "learning_rate": 9.756097560975611e-06,
-      "loss": 2.5758,
-      "mean_token_accuracy": 0.4741927579045296,
-      "step": 280
-    },
-    {
-      "epoch": 0.025289962501090085,
-      "grad_norm": 5.978829860687256,
-      "learning_rate": 1.0104529616724739e-05,
-      "loss": 2.5461,
-      "mean_token_accuracy": 0.47919520139694216,
-      "step": 290
-    },
-    {
-      "epoch": 0.026162030173541465,
-      "grad_norm": 2.7352025508880615,
-      "learning_rate": 1.045296167247387e-05,
-      "loss": 2.5477,
-      "mean_token_accuracy": 0.4759173199534416,
-      "step": 300
-    },
-    {
-      "epoch": 0.026162030173541465,
-      "eval_runtime": 4.1203,
-      "eval_samples_per_second": 266.242,
-      "eval_steps_per_second": 66.743,
-      "step": 300
-    },
-    {
-      "epoch": 0.02703409784599285,
-      "grad_norm": 3.9053685665130615,
-      "learning_rate": 1.0801393728222997e-05,
-      "loss": 2.5406,
-      "mean_token_accuracy": 0.47798434644937515,
-      "step": 310
-    },
-    {
-      "epoch": 0.027906165518444232,
-      "grad_norm": 2.710395574569702,
-      "learning_rate": 1.1149825783972127e-05,
-      "loss": 2.532,
-      "mean_token_accuracy": 0.4715631142258644,
-      "step": 320
-    },
-    {
-      "epoch": 0.028778233190895612,
-      "grad_norm": 2.643364667892456,
-      "learning_rate": 1.1498257839721256e-05,
-      "loss": 2.4617,
-      "mean_token_accuracy": 0.48740215301513673,
-      "step": 330
-    },
-    {
-      "epoch": 0.029650300863346996,
-      "grad_norm": 2.8194141387939453,
-      "learning_rate": 1.1846689895470385e-05,
-      "loss": 2.4797,
-      "mean_token_accuracy": 0.48741438388824465,
-      "step": 340
-    },
-    {
-      "epoch": 0.03052236853579838,
-      "grad_norm": 2.6089999675750732,
-      "learning_rate": 1.2195121951219513e-05,
-      "loss": 2.4398,
-      "mean_token_accuracy": 0.490472112596035,
-      "step": 350
-    },
-    {
-      "epoch": 0.03139443620824976,
-      "grad_norm": 2.952528476715088,
-      "learning_rate": 1.254355400696864e-05,
-      "loss": 2.4477,
-      "mean_token_accuracy": 0.4910836607217789,
-      "step": 360
-    },
-    {
-      "epoch": 0.03226650388070114,
-      "grad_norm": 3.01657772064209,
-      "learning_rate": 1.2891986062717772e-05,
-      "loss": 2.4352,
-      "mean_token_accuracy": 0.4877568528056145,
-      "step": 370
-    },
-    {
-      "epoch": 0.03313857155315252,
-      "grad_norm": 3.2113053798675537,
-      "learning_rate": 1.32404181184669e-05,
-      "loss": 2.4141,
-      "mean_token_accuracy": 0.4919153586030006,
-      "step": 380
-    },
-    {
-      "epoch": 0.034010639225603906,
-      "grad_norm": 2.7614409923553467,
-      "learning_rate": 1.3588850174216028e-05,
-      "loss": 2.4031,
-      "mean_token_accuracy": 0.4976761281490326,
-      "step": 390
-    },
-    {
-      "epoch": 0.03488270689805529,
-      "grad_norm": 2.456890106201172,
-      "learning_rate": 1.393728222996516e-05,
-      "loss": 2.3363,
-      "mean_token_accuracy": 0.5056873753666877,
-      "step": 400
-    },
-    {
-      "epoch": 0.03488270689805529,
-      "eval_runtime": 4.115,
-      "eval_samples_per_second": 266.587,
-      "eval_steps_per_second": 66.829,
-      "step": 400
-    },
-    {
-      "epoch": 0.03575477457050667,
-      "grad_norm": 2.797640323638916,
-      "learning_rate": 1.4285714285714287e-05,
-      "loss": 2.3937,
-      "mean_token_accuracy": 0.4946917802095413,
-      "step": 410
-    },
-    {
-      "epoch": 0.036626842242958056,
-      "grad_norm": 2.794724702835083,
-      "learning_rate": 1.4634146341463415e-05,
-      "loss": 2.3359,
-      "mean_token_accuracy": 0.5056506857275963,
-      "step": 420
-    },
-    {
-      "epoch": 0.03749890991540943,
-      "grad_norm": 2.551732063293457,
-      "learning_rate": 1.4982578397212544e-05,
-      "loss": 2.3539,
-      "mean_token_accuracy": 0.4986179068684578,
-      "step": 430
-    },
-    {
-      "epoch": 0.038370977587860816,
-      "grad_norm": 3.088500738143921,
-      "learning_rate": 1.5331010452961673e-05,
-      "loss": 2.3289,
-      "mean_token_accuracy": 0.5039505869150162,
-      "step": 440
-    },
-    {
-      "epoch": 0.0392430452603122,
-      "grad_norm": 2.715803384780884,
-      "learning_rate": 1.5679442508710803e-05,
-      "loss": 2.2902,
-      "mean_token_accuracy": 0.5098336532711982,
-      "step": 450
-    },
-    {
-      "epoch": 0.04011511293276358,
-      "grad_norm": 2.976975679397583,
-      "learning_rate": 1.6027874564459932e-05,
-      "loss": 2.307,
-      "mean_token_accuracy": 0.5035102799534797,
-      "step": 460
-    },
-    {
-      "epoch": 0.040987180605214966,
-      "grad_norm": 2.3217976093292236,
-      "learning_rate": 1.637630662020906e-05,
-      "loss": 2.3312,
-      "mean_token_accuracy": 0.5040851220488548,
-      "step": 470
-    },
-    {
-      "epoch": 0.04185924827766635,
-      "grad_norm": 2.297640800476074,
-      "learning_rate": 1.672473867595819e-05,
-      "loss": 2.2992,
-      "mean_token_accuracy": 0.5107387527823448,
-      "step": 480
-    },
-    {
-      "epoch": 0.042731315950117726,
-      "grad_norm": 2.3546853065490723,
-      "learning_rate": 1.7073170731707317e-05,
-      "loss": 2.2715,
-      "mean_token_accuracy": 0.5090997964143753,
-      "step": 490
-    },
-    {
-      "epoch": 0.04360338362256911,
-      "grad_norm": 2.614044666290283,
-      "learning_rate": 1.7421602787456446e-05,
-      "loss": 2.3234,
-      "mean_token_accuracy": 0.5036692798137665,
-      "step": 500
-    },
-    {
-      "epoch": 0.04360338362256911,
-      "eval_runtime": 4.1214,
-      "eval_samples_per_second": 266.175,
-      "eval_steps_per_second": 66.726,
-      "step": 500
-    },
-    {
-      "epoch": 0.04447545129502049,
-      "grad_norm": 2.760845184326172,
-      "learning_rate": 1.7770034843205575e-05,
-      "loss": 2.3121,
-      "mean_token_accuracy": 0.5020547926425933,
-      "step": 510
-    },
-    {
-      "epoch": 0.04534751896747188,
-      "grad_norm": 2.4619712829589844,
-      "learning_rate": 1.8118466898954705e-05,
-      "loss": 2.293,
-      "mean_token_accuracy": 0.5061643838882446,
-      "step": 520
-    },
-    {
-      "epoch": 0.04621958663992326,
-      "grad_norm": 2.5899531841278076,
-      "learning_rate": 1.8466898954703834e-05,
-      "loss": 2.318,
-      "mean_token_accuracy": 0.5036325842142105,
-      "step": 530
-    },
-    {
-      "epoch": 0.047091654312374644,
-      "grad_norm": 2.472874641418457,
-      "learning_rate": 1.8815331010452963e-05,
-      "loss": 2.2969,
-      "mean_token_accuracy": 0.5079133987426758,
-      "step": 540
-    },
-    {
-      "epoch": 0.04796372198482602,
-      "grad_norm": 2.673957347869873,
-      "learning_rate": 1.9163763066202093e-05,
-      "loss": 2.2375,
-      "mean_token_accuracy": 0.5135273903608322,
-      "step": 550
-    },
-    {
-      "epoch": 0.048835789657277404,
-      "grad_norm": 2.67699933052063,
-      "learning_rate": 1.9512195121951222e-05,
-      "loss": 2.2824,
-      "mean_token_accuracy": 0.5077177122235298,
-      "step": 560
-    },
-    {
-      "epoch": 0.04970785732972879,
-      "grad_norm": 2.802086353302002,
-      "learning_rate": 1.9860627177700348e-05,
-      "loss": 2.2695,
-      "mean_token_accuracy": 0.510286208987236,
-      "step": 570
-    },
-    {
-      "epoch": 0.05057992500218017,
-      "grad_norm": 2.619718074798584,
-      "learning_rate": 1.9999933277491715e-05,
-      "loss": 2.2477,
-      "mean_token_accuracy": 0.5157045006752015,
-      "step": 580
-    },
-    {
-      "epoch": 0.051451992674631554,
-      "grad_norm": 3.177259683609009,
-      "learning_rate": 1.999952553205438e-05,
-      "loss": 2.2375,
-      "mean_token_accuracy": 0.5156311169266701,
-      "step": 590
-    },
-    {
-      "epoch": 0.05232406034708293,
-      "grad_norm": 2.4455647468566895,
-      "learning_rate": 1.999874712433585e-05,
-      "loss": 2.202,
-      "mean_token_accuracy": 0.5251589983701705,
-      "step": 600
-    },
-    {
-      "epoch": 0.05232406034708293,
-      "eval_runtime": 4.1201,
-      "eval_samples_per_second": 266.253,
-      "eval_steps_per_second": 66.745,
-      "step": 600
-    },
-    {
-      "epoch": 0.053196128019534314,
-      "grad_norm": 2.240347146987915,
-      "learning_rate": 1.999759808319013e-05,
-      "loss": 2.2035,
-      "mean_token_accuracy": 0.5226027399301529,
-      "step": 610
-    },
-    {
-      "epoch": 0.0540681956919857,
-      "grad_norm": 2.3793680667877197,
-      "learning_rate": 1.9996078451209863e-05,
-      "loss": 2.2457,
-      "mean_token_accuracy": 0.5142612457275391,
-      "step": 620
-    },
-    {
-      "epoch": 0.05494026336443708,
-      "grad_norm": 2.522087812423706,
-      "learning_rate": 1.999418828472475e-05,
-      "loss": 2.2344,
-      "mean_token_accuracy": 0.5140533238649369,
-      "step": 630
-    },
-    {
-      "epoch": 0.055812331036888464,
-      "grad_norm": 2.748728036880493,
-      "learning_rate": 1.9991927653799458e-05,
-      "loss": 2.1723,
-      "mean_token_accuracy": 0.5270914882421494,
-      "step": 640
-    },
-    {
-      "epoch": 0.05668439870933985,
-      "grad_norm": 2.198444366455078,
-      "learning_rate": 1.998929664223102e-05,
-      "loss": 2.2137,
-      "mean_token_accuracy": 0.5156433433294296,
-      "step": 650
-    },
-    {
-      "epoch": 0.057556466381791224,
-      "grad_norm": 2.4703140258789062,
-      "learning_rate": 1.9986295347545738e-05,
-      "loss": 2.1867,
-      "mean_token_accuracy": 0.5237402141094207,
-      "step": 660
-    },
-    {
-      "epoch": 0.05842853405424261,
-      "grad_norm": 2.36777663230896,
-      "learning_rate": 1.998292388099557e-05,
-      "loss": 2.2687,
-      "mean_token_accuracy": 0.510102741420269,
-      "step": 670
-    },
-    {
-      "epoch": 0.05930060172669399,
-      "grad_norm": 2.156102180480957,
-      "learning_rate": 1.9979182367553994e-05,
-      "loss": 2.1902,
-      "mean_token_accuracy": 0.5236912921071053,
-      "step": 680
-    },
-    {
-      "epoch": 0.060172669399145375,
-      "grad_norm": 2.489835023880005,
-      "learning_rate": 1.997507094591137e-05,
-      "loss": 2.1945,
-      "mean_token_accuracy": 0.518860076367855,
-      "step": 690
-    },
-    {
-      "epoch": 0.06104473707159676,
-      "grad_norm": 2.5350422859191895,
-      "learning_rate": 1.9970589768469833e-05,
-      "loss": 2.2203,
-      "mean_token_accuracy": 0.5191046923398972,
-      "step": 700
-    },
-    {
-      "epoch": 0.06104473707159676,
-      "eval_runtime": 4.12,
-      "eval_samples_per_second": 266.264,
-      "eval_steps_per_second": 66.748,
-      "step": 700
-    },
-    {
-      "epoch": 0.06191680474404814,
-      "grad_norm": 2.3168063163757324,
-      "learning_rate": 1.996573900133761e-05,
-      "loss": 2.1754,
-      "mean_token_accuracy": 0.5202054798603057,
-      "step": 710
-    },
-    {
-      "epoch": 0.06278887241649952,
-      "grad_norm": 2.149829864501953,
-      "learning_rate": 1.996051882432286e-05,
-      "loss": 2.1664,
-      "mean_token_accuracy": 0.5268835633993149,
-      "step": 720
-    },
-    {
-      "epoch": 0.0636609400889509,
-      "grad_norm": 2.2810912132263184,
-      "learning_rate": 1.995492943092705e-05,
-      "loss": 2.1453,
-      "mean_token_accuracy": 0.525146770477295,
-      "step": 730
-    },
-    {
-      "epoch": 0.06453300776140228,
-      "grad_norm": 2.395927667617798,
-      "learning_rate": 1.9948971028337737e-05,
-      "loss": 2.1797,
-      "mean_token_accuracy": 0.5216242611408234,
-      "step": 740
-    },
-    {
-      "epoch": 0.06540507543385367,
-      "grad_norm": 2.418855905532837,
-      "learning_rate": 1.9942643837420904e-05,
-      "loss": 2.1797,
-      "mean_token_accuracy": 0.5233855187892914,
-      "step": 750
-    },
-    {
-      "epoch": 0.06627714310630504,
-      "grad_norm": 2.637463092803955,
-      "learning_rate": 1.9935948092712792e-05,
-      "loss": 2.1301,
-      "mean_token_accuracy": 0.5318126186728478,
-      "step": 760
-    },
-    {
-      "epoch": 0.06714921077875644,
-      "grad_norm": 2.405441999435425,
-      "learning_rate": 1.992888404241117e-05,
-      "loss": 2.2141,
-      "mean_token_accuracy": 0.5175269037485123,
-      "step": 770
-    },
-    {
-      "epoch": 0.06802127845120781,
-      "grad_norm": 2.151622772216797,
-      "learning_rate": 1.992145194836616e-05,
-      "loss": 2.1691,
-      "mean_token_accuracy": 0.5256971642374992,
-      "step": 780
-    },
-    {
-      "epoch": 0.0688933461236592,
-      "grad_norm": 2.4048643112182617,
-      "learning_rate": 1.9913652086070535e-05,
-      "loss": 2.1648,
-      "mean_token_accuracy": 0.5227862015366554,
-      "step": 790
-    },
-    {
-      "epoch": 0.06976541379611058,
-      "grad_norm": 2.324415683746338,
-      "learning_rate": 1.9905484744649484e-05,
-      "loss": 2.2094,
-      "mean_token_accuracy": 0.5160714268684388,
-      "step": 800
-    },
-    {
-      "epoch": 0.06976541379611058,
-      "eval_runtime": 4.1129,
-      "eval_samples_per_second": 266.722,
-      "eval_steps_per_second": 66.863,
-      "step": 800
-    },
-    {
-      "epoch": 0.07063748146856196,
-      "grad_norm": 2.617950916290283,
-      "learning_rate": 1.989695022684991e-05,
-      "loss": 2.1469,
-      "mean_token_accuracy": 0.5242906078696251,
-      "step": 810
-    },
-    {
-      "epoch": 0.07150954914101335,
-      "grad_norm": 2.1124157905578613,
-      "learning_rate": 1.988804884902921e-05,
-      "loss": 2.1293,
-      "mean_token_accuracy": 0.5324119389057159,
-      "step": 820
-    },
-    {
-      "epoch": 0.07238161681346472,
-      "grad_norm": 2.2512662410736084,
-      "learning_rate": 1.9878780941143538e-05,
-      "loss": 2.1434,
-      "mean_token_accuracy": 0.5242416709661484,
-      "step": 830
-    },
-    {
-      "epoch": 0.07325368448591611,
-      "grad_norm": 2.2564539909362793,
-      "learning_rate": 1.9869146846735576e-05,
-      "loss": 2.116,
-      "mean_token_accuracy": 0.531274464726448,
-      "step": 840
-    },
-    {
-      "epoch": 0.07412575215836749,
-      "grad_norm": 2.0179219245910645,
-      "learning_rate": 1.985914692292182e-05,
-      "loss": 2.1605,
-      "mean_token_accuracy": 0.5246086090803146,
-      "step": 850
-    },
-    {
-      "epoch": 0.07499781983081887,
-      "grad_norm": 1.9747101068496704,
-      "learning_rate": 1.9848781540379312e-05,
-      "loss": 2.1113,
-      "mean_token_accuracy": 0.5351516604423523,
-      "step": 860
-    },
-    {
-      "epoch": 0.07586988750327026,
-      "grad_norm": 2.293004274368286,
-      "learning_rate": 1.983805108333191e-05,
-      "loss": 2.1645,
-      "mean_token_accuracy": 0.5257705494761467,
-      "step": 870
-    },
-    {
-      "epoch": 0.07674195517572163,
-      "grad_norm": 2.3080289363861084,
-      "learning_rate": 1.9826955949536062e-05,
-      "loss": 2.134,
-      "mean_token_accuracy": 0.527433954179287,
-      "step": 880
-    },
-    {
-      "epoch": 0.07761402284817302,
-      "grad_norm": 2.154252767562866,
-      "learning_rate": 1.9815496550266036e-05,
-      "loss": 2.1457,
-      "mean_token_accuracy": 0.5286692768335343,
-      "step": 890
-    },
-    {
-      "epoch": 0.0784860905206244,
-      "grad_norm": 2.330449104309082,
-      "learning_rate": 1.98036733102987e-05,
-      "loss": 2.1754,
-      "mean_token_accuracy": 0.5224702134728432,
-      "step": 900
-    },
-    {
-      "epoch": 0.0784860905206244,
-      "eval_runtime": 4.1175,
-      "eval_samples_per_second": 266.425,
-      "eval_steps_per_second": 66.788,
-      "step": 900
-    },
-    {
-      "epoch": 0.07935815819307579,
-      "grad_norm": 2.0749928951263428,
-      "learning_rate": 1.979148666789775e-05,
-      "loss": 2.1332,
-      "mean_token_accuracy": 0.5318615481257438,
-      "step": 910
-    },
-    {
-      "epoch": 0.08023022586552717,
-      "grad_norm": 1.9350244998931885,
-      "learning_rate": 1.9778937074797494e-05,
-      "loss": 2.1383,
-      "mean_token_accuracy": 0.5262130945920944,
-      "step": 920
-    },
-    {
-      "epoch": 0.08110229353797854,
-      "grad_norm": 1.9555654525756836,
-      "learning_rate": 1.976602499618608e-05,
-      "loss": 2.1172,
-      "mean_token_accuracy": 0.5292074427008628,
-      "step": 930
-    },
-    {
-      "epoch": 0.08197436121042993,
-      "grad_norm": 2.3143930435180664,
-      "learning_rate": 1.9752750910688278e-05,
-      "loss": 2.1285,
-      "mean_token_accuracy": 0.5343566566705704,
-      "step": 940
-    },
-    {
-      "epoch": 0.08284642888288131,
-      "grad_norm": 2.2497706413269043,
-      "learning_rate": 1.9739115310347698e-05,
-      "loss": 2.0887,
-      "mean_token_accuracy": 0.5352372780442238,
-      "step": 950
-    },
-    {
-      "epoch": 0.0837184965553327,
-      "grad_norm": 2.122169256210327,
-      "learning_rate": 1.972511870060861e-05,
-      "loss": 2.1551,
-      "mean_token_accuracy": 0.5229574412107467,
-      "step": 960
-    },
-    {
-      "epoch": 0.08459056422778408,
-      "grad_norm": 2.097722291946411,
-      "learning_rate": 1.9710761600297147e-05,
-      "loss": 2.1336,
-      "mean_token_accuracy": 0.5301736801862716,
-      "step": 970
-    },
-    {
-      "epoch": 0.08546263190023545,
-      "grad_norm": 2.023879051208496,
-      "learning_rate": 1.9696044541602126e-05,
-      "loss": 2.1648,
-      "mean_token_accuracy": 0.5197284772992135,
-      "step": 980
-    },
-    {
-      "epoch": 0.08633469957268684,
-      "grad_norm": 2.1220815181732178,
-      "learning_rate": 1.968096807005528e-05,
-      "loss": 2.1156,
-      "mean_token_accuracy": 0.5341242715716362,
-      "step": 990
-    },
-    {
-      "epoch": 0.08720676724513822,
-      "grad_norm": 1.9110069274902344,
-      "learning_rate": 1.966553274451106e-05,
-      "loss": 2.0816,
-      "mean_token_accuracy": 0.5412548944354058,
-      "step": 1000
-    },
-    {
-      "epoch": 0.08720676724513822,
-      "eval_runtime": 4.125,
-      "eval_samples_per_second": 265.942,
-      "eval_steps_per_second": 66.667,
-      "step": 1000
-    },
-    {
-      "epoch": 0.08807883491758961,
-      "grad_norm": 1.966088891029358,
-      "learning_rate": 1.964973913712591e-05,
-      "loss": 2.1227,
-      "mean_token_accuracy": 0.5298556745052337,
-      "step": 1010
-    },
-    {
-      "epoch": 0.08895090259004099,
-      "grad_norm": 2.2384274005889893,
-      "learning_rate": 1.9633587833337064e-05,
-      "loss": 2.0805,
-      "mean_token_accuracy": 0.5338796466588974,
-      "step": 1020
-    },
-    {
-      "epoch": 0.08982297026249236,
-      "grad_norm": 2.347111225128174,
-      "learning_rate": 1.961707943184083e-05,
-      "loss": 2.118,
-      "mean_token_accuracy": 0.5256604731082917,
-      "step": 1030
-    },
-    {
-      "epoch": 0.09069503793494375,
-      "grad_norm": 2.2390129566192627,
-      "learning_rate": 1.9600214544570432e-05,
-      "loss": 2.0746,
-      "mean_token_accuracy": 0.5383072376251221,
-      "step": 1040
-    },
-    {
-      "epoch": 0.09156710560739513,
-      "grad_norm": 2.306363582611084,
-      "learning_rate": 1.958299379667328e-05,
-      "loss": 2.1098,
-      "mean_token_accuracy": 0.5258683949708939,
-      "step": 1050
-    },
-    {
-      "epoch": 0.09243917327984652,
-      "grad_norm": 1.8623790740966797,
-      "learning_rate": 1.9565417826487835e-05,
-      "loss": 2.0324,
-      "mean_token_accuracy": 0.549840995669365,
-      "step": 1060
-    },
-    {
-      "epoch": 0.0933112409522979,
-      "grad_norm": 2.4428200721740723,
-      "learning_rate": 1.9547487285519922e-05,
-      "loss": 2.1336,
-      "mean_token_accuracy": 0.5295382633805275,
-      "step": 1070
-    },
-    {
-      "epoch": 0.09418330862474929,
-      "grad_norm": 2.0666661262512207,
-      "learning_rate": 1.952920283841861e-05,
-      "loss": 2.1043,
-      "mean_token_accuracy": 0.5329134076833725,
-      "step": 1080
-    },
-    {
-      "epoch": 0.09505537629720066,
-      "grad_norm": 2.1164302825927734,
-      "learning_rate": 1.9510565162951538e-05,
-      "loss": 2.0852,
-      "mean_token_accuracy": 0.532901169359684,
-      "step": 1090
-    },
-    {
-      "epoch": 0.09592744396965204,
-      "grad_norm": 2.1305086612701416,
-      "learning_rate": 1.9491574949979814e-05,
-      "loss": 2.1059,
-      "mean_token_accuracy": 0.5345156580209732,
-      "step": 1100
-    },
-    {
-      "epoch": 0.09592744396965204,
-      "eval_runtime": 4.12,
-      "eval_samples_per_second": 266.26,
-      "eval_steps_per_second": 66.747,
-      "step": 1100
-    },
-    {
-      "epoch": 0.09679951164210343,
-      "grad_norm": 2.1415188312530518,
-      "learning_rate": 1.9472232903432406e-05,
-      "loss": 2.0434,
-      "mean_token_accuracy": 0.5476027473807334,
-      "step": 1110
-    },
-    {
-      "epoch": 0.09767157931455481,
-      "grad_norm": 1.944799542427063,
-      "learning_rate": 1.945253974028004e-05,
-      "loss": 2.0906,
-      "mean_token_accuracy": 0.5327910959720612,
-      "step": 1120
-    },
-    {
-      "epoch": 0.0985436469870062,
-      "grad_norm": 1.9710620641708374,
-      "learning_rate": 1.9432496190508633e-05,
-      "loss": 2.0488,
-      "mean_token_accuracy": 0.540215265750885,
-      "step": 1130
-    },
-    {
-      "epoch": 0.09941571465945757,
-      "grad_norm": 2.0646567344665527,
-      "learning_rate": 1.941210299709222e-05,
-      "loss": 2.0895,
-      "mean_token_accuracy": 0.5337940290570259,
-      "step": 1140
-    },
-    {
-      "epoch": 0.10028778233190895,
-      "grad_norm": 1.8850884437561035,
-      "learning_rate": 1.9391360915965426e-05,
-      "loss": 2.0781,
-      "mean_token_accuracy": 0.5401418745517731,
-      "step": 1150
-    },
-    {
-      "epoch": 0.10115985000436034,
-      "grad_norm": 2.4646894931793213,
-      "learning_rate": 1.9370270715995447e-05,
-      "loss": 2.118,
-      "mean_token_accuracy": 0.5300391420722008,
-      "step": 1160
-    },
-    {
-      "epoch": 0.10203191767681172,
-      "grad_norm": 2.2924323081970215,
-      "learning_rate": 1.934883317895354e-05,
-      "loss": 2.0551,
-      "mean_token_accuracy": 0.5437500029802322,
-      "step": 1170
-    },
-    {
-      "epoch": 0.10290398534926311,
-      "grad_norm": 2.0995631217956543,
-      "learning_rate": 1.932704909948604e-05,
-      "loss": 2.0906,
-      "mean_token_accuracy": 0.5312133073806763,
-      "step": 1180
-    },
-    {
-      "epoch": 0.10377605302171448,
-      "grad_norm": 1.8588402271270752,
-      "learning_rate": 1.930491928508492e-05,
-      "loss": 2.0148,
-      "mean_token_accuracy": 0.5477862060070038,
-      "step": 1190
-    },
-    {
-      "epoch": 0.10464812069416586,
-      "grad_norm": 1.8173161745071411,
-      "learning_rate": 1.9282444556057855e-05,
-      "loss": 2.0582,
-      "mean_token_accuracy": 0.5377079248428345,
-      "step": 1200
-    },
-    {
-      "epoch": 0.10464812069416586,
-      "eval_runtime": 4.1197,
-      "eval_samples_per_second": 266.281,
-      "eval_steps_per_second": 66.752,
-      "step": 1200
-    },
-    {
-      "epoch": 0.10552018836661725,
-      "grad_norm": 2.1144661903381348,
-      "learning_rate": 1.9259625745497803e-05,
-      "loss": 2.0172,
-      "mean_token_accuracy": 0.5475660502910614,
-      "step": 1210
-    },
-    {
-      "epoch": 0.10639225603906863,
-      "grad_norm": 1.9685770273208618,
-      "learning_rate": 1.9236463699252136e-05,
-      "loss": 2.0625,
-      "mean_token_accuracy": 0.5404843434691429,
-      "step": 1220
-    },
-    {
-      "epoch": 0.10726432371152002,
-      "grad_norm": 2.020535945892334,
-      "learning_rate": 1.921295927589127e-05,
-      "loss": 2.018,
-      "mean_token_accuracy": 0.5431017577648163,
-      "step": 1230
-    },
-    {
-      "epoch": 0.1081363913839714,
-      "grad_norm": 2.0250167846679688,
-      "learning_rate": 1.9189113346676878e-05,
-      "loss": 2.0535,
-      "mean_token_accuracy": 0.5410102665424347,
-      "step": 1240
-    },
-    {
-      "epoch": 0.10900845905642279,
-      "grad_norm": 1.9987328052520752,
-      "learning_rate": 1.916492679552954e-05,
-      "loss": 2.0273,
-      "mean_token_accuracy": 0.544092458486557,
-      "step": 1250
-    },
-    {
-      "epoch": 0.10988052672887416,
-      "grad_norm": 2.2367770671844482,
-      "learning_rate": 1.914040051899602e-05,
-      "loss": 2.016,
-      "mean_token_accuracy": 0.5526541113853455,
-      "step": 1260
-    },
-    {
-      "epoch": 0.11075259440132554,
-      "grad_norm": 1.9024274349212646,
-      "learning_rate": 1.9115535426216018e-05,
-      "loss": 2.0187,
-      "mean_token_accuracy": 0.5499999985098839,
-      "step": 1270
-    },
-    {
-      "epoch": 0.11162466207377693,
-      "grad_norm": 1.8777433633804321,
-      "learning_rate": 1.9090332438888458e-05,
-      "loss": 2.0316,
-      "mean_token_accuracy": 0.5445817053318024,
-      "step": 1280
-    },
-    {
-      "epoch": 0.1124967297462283,
-      "grad_norm": 2.614100694656372,
-      "learning_rate": 1.906479249123735e-05,
-      "loss": 2.0172,
-      "mean_token_accuracy": 0.5464285716414452,
-      "step": 1290
-    },
-    {
-      "epoch": 0.1133687974186797,
-      "grad_norm": 1.866713523864746,
-      "learning_rate": 1.9038916529977136e-05,
-      "loss": 2.0664,
-      "mean_token_accuracy": 0.5405577331781387,
-      "step": 1300
-    },
-    {
-      "epoch": 0.1133687974186797,
-      "eval_runtime": 4.1214,
-      "eval_samples_per_second": 266.173,
-      "eval_steps_per_second": 66.725,
-      "step": 1300
-    },
-    {
-      "epoch": 0.11424086509113107,
-      "grad_norm": 1.9683363437652588,
-      "learning_rate": 1.901270551427761e-05,
-      "loss": 2.0328,
-      "mean_token_accuracy": 0.5418052852153779,
-      "step": 1310
-    },
-    {
-      "epoch": 0.11511293276358245,
-      "grad_norm": 1.9454294443130493,
-      "learning_rate": 1.898616041572836e-05,
-      "loss": 2.073,
-      "mean_token_accuracy": 0.5360200613737106,
-      "step": 1320
-    },
-    {
-      "epoch": 0.11598500043603384,
-      "grad_norm": 1.7908854484558105,
-      "learning_rate": 1.8959282218302746e-05,
-      "loss": 1.9727,
-      "mean_token_accuracy": 0.5602617397904396,
-      "step": 1330
-    },
-    {
-      "epoch": 0.11685706810848522,
-      "grad_norm": 1.8223388195037842,
-      "learning_rate": 1.893207191832144e-05,
-      "loss": 2.0109,
-      "mean_token_accuracy": 0.5469178110361099,
-      "step": 1340
-    },
-    {
-      "epoch": 0.1177291357809366,
-      "grad_norm": 1.9757128953933716,
-      "learning_rate": 1.8904530524415483e-05,
-      "loss": 2.0832,
-      "mean_token_accuracy": 0.5333047896623612,
-      "step": 1350
-    },
-    {
-      "epoch": 0.11860120345338798,
-      "grad_norm": 1.9593379497528076,
-      "learning_rate": 1.8876659057488905e-05,
-      "loss": 2.0066,
-      "mean_token_accuracy": 0.5489236772060394,
-      "step": 1360
-    },
-    {
-      "epoch": 0.11947327112583936,
-      "grad_norm": 2.1702773571014404,
-      "learning_rate": 1.8848458550680875e-05,
-      "loss": 2.0063,
-      "mean_token_accuracy": 0.5501223087310791,
-      "step": 1370
-    },
-    {
-      "epoch": 0.12034533879829075,
-      "grad_norm": 1.9332603216171265,
-      "learning_rate": 1.8819930049327412e-05,
-      "loss": 2.0082,
-      "mean_token_accuracy": 0.550574854016304,
-      "step": 1380
-    },
-    {
-      "epoch": 0.12121740647074213,
-      "grad_norm": 1.9806733131408691,
-      "learning_rate": 1.8791074610922624e-05,
-      "loss": 2.0191,
-      "mean_token_accuracy": 0.5514677107334137,
-      "step": 1390
-    },
-    {
-      "epoch": 0.12208947414319352,
-      "grad_norm": 1.9311041831970215,
-      "learning_rate": 1.8761893305079528e-05,
-      "loss": 2.0676,
-      "mean_token_accuracy": 0.534821429848671,
-      "step": 1400
-    },
-    {
-      "epoch": 0.12208947414319352,
-      "eval_runtime": 4.1161,
-      "eval_samples_per_second": 266.518,
-      "eval_steps_per_second": 66.812,
-      "step": 1400
-    },
-    {
-      "epoch": 0.12296154181564489,
-      "grad_norm": 1.9246333837509155,
-      "learning_rate": 1.873238721349038e-05,
-      "loss": 2.0227,
-      "mean_token_accuracy": 0.5452421754598618,
-      "step": 1410
-    },
-    {
-      "epoch": 0.12383360948809628,
-      "grad_norm": 2.2859151363372803,
-      "learning_rate": 1.8702557429886607e-05,
-      "loss": 2.0387,
-      "mean_token_accuracy": 0.5422089040279389,
-      "step": 1420
-    },
-    {
-      "epoch": 0.12470567716054766,
-      "grad_norm": 2.0238237380981445,
-      "learning_rate": 1.8672405059998228e-05,
-      "loss": 1.9758,
-      "mean_token_accuracy": 0.555418285727501,
-      "step": 1430
-    },
-    {
-      "epoch": 0.12557774483299905,
-      "grad_norm": 2.0210776329040527,
-      "learning_rate": 1.8641931221512895e-05,
-      "loss": 2.0535,
-      "mean_token_accuracy": 0.5390533328056335,
-      "step": 1440
-    },
-    {
-      "epoch": 0.12644981250545043,
-      "grad_norm": 2.147141456604004,
-      "learning_rate": 1.8611137044034454e-05,
-      "loss": 2.0227,
-      "mean_token_accuracy": 0.5485078305006027,
-      "step": 1450
-    },
-    {
-      "epoch": 0.1273218801779018,
-      "grad_norm": 1.865098476409912,
-      "learning_rate": 1.858002366904107e-05,
-      "loss": 2.0152,
-      "mean_token_accuracy": 0.5483977526426316,
-      "step": 1460
-    },
-    {
-      "epoch": 0.12819394785035318,
-      "grad_norm": 1.7718040943145752,
-      "learning_rate": 1.854859224984292e-05,
-      "loss": 2.0539,
-      "mean_token_accuracy": 0.5422211334109306,
-      "step": 1470
-    },
-    {
-      "epoch": 0.12906601552280456,
-      "grad_norm": 1.7911052703857422,
-      "learning_rate": 1.851684395153944e-05,
-      "loss": 2.0078,
-      "mean_token_accuracy": 0.5465141892433166,
-      "step": 1480
-    },
-    {
-      "epoch": 0.12993808319525596,
-      "grad_norm": 2.0077836513519287,
-      "learning_rate": 1.8484779950976133e-05,
-      "loss": 2.0879,
-      "mean_token_accuracy": 0.5327910944819451,
-      "step": 1490
-    },
-    {
-      "epoch": 0.13081015086770734,
-      "grad_norm": 1.8197457790374756,
-      "learning_rate": 1.8452401436700954e-05,
-      "loss": 2.0551,
-      "mean_token_accuracy": 0.540288645029068,
-      "step": 1500
-    },
-    {
-      "epoch": 0.13081015086770734,
-      "eval_runtime": 4.1094,
-      "eval_samples_per_second": 266.947,
-      "eval_steps_per_second": 66.919,
-      "step": 1500
-    },
-    {
-      "epoch": 0.1316822185401587,
-      "grad_norm": 1.793343424797058,
-      "learning_rate": 1.8419709608920243e-05,
-      "loss": 2.0637,
-      "mean_token_accuracy": 0.5390900135040283,
-      "step": 1510
-    },
-    {
-      "epoch": 0.1325542862126101,
-      "grad_norm": 1.9090853929519653,
-      "learning_rate": 1.8386705679454243e-05,
-      "loss": 1.9789,
-      "mean_token_accuracy": 0.5530944287776947,
-      "step": 1520
-    },
-    {
-      "epoch": 0.1334263538850615,
-      "grad_norm": 1.8935576677322388,
-      "learning_rate": 1.8353390871692176e-05,
-      "loss": 2.0398,
-      "mean_token_accuracy": 0.5414750516414643,
-      "step": 1530
-    },
-    {
-      "epoch": 0.13429842155751287,
-      "grad_norm": 1.7159792184829712,
-      "learning_rate": 1.8319766420546902e-05,
-      "loss": 2.0242,
-      "mean_token_accuracy": 0.5429182916879653,
-      "step": 1540
-    },
-    {
-      "epoch": 0.13517048922996425,
-      "grad_norm": 1.8322516679763794,
-      "learning_rate": 1.8285833572409135e-05,
-      "loss": 2.016,
-      "mean_token_accuracy": 0.5468688800930976,
-      "step": 1550
-    },
-    {
-      "epoch": 0.13604255690241562,
-      "grad_norm": 1.9483593702316284,
-      "learning_rate": 1.8251593585101243e-05,
-      "loss": 2.0176,
-      "mean_token_accuracy": 0.5414750576019287,
-      "step": 1560
-    },
-    {
-      "epoch": 0.136914624574867,
-      "grad_norm": 1.7946149110794067,
-      "learning_rate": 1.821704772783063e-05,
-      "loss": 1.998,
-      "mean_token_accuracy": 0.5516878694295884,
-      "step": 1570
-    },
-    {
-      "epoch": 0.1377866922473184,
-      "grad_norm": 1.7067538499832153,
-      "learning_rate": 1.818219728114267e-05,
-      "loss": 1.998,
-      "mean_token_accuracy": 0.5487157553434372,
-      "step": 1580
-    },
-    {
-      "epoch": 0.13865875991976978,
-      "grad_norm": 1.660068154335022,
-      "learning_rate": 1.8147043536873275e-05,
-      "loss": 2.0168,
-      "mean_token_accuracy": 0.5465508788824082,
-      "step": 1590
-    },
-    {
-      "epoch": 0.13953082759222116,
-      "grad_norm": 1.8815644979476929,
-      "learning_rate": 1.8111587798100974e-05,
-      "loss": 2.0312,
-      "mean_token_accuracy": 0.5469055712223053,
-      "step": 1600
-    },
-    {
-      "epoch": 0.13953082759222116,
-      "eval_runtime": 4.112,
-      "eval_samples_per_second": 266.781,
-      "eval_steps_per_second": 66.878,
-      "step": 1600
-    },
-    {
-      "epoch": 0.14040289526467253,
-      "grad_norm": 1.8400298357009888,
-      "learning_rate": 1.807583137909862e-05,
-      "loss": 2.0102,
-      "mean_token_accuracy": 0.5490826845169068,
-      "step": 1610
-    },
-    {
-      "epoch": 0.1412749629371239,
-      "grad_norm": 1.9555879831314087,
-      "learning_rate": 1.8039775605284687e-05,
-      "loss": 1.9578,
-      "mean_token_accuracy": 0.5540851205587387,
-      "step": 1620
-    },
-    {
-      "epoch": 0.14214703060957531,
-      "grad_norm": 1.7845934629440308,
-      "learning_rate": 1.800342181317413e-05,
-      "loss": 2.0277,
-      "mean_token_accuracy": 0.5414628148078918,
-      "step": 1630
-    },
-    {
-      "epoch": 0.1430190982820267,
-      "grad_norm": 1.726135015487671,
-      "learning_rate": 1.7966771350328825e-05,
-      "loss": 2.0184,
-      "mean_token_accuracy": 0.5429182916879653,
-      "step": 1640
-    },
-    {
-      "epoch": 0.14389116595447807,
-      "grad_norm": 1.8526731729507446,
-      "learning_rate": 1.7929825575307665e-05,
-      "loss": 2.0297,
-      "mean_token_accuracy": 0.5434075310826302,
-      "step": 1650
-    },
-    {
-      "epoch": 0.14476323362692944,
-      "grad_norm": 1.8522251844406128,
-      "learning_rate": 1.7892585857616144e-05,
-      "loss": 1.9961,
-      "mean_token_accuracy": 0.5488136008381843,
-      "step": 1660
-    },
-    {
-      "epoch": 0.14563530129938082,
-      "grad_norm": 1.9614042043685913,
-      "learning_rate": 1.785505357765563e-05,
-      "loss": 2.0184,
-      "mean_token_accuracy": 0.5433341532945632,
-      "step": 1670
-    },
-    {
-      "epoch": 0.14650736897183222,
-      "grad_norm": 1.5994799137115479,
-      "learning_rate": 1.781723012667218e-05,
-      "loss": 1.9703,
-      "mean_token_accuracy": 0.5564946234226227,
-      "step": 1680
-    },
-    {
-      "epoch": 0.1473794366442836,
-      "grad_norm": 1.7674108743667603,
-      "learning_rate": 1.7779116906704986e-05,
-      "loss": 2.0121,
-      "mean_token_accuracy": 0.5460861057043076,
-      "step": 1690
-    },
-    {
-      "epoch": 0.14825150431673498,
-      "grad_norm": 1.8662148714065552,
-      "learning_rate": 1.7740715330534383e-05,
-      "loss": 1.9922,
-      "mean_token_accuracy": 0.5527397289872169,
-      "step": 1700
-    },
-    {
-      "epoch": 0.14825150431673498,
-      "eval_runtime": 4.1159,
-      "eval_samples_per_second": 266.524,
-      "eval_steps_per_second": 66.813,
-      "step": 1700
-    },
-    {
-      "epoch": 0.14912357198918635,
-      "grad_norm": 1.8501564264297485,
-      "learning_rate": 1.770202682162949e-05,
-      "loss": 1.9559,
-      "mean_token_accuracy": 0.5579867869615555,
-      "step": 1710
-    },
-    {
-      "epoch": 0.14999563966163773,
-      "grad_norm": 1.7436656951904297,
-      "learning_rate": 1.7663052814095447e-05,
-      "loss": 1.9828,
-      "mean_token_accuracy": 0.5485445216298104,
-      "step": 1720
-    },
-    {
-      "epoch": 0.15086770733408914,
-      "grad_norm": 1.8908530473709106,
-      "learning_rate": 1.7623794752620255e-05,
-      "loss": 2.0418,
-      "mean_token_accuracy": 0.5387230917811394,
-      "step": 1730
-    },
-    {
-      "epoch": 0.1517397750065405,
-      "grad_norm": 1.8569004535675049,
-      "learning_rate": 1.7584254092421226e-05,
-      "loss": 2.0219,
-      "mean_token_accuracy": 0.5442514717578888,
-      "step": 1740
-    },
-    {
-      "epoch": 0.1526118426789919,
-      "grad_norm": 1.8880852460861206,
-      "learning_rate": 1.754443229919103e-05,
-      "loss": 1.9867,
-      "mean_token_accuracy": 0.5484833672642708,
-      "step": 1750
-    },
-    {
-      "epoch": 0.15348391035144326,
-      "grad_norm": 1.8610656261444092,
-      "learning_rate": 1.7504330849043373e-05,
-      "loss": 1.9996,
-      "mean_token_accuracy": 0.5491193771362305,
-      "step": 1760
-    },
-    {
-      "epoch": 0.15435597802389464,
-      "grad_norm": 1.8499536514282227,
-      "learning_rate": 1.7463951228458288e-05,
-      "loss": 2.0,
-      "mean_token_accuracy": 0.5466854214668274,
-      "step": 1770
-    },
-    {
-      "epoch": 0.15522804569634605,
-      "grad_norm": 1.6235536336898804,
-      "learning_rate": 1.7423294934227017e-05,
-      "loss": 2.0156,
-      "mean_token_accuracy": 0.543431994318962,
-      "step": 1780
-    },
-    {
-      "epoch": 0.15610011336879742,
-      "grad_norm": 1.7828840017318726,
-      "learning_rate": 1.7382363473396543e-05,
-      "loss": 2.0211,
-      "mean_token_accuracy": 0.5458170279860497,
-      "step": 1790
-    },
-    {
-      "epoch": 0.1569721810412488,
-      "grad_norm": 1.7992655038833618,
-      "learning_rate": 1.734115836321372e-05,
-      "loss": 2.023,
-      "mean_token_accuracy": 0.550684928894043,
-      "step": 1800
-    },
-    {
-      "epoch": 0.1569721810412488,
-      "eval_runtime": 4.1181,
-      "eval_samples_per_second": 266.383,
-      "eval_steps_per_second": 66.778,
-      "step": 1800
-    },
-    {
-      "epoch": 0.15784424871370017,
-      "grad_norm": 1.858206868171692,
-      "learning_rate": 1.7299681131069026e-05,
-      "loss": 1.9414,
-      "mean_token_accuracy": 0.5573997050523758,
-      "step": 1810
-    },
-    {
-      "epoch": 0.15871631638615158,
-      "grad_norm": 1.7538492679595947,
-      "learning_rate": 1.725793331443996e-05,
-      "loss": 1.9512,
-      "mean_token_accuracy": 0.5600171267986298,
-      "step": 1820
-    },
-    {
-      "epoch": 0.15958838405860296,
-      "grad_norm": 1.6790558099746704,
-      "learning_rate": 1.7215916460834048e-05,
-      "loss": 2.0461,
-      "mean_token_accuracy": 0.5415484279394149,
-      "step": 1830
-    },
-    {
-      "epoch": 0.16046045173105433,
-      "grad_norm": 1.7191722393035889,
-      "learning_rate": 1.7173632127731462e-05,
-      "loss": 2.0105,
-      "mean_token_accuracy": 0.5449853226542473,
-      "step": 1840
-    },
-    {
-      "epoch": 0.1613325194035057,
-      "grad_norm": 1.8161389827728271,
-      "learning_rate": 1.7131081882527305e-05,
-      "loss": 1.9742,
-      "mean_token_accuracy": 0.551969176530838,
-      "step": 1850
-    },
-    {
-      "epoch": 0.16220458707595709,
-      "grad_norm": 1.8615176677703857,
-      "learning_rate": 1.708826730247351e-05,
-      "loss": 2.0273,
-      "mean_token_accuracy": 0.5425146818161011,
-      "step": 1860
-    },
-    {
-      "epoch": 0.1630766547484085,
-      "grad_norm": 1.8065239191055298,
-      "learning_rate": 1.704518997462037e-05,
-      "loss": 1.9953,
-      "mean_token_accuracy": 0.5492172211408615,
-      "step": 1870
-    },
-    {
-      "epoch": 0.16394872242085987,
-      "grad_norm": 1.6941074132919312,
-      "learning_rate": 1.7001851495757708e-05,
-      "loss": 1.9531,
-      "mean_token_accuracy": 0.556934928894043,
-      "step": 1880
-    },
-    {
-      "epoch": 0.16482079009331124,
-      "grad_norm": 1.8342607021331787,
-      "learning_rate": 1.6958253472355687e-05,
-      "loss": 2.0273,
-      "mean_token_accuracy": 0.5419398233294487,
-      "step": 1890
-    },
-    {
-      "epoch": 0.16569285776576262,
-      "grad_norm": 1.7665150165557861,
-      "learning_rate": 1.6914397520505267e-05,
-      "loss": 2.0016,
-      "mean_token_accuracy": 0.5444716215133667,
-      "step": 1900
-    },
-    {
-      "epoch": 0.16569285776576262,
-      "eval_runtime": 4.1193,
-      "eval_samples_per_second": 266.308,
-      "eval_steps_per_second": 66.759,
-      "step": 1900
-    },
-    {
-      "epoch": 0.166564925438214,
-      "grad_norm": 1.6462085247039795,
-      "learning_rate": 1.6870285265858298e-05,
-      "loss": 1.9848,
-      "mean_token_accuracy": 0.5466364979743957,
-      "step": 1910
-    },
-    {
-      "epoch": 0.1674369931106654,
-      "grad_norm": 1.724899172782898,
-      "learning_rate": 1.6825918343567257e-05,
-      "loss": 1.9793,
-      "mean_token_accuracy": 0.5526296496391296,
-      "step": 1920
-    },
-    {
-      "epoch": 0.16830906078311678,
-      "grad_norm": 1.6339362859725952,
-      "learning_rate": 1.678129839822463e-05,
-      "loss": 1.9582,
-      "mean_token_accuracy": 0.5575587034225464,
-      "step": 1930
-    },
-    {
-      "epoch": 0.16918112845556815,
-      "grad_norm": 2.127652406692505,
-      "learning_rate": 1.673642708380198e-05,
-      "loss": 1.968,
-      "mean_token_accuracy": 0.5497798383235931,
-      "step": 1940
-    },
-    {
-      "epoch": 0.17005319612801953,
-      "grad_norm": 1.7609293460845947,
-      "learning_rate": 1.6691306063588583e-05,
-      "loss": 1.9875,
-      "mean_token_accuracy": 0.5432974576950074,
-      "step": 1950
-    },
-    {
-      "epoch": 0.1709252638004709,
-      "grad_norm": 1.5910046100616455,
-      "learning_rate": 1.6645937010129837e-05,
-      "loss": 1.9262,
-      "mean_token_accuracy": 0.5619251489639282,
-      "step": 1960
-    },
-    {
-      "epoch": 0.1717973314729223,
-      "grad_norm": 1.6892436742782593,
-      "learning_rate": 1.660032160516522e-05,
-      "loss": 1.9895,
-      "mean_token_accuracy": 0.5465141832828522,
-      "step": 1970
-    },
-    {
-      "epoch": 0.1726693991453737,
-      "grad_norm": 1.7870206832885742,
-      "learning_rate": 1.6554461539565953e-05,
-      "loss": 1.9867,
-      "mean_token_accuracy": 0.5505503922700882,
-      "step": 1980
-    },
-    {
-      "epoch": 0.17354146681782506,
-      "grad_norm": 1.7099195718765259,
-      "learning_rate": 1.650835851327236e-05,
-      "loss": 1.9516,
-      "mean_token_accuracy": 0.5541462808847427,
-      "step": 1990
-    },
-    {
-      "epoch": 0.17441353449027644,
-      "grad_norm": 1.6207865476608276,
-      "learning_rate": 1.6462014235230805e-05,
-      "loss": 1.9676,
-      "mean_token_accuracy": 0.5511864051222801,
-      "step": 2000
-    },
-    {
-      "epoch": 0.17441353449027644,
-      "eval_runtime": 4.1246,
-      "eval_samples_per_second": 265.966,
-      "eval_steps_per_second": 66.673,
-      "step": 2000
-    },
-    {
-      "epoch": 0.17528560216272782,
-      "grad_norm": 1.6009984016418457,
-      "learning_rate": 1.641543042333038e-05,
-      "loss": 1.9285,
-      "mean_token_accuracy": 0.5602739691734314,
-      "step": 2010
-    },
-    {
-      "epoch": 0.17615766983517922,
-      "grad_norm": 1.8085410594940186,
-      "learning_rate": 1.636860880433922e-05,
-      "loss": 2.0191,
-      "mean_token_accuracy": 0.5453767135739327,
-      "step": 2020
-    },
-    {
-      "epoch": 0.1770297375076306,
-      "grad_norm": 1.721328854560852,
-      "learning_rate": 1.632155111384047e-05,
-      "loss": 2.0164,
-      "mean_token_accuracy": 0.5440068513154983,
-      "step": 2030
-    },
-    {
-      "epoch": 0.17790180518008197,
-      "grad_norm": 1.9679734706878662,
-      "learning_rate": 1.6274259096168e-05,
-      "loss": 1.9688,
-      "mean_token_accuracy": 0.5518346399068832,
-      "step": 2040
-    },
-    {
-      "epoch": 0.17877387285253335,
-      "grad_norm": 1.791015625,
-      "learning_rate": 1.622673450434169e-05,
-      "loss": 1.9668,
-      "mean_token_accuracy": 0.5527886480093003,
-      "step": 2050
-    },
-    {
-      "epoch": 0.17964594052498473,
-      "grad_norm": 1.7144931554794312,
-      "learning_rate": 1.6178979100002486e-05,
-      "loss": 1.975,
-      "mean_token_accuracy": 0.5520547926425934,
-      "step": 2060
-    },
-    {
-      "epoch": 0.18051800819743613,
-      "grad_norm": 1.723570704460144,
-      "learning_rate": 1.6130994653347096e-05,
-      "loss": 2.0148,
-      "mean_token_accuracy": 0.5422700583934784,
-      "step": 2070
-    },
-    {
-      "epoch": 0.1813900758698875,
-      "grad_norm": 1.6630723476409912,
-      "learning_rate": 1.6082782943062355e-05,
-      "loss": 1.9926,
-      "mean_token_accuracy": 0.544630628824234,
-      "step": 2080
-    },
-    {
-      "epoch": 0.18226214354233888,
-      "grad_norm": 1.8220481872558594,
-      "learning_rate": 1.6034345756259303e-05,
-      "loss": 1.9195,
-      "mean_token_accuracy": 0.5643224030733108,
-      "step": 2090
-    },
-    {
-      "epoch": 0.18313421121479026,
-      "grad_norm": 1.6514785289764404,
-      "learning_rate": 1.598568488840695e-05,
-      "loss": 1.9707,
-      "mean_token_accuracy": 0.5499633014202118,
-      "step": 2100
-    },
-    {
-      "epoch": 0.18313421121479026,
-      "eval_runtime": 4.122,
-      "eval_samples_per_second": 266.134,
-      "eval_steps_per_second": 66.716,
-      "step": 2100
-    },
-    {
-      "epoch": 0.18400627888724164,
-      "grad_norm": 1.6990513801574707,
-      "learning_rate": 1.5936802143265708e-05,
-      "loss": 1.9875,
-      "mean_token_accuracy": 0.5464285671710968,
-      "step": 2110
-    },
-    {
-      "epoch": 0.18487834655969304,
-      "grad_norm": 1.5384515523910522,
-      "learning_rate": 1.5887699332820527e-05,
-      "loss": 1.9789,
-      "mean_token_accuracy": 0.5475332558155059,
-      "step": 2120
-    },
-    {
-      "epoch": 0.18575041423214442,
-      "grad_norm": 1.6312943696975708,
-      "learning_rate": 1.5838378277213745e-05,
-      "loss": 1.9656,
-      "mean_token_accuracy": 0.553339034318924,
-      "step": 2130
-    },
-    {
-      "epoch": 0.1866224819045958,
-      "grad_norm": 1.5812935829162598,
-      "learning_rate": 1.57888408046776e-05,
-      "loss": 1.975,
-      "mean_token_accuracy": 0.5485200643539428,
-      "step": 2140
-    },
-    {
-      "epoch": 0.18749454957704717,
-      "grad_norm": 1.7396554946899414,
-      "learning_rate": 1.573908875146648e-05,
-      "loss": 1.9523,
-      "mean_token_accuracy": 0.5553326815366745,
-      "step": 2150
-    },
-    {
-      "epoch": 0.18836661724949857,
-      "grad_norm": 1.667148470878601,
-      "learning_rate": 1.5689123961788834e-05,
-      "loss": 1.9777,
-      "mean_token_accuracy": 0.5511986374855041,
-      "step": 2160
-    },
-    {
-      "epoch": 0.18923868492194995,
-      "grad_norm": 1.6402790546417236,
-      "learning_rate": 1.563894828773883e-05,
-      "loss": 1.9578,
-      "mean_token_accuracy": 0.5545376718044281,
-      "step": 2170
-    },
-    {
-      "epoch": 0.19011075259440133,
-      "grad_norm": 1.6035139560699463,
-      "learning_rate": 1.55885635892277e-05,
-      "loss": 2.0395,
-      "mean_token_accuracy": 0.5352739706635475,
-      "step": 2180
-    },
-    {
-      "epoch": 0.1909828202668527,
-      "grad_norm": 1.617301344871521,
-      "learning_rate": 1.5537971733914784e-05,
-      "loss": 1.9992,
-      "mean_token_accuracy": 0.5442392379045486,
-      "step": 2190
-    },
-    {
-      "epoch": 0.19185488793930408,
-      "grad_norm": 1.6773086786270142,
-      "learning_rate": 1.5487174597138314e-05,
-      "loss": 1.9297,
-      "mean_token_accuracy": 0.5667318999767303,
-      "step": 2200
-    },
-    {
-      "epoch": 0.19185488793930408,
-      "eval_runtime": 4.1193,
-      "eval_samples_per_second": 266.31,
-      "eval_steps_per_second": 66.76,
-      "step": 2200
-    },
-    {
-      "epoch": 0.19272695561175548,
-      "grad_norm": 1.5105735063552856,
-      "learning_rate": 1.543617406184589e-05,
-      "loss": 1.9473,
-      "mean_token_accuracy": 0.5579011827707291,
-      "step": 2210
-    },
-    {
-      "epoch": 0.19359902328420686,
-      "grad_norm": 1.6412047147750854,
-      "learning_rate": 1.5384972018524678e-05,
-      "loss": 1.9625,
-      "mean_token_accuracy": 0.5585249498486519,
-      "step": 2220
-    },
-    {
-      "epoch": 0.19447109095665824,
-      "grad_norm": 1.7117334604263306,
-      "learning_rate": 1.5333570365131353e-05,
-      "loss": 1.975,
-      "mean_token_accuracy": 0.5482754409313202,
-      "step": 2230
-    },
-    {
-      "epoch": 0.19534315862910961,
-      "grad_norm": 1.73671293258667,
-      "learning_rate": 1.5281971007021728e-05,
-      "loss": 1.9688,
-      "mean_token_accuracy": 0.5536203473806381,
-      "step": 2240
-    },
-    {
-      "epoch": 0.196215226301561,
-      "grad_norm": 1.5416544675827026,
-      "learning_rate": 1.5230175856880132e-05,
-      "loss": 1.943,
-      "mean_token_accuracy": 0.5545988231897354,
-      "step": 2250
-    },
-    {
-      "epoch": 0.1970872939740124,
-      "grad_norm": 1.4799134731292725,
-      "learning_rate": 1.5178186834648509e-05,
-      "loss": 1.9039,
-      "mean_token_accuracy": 0.5644691824913025,
-      "step": 2260
-    },
-    {
-      "epoch": 0.19795936164646377,
-      "grad_norm": 1.7748204469680786,
-      "learning_rate": 1.5126005867455256e-05,
-      "loss": 1.9094,
-      "mean_token_accuracy": 0.5612279802560807,
-      "step": 2270
-    },
-    {
-      "epoch": 0.19883142931891515,
-      "grad_norm": 1.6312216520309448,
-      "learning_rate": 1.5073634889543778e-05,
-      "loss": 2.018,
-      "mean_token_accuracy": 0.5427837550640107,
-      "step": 2280
-    },
-    {
-      "epoch": 0.19970349699136652,
-      "grad_norm": 1.5463025569915771,
-      "learning_rate": 1.5021075842200796e-05,
-      "loss": 1.9863,
-      "mean_token_accuracy": 0.5487402200698852,
-      "step": 2290
-    },
-    {
-      "epoch": 0.2005755646638179,
-      "grad_norm": 1.7069957256317139,
-      "learning_rate": 1.4968330673684387e-05,
-      "loss": 1.9723,
-      "mean_token_accuracy": 0.551186403632164,
-      "step": 2300
-    },
-    {
-      "epoch": 0.2005755646638179,
-      "eval_runtime": 4.1184,
-      "eval_samples_per_second": 266.368,
-      "eval_steps_per_second": 66.774,
-      "step": 2300
-    },
-    {
-      "epoch": 0.2014476323362693,
-      "grad_norm": 1.634435772895813,
-      "learning_rate": 1.4915401339151769e-05,
-      "loss": 1.9191,
-      "mean_token_accuracy": 0.5595768123865128,
-      "step": 2310
-    },
-    {
-      "epoch": 0.20231970000872068,
-      "grad_norm": 1.6615523099899292,
-      "learning_rate": 1.486228980058682e-05,
-      "loss": 1.9664,
-      "mean_token_accuracy": 0.5524339526891708,
-      "step": 2320
-    },
-    {
-      "epoch": 0.20319176768117206,
-      "grad_norm": 1.5431104898452759,
-      "learning_rate": 1.4808998026727348e-05,
-      "loss": 1.9852,
-      "mean_token_accuracy": 0.5489359050989151,
-      "step": 2330
-    },
-    {
-      "epoch": 0.20406383535362344,
-      "grad_norm": 1.6319235563278198,
-      "learning_rate": 1.4755527992992133e-05,
-      "loss": 1.9211,
-      "mean_token_accuracy": 0.5597602725028992,
-      "step": 2340
-    },
-    {
-      "epoch": 0.2049359030260748,
-      "grad_norm": 1.5469952821731567,
-      "learning_rate": 1.4701881681407684e-05,
-      "loss": 1.9492,
-      "mean_token_accuracy": 0.5565802335739136,
-      "step": 2350
-    },
-    {
-      "epoch": 0.20580797069852622,
-      "grad_norm": 1.657025694847107,
-      "learning_rate": 1.464806108053477e-05,
-      "loss": 1.9508,
-      "mean_token_accuracy": 0.5524339556694031,
-      "step": 2360
-    },
-    {
-      "epoch": 0.2066800383709776,
-      "grad_norm": 1.866478681564331,
-      "learning_rate": 1.4594068185394723e-05,
-      "loss": 1.966,
-      "mean_token_accuracy": 0.5484711349010467,
-      "step": 2370
-    },
-    {
-      "epoch": 0.20755210604342897,
-      "grad_norm": 1.7019178867340088,
-      "learning_rate": 1.4539904997395468e-05,
-      "loss": 1.9852,
-      "mean_token_accuracy": 0.546575340628624,
-      "step": 2380
-    },
-    {
-      "epoch": 0.20842417371588035,
-      "grad_norm": 1.5332332849502563,
-      "learning_rate": 1.448557352425735e-05,
-      "loss": 1.9473,
-      "mean_token_accuracy": 0.5539261221885681,
-      "step": 2390
-    },
-    {
-      "epoch": 0.20929624138833172,
-      "grad_norm": 1.5550459623336792,
-      "learning_rate": 1.44310757799387e-05,
-      "loss": 1.9062,
-      "mean_token_accuracy": 0.5623532325029373,
-      "step": 2400
-    },
-    {
-      "epoch": 0.20929624138833172,
-      "eval_runtime": 4.1118,
-      "eval_samples_per_second": 266.79,
-      "eval_steps_per_second": 66.88,
-      "step": 2400
-    },
-    {
-      "epoch": 0.21016830906078313,
-      "grad_norm": 1.564449429512024,
-      "learning_rate": 1.437641378456119e-05,
-      "loss": 1.9941,
-      "mean_token_accuracy": 0.5486179113388061,
-      "step": 2410
-    },
-    {
-      "epoch": 0.2110403767332345,
-      "grad_norm": 1.5772294998168945,
-      "learning_rate": 1.4321589564334946e-05,
-      "loss": 1.9523,
-      "mean_token_accuracy": 0.554953521490097,
-      "step": 2420
-    },
-    {
-      "epoch": 0.21191244440568588,
-      "grad_norm": 1.638231635093689,
-      "learning_rate": 1.4266605151483444e-05,
-      "loss": 1.9777,
-      "mean_token_accuracy": 0.550954008102417,
-      "step": 2430
-    },
-    {
-      "epoch": 0.21278451207813726,
-      "grad_norm": 1.849936604499817,
-      "learning_rate": 1.4211462584168178e-05,
-      "loss": 1.9352,
-      "mean_token_accuracy": 0.5583170294761658,
-      "step": 2440
-    },
-    {
-      "epoch": 0.21365657975058863,
-      "grad_norm": 1.5940241813659668,
-      "learning_rate": 1.4156163906413113e-05,
-      "loss": 1.9203,
-      "mean_token_accuracy": 0.563759782910347,
-      "step": 2450
-    },
-    {
-      "epoch": 0.21452864742304004,
-      "grad_norm": 1.6488444805145264,
-      "learning_rate": 1.4100711168028906e-05,
-      "loss": 1.9488,
-      "mean_token_accuracy": 0.5550146847963333,
-      "step": 2460
-    },
-    {
-      "epoch": 0.2154007150954914,
-      "grad_norm": 1.7454982995986938,
-      "learning_rate": 1.4045106424536938e-05,
-      "loss": 1.941,
-      "mean_token_accuracy": 0.5565190732479095,
-      "step": 2470
-    },
-    {
-      "epoch": 0.2162727827679428,
-      "grad_norm": 1.5013492107391357,
-      "learning_rate": 1.398935173709311e-05,
-      "loss": 1.9883,
-      "mean_token_accuracy": 0.5474437415599823,
-      "step": 2480
-    },
-    {
-      "epoch": 0.21714485044039417,
-      "grad_norm": 1.7090688943862915,
-      "learning_rate": 1.3933449172411446e-05,
-      "loss": 1.9137,
-      "mean_token_accuracy": 0.5604207456111908,
-      "step": 2490
-    },
-    {
-      "epoch": 0.21801691811284557,
-      "grad_norm": 1.811716079711914,
-      "learning_rate": 1.387740080268748e-05,
-      "loss": 1.9359,
-      "mean_token_accuracy": 0.5603595972061157,
-      "step": 2500
-    },
-    {
-      "epoch": 0.21801691811284557,
-      "eval_runtime": 4.1163,
-      "eval_samples_per_second": 266.498,
-      "eval_steps_per_second": 66.807,
-      "step": 2500
-    },
-    {
-      "epoch": 0.21888898578529695,
-      "grad_norm": 1.6570755243301392,
-      "learning_rate": 1.3821208705521442e-05,
-      "loss": 1.9672,
-      "mean_token_accuracy": 0.5500489264726639,
-      "step": 2510
-    },
-    {
-      "epoch": 0.21976105345774832,
-      "grad_norm": 1.59494149684906,
-      "learning_rate": 1.3764874963841255e-05,
-      "loss": 2.0,
-      "mean_token_accuracy": 0.5461839511990547,
-      "step": 2520
-    },
-    {
-      "epoch": 0.2206331211301997,
-      "grad_norm": 1.5657888650894165,
-      "learning_rate": 1.3708401665825319e-05,
-      "loss": 1.9141,
-      "mean_token_accuracy": 0.5600293546915054,
-      "step": 2530
-    },
-    {
-      "epoch": 0.22150518880265108,
-      "grad_norm": 1.6764097213745117,
-      "learning_rate": 1.36517909048251e-05,
-      "loss": 1.952,
-      "mean_token_accuracy": 0.5537304252386093,
-      "step": 2540
-    },
-    {
-      "epoch": 0.22237725647510248,
-      "grad_norm": 1.884006381034851,
-      "learning_rate": 1.3595044779287543e-05,
-      "loss": 1.9461,
-      "mean_token_accuracy": 0.5545988231897354,
-      "step": 2550
-    },
-    {
-      "epoch": 0.22324932414755386,
-      "grad_norm": 1.7752189636230469,
-      "learning_rate": 1.3538165392677288e-05,
-      "loss": 1.9828,
-      "mean_token_accuracy": 0.5504892364144325,
-      "step": 2560
-    },
-    {
-      "epoch": 0.22412139182000523,
-      "grad_norm": 1.5350412130355835,
-      "learning_rate": 1.3481154853398686e-05,
-      "loss": 1.9719,
-      "mean_token_accuracy": 0.5463674247264863,
-      "step": 2570
-    },
-    {
-      "epoch": 0.2249934594924566,
-      "grad_norm": 1.6891847848892212,
-      "learning_rate": 1.3424015274717665e-05,
-      "loss": 1.943,
-      "mean_token_accuracy": 0.5556629180908204,
-      "step": 2580
-    },
-    {
-      "epoch": 0.225865527164908,
-      "grad_norm": 1.618148684501648,
-      "learning_rate": 1.3366748774683376e-05,
-      "loss": 1.9203,
-      "mean_token_accuracy": 0.5611912995576859,
-      "step": 2590
-    },
-    {
-      "epoch": 0.2267375948373594,
-      "grad_norm": 1.7580416202545166,
-      "learning_rate": 1.3309357476049686e-05,
-      "loss": 1.9187,
-      "mean_token_accuracy": 0.5619985282421112,
-      "step": 2600
-    },
-    {
-      "epoch": 0.2267375948373594,
-      "eval_runtime": 4.1058,
-      "eval_samples_per_second": 267.184,
-      "eval_steps_per_second": 66.979,
-      "step": 2600
-    },
-    {
-      "epoch": 0.22760966250981077,
-      "grad_norm": 1.6931602954864502,
-      "learning_rate": 1.3251843506196508e-05,
-      "loss": 1.9324,
-      "mean_token_accuracy": 0.5586105644702911,
-      "step": 2610
-    },
-    {
-      "epoch": 0.22848173018226214,
-      "grad_norm": 1.6409891843795776,
-      "learning_rate": 1.3194208997050915e-05,
-      "loss": 1.902,
-      "mean_token_accuracy": 0.5670376747846604,
-      "step": 2620
-    },
-    {
-      "epoch": 0.22935379785471352,
-      "grad_norm": 1.6231530904769897,
-      "learning_rate": 1.313645608500814e-05,
-      "loss": 1.9426,
-      "mean_token_accuracy": 0.5576076328754425,
-      "step": 2630
-    },
-    {
-      "epoch": 0.2302258655271649,
-      "grad_norm": 1.6396461725234985,
-      "learning_rate": 1.3078586910852364e-05,
-      "loss": 1.941,
-      "mean_token_accuracy": 0.5553816020488739,
-      "step": 2640
-    },
-    {
-      "epoch": 0.2310979331996163,
-      "grad_norm": 1.56830894947052,
-      "learning_rate": 1.3020603619677378e-05,
-      "loss": 1.9379,
-      "mean_token_accuracy": 0.554855677485466,
-      "step": 2650
-    },
-    {
-      "epoch": 0.23197000087206768,
-      "grad_norm": 1.5743964910507202,
-      "learning_rate": 1.296250836080706e-05,
-      "loss": 1.9309,
-      "mean_token_accuracy": 0.5564823836088181,
-      "step": 2660
-    },
-    {
-      "epoch": 0.23284206854451905,
-      "grad_norm": 1.5447441339492798,
-      "learning_rate": 1.2904303287715702e-05,
-      "loss": 1.957,
-      "mean_token_accuracy": 0.551700097322464,
-      "step": 2670
-    },
-    {
-      "epoch": 0.23371413621697043,
-      "grad_norm": 1.502420425415039,
-      "learning_rate": 1.284599055794819e-05,
-      "loss": 1.9355,
-      "mean_token_accuracy": 0.5543297469615937,
-      "step": 2680
-    },
-    {
-      "epoch": 0.2345862038894218,
-      "grad_norm": 1.5204365253448486,
-      "learning_rate": 1.2787572333040022e-05,
-      "loss": 1.9727,
-      "mean_token_accuracy": 0.5479696780443192,
-      "step": 2690
-    },
-    {
-      "epoch": 0.2354582715618732,
-      "grad_norm": 1.599164366722107,
-      "learning_rate": 1.2729050778437197e-05,
-      "loss": 1.9551,
-      "mean_token_accuracy": 0.551528862118721,
-      "step": 2700
-    },
-    {
-      "epoch": 0.2354582715618732,
-      "eval_runtime": 4.1123,
-      "eval_samples_per_second": 266.76,
-      "eval_steps_per_second": 66.872,
-      "step": 2700
-    },
-    {
-      "epoch": 0.2363303392343246,
-      "grad_norm": 1.5735387802124023,
-      "learning_rate": 1.2670428063415932e-05,
-      "loss": 1.9484,
-      "mean_token_accuracy": 0.5550024449825287,
-      "step": 2710
-    },
-    {
-      "epoch": 0.23720240690677596,
-      "grad_norm": 1.6314678192138672,
-      "learning_rate": 1.2611706361002254e-05,
-      "loss": 1.8867,
-      "mean_token_accuracy": 0.5687133073806763,
-      "step": 2720
-    },
-    {
-      "epoch": 0.23807447457922734,
-      "grad_norm": 1.529001235961914,
-      "learning_rate": 1.2552887847891462e-05,
-      "loss": 1.9777,
-      "mean_token_accuracy": 0.5501712322235107,
-      "step": 2730
-    },
-    {
-      "epoch": 0.23894654225167872,
-      "grad_norm": 1.6241025924682617,
-      "learning_rate": 1.2493974704367427e-05,
-      "loss": 1.9434,
-      "mean_token_accuracy": 0.5532045066356659,
-      "step": 2740
-    },
-    {
-      "epoch": 0.23981860992413012,
-      "grad_norm": 1.5482800006866455,
-      "learning_rate": 1.2434969114221777e-05,
-      "loss": 1.9187,
-      "mean_token_accuracy": 0.5591854244470597,
-      "step": 2750
-    },
-    {
-      "epoch": 0.2406906775965815,
-      "grad_norm": 1.7343766689300537,
-      "learning_rate": 1.237587326467296e-05,
-      "loss": 1.9582,
-      "mean_token_accuracy": 0.5533635020256042,
-      "step": 2760
-    },
-    {
-      "epoch": 0.24156274526903287,
-      "grad_norm": 1.618938684463501,
-      "learning_rate": 1.2316689346285146e-05,
-      "loss": 1.9402,
-      "mean_token_accuracy": 0.5565190792083741,
-      "step": 2770
-    },
-    {
-      "epoch": 0.24243481294148425,
-      "grad_norm": 1.4117664098739624,
-      "learning_rate": 1.2257419552887047e-05,
-      "loss": 1.952,
-      "mean_token_accuracy": 0.5575220078229904,
-      "step": 2780
-    },
-    {
-      "epoch": 0.24330688061393563,
-      "grad_norm": 1.5653223991394043,
-      "learning_rate": 1.2198066081490585e-05,
-      "loss": 1.9504,
-      "mean_token_accuracy": 0.5538527399301529,
-      "step": 2790
-    },
-    {
-      "epoch": 0.24417894828638703,
-      "grad_norm": 1.6690893173217773,
-      "learning_rate": 1.213863113220946e-05,
-      "loss": 1.9074,
-      "mean_token_accuracy": 0.5629647761583328,
-      "step": 2800
-    },
-    {
-      "epoch": 0.24417894828638703,
-      "eval_runtime": 4.1128,
-      "eval_samples_per_second": 266.726,
-      "eval_steps_per_second": 66.864,
-      "step": 2800
-    },
-    {
-      "epoch": 0.2450510159588384,
-      "grad_norm": 1.5818336009979248,
-      "learning_rate": 1.2079116908177592e-05,
-      "loss": 1.9348,
-      "mean_token_accuracy": 0.5590141832828521,
-      "step": 2810
-    },
-    {
-      "epoch": 0.24592308363128978,
-      "grad_norm": 1.4075454473495483,
-      "learning_rate": 1.2019525615467462e-05,
-      "loss": 1.8961,
-      "mean_token_accuracy": 0.5670621335506439,
-      "step": 2820
-    },
-    {
-      "epoch": 0.24679515130374116,
-      "grad_norm": 1.5011941194534302,
-      "learning_rate": 1.1959859463008316e-05,
-      "loss": 1.9508,
-      "mean_token_accuracy": 0.5516511708498001,
-      "step": 2830
-    },
-    {
-      "epoch": 0.24766721897619257,
-      "grad_norm": 1.6102718114852905,
-      "learning_rate": 1.1900120662504315e-05,
-      "loss": 1.9316,
-      "mean_token_accuracy": 0.5560053825378418,
-      "step": 2840
-    },
-    {
-      "epoch": 0.24853928664864394,
-      "grad_norm": 1.4400602579116821,
-      "learning_rate": 1.1840311428352536e-05,
-      "loss": 1.9211,
-      "mean_token_accuracy": 0.5615092992782593,
-      "step": 2850
-    },
-    {
-      "epoch": 0.24941135432109532,
-      "grad_norm": 1.4591666460037231,
-      "learning_rate": 1.1780433977560879e-05,
-      "loss": 1.9113,
-      "mean_token_accuracy": 0.5630748599767685,
-      "step": 2860
-    },
-    {
-      "epoch": 0.2502834219935467,
-      "grad_norm": 1.4346725940704346,
-      "learning_rate": 1.1720490529665904e-05,
-      "loss": 1.9281,
-      "mean_token_accuracy": 0.5584760308265686,
-      "step": 2870
-    },
-    {
-      "epoch": 0.2511554896659981,
-      "grad_norm": 1.5606255531311035,
-      "learning_rate": 1.1660483306650558e-05,
-      "loss": 1.8977,
-      "mean_token_accuracy": 0.5632216215133667,
-      "step": 2880
-    },
-    {
-      "epoch": 0.25202755733844945,
-      "grad_norm": 1.4206461906433105,
-      "learning_rate": 1.160041453286179e-05,
-      "loss": 1.9879,
-      "mean_token_accuracy": 0.5511863976716995,
-      "step": 2890
-    },
-    {
-      "epoch": 0.25289962501090085,
-      "grad_norm": 1.3229713439941406,
-      "learning_rate": 1.154028643492812e-05,
-      "loss": 1.9348,
-      "mean_token_accuracy": 0.5559809178113937,
-      "step": 2900
-    },
-    {
-      "epoch": 0.25289962501090085,
-      "eval_runtime": 4.1089,
-      "eval_samples_per_second": 266.983,
-      "eval_steps_per_second": 66.928,
-      "step": 2900
-    },
-    {
-      "epoch": 0.2537716926833522,
-      "grad_norm": 1.521287441253662,
-      "learning_rate": 1.1480101241677097e-05,
-      "loss": 1.9293,
-      "mean_token_accuracy": 0.5588429570198059,
-      "step": 2910
-    },
-    {
-      "epoch": 0.2546437603558036,
-      "grad_norm": 1.6549524068832397,
-      "learning_rate": 1.1419861184052669e-05,
-      "loss": 1.9027,
-      "mean_token_accuracy": 0.5647627174854278,
-      "step": 2920
-    },
-    {
-      "epoch": 0.255515828028255,
-      "grad_norm": 1.4998723268508911,
-      "learning_rate": 1.1359568495032505e-05,
-      "loss": 1.9645,
-      "mean_token_accuracy": 0.5507460862398148,
-      "step": 2930
-    },
-    {
-      "epoch": 0.25638789570070636,
-      "grad_norm": 1.570664882659912,
-      "learning_rate": 1.1299225409545207e-05,
-      "loss": 1.9078,
-      "mean_token_accuracy": 0.5647836327552795,
-      "step": 2940
-    },
-    {
-      "epoch": 0.25725996337315776,
-      "grad_norm": 1.6389505863189697,
-      "learning_rate": 1.123883416438748e-05,
-      "loss": 1.9387,
-      "mean_token_accuracy": 0.5553693801164628,
-      "step": 2950
-    },
-    {
-      "epoch": 0.2581320310456091,
-      "grad_norm": 1.6411775350570679,
-      "learning_rate": 1.1178396998141206e-05,
-      "loss": 1.907,
-      "mean_token_accuracy": 0.5613380670547485,
-      "step": 2960
-    },
-    {
-      "epoch": 0.2590040987180605,
-      "grad_norm": 1.4937372207641602,
-      "learning_rate": 1.1117916151090469e-05,
-      "loss": 1.8922,
-      "mean_token_accuracy": 0.565655580163002,
-      "step": 2970
-    },
-    {
-      "epoch": 0.2598761663905119,
-      "grad_norm": 1.4831377267837524,
-      "learning_rate": 1.1057393865138513e-05,
-      "loss": 1.9074,
-      "mean_token_accuracy": 0.5579501003026962,
-      "step": 2980
-    },
-    {
-      "epoch": 0.26074823406296327,
-      "grad_norm": 1.6630659103393555,
-      "learning_rate": 1.099683238372464e-05,
-      "loss": 1.9117,
-      "mean_token_accuracy": 0.5608855187892914,
-      "step": 2990
-    },
-    {
-      "epoch": 0.2616203017354147,
-      "grad_norm": 1.5346556901931763,
-      "learning_rate": 1.0936233951741052e-05,
-      "loss": 1.8922,
-      "mean_token_accuracy": 0.564285722374916,
-      "step": 3000
-    },
-    {
-      "epoch": 0.2616203017354147,
-      "eval_runtime": 4.1164,
-      "eval_samples_per_second": 266.497,
-      "eval_steps_per_second": 66.807,
-      "step": 3000
-    },
-    {
-      "epoch": 0.2624923694078661,
-      "grad_norm": 1.4140172004699707,
-      "learning_rate": 1.0875600815449624e-05,
-      "loss": 1.8605,
-      "mean_token_accuracy": 0.5742049902677536,
-      "step": 3010
-    },
-    {
-      "epoch": 0.2633644370803174,
-      "grad_norm": 1.5377106666564941,
-      "learning_rate": 1.081493522239866e-05,
-      "loss": 1.9348,
-      "mean_token_accuracy": 0.5571306258440017,
-      "step": 3020
-    },
-    {
-      "epoch": 0.26423650475276883,
-      "grad_norm": 1.63088059425354,
-      "learning_rate": 1.075423942133957e-05,
-      "loss": 1.8941,
-      "mean_token_accuracy": 0.565998038649559,
-      "step": 3030
-    },
-    {
-      "epoch": 0.2651085724252202,
-      "grad_norm": 1.5577826499938965,
-      "learning_rate": 1.0693515662143505e-05,
-      "loss": 1.9359,
-      "mean_token_accuracy": 0.5578033298254013,
-      "step": 3040
-    },
-    {
-      "epoch": 0.2659806400976716,
-      "grad_norm": 1.5849180221557617,
-      "learning_rate": 1.0632766195717979e-05,
-      "loss": 1.9027,
-      "mean_token_accuracy": 0.5592588096857071,
-      "step": 3050
-    },
-    {
-      "epoch": 0.266852707770123,
-      "grad_norm": 1.5935850143432617,
-      "learning_rate": 1.0571993273923412e-05,
-      "loss": 1.9105,
-      "mean_token_accuracy": 0.5602984398603439,
-      "step": 3060
-    },
-    {
-      "epoch": 0.26772477544257434,
-      "grad_norm": 1.4581326246261597,
-      "learning_rate": 1.0511199149489673e-05,
-      "loss": 1.8953,
-      "mean_token_accuracy": 0.5645914882421493,
-      "step": 3070
-    },
-    {
-      "epoch": 0.26859684311502574,
-      "grad_norm": 1.481928825378418,
-      "learning_rate": 1.0450386075932571e-05,
-      "loss": 1.8996,
-      "mean_token_accuracy": 0.5631971657276154,
-      "step": 3080
-    },
-    {
-      "epoch": 0.2694689107874771,
-      "grad_norm": 1.499147653579712,
-      "learning_rate": 1.0389556307470316e-05,
-      "loss": 1.9266,
-      "mean_token_accuracy": 0.5553082227706909,
-      "step": 3090
-    },
-    {
-      "epoch": 0.2703409784599285,
-      "grad_norm": 1.6154686212539673,
-      "learning_rate": 1.0328712098939968e-05,
-      "loss": 1.973,
-      "mean_token_accuracy": 0.5525929570198059,
-      "step": 3100
-    },
-    {
-      "epoch": 0.2703409784599285,
-      "eval_runtime": 4.1144,
-      "eval_samples_per_second": 266.627,
-      "eval_steps_per_second": 66.839,
-      "step": 3100
-    },
-    {
-      "epoch": 0.2712130461323799,
-      "grad_norm": 1.5398544073104858,
-      "learning_rate": 1.0267855705713854e-05,
-      "loss": 1.884,
-      "mean_token_accuracy": 0.5667196631431579,
-      "step": 3110
-    },
-    {
-      "epoch": 0.27208511380483125,
-      "grad_norm": 1.5201234817504883,
-      "learning_rate": 1.020698938361595e-05,
-      "loss": 1.9309,
-      "mean_token_accuracy": 0.5534001976251602,
-      "step": 3120
-    },
-    {
-      "epoch": 0.27295718147728265,
-      "grad_norm": 1.7075104713439941,
-      "learning_rate": 1.0146115388838293e-05,
-      "loss": 1.8879,
-      "mean_token_accuracy": 0.5639799356460571,
-      "step": 3130
-    },
-    {
-      "epoch": 0.273829249149734,
-      "grad_norm": 1.671292781829834,
-      "learning_rate": 1.0085235977857322e-05,
-      "loss": 1.852,
-      "mean_token_accuracy": 0.572517117857933,
-      "step": 3140
-    },
-    {
-      "epoch": 0.2747013168221854,
-      "grad_norm": 1.5689773559570312,
-      "learning_rate": 1.002435340735024e-05,
-      "loss": 1.9004,
-      "mean_token_accuracy": 0.5610200554132462,
-      "step": 3150
-    },
-    {
-      "epoch": 0.2755733844946368,
-      "grad_norm": 1.6257286071777344,
-      "learning_rate": 9.963469934111374e-06,
-      "loss": 1.8852,
-      "mean_token_accuracy": 0.5662915885448456,
-      "step": 3160
-    },
-    {
-      "epoch": 0.27644545216708816,
-      "grad_norm": 1.5670658349990845,
-      "learning_rate": 9.90258781496851e-06,
-      "loss": 1.9367,
-      "mean_token_accuracy": 0.5574853301048279,
-      "step": 3170
-    },
-    {
-      "epoch": 0.27731751983953956,
-      "grad_norm": 1.557520866394043,
-      "learning_rate": 9.841709306699245e-06,
-      "loss": 1.9227,
-      "mean_token_accuracy": 0.5580357134342193,
-      "step": 3180
-    },
-    {
-      "epoch": 0.2781895875119909,
-      "grad_norm": 1.5760271549224854,
-      "learning_rate": 9.78083666594732e-06,
-      "loss": 1.9379,
-      "mean_token_accuracy": 0.5583842992782593,
-      "step": 3190
-    },
-    {
-      "epoch": 0.2790616551844423,
-      "grad_norm": 1.4098557233810425,
-      "learning_rate": 9.719972149138985e-06,
-      "loss": 1.9082,
-      "mean_token_accuracy": 0.5648972660303115,
-      "step": 3200
-    },
-    {
-      "epoch": 0.2790616551844423,
-      "eval_runtime": 4.1053,
-      "eval_samples_per_second": 267.216,
-      "eval_steps_per_second": 66.987,
-      "step": 3200
-    },
-    {
-      "epoch": 0.2799337228568937,
-      "grad_norm": 1.6750293970108032,
-      "learning_rate": 9.659118012399352e-06,
-      "loss": 1.923,
-      "mean_token_accuracy": 0.5586105644702911,
-      "step": 3210
-    },
-    {
-      "epoch": 0.28080579052934507,
-      "grad_norm": 1.443645715713501,
-      "learning_rate": 9.598276511468763e-06,
-      "loss": 1.8488,
-      "mean_token_accuracy": 0.5714774936437607,
-      "step": 3220
-    },
-    {
-      "epoch": 0.28167785820179647,
-      "grad_norm": 1.6804277896881104,
-      "learning_rate": 9.537449901619174e-06,
-      "loss": 1.9398,
-      "mean_token_accuracy": 0.5536815077066422,
-      "step": 3230
-    },
-    {
-      "epoch": 0.2825499258742478,
-      "grad_norm": 1.5900872945785522,
-      "learning_rate": 9.476640437570562e-06,
-      "loss": 1.9656,
-      "mean_token_accuracy": 0.5522504895925522,
-      "step": 3240
-    },
-    {
-      "epoch": 0.2834219935466992,
-      "grad_norm": 1.4893672466278076,
-      "learning_rate": 9.415850373407342e-06,
-      "loss": 1.9023,
-      "mean_token_accuracy": 0.5624388486146927,
-      "step": 3250
-    },
-    {
-      "epoch": 0.28429406121915063,
-      "grad_norm": 1.6059247255325317,
-      "learning_rate": 9.355081962494815e-06,
-      "loss": 1.8516,
-      "mean_token_accuracy": 0.5750850677490235,
-      "step": 3260
-    },
-    {
-      "epoch": 0.285166128891602,
-      "grad_norm": 1.5298104286193848,
-      "learning_rate": 9.294337457395638e-06,
-      "loss": 1.8766,
-      "mean_token_accuracy": 0.570046478509903,
-      "step": 3270
-    },
-    {
-      "epoch": 0.2860381965640534,
-      "grad_norm": 1.6815232038497925,
-      "learning_rate": 9.233619109786332e-06,
-      "loss": 1.8902,
-      "mean_token_accuracy": 0.5696183949708938,
-      "step": 3280
-    },
-    {
-      "epoch": 0.28691026423650473,
-      "grad_norm": 1.4807692766189575,
-      "learning_rate": 9.172929170373804e-06,
-      "loss": 1.9305,
-      "mean_token_accuracy": 0.5596012711524964,
-      "step": 3290
-    },
-    {
-      "epoch": 0.28778233190895613,
-      "grad_norm": 1.568697214126587,
-      "learning_rate": 9.112269888811934e-06,
-      "loss": 1.9293,
-      "mean_token_accuracy": 0.5572040110826493,
-      "step": 3300
-    },
-    {
-      "epoch": 0.28778233190895613,
-      "eval_runtime": 4.1078,
-      "eval_samples_per_second": 267.056,
-      "eval_steps_per_second": 66.946,
-      "step": 3300
-    },
-    {
-      "epoch": 0.28865439958140754,
-      "grad_norm": 1.5465502738952637,
-      "learning_rate": 9.051643513618176e-06,
-      "loss": 1.8973,
-      "mean_token_accuracy": 0.56588796377182,
-      "step": 3310
-    },
-    {
-      "epoch": 0.2895264672538589,
-      "grad_norm": 1.442423939704895,
-      "learning_rate": 8.99105229209021e-06,
-      "loss": 1.9445,
-      "mean_token_accuracy": 0.5507950097322464,
-      "step": 3320
-    },
-    {
-      "epoch": 0.2903985349263103,
-      "grad_norm": 1.6043237447738647,
-      "learning_rate": 8.930498470222641e-06,
-      "loss": 1.9676,
-      "mean_token_accuracy": 0.5513820856809616,
-      "step": 3330
-    },
-    {
-      "epoch": 0.29127060259876164,
-      "grad_norm": 1.501587152481079,
-      "learning_rate": 8.86998429262374e-06,
-      "loss": 1.9148,
-      "mean_token_accuracy": 0.5600415855646134,
-      "step": 3340
-    },
-    {
-      "epoch": 0.29214267027121305,
-      "grad_norm": 1.4882675409317017,
-      "learning_rate": 8.809512002432252e-06,
-      "loss": 1.868,
-      "mean_token_accuracy": 0.5695083141326904,
-      "step": 3350
-    },
-    {
-      "epoch": 0.29301473794366445,
-      "grad_norm": 1.5801355838775635,
-      "learning_rate": 8.749083841234235e-06,
-      "loss": 1.9359,
-      "mean_token_accuracy": 0.5621697574853897,
-      "step": 3360
-    },
-    {
-      "epoch": 0.2938868056161158,
-      "grad_norm": 1.6375526189804077,
-      "learning_rate": 8.688702048979974e-06,
-      "loss": 1.925,
-      "mean_token_accuracy": 0.5591976523399353,
-      "step": 3370
-    },
-    {
-      "epoch": 0.2947588732885672,
-      "grad_norm": 1.4018661975860596,
-      "learning_rate": 8.628368863900954e-06,
-      "loss": 1.9105,
-      "mean_token_accuracy": 0.5577421754598617,
-      "step": 3380
-    },
-    {
-      "epoch": 0.29563094096101855,
-      "grad_norm": 1.4827812910079956,
-      "learning_rate": 8.568086522426884e-06,
-      "loss": 1.9266,
-      "mean_token_accuracy": 0.5585738748311997,
-      "step": 3390
-    },
-    {
-      "epoch": 0.29650300863346996,
-      "grad_norm": 1.2996610403060913,
-      "learning_rate": 8.507857259102814e-06,
-      "loss": 1.8879,
-      "mean_token_accuracy": 0.5667685896158219,
-      "step": 3400
-    },
-    {
-      "epoch": 0.29650300863346996,
-      "eval_runtime": 4.1056,
-      "eval_samples_per_second": 267.193,
-      "eval_steps_per_second": 66.981,
-      "step": 3400
-    },
-    {
-      "epoch": 0.29737507630592136,
-      "grad_norm": 1.5060701370239258,
-      "learning_rate": 8.447683306506279e-06,
-      "loss": 1.9336,
-      "mean_token_accuracy": 0.5541585147380829,
-      "step": 3410
-    },
-    {
-      "epoch": 0.2982471439783727,
-      "grad_norm": 1.4917477369308472,
-      "learning_rate": 8.387566895164566e-06,
-      "loss": 1.8973,
-      "mean_token_accuracy": 0.5613502949476242,
-      "step": 3420
-    },
-    {
-      "epoch": 0.2991192116508241,
-      "grad_norm": 1.6588315963745117,
-      "learning_rate": 8.327510253472023e-06,
-      "loss": 1.9473,
-      "mean_token_accuracy": 0.5537548959255219,
-      "step": 3430
-    },
-    {
-      "epoch": 0.29999127932327546,
-      "grad_norm": 1.47793447971344,
-      "learning_rate": 8.267515607607458e-06,
-      "loss": 1.8723,
-      "mean_token_accuracy": 0.569019079208374,
-      "step": 3440
-    },
-    {
-      "epoch": 0.30086334699572687,
-      "grad_norm": 1.564195990562439,
-      "learning_rate": 8.207585181451611e-06,
-      "loss": 1.8836,
-      "mean_token_accuracy": 0.5648728013038635,
-      "step": 3450
-    },
-    {
-      "epoch": 0.30173541466817827,
-      "grad_norm": 1.6584264039993286,
-      "learning_rate": 8.147721196504736e-06,
-      "loss": 1.8762,
-      "mean_token_accuracy": 0.5659858137369156,
-      "step": 3460
-    },
-    {
-      "epoch": 0.3026074823406296,
-      "grad_norm": 1.784829020500183,
-      "learning_rate": 8.08792587180424e-06,
-      "loss": 1.927,
-      "mean_token_accuracy": 0.5592588067054749,
-      "step": 3470
-    },
-    {
-      "epoch": 0.303479550013081,
-      "grad_norm": 1.4415394067764282,
-      "learning_rate": 8.028201423842437e-06,
-      "loss": 1.9102,
-      "mean_token_accuracy": 0.5570572435855865,
-      "step": 3480
-    },
-    {
-      "epoch": 0.30435161768553237,
-      "grad_norm": 1.4155422449111938,
-      "learning_rate": 7.96855006648438e-06,
-      "loss": 1.8848,
-      "mean_token_accuracy": 0.5659980446100235,
-      "step": 3490
-    },
-    {
-      "epoch": 0.3052236853579838,
-      "grad_norm": 1.5799750089645386,
-      "learning_rate": 7.908974010885795e-06,
-      "loss": 1.9414,
-      "mean_token_accuracy": 0.5574853241443634,
-      "step": 3500
-    },
-    {
-      "epoch": 0.3052236853579838,
-      "eval_runtime": 4.1026,
-      "eval_samples_per_second": 267.39,
-      "eval_steps_per_second": 67.03,
-      "step": 3500
-    },
-    {
-      "epoch": 0.3060957530304352,
-      "grad_norm": 1.3268228769302368,
-      "learning_rate": 7.849475465411136e-06,
-      "loss": 1.8734,
-      "mean_token_accuracy": 0.5663771957159043,
-      "step": 3510
-    },
-    {
-      "epoch": 0.30696782070288653,
-      "grad_norm": 1.3888028860092163,
-      "learning_rate": 7.790056635551704e-06,
-      "loss": 1.9031,
-      "mean_token_accuracy": 0.562769079208374,
-      "step": 3520
-    },
-    {
-      "epoch": 0.30783988837533793,
-      "grad_norm": 1.4219472408294678,
-      "learning_rate": 7.730719723843903e-06,
-      "loss": 1.9375,
-      "mean_token_accuracy": 0.5600415855646134,
-      "step": 3530
-    },
-    {
-      "epoch": 0.3087119560477893,
-      "grad_norm": 1.4739056825637817,
-      "learning_rate": 7.671466929787598e-06,
-      "loss": 1.8926,
-      "mean_token_accuracy": 0.5643590986728668,
-      "step": 3540
-    },
-    {
-      "epoch": 0.3095840237202407,
-      "grad_norm": 1.5766993761062622,
-      "learning_rate": 7.61230044976458e-06,
-      "loss": 1.9215,
-      "mean_token_accuracy": 0.5593933433294296,
-      "step": 3550
-    },
-    {
-      "epoch": 0.3104560913926921,
-      "grad_norm": 1.4130383729934692,
-      "learning_rate": 7.553222476957157e-06,
-      "loss": 1.9086,
-      "mean_token_accuracy": 0.5614970654249192,
-      "step": 3560
-    },
-    {
-      "epoch": 0.31132815906514344,
-      "grad_norm": 1.540473461151123,
-      "learning_rate": 7.494235201266849e-06,
-      "loss": 1.8969,
-      "mean_token_accuracy": 0.5651418775320053,
-      "step": 3570
-    },
-    {
-      "epoch": 0.31220022673759484,
-      "grad_norm": 1.446691870689392,
-      "learning_rate": 7.435340809233218e-06,
-      "loss": 1.9109,
-      "mean_token_accuracy": 0.5628129333257675,
-      "step": 3580
-    },
-    {
-      "epoch": 0.3130722944100462,
-      "grad_norm": 1.4547874927520752,
-      "learning_rate": 7.376541483952811e-06,
-      "loss": 1.9117,
-      "mean_token_accuracy": 0.5662915885448456,
-      "step": 3590
-    },
-    {
-      "epoch": 0.3139443620824976,
-      "grad_norm": 1.4632607698440552,
-      "learning_rate": 7.3178394049982485e-06,
-      "loss": 1.8863,
-      "mean_token_accuracy": 0.5654354155063629,
-      "step": 3600
-    },
-    {
-      "epoch": 0.3139443620824976,
-      "eval_runtime": 4.1018,
-      "eval_samples_per_second": 267.447,
-      "eval_steps_per_second": 67.045,
-      "step": 3600
-    },
-    {
-      "epoch": 0.314816429754949,
-      "grad_norm": 1.6280646324157715,
-      "learning_rate": 7.259236748337421e-06,
-      "loss": 1.8805,
-      "mean_token_accuracy": 0.5715141892433167,
-      "step": 3610
-    },
-    {
-      "epoch": 0.31568849742740035,
-      "grad_norm": 1.5691324472427368,
-      "learning_rate": 7.20073568625284e-06,
-      "loss": 1.9309,
-      "mean_token_accuracy": 0.5600538164377212,
-      "step": 3620
-    },
-    {
-      "epoch": 0.31656056509985175,
-      "grad_norm": 1.6000590324401855,
-      "learning_rate": 7.1423383872611045e-06,
-      "loss": 1.9258,
-      "mean_token_accuracy": 0.5595278859138488,
-      "step": 3630
-    },
-    {
-      "epoch": 0.31743263277230316,
-      "grad_norm": 1.5215147733688354,
-      "learning_rate": 7.084047016032528e-06,
-      "loss": 1.9176,
-      "mean_token_accuracy": 0.5602739751338959,
-      "step": 3640
-    },
-    {
-      "epoch": 0.3183047004447545,
-      "grad_norm": 1.5490972995758057,
-      "learning_rate": 7.025863733310894e-06,
-      "loss": 1.8602,
-      "mean_token_accuracy": 0.5704990208148957,
-      "step": 3650
-    },
-    {
-      "epoch": 0.3191767681172059,
-      "grad_norm": 1.5251253843307495,
-      "learning_rate": 6.967790695833363e-06,
-      "loss": 1.9211,
-      "mean_token_accuracy": 0.5594178080558777,
-      "step": 3660
-    },
-    {
-      "epoch": 0.32004883578965726,
-      "grad_norm": 1.5089037418365479,
-      "learning_rate": 6.909830056250527e-06,
-      "loss": 1.8621,
-      "mean_token_accuracy": 0.5697883903980255,
-      "step": 3670
-    },
-    {
-      "epoch": 0.32092090346210866,
-      "grad_norm": 1.489286184310913,
-      "learning_rate": 6.851983963046612e-06,
-      "loss": 1.8965,
-      "mean_token_accuracy": 0.5612646728754044,
-      "step": 3680
-    },
-    {
-      "epoch": 0.32179297113456007,
-      "grad_norm": 1.411023497581482,
-      "learning_rate": 6.794254560459843e-06,
-      "loss": 1.8848,
-      "mean_token_accuracy": 0.5651663362979888,
-      "step": 3690
-    },
-    {
-      "epoch": 0.3226650388070114,
-      "grad_norm": 1.4782873392105103,
-      "learning_rate": 6.736643988402958e-06,
-      "loss": 1.893,
-      "mean_token_accuracy": 0.5596257388591767,
-      "step": 3700
-    },
-    {
-      "epoch": 0.3226650388070114,
-      "eval_runtime": 4.109,
-      "eval_samples_per_second": 266.978,
-      "eval_steps_per_second": 66.927,
-      "step": 3700
-    },
-    {
-      "epoch": 0.3235371064794628,
-      "grad_norm": 1.4194209575653076,
-      "learning_rate": 6.679154382383883e-06,
-      "loss": 1.9293,
-      "mean_token_accuracy": 0.5588184893131256,
-      "step": 3710
-    },
-    {
-      "epoch": 0.32440917415191417,
-      "grad_norm": 1.4557321071624756,
-      "learning_rate": 6.621787873426581e-06,
-      "loss": 1.9383,
-      "mean_token_accuracy": 0.5550146788358689,
-      "step": 3720
-    },
-    {
-      "epoch": 0.3252812418243656,
-      "grad_norm": 1.4154465198516846,
-      "learning_rate": 6.564546587992054e-06,
-      "loss": 1.8824,
-      "mean_token_accuracy": 0.5645914882421493,
-      "step": 3730
-    },
-    {
-      "epoch": 0.326153309496817,
-      "grad_norm": 1.416695475578308,
-      "learning_rate": 6.507432647899519e-06,
-      "loss": 1.9559,
-      "mean_token_accuracy": 0.5520181059837341,
-      "step": 3740
-    },
-    {
-      "epoch": 0.3270253771692683,
-      "grad_norm": 1.4744834899902344,
-      "learning_rate": 6.450448170247757e-06,
-      "loss": 1.8687,
-      "mean_token_accuracy": 0.5670009821653366,
-      "step": 3750
-    },
-    {
-      "epoch": 0.32789744484171973,
-      "grad_norm": 1.5527193546295166,
-      "learning_rate": 6.393595267336639e-06,
-      "loss": 1.8965,
-      "mean_token_accuracy": 0.5659001976251602,
-      "step": 3760
-    },
-    {
-      "epoch": 0.3287695125141711,
-      "grad_norm": 1.592223048210144,
-      "learning_rate": 6.3368760465888226e-06,
-      "loss": 1.9238,
-      "mean_token_accuracy": 0.5569593936204911,
-      "step": 3770
-    },
-    {
-      "epoch": 0.3296415801866225,
-      "grad_norm": 1.4952565431594849,
-      "learning_rate": 6.280292610471639e-06,
-      "loss": 1.9148,
-      "mean_token_accuracy": 0.5601149737834931,
-      "step": 3780
-    },
-    {
-      "epoch": 0.3305136478590739,
-      "grad_norm": 1.5938102006912231,
-      "learning_rate": 6.223847056419154e-06,
-      "loss": 1.9199,
-      "mean_token_accuracy": 0.5570327907800674,
-      "step": 3790
-    },
-    {
-      "epoch": 0.33138571553152524,
-      "grad_norm": 1.5577000379562378,
-      "learning_rate": 6.1675414767544285e-06,
-      "loss": 1.9086,
-      "mean_token_accuracy": 0.5583659410476685,
-      "step": 3800
-    },
-    {
-      "epoch": 0.33138571553152524,
-      "eval_runtime": 4.1047,
-      "eval_samples_per_second": 267.253,
-      "eval_steps_per_second": 66.996,
-      "step": 3800
-    },
-    {
-      "epoch": 0.33225778320397664,
-      "grad_norm": 1.4879730939865112,
-      "learning_rate": 6.111377958611948e-06,
-      "loss": 1.8902,
-      "mean_token_accuracy": 0.5641389429569245,
-      "step": 3810
-    },
-    {
-      "epoch": 0.333129850876428,
-      "grad_norm": 1.506752371788025,
-      "learning_rate": 6.055358583860267e-06,
-      "loss": 1.8793,
-      "mean_token_accuracy": 0.5660102754831314,
-      "step": 3820
-    },
-    {
-      "epoch": 0.3340019185488794,
-      "grad_norm": 1.4874769449234009,
-      "learning_rate": 5.99948542902483e-06,
-      "loss": 1.8762,
-      "mean_token_accuracy": 0.5677103757858276,
-      "step": 3830
-    },
-    {
-      "epoch": 0.3348739862213308,
-      "grad_norm": 1.5656108856201172,
-      "learning_rate": 5.943760565211011e-06,
-      "loss": 1.9336,
-      "mean_token_accuracy": 0.5553571343421936,
-      "step": 3840
-    },
-    {
-      "epoch": 0.33574605389378215,
-      "grad_norm": 1.677775263786316,
-      "learning_rate": 5.8881860580273285e-06,
-      "loss": 1.9254,
-      "mean_token_accuracy": 0.558292555809021,
-      "step": 3850
-    },
-    {
-      "epoch": 0.33661812156623355,
-      "grad_norm": 1.461595058441162,
-      "learning_rate": 5.832763967508885e-06,
-      "loss": 1.9734,
-      "mean_token_accuracy": 0.5467954963445664,
-      "step": 3860
-    },
-    {
-      "epoch": 0.3374901892386849,
-      "grad_norm": 1.462888479232788,
-      "learning_rate": 5.777496348041009e-06,
-      "loss": 1.8762,
-      "mean_token_accuracy": 0.5701810151338578,
-      "step": 3870
-    },
-    {
-      "epoch": 0.3383622569111363,
-      "grad_norm": 1.425092101097107,
-      "learning_rate": 5.722385248283092e-06,
-      "loss": 1.9031,
-      "mean_token_accuracy": 0.5621208429336548,
-      "step": 3880
-    },
-    {
-      "epoch": 0.3392343245835877,
-      "grad_norm": 1.4690195322036743,
-      "learning_rate": 5.667432711092651e-06,
-      "loss": 1.9172,
-      "mean_token_accuracy": 0.5556506812572479,
-      "step": 3890
-    },
-    {
-      "epoch": 0.34010639225603906,
-      "grad_norm": 1.5467807054519653,
-      "learning_rate": 5.61264077344962e-06,
-      "loss": 1.9305,
-      "mean_token_accuracy": 0.5574241682887078,
-      "step": 3900
-    },
-    {
-      "epoch": 0.34010639225603906,
-      "eval_runtime": 4.1085,
-      "eval_samples_per_second": 267.005,
-      "eval_steps_per_second": 66.934,
-      "step": 3900
-    },
-    {
-      "epoch": 0.34097845992849046,
-      "grad_norm": 1.5516598224639893,
-      "learning_rate": 5.558011466380824e-06,
-      "loss": 1.8699,
-      "mean_token_accuracy": 0.5630993127822876,
-      "step": 3910
-    },
-    {
-      "epoch": 0.3418505276009418,
-      "grad_norm": 1.448142647743225,
-      "learning_rate": 5.5035468148846926e-06,
-      "loss": 1.9035,
-      "mean_token_accuracy": 0.5598336637020112,
-      "step": 3920
-    },
-    {
-      "epoch": 0.3427225952733932,
-      "grad_norm": 1.3374143838882446,
-      "learning_rate": 5.449248837856224e-06,
-      "loss": 1.8684,
-      "mean_token_accuracy": 0.5692759335041047,
-      "step": 3930
-    },
-    {
-      "epoch": 0.3435946629458446,
-      "grad_norm": 1.5164045095443726,
-      "learning_rate": 5.395119548012112e-06,
-      "loss": 1.932,
-      "mean_token_accuracy": 0.5567636907100677,
-      "step": 3940
-    },
-    {
-      "epoch": 0.34446673061829597,
-      "grad_norm": 1.4611291885375977,
-      "learning_rate": 5.34116095181616e-06,
-      "loss": 1.9012,
-      "mean_token_accuracy": 0.5622798442840576,
-      "step": 3950
-    },
-    {
-      "epoch": 0.3453387982907474,
-      "grad_norm": 1.498421311378479,
-      "learning_rate": 5.287375049404909e-06,
-      "loss": 1.8938,
-      "mean_token_accuracy": 0.5640777856111526,
-      "step": 3960
-    },
-    {
-      "epoch": 0.3462108659631987,
-      "grad_norm": 1.4730089902877808,
-      "learning_rate": 5.233763834513479e-06,
-      "loss": 1.8672,
-      "mean_token_accuracy": 0.5678571403026581,
-      "step": 3970
-    },
-    {
-      "epoch": 0.3470829336356501,
-      "grad_norm": 1.4765514135360718,
-      "learning_rate": 5.180329294401685e-06,
-      "loss": 1.8504,
-      "mean_token_accuracy": 0.5705724060535431,
-      "step": 3980
-    },
-    {
-      "epoch": 0.34795500130810153,
-      "grad_norm": 1.5854777097702026,
-      "learning_rate": 5.127073409780352e-06,
-      "loss": 1.9102,
-      "mean_token_accuracy": 0.5604452058672905,
-      "step": 3990
-    },
-    {
-      "epoch": 0.3488270689805529,
-      "grad_norm": 1.5666085481643677,
-      "learning_rate": 5.0739981547379215e-06,
-      "loss": 1.9187,
-      "mean_token_accuracy": 0.5581457853317261,
-      "step": 4000
-    },
-    {
-      "epoch": 0.3488270689805529,
-      "eval_runtime": 4.108,
-      "eval_samples_per_second": 267.04,
-      "eval_steps_per_second": 66.943,
-      "step": 4000
-    },
-    {
-      "epoch": 0.3496991366530043,
-      "grad_norm": 1.4239704608917236,
-      "learning_rate": 5.02110549666724e-06,
-      "loss": 1.8645,
-      "mean_token_accuracy": 0.5709270983934402,
-      "step": 4010
-    },
-    {
-      "epoch": 0.35057120432545563,
-      "grad_norm": 1.4216145277023315,
-      "learning_rate": 4.968397396192675e-06,
-      "loss": 1.8629,
-      "mean_token_accuracy": 0.5699975579977036,
-      "step": 4020
-    },
-    {
-      "epoch": 0.35144327199790704,
-      "grad_norm": 1.4933892488479614,
-      "learning_rate": 4.91587580709739e-06,
-      "loss": 1.8805,
-      "mean_token_accuracy": 0.5656066536903381,
-      "step": 4030
-    },
-    {
-      "epoch": 0.35231533967035844,
-      "grad_norm": 1.5485926866531372,
-      "learning_rate": 4.863542676250972e-06,
-      "loss": 1.8051,
-      "mean_token_accuracy": 0.5818126231431962,
-      "step": 4040
-    },
-    {
-      "epoch": 0.3531874073428098,
-      "grad_norm": 1.46506929397583,
-      "learning_rate": 4.811399943537223e-06,
-      "loss": 1.9133,
-      "mean_token_accuracy": 0.55625,
-      "step": 4050
-    },
-    {
-      "epoch": 0.3540594750152612,
-      "grad_norm": 1.4883263111114502,
-      "learning_rate": 4.759449541782272e-06,
-      "loss": 1.8926,
-      "mean_token_accuracy": 0.5676736831665039,
-      "step": 4060
-    },
-    {
-      "epoch": 0.35493154268771254,
-      "grad_norm": 1.4316681623458862,
-      "learning_rate": 4.707693396682936e-06,
-      "loss": 1.8789,
-      "mean_token_accuracy": 0.5686521470546723,
-      "step": 4070
-    },
-    {
-      "epoch": 0.35580361036016395,
-      "grad_norm": 1.6278921365737915,
-      "learning_rate": 4.656133426735315e-06,
-      "loss": 1.8848,
-      "mean_token_accuracy": 0.567025437951088,
-      "step": 4080
-    },
-    {
-      "epoch": 0.35667567803261535,
-      "grad_norm": 1.4322975873947144,
-      "learning_rate": 4.604771543163706e-06,
-      "loss": 1.9168,
-      "mean_token_accuracy": 0.5565557777881622,
-      "step": 4090
-    },
-    {
-      "epoch": 0.3575477457050667,
-      "grad_norm": 1.4468016624450684,
-      "learning_rate": 4.5536096498497295e-06,
-      "loss": 1.9434,
-      "mean_token_accuracy": 0.553241191804409,
-      "step": 4100
-    },
-    {
-      "epoch": 0.3575477457050667,
-      "eval_runtime": 4.1031,
-      "eval_samples_per_second": 267.359,
-      "eval_steps_per_second": 67.023,
-      "step": 4100
-    },
-    {
-      "epoch": 0.3584198133775181,
-      "grad_norm": 1.4846084117889404,
-      "learning_rate": 4.502649643261779e-06,
-      "loss": 1.8633,
-      "mean_token_accuracy": 0.5689456850290299,
-      "step": 4110
-    },
-    {
-      "epoch": 0.35929188104996945,
-      "grad_norm": 1.613488793373108,
-      "learning_rate": 4.451893412384707e-06,
-      "loss": 1.941,
-      "mean_token_accuracy": 0.5556629240512848,
-      "step": 4120
-    },
-    {
-      "epoch": 0.36016394872242086,
-      "grad_norm": 1.6922272443771362,
-      "learning_rate": 4.401342838649818e-06,
-      "loss": 1.8359,
-      "mean_token_accuracy": 0.576259782910347,
-      "step": 4130
-    },
-    {
-      "epoch": 0.36103601639487226,
-      "grad_norm": 1.6881250143051147,
-      "learning_rate": 4.350999795865109e-06,
-      "loss": 1.8656,
-      "mean_token_accuracy": 0.5724437385797501,
-      "step": 4140
-    },
-    {
-      "epoch": 0.3619080840673236,
-      "grad_norm": 1.455878496170044,
-      "learning_rate": 4.300866150145837e-06,
-      "loss": 1.898,
-      "mean_token_accuracy": 0.5677959859371186,
-      "step": 4150
-    },
-    {
-      "epoch": 0.362780151739775,
-      "grad_norm": 1.4437832832336426,
-      "learning_rate": 4.250943759845316e-06,
-      "loss": 1.8629,
-      "mean_token_accuracy": 0.5684809237718582,
-      "step": 4160
-    },
-    {
-      "epoch": 0.36365221941222636,
-      "grad_norm": 1.4966909885406494,
-      "learning_rate": 4.201234475486063e-06,
-      "loss": 1.8715,
-      "mean_token_accuracy": 0.5676981419324875,
-      "step": 4170
-    },
-    {
-      "epoch": 0.36452428708467777,
-      "grad_norm": 1.3970211744308472,
-      "learning_rate": 4.1517401396911725e-06,
-      "loss": 1.9258,
-      "mean_token_accuracy": 0.5565313100814819,
-      "step": 4180
-    },
-    {
-      "epoch": 0.36539635475712917,
-      "grad_norm": 1.3464200496673584,
-      "learning_rate": 4.1024625871160325e-06,
-      "loss": 1.8695,
-      "mean_token_accuracy": 0.5691413849592208,
-      "step": 4190
-    },
-    {
-      "epoch": 0.3662684224295805,
-      "grad_norm": 1.4210983514785767,
-      "learning_rate": 4.053403644380321e-06,
-      "loss": 1.8594,
-      "mean_token_accuracy": 0.5667196661233902,
-      "step": 4200
-    },
-    {
-      "epoch": 0.3662684224295805,
-      "eval_runtime": 4.1086,
-      "eval_samples_per_second": 267.004,
-      "eval_steps_per_second": 66.934,
-      "step": 4200
-    },
-    {
-      "epoch": 0.3671404901020319,
-      "grad_norm": 1.4295787811279297,
-      "learning_rate": 4.004565130000277e-06,
-      "loss": 1.8941,
-      "mean_token_accuracy": 0.5629770070314407,
-      "step": 4210
-    },
-    {
-      "epoch": 0.3680125577744833,
-      "grad_norm": 1.4431244134902954,
-      "learning_rate": 3.955948854321321e-06,
-      "loss": 1.8938,
-      "mean_token_accuracy": 0.5641878634691239,
-      "step": 4220
-    },
-    {
-      "epoch": 0.3688846254469347,
-      "grad_norm": 1.4570814371109009,
-      "learning_rate": 3.907556619450909e-06,
-      "loss": 1.8941,
-      "mean_token_accuracy": 0.5639432519674301,
-      "step": 4230
-    },
-    {
-      "epoch": 0.3697566931193861,
-      "grad_norm": 1.5730128288269043,
-      "learning_rate": 3.859390219191775e-06,
-      "loss": 1.8902,
-      "mean_token_accuracy": 0.5635029405355454,
-      "step": 4240
-    },
-    {
-      "epoch": 0.37062876079183743,
-      "grad_norm": 1.4484063386917114,
-      "learning_rate": 3.8114514389754098e-06,
-      "loss": 1.8863,
-      "mean_token_accuracy": 0.5642490148544311,
-      "step": 4250
-    },
-    {
-      "epoch": 0.37150082846428883,
-      "grad_norm": 1.4842562675476074,
-      "learning_rate": 3.7637420557958927e-06,
-      "loss": 1.8777,
-      "mean_token_accuracy": 0.5692514628171921,
-      "step": 4260
-    },
-    {
-      "epoch": 0.3723728961367402,
-      "grad_norm": 1.359551191329956,
-      "learning_rate": 3.7162638381440077e-06,
-      "loss": 1.875,
-      "mean_token_accuracy": 0.5667930454015732,
-      "step": 4270
-    },
-    {
-      "epoch": 0.3732449638091916,
-      "grad_norm": 1.4525691270828247,
-      "learning_rate": 3.6690185459417107e-06,
-      "loss": 1.8914,
-      "mean_token_accuracy": 0.5661203473806381,
-      "step": 4280
-    },
-    {
-      "epoch": 0.374117031481643,
-      "grad_norm": 1.409770131111145,
-      "learning_rate": 3.622007930476865e-06,
-      "loss": 1.8371,
-      "mean_token_accuracy": 0.5788649708032608,
-      "step": 4290
-    },
-    {
-      "epoch": 0.37498909915409434,
-      "grad_norm": 1.468183994293213,
-      "learning_rate": 3.575233734338356e-06,
-      "loss": 1.9273,
-      "mean_token_accuracy": 0.5551247596740723,
-      "step": 4300
-    },
-    {
-      "epoch": 0.37498909915409434,
-      "eval_runtime": 4.1065,
-      "eval_samples_per_second": 267.135,
-      "eval_steps_per_second": 66.966,
-      "step": 4300
-    },
-    {
-      "epoch": 0.37586116682654575,
-      "grad_norm": 1.4434620141983032,
-      "learning_rate": 3.528697691351465e-06,
-      "loss": 1.9254,
-      "mean_token_accuracy": 0.5575464814901352,
-      "step": 4310
-    },
-    {
-      "epoch": 0.37673323449899715,
-      "grad_norm": 1.4546566009521484,
-      "learning_rate": 3.4824015265136278e-06,
-      "loss": 1.9082,
-      "mean_token_accuracy": 0.5626589953899384,
-      "step": 4320
-    },
-    {
-      "epoch": 0.3776053021714485,
-      "grad_norm": 1.4952583312988281,
-      "learning_rate": 3.436346955930472e-06,
-      "loss": 1.8832,
-      "mean_token_accuracy": 0.5677103757858276,
-      "step": 4330
-    },
-    {
-      "epoch": 0.3784773698438999,
-      "grad_norm": 1.6326318979263306,
-      "learning_rate": 3.3905356867522187e-06,
-      "loss": 1.9598,
-      "mean_token_accuracy": 0.551822406053543,
-      "step": 4340
-    },
-    {
-      "epoch": 0.37934943751635125,
-      "grad_norm": 1.5317000150680542,
-      "learning_rate": 3.344969417110391e-06,
-      "loss": 1.9031,
-      "mean_token_accuracy": 0.5641022503376008,
-      "step": 4350
-    },
-    {
-      "epoch": 0.38022150518880266,
-      "grad_norm": 1.435085415840149,
-      "learning_rate": 3.29964983605487e-06,
-      "loss": 1.9367,
-      "mean_token_accuracy": 0.555748525261879,
-      "step": 4360
-    },
-    {
-      "epoch": 0.38109357286125406,
-      "grad_norm": 1.5483328104019165,
-      "learning_rate": 3.2545786234913e-06,
-      "loss": 1.8453,
-      "mean_token_accuracy": 0.5711594849824906,
-      "step": 4370
-    },
-    {
-      "epoch": 0.3819656405337054,
-      "grad_norm": 1.4388127326965332,
-      "learning_rate": 3.2097574501187877e-06,
-      "loss": 1.8969,
-      "mean_token_accuracy": 0.5651296466588974,
-      "step": 4380
-    },
-    {
-      "epoch": 0.3828377082061568,
-      "grad_norm": 1.4018774032592773,
-      "learning_rate": 3.165187977368007e-06,
-      "loss": 1.8625,
-      "mean_token_accuracy": 0.5673434406518936,
-      "step": 4390
-    },
-    {
-      "epoch": 0.38370977587860816,
-      "grad_norm": 1.757211446762085,
-      "learning_rate": 3.120871857339582e-06,
-      "loss": 1.9324,
-      "mean_token_accuracy": 0.5543664380908012,
-      "step": 4400
-    },
-    {
-      "epoch": 0.38370977587860816,
-      "eval_runtime": 4.0917,
-      "eval_samples_per_second": 268.104,
-      "eval_steps_per_second": 67.209,
-      "step": 4400
-    },
-    {
-      "epoch": 0.38458184355105957,
-      "grad_norm": 1.48627769947052,
-      "learning_rate": 3.0768107327428766e-06,
-      "loss": 1.8559,
-      "mean_token_accuracy": 0.5772627204656601,
-      "step": 4410
-    },
-    {
-      "epoch": 0.38545391122351097,
-      "grad_norm": 1.4570212364196777,
-      "learning_rate": 3.033006236835071e-06,
-      "loss": 1.8828,
-      "mean_token_accuracy": 0.5623532295227051,
-      "step": 4420
-    },
-    {
-      "epoch": 0.3863259788959623,
-      "grad_norm": 1.6207778453826904,
-      "learning_rate": 2.9894599933606518e-06,
-      "loss": 1.877,
-      "mean_token_accuracy": 0.5685176163911819,
-      "step": 4430
-    },
-    {
-      "epoch": 0.3871980465684137,
-      "grad_norm": 1.420271635055542,
-      "learning_rate": 2.9461736164911934e-06,
-      "loss": 1.8277,
-      "mean_token_accuracy": 0.5782901138067246,
-      "step": 4440
-    },
-    {
-      "epoch": 0.38807011424086507,
-      "grad_norm": 1.6583901643753052,
-      "learning_rate": 2.903148710765552e-06,
-      "loss": 1.9031,
-      "mean_token_accuracy": 0.5580968677997589,
-      "step": 4450
-    },
-    {
-      "epoch": 0.3889421819133165,
-      "grad_norm": 1.5385041236877441,
-      "learning_rate": 2.8603868710303662e-06,
-      "loss": 1.843,
-      "mean_token_accuracy": 0.5741682916879653,
-      "step": 4460
-    },
-    {
-      "epoch": 0.3898142495857679,
-      "grad_norm": 1.5565297603607178,
-      "learning_rate": 2.8178896823809465e-06,
-      "loss": 1.8918,
-      "mean_token_accuracy": 0.56193737834692,
-      "step": 4470
-    },
-    {
-      "epoch": 0.39068631725821923,
-      "grad_norm": 1.5299102067947388,
-      "learning_rate": 2.7756587201025297e-06,
-      "loss": 1.8813,
-      "mean_token_accuracy": 0.5646159529685975,
-      "step": 4480
-    },
-    {
-      "epoch": 0.39155838493067063,
-      "grad_norm": 1.6014642715454102,
-      "learning_rate": 2.7336955496118666e-06,
-      "loss": 1.9051,
-      "mean_token_accuracy": 0.560237281024456,
-      "step": 4490
-    },
-    {
-      "epoch": 0.392430452603122,
-      "grad_norm": 1.5329208374023438,
-      "learning_rate": 2.692001726399215e-06,
-      "loss": 1.8977,
-      "mean_token_accuracy": 0.5593077346682549,
-      "step": 4500
-    },
-    {
-      "epoch": 0.392430452603122,
-      "eval_runtime": 4.1069,
-      "eval_samples_per_second": 267.113,
-      "eval_steps_per_second": 66.961,
-      "step": 4500
-    },
-    {
-      "epoch": 0.3933025202755734,
-      "grad_norm": 1.498526930809021,
-      "learning_rate": 2.6505787959706607e-06,
-      "loss": 1.8527,
-      "mean_token_accuracy": 0.5704745650291443,
-      "step": 4510
-    },
-    {
-      "epoch": 0.3941745879480248,
-      "grad_norm": 1.463271975517273,
-      "learning_rate": 2.609428293790852e-06,
-      "loss": 1.8305,
-      "mean_token_accuracy": 0.5753424674272537,
-      "step": 4520
-    },
-    {
-      "epoch": 0.39504665562047614,
-      "grad_norm": 1.7243947982788086,
-      "learning_rate": 2.5685517452260566e-06,
-      "loss": 1.9227,
-      "mean_token_accuracy": 0.5585371851921082,
-      "step": 4530
-    },
-    {
-      "epoch": 0.39591872329292754,
-      "grad_norm": 1.3657575845718384,
-      "learning_rate": 2.5279506654876473e-06,
-      "loss": 1.8477,
-      "mean_token_accuracy": 0.5690313071012497,
-      "step": 4540
-    },
-    {
-      "epoch": 0.3967907909653789,
-      "grad_norm": 1.5750689506530762,
-      "learning_rate": 2.487626559575911e-06,
-      "loss": 1.9074,
-      "mean_token_accuracy": 0.5619496077299118,
-      "step": 4550
-    },
-    {
-      "epoch": 0.3976628586378303,
-      "grad_norm": 1.6887179613113403,
-      "learning_rate": 2.4475809222242775e-06,
-      "loss": 1.8961,
-      "mean_token_accuracy": 0.565631115436554,
-      "step": 4560
-    },
-    {
-      "epoch": 0.3985349263102817,
-      "grad_norm": 1.399027943611145,
-      "learning_rate": 2.4078152378439033e-06,
-      "loss": 1.9125,
-      "mean_token_accuracy": 0.5578767091035843,
-      "step": 4570
-    },
-    {
-      "epoch": 0.39940699398273305,
-      "grad_norm": 1.68111252784729,
-      "learning_rate": 2.3683309804686604e-06,
-      "loss": 1.8574,
-      "mean_token_accuracy": 0.5706335574388504,
-      "step": 4580
-    },
-    {
-      "epoch": 0.40027906165518445,
-      "grad_norm": 1.382102370262146,
-      "learning_rate": 2.329129613700478e-06,
-      "loss": 1.9242,
-      "mean_token_accuracy": 0.5523483395576477,
-      "step": 4590
-    },
-    {
-      "epoch": 0.4011511293276358,
-      "grad_norm": 1.4959888458251953,
-      "learning_rate": 2.29021259065511e-06,
-      "loss": 1.8902,
-      "mean_token_accuracy": 0.561656066775322,
-      "step": 4600
-    },
-    {
-      "epoch": 0.4011511293276358,
-      "eval_runtime": 4.11,
-      "eval_samples_per_second": 266.913,
-      "eval_steps_per_second": 66.911,
-      "step": 4600
-    },
-    {
-      "epoch": 0.4020231970000872,
-      "grad_norm": 1.4049890041351318,
-      "learning_rate": 2.251581353908252e-06,
-      "loss": 1.9051,
-      "mean_token_accuracy": 0.5593566566705703,
-      "step": 4610
-    },
-    {
-      "epoch": 0.4028952646725386,
-      "grad_norm": 1.4973894357681274,
-      "learning_rate": 2.2132373354420833e-06,
-      "loss": 1.8988,
-      "mean_token_accuracy": 0.5606775879859924,
-      "step": 4620
-    },
-    {
-      "epoch": 0.40376733234498996,
-      "grad_norm": 1.5170193910598755,
-      "learning_rate": 2.1751819565921774e-06,
-      "loss": 1.9172,
-      "mean_token_accuracy": 0.558891874551773,
-      "step": 4630
-    },
-    {
-      "epoch": 0.40463940001744136,
-      "grad_norm": 1.368288516998291,
-      "learning_rate": 2.137416627994814e-06,
-      "loss": 1.8676,
-      "mean_token_accuracy": 0.5697040140628815,
-      "step": 4640
-    },
-    {
-      "epoch": 0.4055114676898927,
-      "grad_norm": 1.5546929836273193,
-      "learning_rate": 2.0999427495347035e-06,
-      "loss": 1.8969,
-      "mean_token_accuracy": 0.563894322514534,
-      "step": 4650
-    },
-    {
-      "epoch": 0.4063835353623441,
-      "grad_norm": 1.4439570903778076,
-      "learning_rate": 2.0627617102930753e-06,
-      "loss": 1.9043,
-      "mean_token_accuracy": 0.5592098832130432,
-      "step": 4660
-    },
-    {
-      "epoch": 0.4072556030347955,
-      "grad_norm": 1.3387343883514404,
-      "learning_rate": 2.02587488849621e-06,
-      "loss": 1.8578,
-      "mean_token_accuracy": 0.5713796466588974,
-      "step": 4670
-    },
-    {
-      "epoch": 0.40812767070724687,
-      "grad_norm": 1.4414863586425781,
-      "learning_rate": 1.989283651464329e-06,
-      "loss": 1.9059,
-      "mean_token_accuracy": 0.5573385506868362,
-      "step": 4680
-    },
-    {
-      "epoch": 0.4089997383796983,
-      "grad_norm": 1.4976356029510498,
-      "learning_rate": 1.952989355560929e-06,
-      "loss": 1.8551,
-      "mean_token_accuracy": 0.5743762284517289,
-      "step": 4690
-    },
-    {
-      "epoch": 0.4098718060521496,
-      "grad_norm": 1.5303518772125244,
-      "learning_rate": 1.9169933461424928e-06,
-      "loss": 1.9039,
-      "mean_token_accuracy": 0.5582803308963775,
-      "step": 4700
-    },
-    {
-      "epoch": 0.4098718060521496,
-      "eval_runtime": 4.114,
-      "eval_samples_per_second": 266.652,
-      "eval_steps_per_second": 66.845,
-      "step": 4700
-    },
-    {
-      "epoch": 0.410743873724601,
-      "grad_norm": 1.329714059829712,
-      "learning_rate": 1.8812969575086272e-06,
-      "loss": 1.8828,
-      "mean_token_accuracy": 0.5661570459604264,
-      "step": 4710
-    },
-    {
-      "epoch": 0.41161594139705243,
-      "grad_norm": 1.4766762256622314,
-      "learning_rate": 1.8459015128525937e-06,
-      "loss": 1.8074,
-      "mean_token_accuracy": 0.5791462808847427,
-      "step": 4720
-    },
-    {
-      "epoch": 0.4124880090695038,
-      "grad_norm": 1.4960500001907349,
-      "learning_rate": 1.8108083242122764e-06,
-      "loss": 1.8648,
-      "mean_token_accuracy": 0.5667930454015732,
-      "step": 4730
-    },
-    {
-      "epoch": 0.4133600767419552,
-      "grad_norm": 1.515897274017334,
-      "learning_rate": 1.7760186924215239e-06,
-      "loss": 1.8172,
-      "mean_token_accuracy": 0.5795376718044281,
-      "step": 4740
-    },
-    {
-      "epoch": 0.41423214441440653,
-      "grad_norm": 1.44707453250885,
-      "learning_rate": 1.7415339070619586e-06,
-      "loss": 1.8941,
-      "mean_token_accuracy": 0.5631115436553955,
-      "step": 4750
-    },
-    {
-      "epoch": 0.41510421208685794,
-      "grad_norm": 1.5383362770080566,
-      "learning_rate": 1.7073552464151465e-06,
-      "loss": 1.8813,
-      "mean_token_accuracy": 0.5675880581140518,
-      "step": 4760
-    },
-    {
-      "epoch": 0.41597627975930934,
-      "grad_norm": 1.4443554878234863,
-      "learning_rate": 1.6734839774152322e-06,
-      "loss": 1.923,
-      "mean_token_accuracy": 0.5600538164377212,
-      "step": 4770
-    },
-    {
-      "epoch": 0.4168483474317607,
-      "grad_norm": 1.3475011587142944,
-      "learning_rate": 1.6399213556019732e-06,
-      "loss": 1.8531,
-      "mean_token_accuracy": 0.5707314044237137,
-      "step": 4780
-    },
-    {
-      "epoch": 0.4177204151042121,
-      "grad_norm": 1.4956789016723633,
-      "learning_rate": 1.6066686250741904e-06,
-      "loss": 1.8625,
-      "mean_token_accuracy": 0.5720768064260483,
-      "step": 4790
-    },
-    {
-      "epoch": 0.41859248277666344,
-      "grad_norm": 1.4578595161437988,
-      "learning_rate": 1.573727018443667e-06,
-      "loss": 1.8871,
-      "mean_token_accuracy": 0.5650440245866776,
-      "step": 4800
-    },
-    {
-      "epoch": 0.41859248277666344,
-      "eval_runtime": 4.1085,
-      "eval_samples_per_second": 267.009,
-      "eval_steps_per_second": 66.935,
-      "step": 4800
-    },
-    {
-      "epoch": 0.41946455044911485,
-      "grad_norm": 1.5240708589553833,
-      "learning_rate": 1.5410977567894403e-06,
-      "loss": 1.9082,
-      "mean_token_accuracy": 0.5567025423049927,
-      "step": 4810
-    },
-    {
-      "epoch": 0.42033661812156625,
-      "grad_norm": 1.4677386283874512,
-      "learning_rate": 1.5087820496125595e-06,
-      "loss": 1.9078,
-      "mean_token_accuracy": 0.563490703701973,
-      "step": 4820
-    },
-    {
-      "epoch": 0.4212086857940176,
-      "grad_norm": 1.5715758800506592,
-      "learning_rate": 1.4767810947912275e-06,
-      "loss": 1.9012,
-      "mean_token_accuracy": 0.5576076328754425,
-      "step": 4830
-    },
-    {
-      "epoch": 0.422080753466469,
-      "grad_norm": 1.4171923398971558,
-      "learning_rate": 1.4450960785364244e-06,
-      "loss": 1.9105,
-      "mean_token_accuracy": 0.5567759364843369,
-      "step": 4840
-    },
-    {
-      "epoch": 0.42295282113892035,
-      "grad_norm": 1.49934983253479,
-      "learning_rate": 1.4137281753479092e-06,
-      "loss": 1.9234,
-      "mean_token_accuracy": 0.5603106647729874,
-      "step": 4850
-    },
-    {
-      "epoch": 0.42382488881137176,
-      "grad_norm": 1.5522711277008057,
-      "learning_rate": 1.3826785479707128e-06,
-      "loss": 1.9043,
-      "mean_token_accuracy": 0.5598581194877624,
-      "step": 4860
-    },
-    {
-      "epoch": 0.42469695648382316,
-      "grad_norm": 1.5112053155899048,
-      "learning_rate": 1.3519483473520124e-06,
-      "loss": 1.9227,
-      "mean_token_accuracy": 0.5539138972759247,
-      "step": 4870
-    },
-    {
-      "epoch": 0.4255690241562745,
-      "grad_norm": 1.4467791318893433,
-      "learning_rate": 1.3215387125984813e-06,
-      "loss": 1.8898,
-      "mean_token_accuracy": 0.5636863961815834,
-      "step": 4880
-    },
-    {
-      "epoch": 0.4264410918287259,
-      "grad_norm": 1.3098102807998657,
-      "learning_rate": 1.2914507709340596e-06,
-      "loss": 1.8742,
-      "mean_token_accuracy": 0.5659001886844635,
-      "step": 4890
-    },
-    {
-      "epoch": 0.42731315950117726,
-      "grad_norm": 1.4678281545639038,
-      "learning_rate": 1.2616856376581766e-06,
-      "loss": 1.8914,
-      "mean_token_accuracy": 0.5619863063097,
-      "step": 4900
-    },
-    {
-      "epoch": 0.42731315950117726,
-      "eval_runtime": 4.1108,
-      "eval_samples_per_second": 266.856,
-      "eval_steps_per_second": 66.897,
-      "step": 4900
-    },
-    {
-      "epoch": 0.42818522717362867,
-      "grad_norm": 1.418245792388916,
-      "learning_rate": 1.2322444161044e-06,
-      "loss": 1.9062,
-      "mean_token_accuracy": 0.5632705509662628,
-      "step": 4910
-    },
-    {
-      "epoch": 0.4290572948460801,
-      "grad_norm": 1.5337378978729248,
-      "learning_rate": 1.2031281975995467e-06,
-      "loss": 1.8906,
-      "mean_token_accuracy": 0.5617906093597412,
-      "step": 4920
-    },
-    {
-      "epoch": 0.4299293625185314,
-      "grad_norm": 1.6260164976119995,
-      "learning_rate": 1.1743380614232213e-06,
-      "loss": 1.9035,
-      "mean_token_accuracy": 0.5603106617927551,
-      "step": 4930
-    },
-    {
-      "epoch": 0.4308014301909828,
-      "grad_norm": 1.513599157333374,
-      "learning_rate": 1.1458750747678105e-06,
-      "loss": 1.8594,
-      "mean_token_accuracy": 0.5726394325494766,
-      "step": 4940
-    },
-    {
-      "epoch": 0.43167349786343423,
-      "grad_norm": 1.484406590461731,
-      "learning_rate": 1.1177402926989345e-06,
-      "loss": 1.891,
-      "mean_token_accuracy": 0.5662182003259659,
-      "step": 4950
-    },
-    {
-      "epoch": 0.4325455655358856,
-      "grad_norm": 1.408384919166565,
-      "learning_rate": 1.0899347581163222e-06,
-      "loss": 1.8996,
-      "mean_token_accuracy": 0.5644324779510498,
-      "step": 4960
-    },
-    {
-      "epoch": 0.433417633208337,
-      "grad_norm": 1.5410149097442627,
-      "learning_rate": 1.0624595017151685e-06,
-      "loss": 1.9098,
-      "mean_token_accuracy": 0.5581335574388504,
-      "step": 4970
-    },
-    {
-      "epoch": 0.43428970088078833,
-      "grad_norm": 1.6162104606628418,
-      "learning_rate": 1.0353155419479122e-06,
-      "loss": 1.8781,
-      "mean_token_accuracy": 0.5670254439115524,
-      "step": 4980
-    },
-    {
-      "epoch": 0.43516176855323974,
-      "grad_norm": 1.6093658208847046,
-      "learning_rate": 1.0085038849865025e-06,
-      "loss": 1.8883,
-      "mean_token_accuracy": 0.5637108623981476,
-      "step": 4990
-    },
-    {
-      "epoch": 0.43603383622569114,
-      "grad_norm": 1.3822484016418457,
-      "learning_rate": 9.820255246850853e-07,
-      "loss": 1.8945,
-      "mean_token_accuracy": 0.563845407962799,
-      "step": 5000
-    },
-    {
-      "epoch": 0.43603383622569114,
-      "eval_runtime": 4.1102,
-      "eval_samples_per_second": 266.899,
-      "eval_steps_per_second": 66.907,
-      "step": 5000
-    },
-    {
-      "epoch": 0.4369059038981425,
-      "grad_norm": 1.4728370904922485,
-      "learning_rate": 9.55881442543174e-07,
-      "loss": 1.9078,
-      "mean_token_accuracy": 0.5625978440046311,
-      "step": 5010
-    },
-    {
-      "epoch": 0.4377779715705939,
-      "grad_norm": 1.38115656375885,
-      "learning_rate": 9.30072607669259e-07,
-      "loss": 1.9062,
-      "mean_token_accuracy": 0.5597235798835755,
-      "step": 5020
-    },
-    {
-      "epoch": 0.43865003924304524,
-      "grad_norm": 1.4854556322097778,
-      "learning_rate": 9.045999767448988e-07,
-      "loss": 1.882,
-      "mean_token_accuracy": 0.5672089040279389,
-      "step": 5030
-    },
-    {
-      "epoch": 0.43952210691549665,
-      "grad_norm": 1.4074499607086182,
-      "learning_rate": 8.794644939892361e-07,
-      "loss": 1.9191,
-      "mean_token_accuracy": 0.5596624314785004,
-      "step": 5040
-    },
-    {
-      "epoch": 0.44039417458794805,
-      "grad_norm": 1.5734814405441284,
-      "learning_rate": 8.546670911240196e-07,
-      "loss": 1.9074,
-      "mean_token_accuracy": 0.5634173184633255,
-      "step": 5050
-    },
-    {
-      "epoch": 0.4412662422603994,
-      "grad_norm": 1.4320417642593384,
-      "learning_rate": 8.302086873390536e-07,
-      "loss": 1.9172,
-      "mean_token_accuracy": 0.5601027399301529,
-      "step": 5060
-    },
-    {
-      "epoch": 0.4421383099328508,
-      "grad_norm": 1.4159865379333496,
-      "learning_rate": 8.060901892581241e-07,
-      "loss": 1.9051,
-      "mean_token_accuracy": 0.5619251489639282,
-      "step": 5070
-    },
-    {
-      "epoch": 0.44301037760530215,
-      "grad_norm": 1.4995352029800415,
-      "learning_rate": 7.82312490905407e-07,
-      "loss": 1.8742,
-      "mean_token_accuracy": 0.5677103787660599,
-      "step": 5080
-    },
-    {
-      "epoch": 0.44388244527775356,
-      "grad_norm": 1.4121575355529785,
-      "learning_rate": 7.588764736723086e-07,
-      "loss": 1.8945,
-      "mean_token_accuracy": 0.5644447207450867,
-      "step": 5090
-    },
-    {
-      "epoch": 0.44475451295020496,
-      "grad_norm": 1.5248502492904663,
-      "learning_rate": 7.357830062848114e-07,
-      "loss": 1.8672,
-      "mean_token_accuracy": 0.5671966731548309,
-      "step": 5100
-    },
-    {
-      "epoch": 0.44475451295020496,
-      "eval_runtime": 4.1039,
-      "eval_samples_per_second": 267.307,
-      "eval_steps_per_second": 67.01,
-      "step": 5100
-    },
-    {
-      "epoch": 0.4456265806226563,
-      "grad_norm": 1.5539908409118652,
-      "learning_rate": 7.130329447712581e-07,
-      "loss": 1.8086,
-      "mean_token_accuracy": 0.5802103757858277,
-      "step": 5110
-    },
-    {
-      "epoch": 0.4464986482951077,
-      "grad_norm": 1.3804877996444702,
-      "learning_rate": 6.906271324306335e-07,
-      "loss": 1.898,
-      "mean_token_accuracy": 0.5583047866821289,
-      "step": 5120
-    },
-    {
-      "epoch": 0.44737071596755906,
-      "grad_norm": 1.5185437202453613,
-      "learning_rate": 6.685663998012926e-07,
-      "loss": 1.8848,
-      "mean_token_accuracy": 0.5646648734807969,
-      "step": 5130
-    },
-    {
-      "epoch": 0.44824278364001047,
-      "grad_norm": 1.5041149854660034,
-      "learning_rate": 6.468515646301865e-07,
-      "loss": 1.9,
-      "mean_token_accuracy": 0.5624755382537842,
-      "step": 5140
-    },
-    {
-      "epoch": 0.44911485131246187,
-      "grad_norm": 1.5699764490127563,
-      "learning_rate": 6.254834318425363e-07,
-      "loss": 1.898,
-      "mean_token_accuracy": 0.5597235798835755,
-      "step": 5150
-    },
-    {
-      "epoch": 0.4499869189849132,
-      "grad_norm": 1.4059476852416992,
-      "learning_rate": 6.044627935120107e-07,
-      "loss": 1.8836,
-      "mean_token_accuracy": 0.5645670294761658,
-      "step": 5160
-    },
-    {
-      "epoch": 0.4508589866573646,
-      "grad_norm": 1.4172667264938354,
-      "learning_rate": 5.837904288313545e-07,
-      "loss": 1.875,
-      "mean_token_accuracy": 0.5670988261699677,
-      "step": 5170
-    },
-    {
-      "epoch": 0.451731054329816,
-      "grad_norm": 1.5338667631149292,
-      "learning_rate": 5.634671040835104e-07,
-      "loss": 1.9406,
-      "mean_token_accuracy": 0.5525562673807144,
-      "step": 5180
-    },
-    {
-      "epoch": 0.4526031220022674,
-      "grad_norm": 1.732886791229248,
-      "learning_rate": 5.43493572613214e-07,
-      "loss": 1.9387,
-      "mean_token_accuracy": 0.5539505898952484,
-      "step": 5190
-    },
-    {
-      "epoch": 0.4534751896747188,
-      "grad_norm": 1.4870532751083374,
-      "learning_rate": 5.238705747990669e-07,
-      "loss": 1.8348,
-      "mean_token_accuracy": 0.5733610540628433,
-      "step": 5200
-    },
-    {
-      "epoch": 0.4534751896747188,
-      "eval_runtime": 4.1146,
-      "eval_samples_per_second": 266.609,
-      "eval_steps_per_second": 66.835,
-      "step": 5200
-    },
-    {
-      "epoch": 0.45434725734717013,
-      "grad_norm": 1.4174110889434814,
-      "learning_rate": 5.045988380260935e-07,
-      "loss": 1.8582,
-      "mean_token_accuracy": 0.568162915110588,
-      "step": 5210
-    },
-    {
-      "epoch": 0.45521932501962153,
-      "grad_norm": 1.4171583652496338,
-      "learning_rate": 4.856790766587815e-07,
-      "loss": 1.9176,
-      "mean_token_accuracy": 0.5638454034924507,
-      "step": 5220
-    },
-    {
-      "epoch": 0.4560913926920729,
-      "grad_norm": 1.4004004001617432,
-      "learning_rate": 4.6711199201459833e-07,
-      "loss": 1.8934,
-      "mean_token_accuracy": 0.5612157553434372,
-      "step": 5230
-    },
-    {
-      "epoch": 0.4569634603645243,
-      "grad_norm": 1.5932490825653076,
-      "learning_rate": 4.488982723379887e-07,
-      "loss": 1.8562,
-      "mean_token_accuracy": 0.5693126261234284,
-      "step": 5240
-    },
-    {
-      "epoch": 0.4578355280369757,
-      "grad_norm": 1.5290868282318115,
-      "learning_rate": 4.3103859277488056e-07,
-      "loss": 1.8496,
-      "mean_token_accuracy": 0.5724804371595382,
-      "step": 5250
-    },
-    {
-      "epoch": 0.45870759570942704,
-      "grad_norm": 1.4812133312225342,
-      "learning_rate": 4.1353361534763657e-07,
-      "loss": 1.8875,
-      "mean_token_accuracy": 0.5647015601396561,
-      "step": 5260
-    },
-    {
-      "epoch": 0.45957966338187844,
-      "grad_norm": 1.5241347551345825,
-      "learning_rate": 3.963839889305343e-07,
-      "loss": 1.8875,
-      "mean_token_accuracy": 0.56280577480793,
-      "step": 5270
-    },
-    {
-      "epoch": 0.4604517310543298,
-      "grad_norm": 1.5606677532196045,
-      "learning_rate": 3.7959034922569804e-07,
-      "loss": 1.916,
-      "mean_token_accuracy": 0.5564089983701705,
-      "step": 5280
-    },
-    {
-      "epoch": 0.4613237987267812,
-      "grad_norm": 1.390710711479187,
-      "learning_rate": 3.631533187395453e-07,
-      "loss": 1.8949,
-      "mean_token_accuracy": 0.562365460395813,
-      "step": 5290
-    },
-    {
-      "epoch": 0.4621958663992326,
-      "grad_norm": 1.4803236722946167,
-      "learning_rate": 3.470735067597053e-07,
-      "loss": 1.882,
-      "mean_token_accuracy": 0.5627935439348221,
-      "step": 5300
-    },
-    {
-      "epoch": 0.4621958663992326,
-      "eval_runtime": 4.1049,
-      "eval_samples_per_second": 267.244,
-      "eval_steps_per_second": 66.994,
-      "step": 5300
-    },
-    {
-      "epoch": 0.46306793407168395,
-      "grad_norm": 1.5395022630691528,
-      "learning_rate": 3.313515093324393e-07,
-      "loss": 1.8574,
-      "mean_token_accuracy": 0.5721135020256043,
-      "step": 5310
-    },
-    {
-      "epoch": 0.46394000174413536,
-      "grad_norm": 1.4910304546356201,
-      "learning_rate": 3.1598790924053936e-07,
-      "loss": 1.8723,
-      "mean_token_accuracy": 0.5678204536437989,
-      "step": 5320
-    },
-    {
-      "epoch": 0.4648120694165867,
-      "grad_norm": 1.573488712310791,
-      "learning_rate": 3.009832759817344e-07,
-      "loss": 1.9164,
-      "mean_token_accuracy": 0.5573263168334961,
-      "step": 5330
-    },
-    {
-      "epoch": 0.4656841370890381,
-      "grad_norm": 1.3606470823287964,
-      "learning_rate": 2.8633816574757166e-07,
-      "loss": 1.8629,
-      "mean_token_accuracy": 0.5678938329219818,
-      "step": 5340
-    },
-    {
-      "epoch": 0.4665562047614895,
-      "grad_norm": 1.450344204902649,
-      "learning_rate": 2.720531214028055e-07,
-      "loss": 1.8523,
-      "mean_token_accuracy": 0.5726761221885681,
-      "step": 5350
-    },
-    {
-      "epoch": 0.46742827243394086,
-      "grad_norm": 1.5320910215377808,
-      "learning_rate": 2.5812867246527207e-07,
-      "loss": 1.866,
-      "mean_token_accuracy": 0.5658512741327286,
-      "step": 5360
-    },
-    {
-      "epoch": 0.46830034010639227,
-      "grad_norm": 1.4426729679107666,
-      "learning_rate": 2.445653350862609e-07,
-      "loss": 1.9008,
-      "mean_token_accuracy": 0.5599437385797501,
-      "step": 5370
-    },
-    {
-      "epoch": 0.4691724077788436,
-      "grad_norm": 1.478639841079712,
-      "learning_rate": 2.3136361203138668e-07,
-      "loss": 1.9012,
-      "mean_token_accuracy": 0.5642245531082153,
-      "step": 5380
-    },
-    {
-      "epoch": 0.470044475451295,
-      "grad_norm": 1.4998114109039307,
-      "learning_rate": 2.1852399266194312e-07,
-      "loss": 1.8883,
-      "mean_token_accuracy": 0.5647871851921081,
-      "step": 5390
-    },
-    {
-      "epoch": 0.4709165431237464,
-      "grad_norm": 1.3921449184417725,
-      "learning_rate": 2.0604695291677523e-07,
-      "loss": 1.8711,
-      "mean_token_accuracy": 0.5666218250989914,
-      "step": 5400
-    },
-    {
-      "epoch": 0.4709165431237464,
-      "eval_runtime": 4.1108,
-      "eval_samples_per_second": 266.857,
-      "eval_steps_per_second": 66.897,
-      "step": 5400
-    },
-    {
-      "epoch": 0.47178861079619777,
-      "grad_norm": 1.5820482969284058,
-      "learning_rate": 1.9393295529462674e-07,
-      "loss": 1.8715,
-      "mean_token_accuracy": 0.5660225063562393,
-      "step": 5410
-    },
-    {
-      "epoch": 0.4726606784686492,
-      "grad_norm": 1.3940504789352417,
-      "learning_rate": 1.8218244883700386e-07,
-      "loss": 1.9039,
-      "mean_token_accuracy": 0.5611545979976654,
-      "step": 5420
-    },
-    {
-      "epoch": 0.4735327461411005,
-      "grad_norm": 1.3857200145721436,
-      "learning_rate": 1.7079586911152413e-07,
-      "loss": 1.8703,
-      "mean_token_accuracy": 0.5653131127357482,
-      "step": 5430
-    },
-    {
-      "epoch": 0.47440481381355193,
-      "grad_norm": 1.3877339363098145,
-      "learning_rate": 1.597736381957782e-07,
-      "loss": 1.8613,
-      "mean_token_accuracy": 0.5747431516647339,
-      "step": 5440
-    },
-    {
-      "epoch": 0.47527688148600333,
-      "grad_norm": 1.4594693183898926,
-      "learning_rate": 1.4911616466167345e-07,
-      "loss": 1.9023,
-      "mean_token_accuracy": 0.558867409825325,
-      "step": 5450
-    },
-    {
-      "epoch": 0.4761489491584547,
-      "grad_norm": 1.3865221738815308,
-      "learning_rate": 1.3882384356030066e-07,
-      "loss": 1.9434,
-      "mean_token_accuracy": 0.5540484338998795,
-      "step": 5460
-    },
-    {
-      "epoch": 0.4770210168309061,
-      "grad_norm": 1.4105465412139893,
-      "learning_rate": 1.2889705640728445e-07,
-      "loss": 1.848,
-      "mean_token_accuracy": 0.5747798383235931,
-      "step": 5470
-    },
-    {
-      "epoch": 0.47789308450335743,
-      "grad_norm": 1.4138177633285522,
-      "learning_rate": 1.1933617116863805e-07,
-      "loss": 1.9055,
-      "mean_token_accuracy": 0.5603106617927551,
-      "step": 5480
-    },
-    {
-      "epoch": 0.47876515217580884,
-      "grad_norm": 1.5450538396835327,
-      "learning_rate": 1.1014154224713302e-07,
-      "loss": 1.8957,
-      "mean_token_accuracy": 0.5671232879161835,
-      "step": 5490
-    },
-    {
-      "epoch": 0.47963721984826024,
-      "grad_norm": 1.4478673934936523,
-      "learning_rate": 1.0131351046915094e-07,
-      "loss": 1.8379,
-      "mean_token_accuracy": 0.5717954963445664,
-      "step": 5500
-    },
-    {
-      "epoch": 0.47963721984826024,
-      "eval_runtime": 4.1083,
-      "eval_samples_per_second": 267.02,
-      "eval_steps_per_second": 66.937,
-      "step": 5500
-    },
-    {
-      "epoch": 0.4805092875207116,
-      "grad_norm": 1.3963549137115479,
-      "learning_rate": 9.285240307206123e-08,
-      "loss": 1.8922,
-      "mean_token_accuracy": 0.5628913849592209,
-      "step": 5510
-    },
-    {
-      "epoch": 0.481381355193163,
-      "grad_norm": 1.4796730279922485,
-      "learning_rate": 8.475853369207753e-08,
-      "loss": 1.9172,
-      "mean_token_accuracy": 0.559552350640297,
-      "step": 5520
-    },
-    {
-      "epoch": 0.48225342286561435,
-      "grad_norm": 1.4058685302734375,
-      "learning_rate": 7.703220235264708e-08,
-      "loss": 1.8945,
-      "mean_token_accuracy": 0.5631971627473831,
-      "step": 5530
-    },
-    {
-      "epoch": 0.48312549053806575,
-      "grad_norm": 1.4438725709915161,
-      "learning_rate": 6.967369545331615e-08,
-      "loss": 1.8973,
-      "mean_token_accuracy": 0.5618884593248368,
-      "step": 5540
-    },
-    {
-      "epoch": 0.48399755821051715,
-      "grad_norm": 1.3665748834609985,
-      "learning_rate": 6.26832857591242e-08,
-      "loss": 1.8797,
-      "mean_token_accuracy": 0.5634662449359894,
-      "step": 5550
-    },
-    {
-      "epoch": 0.4848696258829685,
-      "grad_norm": 1.4306751489639282,
-      "learning_rate": 5.606123239048522e-08,
-      "loss": 1.8824,
-      "mean_token_accuracy": 0.5639339298009872,
-      "step": 5560
-    },
-    {
-      "epoch": 0.4857416935554199,
-      "grad_norm": 1.439139485359192,
-      "learning_rate": 4.9807780813586615e-08,
-      "loss": 1.8562,
-      "mean_token_accuracy": 0.5706457942724228,
-      "step": 5570
-    },
-    {
-      "epoch": 0.48661376122787126,
-      "grad_norm": 1.441157341003418,
-      "learning_rate": 4.392316283128861e-08,
-      "loss": 1.827,
-      "mean_token_accuracy": 0.5745107680559158,
-      "step": 5580
-    },
-    {
-      "epoch": 0.48748582890032266,
-      "grad_norm": 1.5141595602035522,
-      "learning_rate": 3.840759657453452e-08,
-      "loss": 1.8984,
-      "mean_token_accuracy": 0.5596257299184799,
-      "step": 5590
-    },
-    {
-      "epoch": 0.48835789657277406,
-      "grad_norm": 1.441131830215454,
-      "learning_rate": 3.326128649426053e-08,
-      "loss": 1.9004,
-      "mean_token_accuracy": 0.5624877750873566,
-      "step": 5600
-    },
-    {
-      "epoch": 0.48835789657277406,
-      "eval_runtime": 4.106,
-      "eval_samples_per_second": 267.168,
-      "eval_steps_per_second": 66.975,
-      "step": 5600
-    },
-    {
-      "epoch": 0.4892299642452254,
-      "grad_norm": 1.3957711458206177,
-      "learning_rate": 2.8484423353822842e-08,
-      "loss": 1.8438,
-      "mean_token_accuracy": 0.5737646847963334,
-      "step": 5610
-    },
-    {
-      "epoch": 0.4901020319176768,
-      "grad_norm": 1.561988115310669,
-      "learning_rate": 2.4077184221920068e-08,
-      "loss": 1.8184,
-      "mean_token_accuracy": 0.5775073438882827,
-      "step": 5620
-    },
-    {
-      "epoch": 0.4909740995901282,
-      "grad_norm": 1.4194732904434204,
-      "learning_rate": 2.003973246603508e-08,
-      "loss": 1.8398,
-      "mean_token_accuracy": 0.5733365982770919,
-      "step": 5630
-    },
-    {
-      "epoch": 0.49184616726257957,
-      "grad_norm": 1.5771911144256592,
-      "learning_rate": 1.637221774637765e-08,
-      "loss": 1.9125,
-      "mean_token_accuracy": 0.5587940245866776,
-      "step": 5640
-    },
-    {
-      "epoch": 0.492718234935031,
-      "grad_norm": 1.5131736993789673,
-      "learning_rate": 1.3074776010334466e-08,
-      "loss": 1.85,
-      "mean_token_accuracy": 0.5695205390453338,
-      "step": 5650
-    },
-    {
-      "epoch": 0.4935903026074823,
-      "grad_norm": 1.4102033376693726,
-      "learning_rate": 1.0147529487432028e-08,
-      "loss": 1.8414,
-      "mean_token_accuracy": 0.5746453076601028,
-      "step": 5660
-    },
-    {
-      "epoch": 0.4944623702799337,
-      "grad_norm": 1.5022151470184326,
-      "learning_rate": 7.590586684805834e-09,
-      "loss": 1.9211,
-      "mean_token_accuracy": 0.5547822833061218,
-      "step": 5670
-    },
-    {
-      "epoch": 0.49533443795238513,
-      "grad_norm": 1.432579517364502,
-      "learning_rate": 5.4040423831802635e-09,
-      "loss": 1.8719,
-      "mean_token_accuracy": 0.5684197664260864,
-      "step": 5680
-    },
-    {
-      "epoch": 0.4962065056248365,
-      "grad_norm": 1.4906742572784424,
-      "learning_rate": 3.587977633348061e-09,
-      "loss": 1.8551,
-      "mean_token_accuracy": 0.5734589099884033,
-      "step": 5690
-    },
-    {
-      "epoch": 0.4970785732972879,
-      "grad_norm": 1.3798710107803345,
-      "learning_rate": 2.1424597531749524e-09,
-      "loss": 1.9027,
-      "mean_token_accuracy": 0.5608488231897354,
-      "step": 5700
-    },
-    {
-      "epoch": 0.4970785732972879,
-      "eval_runtime": 4.1089,
-      "eval_samples_per_second": 266.979,
-      "eval_steps_per_second": 66.927,
-      "step": 5700
-    },
-    {
-      "epoch": 0.49795064096973923,
-      "grad_norm": 1.3952172994613647,
-      "learning_rate": 1.0675423250994244e-09,
-      "loss": 1.8711,
-      "mean_token_accuracy": 0.5696795523166657,
-      "step": 5710
-    },
-    {
-      "epoch": 0.49882270864219064,
-      "grad_norm": 1.5160133838653564,
-      "learning_rate": 3.6326519414431327e-10,
-      "loss": 1.8867,
-      "mean_token_accuracy": 0.5646648645401001,
-      "step": 5720
-    },
-    {
-      "epoch": 0.49969477631464204,
-      "grad_norm": 1.4322092533111572,
-      "learning_rate": 2.965446644798142e-11,
-      "loss": 1.9,
-      "mean_token_accuracy": 0.561533761024475,
-      "step": 5730
-    },
-    {
-      "epoch": 0.5000436033836225,
-      "mean_token_accuracy": 0.5637230947613716,
-      "step": 5734,
+      "epoch": 0.005057992500218017,
+      "mean_token_accuracy": 0.4595768079161644,
+      "step": 58,
       "total_flos": 0.0,
-      "train_loss": 2.016458841995117,
-      "train_runtime": 997.6055,
-      "train_samples_per_second": 91.954,
-      "train_steps_per_second": 5.748
+      "train_loss": 2.839574353448276,
+      "train_runtime": 8.9764,
+      "train_samples_per_second": 102.195,
+      "train_steps_per_second": 6.461
     }
   ],
   "logging_steps": 10,
-  "max_steps": 5734,
+  "max_steps": 58,
   "num_input_tokens_seen": 0,
   "num_train_epochs": 1,
   "save_steps": 500,