diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5026 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.5000436033836225,
+  "eval_steps": 100,
+  "global_step": 5734,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0008720676724513823,
+      "grad_norm": 12.805307388305664,
+      "learning_rate": 3.4843205574912896e-07,
+      "loss": 7.9406,
+      "mean_token_accuracy": 0.02709148731082678,
+      "step": 10
+    },
+    {
+      "epoch": 0.0017441353449027645,
+      "grad_norm": 11.99422550201416,
+      "learning_rate": 6.968641114982579e-07,
+      "loss": 7.9062,
+      "mean_token_accuracy": 0.029146282467991114,
+      "step": 20
+    },
+    {
+      "epoch": 0.002616203017354147,
+      "grad_norm": 10.550246238708496,
+      "learning_rate": 1.045296167247387e-06,
+      "loss": 7.9875,
+      "mean_token_accuracy": 0.027421722188591957,
+      "step": 30
+    },
+    {
+      "epoch": 0.003488270689805529,
+      "grad_norm": 12.254673957824707,
+      "learning_rate": 1.3937282229965158e-06,
+      "loss": 7.8672,
+      "mean_token_accuracy": 0.028926125168800353,
+      "step": 40
+    },
+    {
+      "epoch": 0.004360338362256911,
+      "grad_norm": 10.699244499206543,
+      "learning_rate": 1.742160278745645e-06,
+      "loss": 7.8734,
+      "mean_token_accuracy": 0.028449119441211222,
+      "step": 50
+    },
+    {
+      "epoch": 0.005232406034708294,
+      "grad_norm": 11.096329689025879,
+      "learning_rate": 2.090592334494774e-06,
+      "loss": 7.8344,
+      "mean_token_accuracy": 0.03172700610011816,
+      "step": 60
+    },
+    {
+      "epoch": 0.0061044737071596755,
+      "grad_norm": 9.731986999511719,
+      "learning_rate": 2.4390243902439027e-06,
+      "loss": 7.6797,
+      "mean_token_accuracy": 0.03680283790454268,
+      "step": 70
+    },
+    {
+      "epoch": 0.006976541379611058,
+      "grad_norm": 9.390852928161621,
+      "learning_rate": 2.7874564459930316e-06,
+      "loss": 7.5891,
+      "mean_token_accuracy": 0.04033757336437702,
+      "step": 80
+    },
+    {
+      "epoch": 0.00784860905206244,
+      "grad_norm": 6.090016841888428,
+      "learning_rate": 3.13588850174216e-06,
+      "loss": 7.4156,
+      "mean_token_accuracy": 0.047504892013967036,
+      "step": 90
+    },
+    {
+      "epoch": 0.008720676724513822,
+      "grad_norm": 7.438959121704102,
+      "learning_rate": 3.48432055749129e-06,
+      "loss": 7.2484,
+      "mean_token_accuracy": 0.049645303003489974,
+      "step": 100
+    },
+    {
+      "epoch": 0.008720676724513822,
+      "eval_runtime": 4.1519,
+      "eval_samples_per_second": 264.214,
+      "eval_steps_per_second": 66.234,
+      "step": 100
+    },
+    {
+      "epoch": 0.009592744396965204,
+      "grad_norm": 8.58280086517334,
+      "learning_rate": 3.832752613240418e-06,
+      "loss": 7.2125,
+      "mean_token_accuracy": 0.05371819958090782,
+      "step": 110
+    },
+    {
+      "epoch": 0.010464812069416587,
+      "grad_norm": 9.017817497253418,
+      "learning_rate": 4.181184668989548e-06,
+      "loss": 7.1484,
+      "mean_token_accuracy": 0.05711839552968741,
+      "step": 120
+    },
+    {
+      "epoch": 0.01133687974186797,
+      "grad_norm": 6.21274995803833,
+      "learning_rate": 4.529616724738676e-06,
+      "loss": 6.9734,
+      "mean_token_accuracy": 0.05999266132712364,
+      "step": 130
+    },
+    {
+      "epoch": 0.012208947414319351,
+      "grad_norm": 12.326318740844727,
+      "learning_rate": 4.8780487804878055e-06,
+      "loss": 6.8844,
+      "mean_token_accuracy": 0.05292318966239691,
+      "step": 140
+    },
+    {
+      "epoch": 0.013081015086770733,
+      "grad_norm": 5.085020542144775,
+      "learning_rate": 5.226480836236935e-06,
+      "loss": 6.7891,
+      "mean_token_accuracy": 0.05583414919674397,
+      "step": 150
+    },
+    {
+      "epoch": 0.013953082759222116,
+      "grad_norm": 5.138378620147705,
+      "learning_rate": 5.574912891986063e-06,
+      "loss": 6.7141,
+      "mean_token_accuracy": 0.05869618393480778,
+      "step": 160
+    },
+    {
+      "epoch": 0.014825150431673498,
+      "grad_norm": 5.108981132507324,
+      "learning_rate": 5.923344947735193e-06,
+      "loss": 6.7766,
+      "mean_token_accuracy": 0.06010273937135935,
+      "step": 170
+    },
+    {
+      "epoch": 0.01569721810412488,
+      "grad_norm": 5.037508964538574,
+      "learning_rate": 6.27177700348432e-06,
+      "loss": 6.6297,
+      "mean_token_accuracy": 0.060934442281723025,
+      "step": 180
+    },
+    {
+      "epoch": 0.01656928577657626,
+      "grad_norm": 4.582004070281982,
+      "learning_rate": 6.62020905923345e-06,
+      "loss": 6.5812,
+      "mean_token_accuracy": 0.06325831692665815,
+      "step": 190
+    },
+    {
+      "epoch": 0.017441353449027645,
+      "grad_norm": 3.924436092376709,
+      "learning_rate": 6.96864111498258e-06,
+      "loss": 6.5187,
+      "mean_token_accuracy": 0.06273238714784383,
+      "step": 200
+    },
+    {
+      "epoch": 0.017441353449027645,
+      "eval_runtime": 4.1371,
+      "eval_samples_per_second": 265.159,
+      "eval_steps_per_second": 66.471,
+      "step": 200
+    },
+    {
+      "epoch": 0.018313421121479028,
+      "grad_norm": 6.232868194580078,
+      "learning_rate": 7.317073170731707e-06,
+      "loss": 6.5016,
+      "mean_token_accuracy": 0.06371086072176695,
+      "step": 210
+    },
+    {
+      "epoch": 0.019185488793930408,
+      "grad_norm": 8.982894897460938,
+      "learning_rate": 7.665505226480837e-06,
+      "loss": 6.3937,
+      "mean_token_accuracy": 0.06835861038416624,
+      "step": 220
+    },
+    {
+      "epoch": 0.02005755646638179,
+      "grad_norm": 4.787600040435791,
+      "learning_rate": 8.013937282229966e-06,
+      "loss": 6.3688,
+      "mean_token_accuracy": 0.06887230966240168,
+      "step": 230
+    },
+    {
+      "epoch": 0.020929624138833175,
+      "grad_norm": 4.649842739105225,
+      "learning_rate": 8.362369337979095e-06,
+      "loss": 6.4219,
+      "mean_token_accuracy": 0.06467710342258215,
+      "step": 240
+    },
+    {
+      "epoch": 0.021801691811284555,
+      "grad_norm": 4.994523048400879,
+      "learning_rate": 8.710801393728223e-06,
+      "loss": 6.3141,
+      "mean_token_accuracy": 0.06988747492432594,
+      "step": 250
+    },
+    {
+      "epoch": 0.02267375948373594,
+      "grad_norm": 6.446099758148193,
+      "learning_rate": 9.059233449477352e-06,
+      "loss": 6.3406,
+      "mean_token_accuracy": 0.07202788610011339,
+      "step": 260
+    },
+    {
+      "epoch": 0.023545827156187322,
+      "grad_norm": 7.308241844177246,
+      "learning_rate": 9.407665505226482e-06,
+      "loss": 6.25,
+      "mean_token_accuracy": 0.07442514672875404,
+      "step": 270
+    },
+    {
+      "epoch": 0.024417894828638702,
+      "grad_norm": 6.441298484802246,
+      "learning_rate": 9.756097560975611e-06,
+      "loss": 6.2703,
+      "mean_token_accuracy": 0.07792319059371948,
+      "step": 280
+    },
+    {
+      "epoch": 0.025289962501090085,
+      "grad_norm": 12.88493824005127,
+      "learning_rate": 1.0104529616724739e-05,
+      "loss": 6.2531,
+      "mean_token_accuracy": 0.07635763306170702,
+      "step": 290
+    },
+    {
+      "epoch": 0.026162030173541465,
+      "grad_norm": 9.991170883178711,
+      "learning_rate": 1.045296167247387e-05,
+      "loss": 6.1781,
+      "mean_token_accuracy": 0.07740949131548405,
+      "step": 300
+    },
+    {
+      "epoch": 0.026162030173541465,
+      "eval_runtime": 4.1328,
+      "eval_samples_per_second": 265.438,
+      "eval_steps_per_second": 66.541,
+      "step": 300
+    },
+    {
+      "epoch": 0.02703409784599285,
+      "grad_norm": 8.042091369628906,
+      "learning_rate": 1.0801393728222997e-05,
+      "loss": 6.2328,
+      "mean_token_accuracy": 0.07506115455180407,
+      "step": 310
+    },
+    {
+      "epoch": 0.027906165518444232,
+      "grad_norm": 8.354701042175293,
+      "learning_rate": 1.1149825783972127e-05,
+      "loss": 6.1344,
+      "mean_token_accuracy": 0.08287671245634556,
+      "step": 320
+    },
+    {
+      "epoch": 0.028778233190895612,
+      "grad_norm": 6.962996482849121,
+      "learning_rate": 1.1498257839721256e-05,
+      "loss": 6.1297,
+      "mean_token_accuracy": 0.07956213317811489,
+      "step": 330
+    },
+    {
+      "epoch": 0.029650300863346996,
+      "grad_norm": 6.655036926269531,
+      "learning_rate": 1.1846689895470385e-05,
+      "loss": 6.075,
+      "mean_token_accuracy": 0.09151174090802669,
+      "step": 340
+    },
+    {
+      "epoch": 0.03052236853579838,
+      "grad_norm": 7.6797027587890625,
+      "learning_rate": 1.2195121951219513e-05,
+      "loss": 6.0953,
+      "mean_token_accuracy": 0.09633072409778834,
+      "step": 350
+    },
+    {
+      "epoch": 0.03139443620824976,
+      "grad_norm": 11.0787992477417,
+      "learning_rate": 1.254355400696864e-05,
+      "loss": 6.1141,
+      "mean_token_accuracy": 0.10138209424912929,
+      "step": 360
+    },
+    {
+      "epoch": 0.03226650388070114,
+      "grad_norm": 9.738404273986816,
+      "learning_rate": 1.2891986062717772e-05,
+      "loss": 6.1031,
+      "mean_token_accuracy": 0.0869863012805581,
+      "step": 370
+    },
+    {
+      "epoch": 0.03313857155315252,
+      "grad_norm": 11.745001792907715,
+      "learning_rate": 1.32404181184669e-05,
+      "loss": 6.0656,
+      "mean_token_accuracy": 0.08767123371362687,
+      "step": 380
+    },
+    {
+      "epoch": 0.034010639225603906,
+      "grad_norm": 10.759794235229492,
+      "learning_rate": 1.3588850174216028e-05,
+      "loss": 5.95,
+      "mean_token_accuracy": 0.1073140885680914,
+      "step": 390
+    },
+    {
+      "epoch": 0.03488270689805529,
+      "grad_norm": 11.453364372253418,
+      "learning_rate": 1.393728222996516e-05,
+      "loss": 5.8797,
+      "mean_token_accuracy": 0.11714774928987026,
+      "step": 400
+    },
+    {
+      "epoch": 0.03488270689805529,
+      "eval_runtime": 4.1324,
+      "eval_samples_per_second": 265.463,
+      "eval_steps_per_second": 66.547,
+      "step": 400
+    },
+    {
+      "epoch": 0.03575477457050667,
+      "grad_norm": 10.520788192749023,
+      "learning_rate": 1.4285714285714287e-05,
+      "loss": 5.9484,
+      "mean_token_accuracy": 0.10590753369033337,
+      "step": 410
+    },
+    {
+      "epoch": 0.036626842242958056,
+      "grad_norm": 9.388611793518066,
+      "learning_rate": 1.4634146341463415e-05,
+      "loss": 5.8266,
+      "mean_token_accuracy": 0.11970401257276535,
+      "step": 420
+    },
+    {
+      "epoch": 0.03749890991540943,
+      "grad_norm": 9.268442153930664,
+      "learning_rate": 1.4982578397212544e-05,
+      "loss": 5.8312,
+      "mean_token_accuracy": 0.11612035147845745,
+      "step": 430
+    },
+    {
+      "epoch": 0.038370977587860816,
+      "grad_norm": 9.636759757995605,
+      "learning_rate": 1.5331010452961673e-05,
+      "loss": 5.7812,
+      "mean_token_accuracy": 0.12024217247962951,
+      "step": 440
+    },
+    {
+      "epoch": 0.0392430452603122,
+      "grad_norm": 9.492960929870605,
+      "learning_rate": 1.5679442508710803e-05,
+      "loss": 5.775,
+      "mean_token_accuracy": 0.1205479446798563,
+      "step": 450
+    },
+    {
+      "epoch": 0.04011511293276358,
+      "grad_norm": 12.608254432678223,
+      "learning_rate": 1.6027874564459932e-05,
+      "loss": 5.7266,
+      "mean_token_accuracy": 0.12345890402793884,
+      "step": 460
+    },
+    {
+      "epoch": 0.040987180605214966,
+      "grad_norm": 11.828970909118652,
+      "learning_rate": 1.637630662020906e-05,
+      "loss": 5.6984,
+      "mean_token_accuracy": 0.13238747492432595,
+      "step": 470
+    },
+    {
+      "epoch": 0.04185924827766635,
+      "grad_norm": 9.523595809936523,
+      "learning_rate": 1.672473867595819e-05,
+      "loss": 5.6797,
+      "mean_token_accuracy": 0.13346379697322847,
+      "step": 480
+    },
+    {
+      "epoch": 0.042731315950117726,
+      "grad_norm": 9.293509483337402,
+      "learning_rate": 1.7073170731707317e-05,
+      "loss": 5.6125,
+      "mean_token_accuracy": 0.1267490204423666,
+      "step": 490
+    },
+    {
+      "epoch": 0.04360338362256911,
+      "grad_norm": 8.775671005249023,
+      "learning_rate": 1.7421602787456446e-05,
+      "loss": 5.6859,
+      "mean_token_accuracy": 0.12254158593714237,
+      "step": 500
+    },
+    {
+      "epoch": 0.04360338362256911,
+      "eval_runtime": 4.138,
+      "eval_samples_per_second": 265.106,
+      "eval_steps_per_second": 66.458,
+      "step": 500
+    },
+    {
+      "epoch": 0.04447545129502049,
+      "grad_norm": 10.688274383544922,
+      "learning_rate": 1.7770034843205575e-05,
+      "loss": 5.5703,
+      "mean_token_accuracy": 0.135371820256114,
+      "step": 510
+    },
+    {
+      "epoch": 0.04534751896747188,
+      "grad_norm": 10.328093528747559,
+      "learning_rate": 1.8118466898954705e-05,
+      "loss": 5.5297,
+      "mean_token_accuracy": 0.13945694863796235,
+      "step": 520
+    },
+    {
+      "epoch": 0.04621958663992326,
+      "grad_norm": 13.896836280822754,
+      "learning_rate": 1.8466898954703834e-05,
+      "loss": 5.5625,
+      "mean_token_accuracy": 0.13023483343422412,
+      "step": 530
+    },
+    {
+      "epoch": 0.047091654312374644,
+      "grad_norm": 9.42554759979248,
+      "learning_rate": 1.8815331010452963e-05,
+      "loss": 5.5469,
+      "mean_token_accuracy": 0.1324975538998842,
+      "step": 540
+    },
+    {
+      "epoch": 0.04796372198482602,
+      "grad_norm": 13.343840599060059,
+      "learning_rate": 1.9163763066202093e-05,
+      "loss": 5.4156,
+      "mean_token_accuracy": 0.13824608661234378,
+      "step": 550
+    },
+    {
+      "epoch": 0.048835789657277404,
+      "grad_norm": 10.667329788208008,
+      "learning_rate": 1.9512195121951222e-05,
+      "loss": 5.4641,
+      "mean_token_accuracy": 0.13923679143190384,
+      "step": 560
+    },
+    {
+      "epoch": 0.04970785732972879,
+      "grad_norm": 9.04549789428711,
+      "learning_rate": 1.9860627177700348e-05,
+      "loss": 5.3828,
+      "mean_token_accuracy": 0.14715019688010217,
+      "step": 570
+    },
+    {
+      "epoch": 0.05057992500218017,
+      "grad_norm": 11.892061233520508,
+      "learning_rate": 1.9999933277491715e-05,
+      "loss": 5.2531,
+      "mean_token_accuracy": 0.15702054798603057,
+      "step": 580
+    },
+    {
+      "epoch": 0.051451992674631554,
+      "grad_norm": 8.392095565795898,
+      "learning_rate": 1.999952553205438e-05,
+      "loss": 5.2562,
+      "mean_token_accuracy": 0.15291095860302448,
+      "step": 590
+    },
+    {
+      "epoch": 0.05232406034708293,
+      "grad_norm": 9.140569686889648,
+      "learning_rate": 1.999874712433585e-05,
+      "loss": 5.175,
+      "mean_token_accuracy": 0.16651174128055574,
+      "step": 600
+    },
+    {
+      "epoch": 0.05232406034708293,
+      "eval_runtime": 4.1345,
+      "eval_samples_per_second": 265.329,
+      "eval_steps_per_second": 66.514,
+      "step": 600
+    },
+    {
+      "epoch": 0.053196128019534314,
+      "grad_norm": 7.695169925689697,
+      "learning_rate": 1.999759808319013e-05,
+      "loss": 5.1359,
+      "mean_token_accuracy": 0.16329500935971736,
+      "step": 610
+    },
+    {
+      "epoch": 0.0540681956919857,
+      "grad_norm": 8.865972518920898,
+      "learning_rate": 1.9996078451209863e-05,
+      "loss": 5.2344,
+      "mean_token_accuracy": 0.1484344407916069,
+      "step": 620
+    },
+    {
+      "epoch": 0.05494026336443708,
+      "grad_norm": 8.918686866760254,
+      "learning_rate": 1.999418828472475e-05,
+      "loss": 5.1109,
+      "mean_token_accuracy": 0.1650440316647291,
+      "step": 630
+    },
+    {
+      "epoch": 0.055812331036888464,
+      "grad_norm": 9.925086975097656,
+      "learning_rate": 1.9991927653799458e-05,
+      "loss": 5.1516,
+      "mean_token_accuracy": 0.15656800717115402,
+      "step": 640
+    },
+    {
+      "epoch": 0.05668439870933985,
+      "grad_norm": 8.371522903442383,
+      "learning_rate": 1.998929664223102e-05,
+      "loss": 5.0891,
+      "mean_token_accuracy": 0.15861056968569756,
+      "step": 650
+    },
+    {
+      "epoch": 0.057556466381791224,
+      "grad_norm": 9.309918403625488,
+      "learning_rate": 1.9986295347545738e-05,
+      "loss": 4.9969,
+      "mean_token_accuracy": 0.17133072391152382,
+      "step": 660
+    },
+    {
+      "epoch": 0.05842853405424261,
+      "grad_norm": 7.434914588928223,
+      "learning_rate": 1.998292388099557e-05,
+      "loss": 5.0656,
+      "mean_token_accuracy": 0.16520303413271903,
+      "step": 670
+    },
+    {
+      "epoch": 0.05930060172669399,
+      "grad_norm": 8.671019554138184,
+      "learning_rate": 1.9979182367553994e-05,
+      "loss": 4.9297,
+      "mean_token_accuracy": 0.17629647701978685,
+      "step": 680
+    },
+    {
+      "epoch": 0.060172669399145375,
+      "grad_norm": 7.99770450592041,
+      "learning_rate": 1.997507094591137e-05,
+      "loss": 4.8984,
+      "mean_token_accuracy": 0.17396037131547928,
+      "step": 690
+    },
+    {
+      "epoch": 0.06104473707159676,
+      "grad_norm": 8.327537536621094,
+      "learning_rate": 1.9970589768469833e-05,
+      "loss": 4.9406,
+      "mean_token_accuracy": 0.17912181913852693,
+      "step": 700
+    },
+    {
+      "epoch": 0.06104473707159676,
+      "eval_runtime": 4.1319,
+      "eval_samples_per_second": 265.498,
+      "eval_steps_per_second": 66.556,
+      "step": 700
+    },
+    {
+      "epoch": 0.06191680474404814,
+      "grad_norm": 6.633450031280518,
+      "learning_rate": 1.996573900133761e-05,
+      "loss": 4.7891,
+      "mean_token_accuracy": 0.17581947073340415,
+      "step": 710
+    },
+    {
+      "epoch": 0.06278887241649952,
+      "grad_norm": 8.022377014160156,
+      "learning_rate": 1.996051882432286e-05,
+      "loss": 4.7406,
+      "mean_token_accuracy": 0.19094911962747574,
+      "step": 720
+    },
+    {
+      "epoch": 0.0636609400889509,
+      "grad_norm": 7.844242095947266,
+      "learning_rate": 1.995492943092705e-05,
+      "loss": 4.6953,
+      "mean_token_accuracy": 0.18763453960418702,
+      "step": 730
+    },
+    {
+      "epoch": 0.06453300776140228,
+      "grad_norm": 6.448212623596191,
+      "learning_rate": 1.9948971028337737e-05,
+      "loss": 4.6922,
+      "mean_token_accuracy": 0.19351761266589165,
+      "step": 740
+    },
+    {
+      "epoch": 0.06540507543385367,
+      "grad_norm": 6.086342811584473,
+      "learning_rate": 1.9942643837420904e-05,
+      "loss": 4.6172,
+      "mean_token_accuracy": 0.20698385685682297,
+      "step": 750
+    },
+    {
+      "epoch": 0.06627714310630504,
+      "grad_norm": 7.189157009124756,
+      "learning_rate": 1.9935948092712792e-05,
+      "loss": 4.5062,
+      "mean_token_accuracy": 0.21434687003493308,
+      "step": 760
+    },
+    {
+      "epoch": 0.06714921077875644,
+      "grad_norm": 7.056966781616211,
+      "learning_rate": 1.992888404241117e-05,
+      "loss": 4.6781,
+      "mean_token_accuracy": 0.18928571343421935,
+      "step": 770
+    },
+    {
+      "epoch": 0.06802127845120781,
+      "grad_norm": 6.542293071746826,
+      "learning_rate": 1.992145194836616e-05,
+      "loss": 4.5484,
+      "mean_token_accuracy": 0.20753424614667892,
+      "step": 780
+    },
+    {
+      "epoch": 0.0688933461236592,
+      "grad_norm": 6.556288719177246,
+      "learning_rate": 1.9913652086070535e-05,
+      "loss": 4.4891,
+      "mean_token_accuracy": 0.20863502994179725,
+      "step": 790
+    },
+    {
+      "epoch": 0.06976541379611058,
+      "grad_norm": 7.45630407333374,
+      "learning_rate": 1.9905484744649484e-05,
+      "loss": 4.5062,
+      "mean_token_accuracy": 0.20621330812573432,
+      "step": 800
+    },
+    {
+      "epoch": 0.06976541379611058,
+      "eval_runtime": 4.1297,
+      "eval_samples_per_second": 265.635,
+      "eval_steps_per_second": 66.59,
+      "step": 800
+    },
+    {
+      "epoch": 0.07063748146856196,
+      "grad_norm": 6.321140766143799,
+      "learning_rate": 1.989695022684991e-05,
+      "loss": 4.3914,
+      "mean_token_accuracy": 0.21566780731081964,
+      "step": 810
+    },
+    {
+      "epoch": 0.07150954914101335,
+      "grad_norm": 6.933566093444824,
+      "learning_rate": 1.988804884902921e-05,
+      "loss": 4.343,
+      "mean_token_accuracy": 0.23220401257276535,
+      "step": 820
+    },
+    {
+      "epoch": 0.07238161681346472,
+      "grad_norm": 6.512579441070557,
+      "learning_rate": 1.9878780941143538e-05,
+      "loss": 4.3508,
+      "mean_token_accuracy": 0.21815068647265434,
+      "step": 830
+    },
+    {
+      "epoch": 0.07325368448591611,
+      "grad_norm": 6.477738857269287,
+      "learning_rate": 1.9869146846735576e-05,
+      "loss": 4.2539,
+      "mean_token_accuracy": 0.2343199595808983,
+      "step": 840
+    },
+    {
+      "epoch": 0.07412575215836749,
+      "grad_norm": 5.379225730895996,
+      "learning_rate": 1.985914692292182e-05,
+      "loss": 4.2922,
+      "mean_token_accuracy": 0.23241193890571593,
+      "step": 850
+    },
+    {
+      "epoch": 0.07499781983081887,
+      "grad_norm": 5.749763488769531,
+      "learning_rate": 1.9848781540379312e-05,
+      "loss": 4.2625,
+      "mean_token_accuracy": 0.241169273853302,
+      "step": 860
+    },
+    {
+      "epoch": 0.07586988750327026,
+      "grad_norm": 5.9795732498168945,
+      "learning_rate": 1.983805108333191e-05,
+      "loss": 4.2883,
+      "mean_token_accuracy": 0.238918786495924,
+      "step": 870
+    },
+    {
+      "epoch": 0.07674195517572163,
+      "grad_norm": 5.839418888092041,
+      "learning_rate": 1.9826955949536062e-05,
+      "loss": 4.1148,
+      "mean_token_accuracy": 0.24826320856809617,
+      "step": 880
+    },
+    {
+      "epoch": 0.07761402284817302,
+      "grad_norm": 5.555854797363281,
+      "learning_rate": 1.9815496550266036e-05,
+      "loss": 4.1461,
+      "mean_token_accuracy": 0.2496330738067627,
+      "step": 890
+    },
+    {
+      "epoch": 0.0784860905206244,
+      "grad_norm": 6.0836005210876465,
+      "learning_rate": 1.98036733102987e-05,
+      "loss": 4.2266,
+      "mean_token_accuracy": 0.23541686460375785,
+      "step": 900
+    },
+    {
+      "epoch": 0.0784860905206244,
+      "eval_runtime": 4.1312,
+      "eval_samples_per_second": 265.543,
+      "eval_steps_per_second": 66.567,
+      "step": 900
+    },
+    {
+      "epoch": 0.07935815819307579,
+      "grad_norm": 5.527903079986572,
+      "learning_rate": 1.979148666789775e-05,
+      "loss": 4.1281,
+      "mean_token_accuracy": 0.2536692745983601,
+      "step": 910
+    },
+    {
+      "epoch": 0.08023022586552717,
+      "grad_norm": 4.603416919708252,
+      "learning_rate": 1.9778937074797494e-05,
+      "loss": 4.0227,
+      "mean_token_accuracy": 0.2580249719321728,
+      "step": 920
+    },
+    {
+      "epoch": 0.08110229353797854,
+      "grad_norm": 5.568511962890625,
+      "learning_rate": 1.976602499618608e-05,
+      "loss": 3.9992,
+      "mean_token_accuracy": 0.26352739706635475,
+      "step": 930
+    },
+    {
+      "epoch": 0.08197436121042993,
+      "grad_norm": 6.003087043762207,
+      "learning_rate": 1.9752750910688278e-05,
+      "loss": 4.0109,
+      "mean_token_accuracy": 0.2687255389988422,
+      "step": 940
+    },
+    {
+      "epoch": 0.08284642888288131,
+      "grad_norm": 5.8560357093811035,
+      "learning_rate": 1.9739115310347698e-05,
+      "loss": 3.9531,
+      "mean_token_accuracy": 0.2695939376950264,
+      "step": 950
+    },
+    {
+      "epoch": 0.0837184965553327,
+      "grad_norm": 6.488517761230469,
+      "learning_rate": 1.972511870060861e-05,
+      "loss": 3.9805,
+      "mean_token_accuracy": 0.2643835633993149,
+      "step": 960
+    },
+    {
+      "epoch": 0.08459056422778408,
+      "grad_norm": 5.626370429992676,
+      "learning_rate": 1.9710761600297147e-05,
+      "loss": 3.9367,
+      "mean_token_accuracy": 0.27651663646101954,
+      "step": 970
+    },
+    {
+      "epoch": 0.08546263190023545,
+      "grad_norm": 5.2302937507629395,
+      "learning_rate": 1.9696044541602126e-05,
+      "loss": 3.9945,
+      "mean_token_accuracy": 0.2608365938067436,
+      "step": 980
+    },
+    {
+      "epoch": 0.08633469957268684,
+      "grad_norm": 5.086008548736572,
+      "learning_rate": 1.968096807005528e-05,
+      "loss": 3.8375,
+      "mean_token_accuracy": 0.2929182954132557,
+      "step": 990
+    },
+    {
+      "epoch": 0.08720676724513822,
+      "grad_norm": 4.8101301193237305,
+      "learning_rate": 1.966553274451106e-05,
+      "loss": 3.7711,
+      "mean_token_accuracy": 0.2986668303608894,
+      "step": 1000
+    },
+    {
+      "epoch": 0.08720676724513822,
+      "eval_runtime": 4.1333,
+      "eval_samples_per_second": 265.405,
+      "eval_steps_per_second": 66.533,
+      "step": 1000
+    },
+    {
+      "epoch": 0.08807883491758961,
+      "grad_norm": 4.364314556121826,
+      "learning_rate": 1.964973913712591e-05,
+      "loss": 3.8,
+      "mean_token_accuracy": 0.2898727975785732,
+      "step": 1010
+    },
+    {
+      "epoch": 0.08895090259004099,
+      "grad_norm": 5.133767604827881,
+      "learning_rate": 1.9633587833337064e-05,
+      "loss": 3.7023,
+      "mean_token_accuracy": 0.29648972749710084,
+      "step": 1020
+    },
+    {
+      "epoch": 0.08982297026249236,
+      "grad_norm": 4.910743713378906,
+      "learning_rate": 1.961707943184083e-05,
+      "loss": 3.6938,
+      "mean_token_accuracy": 0.29639187902212144,
+      "step": 1030
+    },
+    {
+      "epoch": 0.09069503793494375,
+      "grad_norm": 4.70925760269165,
+      "learning_rate": 1.9600214544570432e-05,
+      "loss": 3.6437,
+      "mean_token_accuracy": 0.31215753555297854,
+      "step": 1040
+    },
+    {
+      "epoch": 0.09156710560739513,
+      "grad_norm": 5.305742263793945,
+      "learning_rate": 1.958299379667328e-05,
+      "loss": 3.6883,
+      "mean_token_accuracy": 0.29545009732246397,
+      "step": 1050
+    },
+    {
+      "epoch": 0.09243917327984652,
+      "grad_norm": 4.568931579589844,
+      "learning_rate": 1.9565417826487835e-05,
+      "loss": 3.5703,
+      "mean_token_accuracy": 0.32718933522701266,
+      "step": 1060
+    },
+    {
+      "epoch": 0.0933112409522979,
+      "grad_norm": 4.4945878982543945,
+      "learning_rate": 1.9547487285519922e-05,
+      "loss": 3.7023,
+      "mean_token_accuracy": 0.3070924893021584,
+      "step": 1070
+    },
+    {
+      "epoch": 0.09418330862474929,
+      "grad_norm": 5.122004985809326,
+      "learning_rate": 1.952920283841861e-05,
+      "loss": 3.6258,
+      "mean_token_accuracy": 0.3140288650989532,
+      "step": 1080
+    },
+    {
+      "epoch": 0.09505537629720066,
+      "grad_norm": 4.557742595672607,
+      "learning_rate": 1.9510565162951538e-05,
+      "loss": 3.5555,
+      "mean_token_accuracy": 0.31724559515714645,
+      "step": 1090
+    },
+    {
+      "epoch": 0.09592744396965204,
+      "grad_norm": 4.375725269317627,
+      "learning_rate": 1.9491574949979814e-05,
+      "loss": 3.5984,
+      "mean_token_accuracy": 0.3202910989522934,
+      "step": 1100
+    },
+    {
+      "epoch": 0.09592744396965204,
+      "eval_runtime": 4.1261,
+      "eval_samples_per_second": 265.87,
+      "eval_steps_per_second": 66.649,
+      "step": 1100
+    },
+    {
+      "epoch": 0.09679951164210343,
+      "grad_norm": 4.4627580642700195,
+      "learning_rate": 1.9472232903432406e-05,
+      "loss": 3.532,
+      "mean_token_accuracy": 0.3335127204656601,
+      "step": 1110
+    },
+    {
+      "epoch": 0.09767157931455481,
+      "grad_norm": 4.1603922843933105,
+      "learning_rate": 1.945253974028004e-05,
+      "loss": 3.5086,
+      "mean_token_accuracy": 0.33082191944122313,
+      "step": 1120
+    },
+    {
+      "epoch": 0.0985436469870062,
+      "grad_norm": 4.045429229736328,
+      "learning_rate": 1.9432496190508633e-05,
+      "loss": 3.4445,
+      "mean_token_accuracy": 0.33649706095457077,
+      "step": 1130
+    },
+    {
+      "epoch": 0.09941571465945757,
+      "grad_norm": 4.346405982971191,
+      "learning_rate": 1.941210299709222e-05,
+      "loss": 3.4883,
+      "mean_token_accuracy": 0.3298556759953499,
+      "step": 1140
+    },
+    {
+      "epoch": 0.10028778233190895,
+      "grad_norm": 3.910358190536499,
+      "learning_rate": 1.9391360915965426e-05,
+      "loss": 3.4234,
+      "mean_token_accuracy": 0.3448018580675125,
+      "step": 1150
+    },
+    {
+      "epoch": 0.10115985000436034,
+      "grad_norm": 4.750843524932861,
+      "learning_rate": 1.9370270715995447e-05,
+      "loss": 3.5266,
+      "mean_token_accuracy": 0.3246086061000824,
+      "step": 1160
+    },
+    {
+      "epoch": 0.10203191767681172,
+      "grad_norm": 4.294528007507324,
+      "learning_rate": 1.934883317895354e-05,
+      "loss": 3.3664,
+      "mean_token_accuracy": 0.35226272195577624,
+      "step": 1170
+    },
+    {
+      "epoch": 0.10290398534926311,
+      "grad_norm": 4.865283489227295,
+      "learning_rate": 1.932704909948604e-05,
+      "loss": 3.4406,
+      "mean_token_accuracy": 0.3327666327357292,
+      "step": 1180
+    },
+    {
+      "epoch": 0.10377605302171448,
+      "grad_norm": 4.08066463470459,
+      "learning_rate": 1.930491928508492e-05,
+      "loss": 3.3445,
+      "mean_token_accuracy": 0.3489603757858276,
+      "step": 1190
+    },
+    {
+      "epoch": 0.10464812069416586,
+      "grad_norm": 4.122050762176514,
+      "learning_rate": 1.9282444556057855e-05,
+      "loss": 3.3273,
+      "mean_token_accuracy": 0.3499755382537842,
+      "step": 1200
+    },
+    {
+      "epoch": 0.10464812069416586,
+      "eval_runtime": 4.1289,
+      "eval_samples_per_second": 265.686,
+      "eval_steps_per_second": 66.603,
+      "step": 1200
+    },
+    {
+      "epoch": 0.10552018836661725,
+      "grad_norm": 4.377065181732178,
+      "learning_rate": 1.9259625745497803e-05,
+      "loss": 3.2516,
+      "mean_token_accuracy": 0.3601638913154602,
+      "step": 1210
+    },
+    {
+      "epoch": 0.10639225603906863,
+      "grad_norm": 4.147984504699707,
+      "learning_rate": 1.9236463699252136e-05,
+      "loss": 3.3523,
+      "mean_token_accuracy": 0.3502446189522743,
+      "step": 1220
+    },
+    {
+      "epoch": 0.10726432371152002,
+      "grad_norm": 3.8092539310455322,
+      "learning_rate": 1.921295927589127e-05,
+      "loss": 3.2445,
+      "mean_token_accuracy": 0.3596868857741356,
+      "step": 1230
+    },
+    {
+      "epoch": 0.1081363913839714,
+      "grad_norm": 4.0484113693237305,
+      "learning_rate": 1.9189113346676878e-05,
+      "loss": 3.2758,
+      "mean_token_accuracy": 0.3600293517112732,
+      "step": 1240
+    },
+    {
+      "epoch": 0.10900845905642279,
+      "grad_norm": 4.4944233894348145,
+      "learning_rate": 1.916492679552954e-05,
+      "loss": 3.2563,
+      "mean_token_accuracy": 0.35364481061697006,
+      "step": 1250
+    },
+    {
+      "epoch": 0.10988052672887416,
+      "grad_norm": 3.984053373336792,
+      "learning_rate": 1.914040051899602e-05,
+      "loss": 3.2687,
+      "mean_token_accuracy": 0.3680406078696251,
+      "step": 1260
+    },
+    {
+      "epoch": 0.11075259440132554,
+      "grad_norm": 3.9245243072509766,
+      "learning_rate": 1.9115535426216018e-05,
+      "loss": 3.2164,
+      "mean_token_accuracy": 0.37181995809078217,
+      "step": 1270
+    },
+    {
+      "epoch": 0.11162466207377693,
+      "grad_norm": 4.007757186889648,
+      "learning_rate": 1.9090332438888458e-05,
+      "loss": 3.2258,
+      "mean_token_accuracy": 0.371012718975544,
+      "step": 1280
+    },
+    {
+      "epoch": 0.1124967297462283,
+      "grad_norm": 3.6724660396575928,
+      "learning_rate": 1.906479249123735e-05,
+      "loss": 3.1961,
+      "mean_token_accuracy": 0.3675024464726448,
+      "step": 1290
+    },
+    {
+      "epoch": 0.1133687974186797,
+      "grad_norm": 3.796718120574951,
+      "learning_rate": 1.9038916529977136e-05,
+      "loss": 3.2742,
+      "mean_token_accuracy": 0.364249025285244,
+      "step": 1300
+    },
+    {
+      "epoch": 0.1133687974186797,
+      "eval_runtime": 4.1283,
+      "eval_samples_per_second": 265.725,
+      "eval_steps_per_second": 66.613,
+      "step": 1300
+    },
+    {
+      "epoch": 0.11424086509113107,
+      "grad_norm": 3.9330341815948486,
+      "learning_rate": 1.901270551427761e-05,
+      "loss": 3.1664,
+      "mean_token_accuracy": 0.3738380640745163,
+      "step": 1310
+    },
+    {
+      "epoch": 0.11511293276358245,
+      "grad_norm": 3.5709919929504395,
+      "learning_rate": 1.898616041572836e-05,
+      "loss": 3.2211,
+      "mean_token_accuracy": 0.36432240456342696,
+      "step": 1320
+    },
+    {
+      "epoch": 0.11598500043603384,
+      "grad_norm": 3.698350191116333,
+      "learning_rate": 1.8959282218302746e-05,
+      "loss": 3.0789,
+      "mean_token_accuracy": 0.39012965112924575,
+      "step": 1330
+    },
+    {
+      "epoch": 0.11685706810848522,
+      "grad_norm": 3.5991384983062744,
+      "learning_rate": 1.893207191832144e-05,
+      "loss": 3.1117,
+      "mean_token_accuracy": 0.3800391390919685,
+      "step": 1340
+    },
+    {
+      "epoch": 0.1177291357809366,
+      "grad_norm": 3.959972620010376,
+      "learning_rate": 1.8904530524415483e-05,
+      "loss": 3.2102,
+      "mean_token_accuracy": 0.36545988619327546,
+      "step": 1350
+    },
+    {
+      "epoch": 0.11860120345338798,
+      "grad_norm": 3.513537645339966,
+      "learning_rate": 1.8876659057488905e-05,
+      "loss": 3.1094,
+      "mean_token_accuracy": 0.3847602754831314,
+      "step": 1360
+    },
+    {
+      "epoch": 0.11947327112583936,
+      "grad_norm": 3.834674596786499,
+      "learning_rate": 1.8848458550680875e-05,
+      "loss": 3.0852,
+      "mean_token_accuracy": 0.3872553795576096,
+      "step": 1370
+    },
+    {
+      "epoch": 0.12034533879829075,
+      "grad_norm": 3.618102550506592,
+      "learning_rate": 1.8819930049327412e-05,
+      "loss": 3.0797,
+      "mean_token_accuracy": 0.3904965788125992,
+      "step": 1380
+    },
+    {
+      "epoch": 0.12121740647074213,
+      "grad_norm": 3.6198570728302,
+      "learning_rate": 1.8791074610922624e-05,
+      "loss": 3.0875,
+      "mean_token_accuracy": 0.39086350202560427,
+      "step": 1390
+    },
+    {
+      "epoch": 0.12208947414319352,
+      "grad_norm": 3.8854329586029053,
+      "learning_rate": 1.8761893305079528e-05,
+      "loss": 3.1414,
+      "mean_token_accuracy": 0.37323874682188035,
+      "step": 1400
+    },
+    {
+      "epoch": 0.12208947414319352,
+      "eval_runtime": 4.1348,
+      "eval_samples_per_second": 265.309,
+      "eval_steps_per_second": 66.509,
+      "step": 1400
+    },
+    {
+      "epoch": 0.12296154181564489,
+      "grad_norm": 4.041639804840088,
+      "learning_rate": 1.873238721349038e-05,
+      "loss": 3.0742,
+      "mean_token_accuracy": 0.3837573379278183,
+      "step": 1410
+    },
+    {
+      "epoch": 0.12383360948809628,
+      "grad_norm": 4.237380027770996,
+      "learning_rate": 1.8702557429886607e-05,
+      "loss": 3.0914,
+      "mean_token_accuracy": 0.3820450097322464,
+      "step": 1420
+    },
+    {
+      "epoch": 0.12470567716054766,
+      "grad_norm": 4.059397220611572,
+      "learning_rate": 1.8672405059998228e-05,
+      "loss": 2.9609,
+      "mean_token_accuracy": 0.40667808055877686,
+      "step": 1430
+    },
+    {
+      "epoch": 0.12557774483299905,
+      "grad_norm": 3.358105182647705,
+      "learning_rate": 1.8641931221512895e-05,
+      "loss": 3.0742,
+      "mean_token_accuracy": 0.3837695673108101,
+      "step": 1440
+    },
+    {
+      "epoch": 0.12644981250545043,
+      "grad_norm": 3.401979446411133,
+      "learning_rate": 1.8611137044034454e-05,
+      "loss": 3.0383,
+      "mean_token_accuracy": 0.39514432698488233,
+      "step": 1450
+    },
+    {
+      "epoch": 0.1273218801779018,
+      "grad_norm": 3.436000108718872,
+      "learning_rate": 1.858002366904107e-05,
+      "loss": 3.0211,
+      "mean_token_accuracy": 0.3948630094528198,
+      "step": 1460
+    },
+    {
+      "epoch": 0.12819394785035318,
+      "grad_norm": 3.341808795928955,
+      "learning_rate": 1.854859224984292e-05,
+      "loss": 3.0477,
+      "mean_token_accuracy": 0.3904598832130432,
+      "step": 1470
+    },
+    {
+      "epoch": 0.12906601552280456,
+      "grad_norm": 3.333475112915039,
+      "learning_rate": 1.851684395153944e-05,
+      "loss": 2.9945,
+      "mean_token_accuracy": 0.39763942956924436,
+      "step": 1480
+    },
+    {
+      "epoch": 0.12993808319525596,
+      "grad_norm": 3.9498677253723145,
+      "learning_rate": 1.8484779950976133e-05,
+      "loss": 3.0828,
+      "mean_token_accuracy": 0.3841120347380638,
+      "step": 1490
+    },
+    {
+      "epoch": 0.13081015086770734,
+      "grad_norm": 3.5516517162323,
+      "learning_rate": 1.8452401436700954e-05,
+      "loss": 3.0523,
+      "mean_token_accuracy": 0.3902886524796486,
+      "step": 1500
+    },
+    {
+      "epoch": 0.13081015086770734,
+      "eval_runtime": 4.1275,
+      "eval_samples_per_second": 265.781,
+      "eval_steps_per_second": 66.627,
+      "step": 1500
+    },
+    {
+      "epoch": 0.1316822185401587,
+      "grad_norm": 3.62166690826416,
+      "learning_rate": 1.8419709608920243e-05,
+      "loss": 3.0422,
+      "mean_token_accuracy": 0.39096135199069976,
+      "step": 1510
+    },
+    {
+      "epoch": 0.1325542862126101,
+      "grad_norm": 4.141049385070801,
+      "learning_rate": 1.8386705679454243e-05,
+      "loss": 2.9211,
+      "mean_token_accuracy": 0.4110078290104866,
+      "step": 1520
+    },
+    {
+      "epoch": 0.1334263538850615,
+      "grad_norm": 3.757030487060547,
+      "learning_rate": 1.8353390871692176e-05,
+      "loss": 3.0242,
+      "mean_token_accuracy": 0.3907778859138489,
+      "step": 1530
+    },
+    {
+      "epoch": 0.13429842155751287,
+      "grad_norm": 3.1740000247955322,
+      "learning_rate": 1.8319766420546902e-05,
+      "loss": 2.9727,
+      "mean_token_accuracy": 0.4011374771595001,
+      "step": 1540
+    },
+    {
+      "epoch": 0.13517048922996425,
+      "grad_norm": 3.0472218990325928,
+      "learning_rate": 1.8285833572409135e-05,
+      "loss": 2.9555,
+      "mean_token_accuracy": 0.4009907066822052,
+      "step": 1550
+    },
+    {
+      "epoch": 0.13604255690241562,
+      "grad_norm": 3.645477533340454,
+      "learning_rate": 1.8251593585101243e-05,
+      "loss": 2.9727,
+      "mean_token_accuracy": 0.39836105704307556,
+      "step": 1560
+    },
+    {
+      "epoch": 0.136914624574867,
+      "grad_norm": 3.587836503982544,
+      "learning_rate": 1.821704772783063e-05,
+      "loss": 2.918,
+      "mean_token_accuracy": 0.41287915855646135,
+      "step": 1570
+    },
+    {
+      "epoch": 0.1377866922473184,
+      "grad_norm": 3.234576940536499,
+      "learning_rate": 1.818219728114267e-05,
+      "loss": 2.9281,
+      "mean_token_accuracy": 0.40916095674037933,
+      "step": 1580
+    },
+    {
+      "epoch": 0.13865875991976978,
+      "grad_norm": 3.24845027923584,
+      "learning_rate": 1.8147043536873275e-05,
+      "loss": 2.9125,
+      "mean_token_accuracy": 0.4080357149243355,
+      "step": 1590
+    },
+    {
+      "epoch": 0.13953082759222116,
+      "grad_norm": 3.2831177711486816,
+      "learning_rate": 1.8111587798100974e-05,
+      "loss": 2.9648,
+      "mean_token_accuracy": 0.4068860083818436,
+      "step": 1600
+    },
+    {
+      "epoch": 0.13953082759222116,
+      "eval_runtime": 4.1326,
+      "eval_samples_per_second": 265.449,
+      "eval_steps_per_second": 66.544,
+      "step": 1600
+    },
+    {
+      "epoch": 0.14040289526467253,
+      "grad_norm": 3.7048988342285156,
+      "learning_rate": 1.807583137909862e-05,
+      "loss": 2.9148,
+      "mean_token_accuracy": 0.413246089220047,
+      "step": 1610
+    },
+    {
+      "epoch": 0.1412749629371239,
+      "grad_norm": 3.4570460319519043,
+      "learning_rate": 1.8039775605284687e-05,
+      "loss": 2.8078,
+      "mean_token_accuracy": 0.4237279832363129,
+      "step": 1620
+    },
+    {
+      "epoch": 0.14214703060957531,
+      "grad_norm": 3.099559783935547,
+      "learning_rate": 1.800342181317413e-05,
+      "loss": 2.918,
+      "mean_token_accuracy": 0.4077666312456131,
+      "step": 1630
+    },
+    {
+      "epoch": 0.1430190982820267,
+      "grad_norm": 3.0252482891082764,
+      "learning_rate": 1.7966771350328825e-05,
+      "loss": 2.8844,
+      "mean_token_accuracy": 0.4083292558789253,
+      "step": 1640
+    },
+    {
+      "epoch": 0.14389116595447807,
+      "grad_norm": 3.1384706497192383,
+      "learning_rate": 1.7929825575307665e-05,
+      "loss": 2.918,
+      "mean_token_accuracy": 0.4091854229569435,
+      "step": 1650
+    },
+    {
+      "epoch": 0.14476323362692944,
+      "grad_norm": 3.6761953830718994,
+      "learning_rate": 1.7892585857616144e-05,
+      "loss": 2.8633,
+      "mean_token_accuracy": 0.41793052554130555,
+      "step": 1660
+    },
+    {
+      "epoch": 0.14563530129938082,
+      "grad_norm": 3.1798853874206543,
+      "learning_rate": 1.785505357765563e-05,
+      "loss": 2.8742,
+      "mean_token_accuracy": 0.41296477168798446,
+      "step": 1670
+    },
+    {
+      "epoch": 0.14650736897183222,
+      "grad_norm": 2.878243923187256,
+      "learning_rate": 1.781723012667218e-05,
+      "loss": 2.8141,
+      "mean_token_accuracy": 0.42703033089637754,
+      "step": 1680
+    },
+    {
+      "epoch": 0.1473794366442836,
+      "grad_norm": 3.0460922718048096,
+      "learning_rate": 1.7779116906704986e-05,
+      "loss": 2.8719,
+      "mean_token_accuracy": 0.41503180265426637,
+      "step": 1690
+    },
+    {
+      "epoch": 0.14825150431673498,
+      "grad_norm": 3.0491960048675537,
+      "learning_rate": 1.7740715330534383e-05,
+      "loss": 2.8453,
+      "mean_token_accuracy": 0.42374021410942075,
+      "step": 1700
+    },
+    {
+      "epoch": 0.14825150431673498,
+      "eval_runtime": 4.1292,
+      "eval_samples_per_second": 265.67,
+      "eval_steps_per_second": 66.599,
+      "step": 1700
+    },
+    {
+      "epoch": 0.14912357198918635,
+      "grad_norm": 3.0250508785247803,
+      "learning_rate": 1.770202682162949e-05,
+      "loss": 2.7922,
+      "mean_token_accuracy": 0.4278742656111717,
+      "step": 1710
+    },
+    {
+      "epoch": 0.14999563966163773,
+      "grad_norm": 2.9240477085113525,
+      "learning_rate": 1.7663052814095447e-05,
+      "loss": 2.7781,
+      "mean_token_accuracy": 0.4257338553667068,
+      "step": 1720
+    },
+    {
+      "epoch": 0.15086770733408914,
+      "grad_norm": 3.1178035736083984,
+      "learning_rate": 1.7623794752620255e-05,
+      "loss": 2.8961,
+      "mean_token_accuracy": 0.4085371822118759,
+      "step": 1730
+    },
+    {
+      "epoch": 0.1517397750065405,
+      "grad_norm": 3.2408981323242188,
+      "learning_rate": 1.7584254092421226e-05,
+      "loss": 2.8453,
+      "mean_token_accuracy": 0.4179305285215378,
+      "step": 1740
+    },
+    {
+      "epoch": 0.1526118426789919,
+      "grad_norm": 3.1920886039733887,
+      "learning_rate": 1.754443229919103e-05,
+      "loss": 2.8078,
+      "mean_token_accuracy": 0.4238502934575081,
+      "step": 1750
+    },
+    {
+      "epoch": 0.15348391035144326,
+      "grad_norm": 3.2301573753356934,
+      "learning_rate": 1.7504330849043373e-05,
+      "loss": 2.8148,
+      "mean_token_accuracy": 0.4219178080558777,
+      "step": 1760
+    },
+    {
+      "epoch": 0.15435597802389464,
+      "grad_norm": 2.9890947341918945,
+      "learning_rate": 1.7463951228458288e-05,
+      "loss": 2.8078,
+      "mean_token_accuracy": 0.42202788293361665,
+      "step": 1770
+    },
+    {
+      "epoch": 0.15522804569634605,
+      "grad_norm": 2.8763387203216553,
+      "learning_rate": 1.7423294934227017e-05,
+      "loss": 2.8242,
+      "mean_token_accuracy": 0.41895792335271836,
+      "step": 1780
+    },
+    {
+      "epoch": 0.15610011336879742,
+      "grad_norm": 2.7581729888916016,
+      "learning_rate": 1.7382363473396543e-05,
+      "loss": 2.8328,
+      "mean_token_accuracy": 0.4259295478463173,
+      "step": 1790
+    },
+    {
+      "epoch": 0.1569721810412488,
+      "grad_norm": 2.963263988494873,
+      "learning_rate": 1.734115836321372e-05,
+      "loss": 2.807,
+      "mean_token_accuracy": 0.43183708637952806,
+      "step": 1800
+    },
+    {
+      "epoch": 0.1569721810412488,
+      "eval_runtime": 4.1274,
+      "eval_samples_per_second": 265.783,
+      "eval_steps_per_second": 66.627,
+      "step": 1800
+    },
+    {
+      "epoch": 0.15784424871370017,
+      "grad_norm": 2.855430841445923,
+      "learning_rate": 1.7299681131069026e-05,
+      "loss": 2.693,
+      "mean_token_accuracy": 0.4404476538300514,
+      "step": 1810
+    },
+    {
+      "epoch": 0.15871631638615158,
+      "grad_norm": 3.119471788406372,
+      "learning_rate": 1.725793331443996e-05,
+      "loss": 2.7398,
+      "mean_token_accuracy": 0.4388209402561188,
+      "step": 1820
+    },
+    {
+      "epoch": 0.15958838405860296,
+      "grad_norm": 2.793922185897827,
+      "learning_rate": 1.7215916460834048e-05,
+      "loss": 2.843,
+      "mean_token_accuracy": 0.42108610570430755,
+      "step": 1830
+    },
+    {
+      "epoch": 0.16046045173105433,
+      "grad_norm": 2.9096503257751465,
+      "learning_rate": 1.7173632127731462e-05,
+      "loss": 2.7836,
+      "mean_token_accuracy": 0.4271893367171288,
+      "step": 1840
+    },
+    {
+      "epoch": 0.1613325194035057,
+      "grad_norm": 3.204503297805786,
+      "learning_rate": 1.7131081882527305e-05,
+      "loss": 2.7602,
+      "mean_token_accuracy": 0.4297088995575905,
+      "step": 1850
+    },
+    {
+      "epoch": 0.16220458707595709,
+      "grad_norm": 3.056335687637329,
+      "learning_rate": 1.708826730247351e-05,
+      "loss": 2.7836,
+      "mean_token_accuracy": 0.42745841294527054,
+      "step": 1860
+    },
+    {
+      "epoch": 0.1630766547484085,
+      "grad_norm": 3.0003669261932373,
+      "learning_rate": 1.704518997462037e-05,
+      "loss": 2.7781,
+      "mean_token_accuracy": 0.4310176149010658,
+      "step": 1870
+    },
+    {
+      "epoch": 0.16394872242085987,
+      "grad_norm": 2.989548683166504,
+      "learning_rate": 1.7001851495757708e-05,
+      "loss": 2.7,
+      "mean_token_accuracy": 0.4441413924098015,
+      "step": 1880
+    },
+    {
+      "epoch": 0.16482079009331124,
+      "grad_norm": 2.8589212894439697,
+      "learning_rate": 1.6958253472355687e-05,
+      "loss": 2.7969,
+      "mean_token_accuracy": 0.428583662211895,
+      "step": 1890
+    },
+    {
+      "epoch": 0.16569285776576262,
+      "grad_norm": 3.0321006774902344,
+      "learning_rate": 1.6914397520505267e-05,
+      "loss": 2.775,
+      "mean_token_accuracy": 0.42777641862630844,
+      "step": 1900
+    },
+    {
+      "epoch": 0.16569285776576262,
+      "eval_runtime": 4.1313,
+      "eval_samples_per_second": 265.534,
+      "eval_steps_per_second": 66.565,
+      "step": 1900
+    },
+    {
+      "epoch": 0.166564925438214,
+      "grad_norm": 3.001772880554199,
+      "learning_rate": 1.6870285265858298e-05,
+      "loss": 2.7281,
+      "mean_token_accuracy": 0.4312989264726639,
+      "step": 1910
+    },
+    {
+      "epoch": 0.1674369931106654,
+      "grad_norm": 2.932978630065918,
+      "learning_rate": 1.6825918343567257e-05,
+      "loss": 2.7188,
+      "mean_token_accuracy": 0.43893101513385774,
+      "step": 1920
+    },
+    {
+      "epoch": 0.16830906078311678,
+      "grad_norm": 2.9124691486358643,
+      "learning_rate": 1.678129839822463e-05,
+      "loss": 2.7078,
+      "mean_token_accuracy": 0.4415362000465393,
+      "step": 1930
+    },
+    {
+      "epoch": 0.16918112845556815,
+      "grad_norm": 3.4372551441192627,
+      "learning_rate": 1.673642708380198e-05,
+      "loss": 2.6633,
+      "mean_token_accuracy": 0.44169520288705827,
+      "step": 1940
+    },
+    {
+      "epoch": 0.17005319612801953,
+      "grad_norm": 3.0920894145965576,
+      "learning_rate": 1.6691306063588583e-05,
+      "loss": 2.7062,
+      "mean_token_accuracy": 0.43405088037252426,
+      "step": 1950
+    },
+    {
+      "epoch": 0.1709252638004709,
+      "grad_norm": 2.7911059856414795,
+      "learning_rate": 1.6645937010129837e-05,
+      "loss": 2.6176,
+      "mean_token_accuracy": 0.4528008818626404,
+      "step": 1960
+    },
+    {
+      "epoch": 0.1717973314729223,
+      "grad_norm": 2.8475165367126465,
+      "learning_rate": 1.660032160516522e-05,
+      "loss": 2.6953,
+      "mean_token_accuracy": 0.4378180027008057,
+      "step": 1970
+    },
+    {
+      "epoch": 0.1726693991453737,
+      "grad_norm": 2.9752726554870605,
+      "learning_rate": 1.6554461539565953e-05,
+      "loss": 2.7117,
+      "mean_token_accuracy": 0.43961594849824903,
+      "step": 1980
+    },
+    {
+      "epoch": 0.17354146681782506,
+      "grad_norm": 2.8593695163726807,
+      "learning_rate": 1.650835851327236e-05,
+      "loss": 2.6562,
+      "mean_token_accuracy": 0.4452421724796295,
+      "step": 1990
+    },
+    {
+      "epoch": 0.17441353449027644,
+      "grad_norm": 2.9197885990142822,
+      "learning_rate": 1.6462014235230805e-05,
+      "loss": 2.6602,
+      "mean_token_accuracy": 0.4435909941792488,
+      "step": 2000
+    },
+    {
+      "epoch": 0.17441353449027644,
+      "eval_runtime": 4.1313,
+      "eval_samples_per_second": 265.534,
+      "eval_steps_per_second": 66.565,
+      "step": 2000
+    },
+    {
+      "epoch": 0.17528560216272782,
+      "grad_norm": 2.9950215816497803,
+      "learning_rate": 1.641543042333038e-05,
+      "loss": 2.6203,
+      "mean_token_accuracy": 0.4521281778812408,
+      "step": 2010
+    },
+    {
+      "epoch": 0.17615766983517922,
+      "grad_norm": 2.9370808601379395,
+      "learning_rate": 1.636860880433922e-05,
+      "loss": 2.7664,
+      "mean_token_accuracy": 0.4317881613969803,
+      "step": 2020
+    },
+    {
+      "epoch": 0.1770297375076306,
+      "grad_norm": 2.8373124599456787,
+      "learning_rate": 1.632155111384047e-05,
+      "loss": 2.7141,
+      "mean_token_accuracy": 0.43976271748542783,
+      "step": 2030
+    },
+    {
+      "epoch": 0.17790180518008197,
+      "grad_norm": 2.8195762634277344,
+      "learning_rate": 1.6274259096168e-05,
+      "loss": 2.7094,
+      "mean_token_accuracy": 0.4360934436321259,
+      "step": 2040
+    },
+    {
+      "epoch": 0.17877387285253335,
+      "grad_norm": 2.7108118534088135,
+      "learning_rate": 1.622673450434169e-05,
+      "loss": 2.6523,
+      "mean_token_accuracy": 0.44914383441209793,
+      "step": 2050
+    },
+    {
+      "epoch": 0.17964594052498473,
+      "grad_norm": 2.8289873600006104,
+      "learning_rate": 1.6178979100002486e-05,
+      "loss": 2.675,
+      "mean_token_accuracy": 0.4425146758556366,
+      "step": 2060
+    },
+    {
+      "epoch": 0.18051800819743613,
+      "grad_norm": 2.755321502685547,
+      "learning_rate": 1.6130994653347096e-05,
+      "loss": 2.725,
+      "mean_token_accuracy": 0.4364603698253632,
+      "step": 2070
+    },
+    {
+      "epoch": 0.1813900758698875,
+      "grad_norm": 2.763747453689575,
+      "learning_rate": 1.6082782943062355e-05,
+      "loss": 2.6883,
+      "mean_token_accuracy": 0.43917563557624817,
+      "step": 2080
+    },
+    {
+      "epoch": 0.18226214354233888,
+      "grad_norm": 3.3703789710998535,
+      "learning_rate": 1.6034345756259303e-05,
+      "loss": 2.5836,
+      "mean_token_accuracy": 0.4629647731781006,
+      "step": 2090
+    },
+    {
+      "epoch": 0.18313421121479026,
+      "grad_norm": 2.9087443351745605,
+      "learning_rate": 1.598568488840695e-05,
+      "loss": 2.6602,
+      "mean_token_accuracy": 0.44581702500581744,
+      "step": 2100
+    },
+    {
+      "epoch": 0.18313421121479026,
+      "eval_runtime": 4.1314,
+      "eval_samples_per_second": 265.53,
+      "eval_steps_per_second": 66.564,
+      "step": 2100
+    },
+    {
+      "epoch": 0.18400627888724164,
+      "grad_norm": 2.908827781677246,
+      "learning_rate": 1.5936802143265708e-05,
+      "loss": 2.693,
+      "mean_token_accuracy": 0.4390166327357292,
+      "step": 2110
+    },
+    {
+      "epoch": 0.18487834655969304,
+      "grad_norm": 2.638753890991211,
+      "learning_rate": 1.5887699332820527e-05,
+      "loss": 2.6516,
+      "mean_token_accuracy": 0.4434474125504494,
+      "step": 2120
+    },
+    {
+      "epoch": 0.18575041423214442,
+      "grad_norm": 2.619760751724243,
+      "learning_rate": 1.5838378277213745e-05,
+      "loss": 2.6445,
+      "mean_token_accuracy": 0.44809197783470156,
+      "step": 2130
+    },
+    {
+      "epoch": 0.1866224819045958,
+      "grad_norm": 2.7272400856018066,
+      "learning_rate": 1.57888408046776e-05,
+      "loss": 2.6234,
+      "mean_token_accuracy": 0.44898483753204343,
+      "step": 2140
+    },
+    {
+      "epoch": 0.18749454957704717,
+      "grad_norm": 2.7914986610412598,
+      "learning_rate": 1.573908875146648e-05,
+      "loss": 2.5961,
+      "mean_token_accuracy": 0.4530821919441223,
+      "step": 2150
+    },
+    {
+      "epoch": 0.18836661724949857,
+      "grad_norm": 2.588865280151367,
+      "learning_rate": 1.5689123961788834e-05,
+      "loss": 2.6352,
+      "mean_token_accuracy": 0.4493517607450485,
+      "step": 2160
+    },
+    {
+      "epoch": 0.18923868492194995,
+      "grad_norm": 2.6424171924591064,
+      "learning_rate": 1.563894828773883e-05,
+      "loss": 2.6109,
+      "mean_token_accuracy": 0.45457435846328736,
+      "step": 2170
+    },
+    {
+      "epoch": 0.19011075259440133,
+      "grad_norm": 2.7493107318878174,
+      "learning_rate": 1.55885635892277e-05,
+      "loss": 2.7016,
+      "mean_token_accuracy": 0.4329256385564804,
+      "step": 2180
+    },
+    {
+      "epoch": 0.1909828202668527,
+      "grad_norm": 2.6690175533294678,
+      "learning_rate": 1.5537971733914784e-05,
+      "loss": 2.6586,
+      "mean_token_accuracy": 0.44171966463327406,
+      "step": 2190
+    },
+    {
+      "epoch": 0.19185488793930408,
+      "grad_norm": 2.6221401691436768,
+      "learning_rate": 1.5487174597138314e-05,
+      "loss": 2.5656,
+      "mean_token_accuracy": 0.46923923641443255,
+      "step": 2200
+    },
+    {
+      "epoch": 0.19185488793930408,
+      "eval_runtime": 4.1333,
+      "eval_samples_per_second": 265.403,
+      "eval_steps_per_second": 66.532,
+      "step": 2200
+    },
+    {
+      "epoch": 0.19272695561175548,
+      "grad_norm": 2.4270994663238525,
+      "learning_rate": 1.543617406184589e-05,
+      "loss": 2.6078,
+      "mean_token_accuracy": 0.4553938314318657,
+      "step": 2210
+    },
+    {
+      "epoch": 0.19359902328420686,
+      "grad_norm": 2.664708375930786,
+      "learning_rate": 1.5384972018524678e-05,
+      "loss": 2.6141,
+      "mean_token_accuracy": 0.4547333657741547,
+      "step": 2220
+    },
+    {
+      "epoch": 0.19447109095665824,
+      "grad_norm": 2.762800693511963,
+      "learning_rate": 1.5333570365131353e-05,
+      "loss": 2.6328,
+      "mean_token_accuracy": 0.44379892498254775,
+      "step": 2230
+    },
+    {
+      "epoch": 0.19534315862910961,
+      "grad_norm": 2.7636358737945557,
+      "learning_rate": 1.5281971007021728e-05,
+      "loss": 2.6039,
+      "mean_token_accuracy": 0.452837572991848,
+      "step": 2240
+    },
+    {
+      "epoch": 0.196215226301561,
+      "grad_norm": 2.8070192337036133,
+      "learning_rate": 1.5230175856880132e-05,
+      "loss": 2.5617,
+      "mean_token_accuracy": 0.4596868857741356,
+      "step": 2250
+    },
+    {
+      "epoch": 0.1970872939740124,
+      "grad_norm": 2.8331243991851807,
+      "learning_rate": 1.5178186834648509e-05,
+      "loss": 2.5438,
+      "mean_token_accuracy": 0.4668175160884857,
+      "step": 2260
+    },
+    {
+      "epoch": 0.19795936164646377,
+      "grad_norm": 2.7152249813079834,
+      "learning_rate": 1.5126005867455256e-05,
+      "loss": 2.5109,
+      "mean_token_accuracy": 0.46671966910362245,
+      "step": 2270
+    },
+    {
+      "epoch": 0.19883142931891515,
+      "grad_norm": 2.550647258758545,
+      "learning_rate": 1.5073634889543778e-05,
+      "loss": 2.6805,
+      "mean_token_accuracy": 0.4403742626309395,
+      "step": 2280
+    },
+    {
+      "epoch": 0.19970349699136652,
+      "grad_norm": 2.3564510345458984,
+      "learning_rate": 1.5021075842200796e-05,
+      "loss": 2.6352,
+      "mean_token_accuracy": 0.44922945499420164,
+      "step": 2290
+    },
+    {
+      "epoch": 0.2005755646638179,
+      "grad_norm": 2.606297492980957,
+      "learning_rate": 1.4968330673684387e-05,
+      "loss": 2.6141,
+      "mean_token_accuracy": 0.45472113490104676,
+      "step": 2300
+    },
+    {
+      "epoch": 0.2005755646638179,
+      "eval_runtime": 4.1202,
+      "eval_samples_per_second": 266.248,
+      "eval_steps_per_second": 66.744,
+      "step": 2300
+    },
+    {
+      "epoch": 0.2014476323362693,
+      "grad_norm": 2.5705673694610596,
+      "learning_rate": 1.4915401339151769e-05,
+      "loss": 2.5227,
+      "mean_token_accuracy": 0.4659980431199074,
+      "step": 2310
+    },
+    {
+      "epoch": 0.20231970000872068,
+      "grad_norm": 2.6135408878326416,
+      "learning_rate": 1.486228980058682e-05,
+      "loss": 2.5906,
+      "mean_token_accuracy": 0.45435420870780946,
+      "step": 2320
+    },
+    {
+      "epoch": 0.20319176768117206,
+      "grad_norm": 2.3405566215515137,
+      "learning_rate": 1.4808998026727348e-05,
+      "loss": 2.6133,
+      "mean_token_accuracy": 0.4514187902212143,
+      "step": 2330
+    },
+    {
+      "epoch": 0.20406383535362344,
+      "grad_norm": 2.586230516433716,
+      "learning_rate": 1.4755527992992133e-05,
+      "loss": 2.5352,
+      "mean_token_accuracy": 0.4640166342258453,
+      "step": 2340
+    },
+    {
+      "epoch": 0.2049359030260748,
+      "grad_norm": 2.503904104232788,
+      "learning_rate": 1.4701881681407684e-05,
+      "loss": 2.5695,
+      "mean_token_accuracy": 0.4587695673108101,
+      "step": 2350
+    },
+    {
+      "epoch": 0.20580797069852622,
+      "grad_norm": 2.3526577949523926,
+      "learning_rate": 1.464806108053477e-05,
+      "loss": 2.5508,
+      "mean_token_accuracy": 0.4621942266821861,
+      "step": 2360
+    },
+    {
+      "epoch": 0.2066800383709776,
+      "grad_norm": 2.5566749572753906,
+      "learning_rate": 1.4594068185394723e-05,
+      "loss": 2.5656,
+      "mean_token_accuracy": 0.4573018580675125,
+      "step": 2370
+    },
+    {
+      "epoch": 0.20755210604342897,
+      "grad_norm": 2.587758779525757,
+      "learning_rate": 1.4539904997395468e-05,
+      "loss": 2.5734,
+      "mean_token_accuracy": 0.45736301243305205,
+      "step": 2380
+    },
+    {
+      "epoch": 0.20842417371588035,
+      "grad_norm": 2.557659864425659,
+      "learning_rate": 1.448557352425735e-05,
+      "loss": 2.5828,
+      "mean_token_accuracy": 0.45392612367868423,
+      "step": 2390
+    },
+    {
+      "epoch": 0.20929624138833172,
+      "grad_norm": 2.566291570663452,
+      "learning_rate": 1.44310757799387e-05,
+      "loss": 2.5023,
+      "mean_token_accuracy": 0.47142857015132905,
+      "step": 2400
+    },
+    {
+      "epoch": 0.20929624138833172,
+      "eval_runtime": 4.1208,
+      "eval_samples_per_second": 266.208,
+      "eval_steps_per_second": 66.734,
+      "step": 2400
+    },
+    {
+      "epoch": 0.21016830906078313,
+      "grad_norm": 2.5104401111602783,
+      "learning_rate": 1.437641378456119e-05,
+      "loss": 2.632,
+      "mean_token_accuracy": 0.4517857074737549,
+      "step": 2410
+    },
+    {
+      "epoch": 0.2110403767332345,
+      "grad_norm": 2.5603201389312744,
+      "learning_rate": 1.4321589564334946e-05,
+      "loss": 2.5672,
+      "mean_token_accuracy": 0.4601761266589165,
+      "step": 2420
+    },
+    {
+      "epoch": 0.21191244440568588,
+      "grad_norm": 2.5465004444122314,
+      "learning_rate": 1.4266605151483444e-05,
+      "loss": 2.5977,
+      "mean_token_accuracy": 0.4545865923166275,
+      "step": 2430
+    },
+    {
+      "epoch": 0.21278451207813726,
+      "grad_norm": 2.701889753341675,
+      "learning_rate": 1.4211462584168178e-05,
+      "loss": 2.5516,
+      "mean_token_accuracy": 0.4641756370663643,
+      "step": 2440
+    },
+    {
+      "epoch": 0.21365657975058863,
+      "grad_norm": 2.547666549682617,
+      "learning_rate": 1.4156163906413113e-05,
+      "loss": 2.4992,
+      "mean_token_accuracy": 0.47222358286380767,
+      "step": 2450
+    },
+    {
+      "epoch": 0.21452864742304004,
+      "grad_norm": 2.40474009513855,
+      "learning_rate": 1.4100711168028906e-05,
+      "loss": 2.5402,
+      "mean_token_accuracy": 0.46253668814897536,
+      "step": 2460
+    },
+    {
+      "epoch": 0.2154007150954914,
+      "grad_norm": 2.695638656616211,
+      "learning_rate": 1.4045106424536938e-05,
+      "loss": 2.5492,
+      "mean_token_accuracy": 0.46247553676366804,
+      "step": 2470
+    },
+    {
+      "epoch": 0.2162727827679428,
+      "grad_norm": 2.4520533084869385,
+      "learning_rate": 1.398935173709311e-05,
+      "loss": 2.5992,
+      "mean_token_accuracy": 0.4556384563446045,
+      "step": 2480
+    },
+    {
+      "epoch": 0.21714485044039417,
+      "grad_norm": 2.4459166526794434,
+      "learning_rate": 1.3933449172411446e-05,
+      "loss": 2.4758,
+      "mean_token_accuracy": 0.47353229075670245,
+      "step": 2490
+    },
+    {
+      "epoch": 0.21801691811284557,
+      "grad_norm": 2.517346143722534,
+      "learning_rate": 1.387740080268748e-05,
+      "loss": 2.543,
+      "mean_token_accuracy": 0.4653864920139313,
+      "step": 2500
+    },
+    {
+      "epoch": 0.21801691811284557,
+      "eval_runtime": 4.1166,
+      "eval_samples_per_second": 266.482,
+      "eval_steps_per_second": 66.803,
+      "step": 2500
+    },
+    {
+      "epoch": 0.21888898578529695,
+      "grad_norm": 2.487534999847412,
+      "learning_rate": 1.3821208705521442e-05,
+      "loss": 2.5711,
+      "mean_token_accuracy": 0.45908757448196413,
+      "step": 2510
+    },
+    {
+      "epoch": 0.21976105345774832,
+      "grad_norm": 2.585965871810913,
+      "learning_rate": 1.3764874963841255e-05,
+      "loss": 2.6031,
+      "mean_token_accuracy": 0.4523483380675316,
+      "step": 2520
+    },
+    {
+      "epoch": 0.2206331211301997,
+      "grad_norm": 2.908734083175659,
+      "learning_rate": 1.3708401665825319e-05,
+      "loss": 2.4742,
+      "mean_token_accuracy": 0.47223581522703173,
+      "step": 2530
+    },
+    {
+      "epoch": 0.22150518880265108,
+      "grad_norm": 2.5196869373321533,
+      "learning_rate": 1.36517909048251e-05,
+      "loss": 2.5266,
+      "mean_token_accuracy": 0.4666952043771744,
+      "step": 2540
+    },
+    {
+      "epoch": 0.22237725647510248,
+      "grad_norm": 2.570237874984741,
+      "learning_rate": 1.3595044779287543e-05,
+      "loss": 2.5297,
+      "mean_token_accuracy": 0.4622186854481697,
+      "step": 2550
+    },
+    {
+      "epoch": 0.22324932414755386,
+      "grad_norm": 2.757084608078003,
+      "learning_rate": 1.3538165392677288e-05,
+      "loss": 2.5758,
+      "mean_token_accuracy": 0.4610322922468185,
+      "step": 2560
+    },
+    {
+      "epoch": 0.22412139182000523,
+      "grad_norm": 2.5194716453552246,
+      "learning_rate": 1.3481154853398686e-05,
+      "loss": 2.5492,
+      "mean_token_accuracy": 0.4552470624446869,
+      "step": 2570
+    },
+    {
+      "epoch": 0.2249934594924566,
+      "grad_norm": 2.566094160079956,
+      "learning_rate": 1.3424015274717665e-05,
+      "loss": 2.5172,
+      "mean_token_accuracy": 0.464469176530838,
+      "step": 2580
+    },
+    {
+      "epoch": 0.225865527164908,
+      "grad_norm": 2.5102264881134033,
+      "learning_rate": 1.3366748774683376e-05,
+      "loss": 2.4922,
+      "mean_token_accuracy": 0.4739726006984711,
+      "step": 2590
+    },
+    {
+      "epoch": 0.2267375948373594,
+      "grad_norm": 2.4986090660095215,
+      "learning_rate": 1.3309357476049686e-05,
+      "loss": 2.4969,
+      "mean_token_accuracy": 0.47189334481954576,
+      "step": 2600
+    },
+    {
+      "epoch": 0.2267375948373594,
+      "eval_runtime": 4.1205,
+      "eval_samples_per_second": 266.23,
+      "eval_steps_per_second": 66.739,
+      "step": 2600
+    },
+    {
+      "epoch": 0.22760966250981077,
+      "grad_norm": 2.406562089920044,
+      "learning_rate": 1.3251843506196508e-05,
+      "loss": 2.5078,
+      "mean_token_accuracy": 0.46998532563447953,
+      "step": 2610
+    },
+    {
+      "epoch": 0.22848173018226214,
+      "grad_norm": 2.538019895553589,
+      "learning_rate": 1.3194208997050915e-05,
+      "loss": 2.4813,
+      "mean_token_accuracy": 0.47781310975551605,
+      "step": 2620
+    },
+    {
+      "epoch": 0.22935379785471352,
+      "grad_norm": 2.3137576580047607,
+      "learning_rate": 1.313645608500814e-05,
+      "loss": 2.5078,
+      "mean_token_accuracy": 0.47041340470314025,
+      "step": 2630
+    },
+    {
+      "epoch": 0.2302258655271649,
+      "grad_norm": 2.411092758178711,
+      "learning_rate": 1.3078586910852364e-05,
+      "loss": 2.5125,
+      "mean_token_accuracy": 0.4682729944586754,
+      "step": 2640
+    },
+    {
+      "epoch": 0.2310979331996163,
+      "grad_norm": 2.3885931968688965,
+      "learning_rate": 1.3020603619677378e-05,
+      "loss": 2.4836,
+      "mean_token_accuracy": 0.4687133103609085,
+      "step": 2650
+    },
+    {
+      "epoch": 0.23197000087206768,
+      "grad_norm": 2.590830087661743,
+      "learning_rate": 1.296250836080706e-05,
+      "loss": 2.4719,
+      "mean_token_accuracy": 0.4720523491501808,
+      "step": 2660
+    },
+    {
+      "epoch": 0.23284206854451905,
+      "grad_norm": 2.3284337520599365,
+      "learning_rate": 1.2904303287715702e-05,
+      "loss": 2.5031,
+      "mean_token_accuracy": 0.4667808249592781,
+      "step": 2670
+    },
+    {
+      "epoch": 0.23371413621697043,
+      "grad_norm": 2.471567392349243,
+      "learning_rate": 1.284599055794819e-05,
+      "loss": 2.4922,
+      "mean_token_accuracy": 0.4702177092432976,
+      "step": 2680
+    },
+    {
+      "epoch": 0.2345862038894218,
+      "grad_norm": 2.3384742736816406,
+      "learning_rate": 1.2787572333040022e-05,
+      "loss": 2.5328,
+      "mean_token_accuracy": 0.4617294520139694,
+      "step": 2690
+    },
+    {
+      "epoch": 0.2354582715618732,
+      "grad_norm": 2.604320526123047,
+      "learning_rate": 1.2729050778437197e-05,
+      "loss": 2.5016,
+      "mean_token_accuracy": 0.4691780835390091,
+      "step": 2700
+    },
+    {
+      "epoch": 0.2354582715618732,
+      "eval_runtime": 4.1368,
+      "eval_samples_per_second": 265.179,
+      "eval_steps_per_second": 66.476,
+      "step": 2700
+    },
+    {
+      "epoch": 0.2363303392343246,
+      "grad_norm": 2.2211201190948486,
+      "learning_rate": 1.2670428063415932e-05,
+      "loss": 2.5227,
+      "mean_token_accuracy": 0.4676492169499397,
+      "step": 2710
+    },
+    {
+      "epoch": 0.23720240690677596,
+      "grad_norm": 2.4421393871307373,
+      "learning_rate": 1.2611706361002254e-05,
+      "loss": 2.425,
+      "mean_token_accuracy": 0.48572651743888856,
+      "step": 2720
+    },
+    {
+      "epoch": 0.23807447457922734,
+      "grad_norm": 2.3915984630584717,
+      "learning_rate": 1.2552887847891462e-05,
+      "loss": 2.5406,
+      "mean_token_accuracy": 0.46364970654249194,
+      "step": 2730
+    },
+    {
+      "epoch": 0.23894654225167872,
+      "grad_norm": 2.4607255458831787,
+      "learning_rate": 1.2493974704367427e-05,
+      "loss": 2.4836,
+      "mean_token_accuracy": 0.47026663422584536,
+      "step": 2740
+    },
+    {
+      "epoch": 0.23981860992413012,
+      "grad_norm": 2.2948453426361084,
+      "learning_rate": 1.2434969114221777e-05,
+      "loss": 2.4699,
+      "mean_token_accuracy": 0.47407044768333434,
+      "step": 2750
+    },
+    {
+      "epoch": 0.2406906775965815,
+      "grad_norm": 2.5067508220672607,
+      "learning_rate": 1.237587326467296e-05,
+      "loss": 2.5078,
+      "mean_token_accuracy": 0.46873776614665985,
+      "step": 2760
+    },
+    {
+      "epoch": 0.24156274526903287,
+      "grad_norm": 2.3885128498077393,
+      "learning_rate": 1.2316689346285146e-05,
+      "loss": 2.4922,
+      "mean_token_accuracy": 0.471991191804409,
+      "step": 2770
+    },
+    {
+      "epoch": 0.24243481294148425,
+      "grad_norm": 2.1619558334350586,
+      "learning_rate": 1.2257419552887047e-05,
+      "loss": 2.507,
+      "mean_token_accuracy": 0.47196673452854154,
+      "step": 2780
+    },
+    {
+      "epoch": 0.24330688061393563,
+      "grad_norm": 2.2310330867767334,
+      "learning_rate": 1.2198066081490585e-05,
+      "loss": 2.5102,
+      "mean_token_accuracy": 0.4685543060302734,
+      "step": 2790
+    },
+    {
+      "epoch": 0.24417894828638703,
+      "grad_norm": 2.5117921829223633,
+      "learning_rate": 1.213863113220946e-05,
+      "loss": 2.4617,
+      "mean_token_accuracy": 0.4767857104539871,
+      "step": 2800
+    },
+    {
+      "epoch": 0.24417894828638703,
+      "eval_runtime": 4.1246,
+      "eval_samples_per_second": 265.966,
+      "eval_steps_per_second": 66.673,
+      "step": 2800
+    },
+    {
+      "epoch": 0.2450510159588384,
+      "grad_norm": 2.364210367202759,
+      "learning_rate": 1.2079116908177592e-05,
+      "loss": 2.4836,
+      "mean_token_accuracy": 0.4738502994179726,
+      "step": 2810
+    },
+    {
+      "epoch": 0.24592308363128978,
+      "grad_norm": 2.2354042530059814,
+      "learning_rate": 1.2019525615467462e-05,
+      "loss": 2.4422,
+      "mean_token_accuracy": 0.48214285671710966,
+      "step": 2820
+    },
+    {
+      "epoch": 0.24679515130374116,
+      "grad_norm": 2.3318979740142822,
+      "learning_rate": 1.1959859463008316e-05,
+      "loss": 2.4828,
+      "mean_token_accuracy": 0.46909246444702146,
+      "step": 2830
+    },
+    {
+      "epoch": 0.24766721897619257,
+      "grad_norm": 2.548313617706299,
+      "learning_rate": 1.1900120662504315e-05,
+      "loss": 2.4641,
+      "mean_token_accuracy": 0.4708537206053734,
+      "step": 2840
+    },
+    {
+      "epoch": 0.24853928664864394,
+      "grad_norm": 2.2255141735076904,
+      "learning_rate": 1.1840311428352536e-05,
+      "loss": 2.4648,
+      "mean_token_accuracy": 0.47885274291038515,
+      "step": 2850
+    },
+    {
+      "epoch": 0.24941135432109532,
+      "grad_norm": 2.438560724258423,
+      "learning_rate": 1.1780433977560879e-05,
+      "loss": 2.4602,
+      "mean_token_accuracy": 0.4786081239581108,
+      "step": 2860
+    },
+    {
+      "epoch": 0.2502834219935467,
+      "grad_norm": 2.4244706630706787,
+      "learning_rate": 1.1720490529665904e-05,
+      "loss": 2.482,
+      "mean_token_accuracy": 0.4734466731548309,
+      "step": 2870
+    },
+    {
+      "epoch": 0.2511554896659981,
+      "grad_norm": 2.156982183456421,
+      "learning_rate": 1.1660483306650558e-05,
+      "loss": 2.4125,
+      "mean_token_accuracy": 0.48068737536668776,
+      "step": 2880
+    },
+    {
+      "epoch": 0.25202755733844945,
+      "grad_norm": 2.1535656452178955,
+      "learning_rate": 1.160041453286179e-05,
+      "loss": 2.55,
+      "mean_token_accuracy": 0.46462817788124083,
+      "step": 2890
+    },
+    {
+      "epoch": 0.25289962501090085,
+      "grad_norm": 2.2408533096313477,
+      "learning_rate": 1.154028643492812e-05,
+      "loss": 2.468,
+      "mean_token_accuracy": 0.47481653839349747,
+      "step": 2900
+    },
+    {
+      "epoch": 0.25289962501090085,
+      "eval_runtime": 4.1272,
+      "eval_samples_per_second": 265.796,
+      "eval_steps_per_second": 66.631,
+      "step": 2900
+    },
+    {
+      "epoch": 0.2537716926833522,
+      "grad_norm": 2.162675380706787,
+      "learning_rate": 1.1480101241677097e-05,
+      "loss": 2.4453,
+      "mean_token_accuracy": 0.47963551580905917,
+      "step": 2910
+    },
+    {
+      "epoch": 0.2546437603558036,
+      "grad_norm": 2.3737261295318604,
+      "learning_rate": 1.1419861184052669e-05,
+      "loss": 2.4297,
+      "mean_token_accuracy": 0.4839041084051132,
+      "step": 2920
+    },
+    {
+      "epoch": 0.255515828028255,
+      "grad_norm": 2.3079073429107666,
+      "learning_rate": 1.1359568495032505e-05,
+      "loss": 2.4961,
+      "mean_token_accuracy": 0.4703522503376007,
+      "step": 2930
+    },
+    {
+      "epoch": 0.25638789570070636,
+      "grad_norm": 2.304546594619751,
+      "learning_rate": 1.1299225409545207e-05,
+      "loss": 2.4383,
+      "mean_token_accuracy": 0.4810631603002548,
+      "step": 2940
+    },
+    {
+      "epoch": 0.25725996337315776,
+      "grad_norm": 2.3197596073150635,
+      "learning_rate": 1.123883416438748e-05,
+      "loss": 2.4602,
+      "mean_token_accuracy": 0.4756849333643913,
+      "step": 2950
+    },
+    {
+      "epoch": 0.2581320310456091,
+      "grad_norm": 2.6116833686828613,
+      "learning_rate": 1.1178396998141206e-05,
+      "loss": 2.4445,
+      "mean_token_accuracy": 0.4768835574388504,
+      "step": 2960
+    },
+    {
+      "epoch": 0.2590040987180605,
+      "grad_norm": 2.2649593353271484,
+      "learning_rate": 1.1117916151090469e-05,
+      "loss": 2.4164,
+      "mean_token_accuracy": 0.4851516604423523,
+      "step": 2970
+    },
+    {
+      "epoch": 0.2598761663905119,
+      "grad_norm": 2.3760316371917725,
+      "learning_rate": 1.1057393865138513e-05,
+      "loss": 2.4305,
+      "mean_token_accuracy": 0.4774706453084946,
+      "step": 2980
+    },
+    {
+      "epoch": 0.26074823406296327,
+      "grad_norm": 2.4932713508605957,
+      "learning_rate": 1.099683238372464e-05,
+      "loss": 2.4379,
+      "mean_token_accuracy": 0.48110322952270507,
+      "step": 2990
+    },
+    {
+      "epoch": 0.2616203017354147,
+      "grad_norm": 2.40559458732605,
+      "learning_rate": 1.0936233951741052e-05,
+      "loss": 2.407,
+      "mean_token_accuracy": 0.48287671208381655,
+      "step": 3000
+    },
+    {
+      "epoch": 0.2616203017354147,
+      "eval_runtime": 4.1296,
+      "eval_samples_per_second": 265.642,
+      "eval_steps_per_second": 66.592,
+      "step": 3000
+    },
+    {
+      "epoch": 0.2624923694078661,
+      "grad_norm": 2.17433762550354,
+      "learning_rate": 1.0875600815449624e-05,
+      "loss": 2.375,
+      "mean_token_accuracy": 0.4947896286845207,
+      "step": 3010
+    },
+    {
+      "epoch": 0.2633644370803174,
+      "grad_norm": 2.4722836017608643,
+      "learning_rate": 1.081493522239866e-05,
+      "loss": 2.4648,
+      "mean_token_accuracy": 0.47459638118743896,
+      "step": 3020
+    },
+    {
+      "epoch": 0.26423650475276883,
+      "grad_norm": 2.380868673324585,
+      "learning_rate": 1.075423942133957e-05,
+      "loss": 2.4008,
+      "mean_token_accuracy": 0.4884784698486328,
+      "step": 3030
+    },
+    {
+      "epoch": 0.2651085724252202,
+      "grad_norm": 2.2299351692199707,
+      "learning_rate": 1.0693515662143505e-05,
+      "loss": 2.457,
+      "mean_token_accuracy": 0.4779965773224831,
+      "step": 3040
+    },
+    {
+      "epoch": 0.2659806400976716,
+      "grad_norm": 2.31659197807312,
+      "learning_rate": 1.0632766195717979e-05,
+      "loss": 2.4203,
+      "mean_token_accuracy": 0.4778620317578316,
+      "step": 3050
+    },
+    {
+      "epoch": 0.266852707770123,
+      "grad_norm": 2.218075752258301,
+      "learning_rate": 1.0571993273923412e-05,
+      "loss": 2.4273,
+      "mean_token_accuracy": 0.4800146758556366,
+      "step": 3060
+    },
+    {
+      "epoch": 0.26772477544257434,
+      "grad_norm": 2.0592129230499268,
+      "learning_rate": 1.0511199149489673e-05,
+      "loss": 2.4133,
+      "mean_token_accuracy": 0.48199608474969863,
+      "step": 3070
+    },
+    {
+      "epoch": 0.26859684311502574,
+      "grad_norm": 2.2392261028289795,
+      "learning_rate": 1.0450386075932571e-05,
+      "loss": 2.4152,
+      "mean_token_accuracy": 0.48124999552965164,
+      "step": 3080
+    },
+    {
+      "epoch": 0.2694689107874771,
+      "grad_norm": 2.229623317718506,
+      "learning_rate": 1.0389556307470316e-05,
+      "loss": 2.4352,
+      "mean_token_accuracy": 0.47666340619325637,
+      "step": 3090
+    },
+    {
+      "epoch": 0.2703409784599285,
+      "grad_norm": 2.517035722732544,
+      "learning_rate": 1.0328712098939968e-05,
+      "loss": 2.5031,
+      "mean_token_accuracy": 0.47388698607683183,
+      "step": 3100
+    },
+    {
+      "epoch": 0.2703409784599285,
+      "eval_runtime": 4.1258,
+      "eval_samples_per_second": 265.885,
+      "eval_steps_per_second": 66.653,
+      "step": 3100
+    },
+    {
+      "epoch": 0.2712130461323799,
+      "grad_norm": 2.2601687908172607,
+      "learning_rate": 1.0267855705713854e-05,
+      "loss": 2.3766,
+      "mean_token_accuracy": 0.48773238658905027,
+      "step": 3110
+    },
+    {
+      "epoch": 0.27208511380483125,
+      "grad_norm": 2.390767812728882,
+      "learning_rate": 1.020698938361595e-05,
+      "loss": 2.4531,
+      "mean_token_accuracy": 0.4725905120372772,
+      "step": 3120
+    },
+    {
+      "epoch": 0.27295718147728265,
+      "grad_norm": 2.3313021659851074,
+      "learning_rate": 1.0146115388838293e-05,
+      "loss": 2.3766,
+      "mean_token_accuracy": 0.4869129091501236,
+      "step": 3130
+    },
+    {
+      "epoch": 0.273829249149734,
+      "grad_norm": 2.6201083660125732,
+      "learning_rate": 1.0085235977857322e-05,
+      "loss": 2.3461,
+      "mean_token_accuracy": 0.4963551893830299,
+      "step": 3140
+    },
+    {
+      "epoch": 0.2747013168221854,
+      "grad_norm": 2.2642321586608887,
+      "learning_rate": 1.002435340735024e-05,
+      "loss": 2.3953,
+      "mean_token_accuracy": 0.482399708032608,
+      "step": 3150
+    },
+    {
+      "epoch": 0.2755733844946368,
+      "grad_norm": 2.377150297164917,
+      "learning_rate": 9.963469934111374e-06,
+      "loss": 2.368,
+      "mean_token_accuracy": 0.4929549887776375,
+      "step": 3160
+    },
+    {
+      "epoch": 0.27644545216708816,
+      "grad_norm": 2.124990463256836,
+      "learning_rate": 9.90258781496851e-06,
+      "loss": 2.4477,
+      "mean_token_accuracy": 0.47852250635623933,
+      "step": 3170
+    },
+    {
+      "epoch": 0.27731751983953956,
+      "grad_norm": 2.2240705490112305,
+      "learning_rate": 9.841709306699245e-06,
+      "loss": 2.4398,
+      "mean_token_accuracy": 0.4784246563911438,
+      "step": 3180
+    },
+    {
+      "epoch": 0.2781895875119909,
+      "grad_norm": 2.230926990509033,
+      "learning_rate": 9.78083666594732e-06,
+      "loss": 2.4586,
+      "mean_token_accuracy": 0.4777524918317795,
+      "step": 3190
+    },
+    {
+      "epoch": 0.2790616551844423,
+      "grad_norm": 2.1134297847747803,
+      "learning_rate": 9.719972149138985e-06,
+      "loss": 2.4109,
+      "mean_token_accuracy": 0.48686399310827255,
+      "step": 3200
+    },
+    {
+      "epoch": 0.2790616551844423,
+      "eval_runtime": 4.1351,
+      "eval_samples_per_second": 265.293,
+      "eval_steps_per_second": 66.505,
+      "step": 3200
+    },
+    {
+      "epoch": 0.2799337228568937,
+      "grad_norm": 2.3881607055664062,
+      "learning_rate": 9.659118012399352e-06,
+      "loss": 2.4234,
+      "mean_token_accuracy": 0.47910958975553514,
+      "step": 3210
+    },
+    {
+      "epoch": 0.28080579052934507,
+      "grad_norm": 2.328941583633423,
+      "learning_rate": 9.598276511468763e-06,
+      "loss": 2.3266,
+      "mean_token_accuracy": 0.49437377601861954,
+      "step": 3220
+    },
+    {
+      "epoch": 0.28167785820179647,
+      "grad_norm": 2.4462809562683105,
+      "learning_rate": 9.537449901619174e-06,
+      "loss": 2.418,
+      "mean_token_accuracy": 0.47892612367868426,
+      "step": 3230
+    },
+    {
+      "epoch": 0.2825499258742478,
+      "grad_norm": 2.4129433631896973,
+      "learning_rate": 9.476640437570562e-06,
+      "loss": 2.5,
+      "mean_token_accuracy": 0.47085371911525725,
+      "step": 3240
+    },
+    {
+      "epoch": 0.2834219935466992,
+      "grad_norm": 2.2412469387054443,
+      "learning_rate": 9.415850373407342e-06,
+      "loss": 2.3937,
+      "mean_token_accuracy": 0.48611790835857394,
+      "step": 3250
+    },
+    {
+      "epoch": 0.28429406121915063,
+      "grad_norm": 2.274876832962036,
+      "learning_rate": 9.355081962494815e-06,
+      "loss": 2.3258,
+      "mean_token_accuracy": 0.49866669028997423,
+      "step": 3260
+    },
+    {
+      "epoch": 0.285166128891602,
+      "grad_norm": 2.426975727081299,
+      "learning_rate": 9.294337457395638e-06,
+      "loss": 2.3586,
+      "mean_token_accuracy": 0.4943248525261879,
+      "step": 3270
+    },
+    {
+      "epoch": 0.2860381965640534,
+      "grad_norm": 2.616424560546875,
+      "learning_rate": 9.233619109786332e-06,
+      "loss": 2.3852,
+      "mean_token_accuracy": 0.49368884563446047,
+      "step": 3280
+    },
+    {
+      "epoch": 0.28691026423650473,
+      "grad_norm": 2.2738358974456787,
+      "learning_rate": 9.172929170373804e-06,
+      "loss": 2.4328,
+      "mean_token_accuracy": 0.4826076328754425,
+      "step": 3290
+    },
+    {
+      "epoch": 0.28778233190895613,
+      "grad_norm": 2.235522508621216,
+      "learning_rate": 9.112269888811934e-06,
+      "loss": 2.4297,
+      "mean_token_accuracy": 0.47972113341093064,
+      "step": 3300
+    },
+    {
+      "epoch": 0.28778233190895613,
+      "eval_runtime": 4.1164,
+      "eval_samples_per_second": 266.497,
+      "eval_steps_per_second": 66.806,
+      "step": 3300
+    },
+    {
+      "epoch": 0.28865439958140754,
+      "grad_norm": 2.2078049182891846,
+      "learning_rate": 9.051643513618176e-06,
+      "loss": 2.3875,
+      "mean_token_accuracy": 0.48949363827705383,
+      "step": 3310
+    },
+    {
+      "epoch": 0.2895264672538589,
+      "grad_norm": 2.2486748695373535,
+      "learning_rate": 8.99105229209021e-06,
+      "loss": 2.4516,
+      "mean_token_accuracy": 0.4734711334109306,
+      "step": 3320
+    },
+    {
+      "epoch": 0.2903985349263103,
+      "grad_norm": 2.2248547077178955,
+      "learning_rate": 8.930498470222641e-06,
+      "loss": 2.4797,
+      "mean_token_accuracy": 0.4729818940162659,
+      "step": 3330
+    },
+    {
+      "epoch": 0.29127060259876164,
+      "grad_norm": 2.356717109680176,
+      "learning_rate": 8.86998429262374e-06,
+      "loss": 2.3867,
+      "mean_token_accuracy": 0.4878180056810379,
+      "step": 3340
+    },
+    {
+      "epoch": 0.29214267027121305,
+      "grad_norm": 2.064194440841675,
+      "learning_rate": 8.809512002432252e-06,
+      "loss": 2.3586,
+      "mean_token_accuracy": 0.4933463826775551,
+      "step": 3350
+    },
+    {
+      "epoch": 0.29301473794366445,
+      "grad_norm": 2.367587089538574,
+      "learning_rate": 8.749083841234235e-06,
+      "loss": 2.4414,
+      "mean_token_accuracy": 0.4839774936437607,
+      "step": 3360
+    },
+    {
+      "epoch": 0.2938868056161158,
+      "grad_norm": 2.2674319744110107,
+      "learning_rate": 8.688702048979974e-06,
+      "loss": 2.4281,
+      "mean_token_accuracy": 0.4836839586496353,
+      "step": 3370
+    },
+    {
+      "epoch": 0.2947588732885672,
+      "grad_norm": 2.0621237754821777,
+      "learning_rate": 8.628368863900954e-06,
+      "loss": 2.3891,
+      "mean_token_accuracy": 0.4839285686612129,
+      "step": 3380
+    },
+    {
+      "epoch": 0.29563094096101855,
+      "grad_norm": 2.0860073566436768,
+      "learning_rate": 8.568086522426884e-06,
+      "loss": 2.4141,
+      "mean_token_accuracy": 0.48272994011640546,
+      "step": 3390
+    },
+    {
+      "epoch": 0.29650300863346996,
+      "grad_norm": 1.9796204566955566,
+      "learning_rate": 8.507857259102814e-06,
+      "loss": 2.3805,
+      "mean_token_accuracy": 0.49032534658908844,
+      "step": 3400
+    },
+    {
+      "epoch": 0.29650300863346996,
+      "eval_runtime": 4.1278,
+      "eval_samples_per_second": 265.762,
+      "eval_steps_per_second": 66.622,
+      "step": 3400
+    },
+    {
+      "epoch": 0.29737507630592136,
+      "grad_norm": 2.121955156326294,
+      "learning_rate": 8.447683306506279e-06,
+      "loss": 2.4219,
+      "mean_token_accuracy": 0.4785102754831314,
+      "step": 3410
+    },
+    {
+      "epoch": 0.2982471439783727,
+      "grad_norm": 2.1485984325408936,
+      "learning_rate": 8.387566895164566e-06,
+      "loss": 2.3625,
+      "mean_token_accuracy": 0.48655821681022643,
+      "step": 3420
+    },
+    {
+      "epoch": 0.2991192116508241,
+      "grad_norm": 2.2083539962768555,
+      "learning_rate": 8.327510253472023e-06,
+      "loss": 2.4484,
+      "mean_token_accuracy": 0.4772749498486519,
+      "step": 3430
+    },
+    {
+      "epoch": 0.29999127932327546,
+      "grad_norm": 2.421006679534912,
+      "learning_rate": 8.267515607607458e-06,
+      "loss": 2.3602,
+      "mean_token_accuracy": 0.49372553676366804,
+      "step": 3440
+    },
+    {
+      "epoch": 0.30086334699572687,
+      "grad_norm": 2.2537267208099365,
+      "learning_rate": 8.207585181451611e-06,
+      "loss": 2.3539,
+      "mean_token_accuracy": 0.4897137925028801,
+      "step": 3450
+    },
+    {
+      "epoch": 0.30173541466817827,
+      "grad_norm": 2.2362277507781982,
+      "learning_rate": 8.147721196504736e-06,
+      "loss": 2.3469,
+      "mean_token_accuracy": 0.4916218236088753,
+      "step": 3460
+    },
+    {
+      "epoch": 0.3026074823406296,
+      "grad_norm": 2.214308500289917,
+      "learning_rate": 8.08792587180424e-06,
+      "loss": 2.4312,
+      "mean_token_accuracy": 0.4826320931315422,
+      "step": 3470
+    },
+    {
+      "epoch": 0.303479550013081,
+      "grad_norm": 2.031965494155884,
+      "learning_rate": 8.028201423842437e-06,
+      "loss": 2.3883,
+      "mean_token_accuracy": 0.4820817068219185,
+      "step": 3480
+    },
+    {
+      "epoch": 0.30435161768553237,
+      "grad_norm": 2.043548345565796,
+      "learning_rate": 7.96855006648438e-06,
+      "loss": 2.3687,
+      "mean_token_accuracy": 0.490300877392292,
+      "step": 3490
+    },
+    {
+      "epoch": 0.3052236853579838,
+      "grad_norm": 2.2613983154296875,
+      "learning_rate": 7.908974010885795e-06,
+      "loss": 2.4477,
+      "mean_token_accuracy": 0.48006359934806825,
+      "step": 3500
+    },
+    {
+      "epoch": 0.3052236853579838,
+      "eval_runtime": 4.139,
+      "eval_samples_per_second": 265.042,
+      "eval_steps_per_second": 66.442,
+      "step": 3500
+    },
+    {
+      "epoch": 0.3060957530304352,
+      "grad_norm": 2.0631442070007324,
+      "learning_rate": 7.849475465411136e-06,
+      "loss": 2.343,
+      "mean_token_accuracy": 0.49260029643774034,
+      "step": 3510
+    },
+    {
+      "epoch": 0.30696782070288653,
+      "grad_norm": 2.1512322425842285,
+      "learning_rate": 7.790056635551704e-06,
+      "loss": 2.3859,
+      "mean_token_accuracy": 0.4881727024912834,
+      "step": 3520
+    },
+    {
+      "epoch": 0.30783988837533793,
+      "grad_norm": 2.205146312713623,
+      "learning_rate": 7.730719723843903e-06,
+      "loss": 2.425,
+      "mean_token_accuracy": 0.48297456055879595,
+      "step": 3530
+    },
+    {
+      "epoch": 0.3087119560477893,
+      "grad_norm": 2.0500307083129883,
+      "learning_rate": 7.671466929787598e-06,
+      "loss": 2.3531,
+      "mean_token_accuracy": 0.49005626142024994,
+      "step": 3540
+    },
+    {
+      "epoch": 0.3095840237202407,
+      "grad_norm": 2.2774555683135986,
+      "learning_rate": 7.61230044976458e-06,
+      "loss": 2.4227,
+      "mean_token_accuracy": 0.4837206467986107,
+      "step": 3550
+    },
+    {
+      "epoch": 0.3104560913926921,
+      "grad_norm": 2.146467685699463,
+      "learning_rate": 7.553222476957157e-06,
+      "loss": 2.4066,
+      "mean_token_accuracy": 0.4871453046798706,
+      "step": 3560
+    },
+    {
+      "epoch": 0.31132815906514344,
+      "grad_norm": 2.130199432373047,
+      "learning_rate": 7.494235201266849e-06,
+      "loss": 2.3703,
+      "mean_token_accuracy": 0.4913405105471611,
+      "step": 3570
+    },
+    {
+      "epoch": 0.31220022673759484,
+      "grad_norm": 2.045860528945923,
+      "learning_rate": 7.435340809233218e-06,
+      "loss": 2.3805,
+      "mean_token_accuracy": 0.49107863306999205,
+      "step": 3580
+    },
+    {
+      "epoch": 0.3130722944100462,
+      "grad_norm": 2.120117664337158,
+      "learning_rate": 7.376541483952811e-06,
+      "loss": 2.4129,
+      "mean_token_accuracy": 0.48899217396974565,
+      "step": 3590
+    },
+    {
+      "epoch": 0.3139443620824976,
+      "grad_norm": 2.115574836730957,
+      "learning_rate": 7.3178394049982485e-06,
+      "loss": 2.3492,
+      "mean_token_accuracy": 0.49445939213037493,
+      "step": 3600
+    },
+    {
+      "epoch": 0.3139443620824976,
+      "eval_runtime": 4.1276,
+      "eval_samples_per_second": 265.769,
+      "eval_steps_per_second": 66.624,
+      "step": 3600
+    },
+    {
+      "epoch": 0.314816429754949,
+      "grad_norm": 2.1968657970428467,
+      "learning_rate": 7.259236748337421e-06,
+      "loss": 2.3672,
+      "mean_token_accuracy": 0.4955968618392944,
+      "step": 3610
+    },
+    {
+      "epoch": 0.31568849742740035,
+      "grad_norm": 2.1085331439971924,
+      "learning_rate": 7.20073568625284e-06,
+      "loss": 2.4242,
+      "mean_token_accuracy": 0.4839285746216774,
+      "step": 3620
+    },
+    {
+      "epoch": 0.31656056509985175,
+      "grad_norm": 2.282261371612549,
+      "learning_rate": 7.1423383872611045e-06,
+      "loss": 2.4078,
+      "mean_token_accuracy": 0.4873287633061409,
+      "step": 3630
+    },
+    {
+      "epoch": 0.31743263277230316,
+      "grad_norm": 2.081146001815796,
+      "learning_rate": 7.084047016032528e-06,
+      "loss": 2.4125,
+      "mean_token_accuracy": 0.4849804311990738,
+      "step": 3640
+    },
+    {
+      "epoch": 0.3183047004447545,
+      "grad_norm": 2.186483860015869,
+      "learning_rate": 7.025863733310894e-06,
+      "loss": 2.3113,
+      "mean_token_accuracy": 0.49919275790452955,
+      "step": 3650
+    },
+    {
+      "epoch": 0.3191767681172059,
+      "grad_norm": 2.1311113834381104,
+      "learning_rate": 6.967790695833363e-06,
+      "loss": 2.418,
+      "mean_token_accuracy": 0.480821917951107,
+      "step": 3660
+    },
+    {
+      "epoch": 0.32004883578965726,
+      "grad_norm": 2.197826862335205,
+      "learning_rate": 6.909830056250527e-06,
+      "loss": 2.3242,
+      "mean_token_accuracy": 0.4950107797980309,
+      "step": 3670
+    },
+    {
+      "epoch": 0.32092090346210866,
+      "grad_norm": 2.15081787109375,
+      "learning_rate": 6.851983963046612e-06,
+      "loss": 2.3578,
+      "mean_token_accuracy": 0.4879647806286812,
+      "step": 3680
+    },
+    {
+      "epoch": 0.32179297113456007,
+      "grad_norm": 2.2143335342407227,
+      "learning_rate": 6.794254560459843e-06,
+      "loss": 2.3523,
+      "mean_token_accuracy": 0.4926002904772758,
+      "step": 3690
+    },
+    {
+      "epoch": 0.3226650388070114,
+      "grad_norm": 2.1243338584899902,
+      "learning_rate": 6.736643988402958e-06,
+      "loss": 2.3445,
+      "mean_token_accuracy": 0.491523976624012,
+      "step": 3700
+    },
+    {
+      "epoch": 0.3226650388070114,
+      "eval_runtime": 4.1266,
+      "eval_samples_per_second": 265.835,
+      "eval_steps_per_second": 66.641,
+      "step": 3700
+    },
+    {
+      "epoch": 0.3235371064794628,
+      "grad_norm": 2.232715129852295,
+      "learning_rate": 6.679154382383883e-06,
+      "loss": 2.4125,
+      "mean_token_accuracy": 0.48493150621652603,
+      "step": 3710
+    },
+    {
+      "epoch": 0.32440917415191417,
+      "grad_norm": 2.0074570178985596,
+      "learning_rate": 6.621787873426581e-06,
+      "loss": 2.4117,
+      "mean_token_accuracy": 0.48068737834692,
+      "step": 3720
+    },
+    {
+      "epoch": 0.3252812418243656,
+      "grad_norm": 2.1245803833007812,
+      "learning_rate": 6.564546587992054e-06,
+      "loss": 2.334,
+      "mean_token_accuracy": 0.49477739483118055,
+      "step": 3730
+    },
+    {
+      "epoch": 0.326153309496817,
+      "grad_norm": 2.26165771484375,
+      "learning_rate": 6.507432647899519e-06,
+      "loss": 2.4367,
+      "mean_token_accuracy": 0.47766634076833725,
+      "step": 3740
+    },
+    {
+      "epoch": 0.3270253771692683,
+      "grad_norm": 2.1744251251220703,
+      "learning_rate": 6.450448170247757e-06,
+      "loss": 2.3188,
+      "mean_token_accuracy": 0.4967098832130432,
+      "step": 3750
+    },
+    {
+      "epoch": 0.32789744484171973,
+      "grad_norm": 2.172355890274048,
+      "learning_rate": 6.393595267336639e-06,
+      "loss": 2.3609,
+      "mean_token_accuracy": 0.4897871851921082,
+      "step": 3760
+    },
+    {
+      "epoch": 0.3287695125141711,
+      "grad_norm": 2.1196014881134033,
+      "learning_rate": 6.3368760465888226e-06,
+      "loss": 2.3984,
+      "mean_token_accuracy": 0.48364725708961487,
+      "step": 3770
+    },
+    {
+      "epoch": 0.3296415801866225,
+      "grad_norm": 2.280076265335083,
+      "learning_rate": 6.280292610471639e-06,
+      "loss": 2.4102,
+      "mean_token_accuracy": 0.48471134901046753,
+      "step": 3780
+    },
+    {
+      "epoch": 0.3305136478590739,
+      "grad_norm": 2.1347899436950684,
+      "learning_rate": 6.223847056419154e-06,
+      "loss": 2.3781,
+      "mean_token_accuracy": 0.4850171208381653,
+      "step": 3790
+    },
+    {
+      "epoch": 0.33138571553152524,
+      "grad_norm": 2.0589747428894043,
+      "learning_rate": 6.1675414767544285e-06,
+      "loss": 2.3773,
+      "mean_token_accuracy": 0.48522505164146423,
+      "step": 3800
+    },
+    {
+      "epoch": 0.33138571553152524,
+      "eval_runtime": 4.1359,
+      "eval_samples_per_second": 265.237,
+      "eval_steps_per_second": 66.491,
+      "step": 3800
+    },
+    {
+      "epoch": 0.33225778320397664,
+      "grad_norm": 2.307960033416748,
+      "learning_rate": 6.111377958611948e-06,
+      "loss": 2.3523,
+      "mean_token_accuracy": 0.49088796228170395,
+      "step": 3810
+    },
+    {
+      "epoch": 0.333129850876428,
+      "grad_norm": 2.280006170272827,
+      "learning_rate": 6.055358583860267e-06,
+      "loss": 2.3648,
+      "mean_token_accuracy": 0.48813600838184357,
+      "step": 3820
+    },
+    {
+      "epoch": 0.3340019185488794,
+      "grad_norm": 2.254883289337158,
+      "learning_rate": 5.99948542902483e-06,
+      "loss": 2.3293,
+      "mean_token_accuracy": 0.49653864800930025,
+      "step": 3830
+    },
+    {
+      "epoch": 0.3348739862213308,
+      "grad_norm": 1.9811010360717773,
+      "learning_rate": 5.943760565211011e-06,
+      "loss": 2.4008,
+      "mean_token_accuracy": 0.481568007171154,
+      "step": 3840
+    },
+    {
+      "epoch": 0.33574605389378215,
+      "grad_norm": 2.3965861797332764,
+      "learning_rate": 5.8881860580273285e-06,
+      "loss": 2.4039,
+      "mean_token_accuracy": 0.485922209918499,
+      "step": 3850
+    },
+    {
+      "epoch": 0.33661812156623355,
+      "grad_norm": 2.1080493927001953,
+      "learning_rate": 5.832763967508885e-06,
+      "loss": 2.4672,
+      "mean_token_accuracy": 0.4722602769732475,
+      "step": 3860
+    },
+    {
+      "epoch": 0.3374901892386849,
+      "grad_norm": 2.0131046772003174,
+      "learning_rate": 5.777496348041009e-06,
+      "loss": 2.3266,
+      "mean_token_accuracy": 0.5004647761583328,
+      "step": 3870
+    },
+    {
+      "epoch": 0.3383622569111363,
+      "grad_norm": 2.074880361557007,
+      "learning_rate": 5.722385248283092e-06,
+      "loss": 2.366,
+      "mean_token_accuracy": 0.49074119329452515,
+      "step": 3880
+    },
+    {
+      "epoch": 0.3392343245835877,
+      "grad_norm": 2.1542370319366455,
+      "learning_rate": 5.667432711092651e-06,
+      "loss": 2.3734,
+      "mean_token_accuracy": 0.48516389429569245,
+      "step": 3890
+    },
+    {
+      "epoch": 0.34010639225603906,
+      "grad_norm": 2.1824331283569336,
+      "learning_rate": 5.61264077344962e-06,
+      "loss": 2.4047,
+      "mean_token_accuracy": 0.4858365938067436,
+      "step": 3900
+    },
+    {
+      "epoch": 0.34010639225603906,
+      "eval_runtime": 4.1271,
+      "eval_samples_per_second": 265.807,
+      "eval_steps_per_second": 66.633,
+      "step": 3900
+    },
+    {
+      "epoch": 0.34097845992849046,
+      "grad_norm": 2.2174854278564453,
+      "learning_rate": 5.558011466380824e-06,
+      "loss": 2.332,
+      "mean_token_accuracy": 0.4895425647497177,
+      "step": 3910
+    },
+    {
+      "epoch": 0.3418505276009418,
+      "grad_norm": 2.2615621089935303,
+      "learning_rate": 5.5035468148846926e-06,
+      "loss": 2.3547,
+      "mean_token_accuracy": 0.4897382602095604,
+      "step": 3920
+    },
+    {
+      "epoch": 0.3427225952733932,
+      "grad_norm": 1.976144552230835,
+      "learning_rate": 5.449248837856224e-06,
+      "loss": 2.3344,
+      "mean_token_accuracy": 0.49690557420253756,
+      "step": 3930
+    },
+    {
+      "epoch": 0.3435946629458446,
+      "grad_norm": 2.3247182369232178,
+      "learning_rate": 5.395119548012112e-06,
+      "loss": 2.4062,
+      "mean_token_accuracy": 0.4821917787194252,
+      "step": 3940
+    },
+    {
+      "epoch": 0.34446673061829597,
+      "grad_norm": 2.0668063163757324,
+      "learning_rate": 5.34116095181616e-06,
+      "loss": 2.3508,
+      "mean_token_accuracy": 0.49102250188589097,
+      "step": 3950
+    },
+    {
+      "epoch": 0.3453387982907474,
+      "grad_norm": 2.0769951343536377,
+      "learning_rate": 5.287375049404909e-06,
+      "loss": 2.3492,
+      "mean_token_accuracy": 0.49485077410936357,
+      "step": 3960
+    },
+    {
+      "epoch": 0.3462108659631987,
+      "grad_norm": 2.0995266437530518,
+      "learning_rate": 5.233763834513479e-06,
+      "loss": 2.3102,
+      "mean_token_accuracy": 0.49756604582071307,
+      "step": 3970
+    },
+    {
+      "epoch": 0.3470829336356501,
+      "grad_norm": 2.2599644660949707,
+      "learning_rate": 5.180329294401685e-06,
+      "loss": 2.2895,
+      "mean_token_accuracy": 0.5033390417695045,
+      "step": 3980
+    },
+    {
+      "epoch": 0.34795500130810153,
+      "grad_norm": 2.1868715286254883,
+      "learning_rate": 5.127073409780352e-06,
+      "loss": 2.3727,
+      "mean_token_accuracy": 0.48818493634462357,
+      "step": 3990
+    },
+    {
+      "epoch": 0.3488270689805529,
+      "grad_norm": 2.258998155593872,
+      "learning_rate": 5.0739981547379215e-06,
+      "loss": 2.3609,
+      "mean_token_accuracy": 0.4893101766705513,
+      "step": 4000
+    },
+    {
+      "epoch": 0.3488270689805529,
+      "eval_runtime": 4.1316,
+      "eval_samples_per_second": 265.513,
+      "eval_steps_per_second": 66.56,
+      "step": 4000
+    },
+    {
+      "epoch": 0.3496991366530043,
+      "grad_norm": 2.1525726318359375,
+      "learning_rate": 5.02110549666724e-06,
+      "loss": 2.3207,
+      "mean_token_accuracy": 0.49832436740398406,
+      "step": 4010
+    },
+    {
+      "epoch": 0.35057120432545563,
+      "grad_norm": 2.147263765335083,
+      "learning_rate": 4.968397396192675e-06,
+      "loss": 2.309,
+      "mean_token_accuracy": 0.49854451715946196,
+      "step": 4020
+    },
+    {
+      "epoch": 0.35144327199790704,
+      "grad_norm": 2.076788902282715,
+      "learning_rate": 4.91587580709739e-06,
+      "loss": 2.357,
+      "mean_token_accuracy": 0.49258806109428405,
+      "step": 4030
+    },
+    {
+      "epoch": 0.35231533967035844,
+      "grad_norm": 2.206960678100586,
+      "learning_rate": 4.863542676250972e-06,
+      "loss": 2.2434,
+      "mean_token_accuracy": 0.5102005824446678,
+      "step": 4040
+    },
+    {
+      "epoch": 0.3531874073428098,
+      "grad_norm": 2.139472723007202,
+      "learning_rate": 4.811399943537223e-06,
+      "loss": 2.3641,
+      "mean_token_accuracy": 0.48745107650756836,
+      "step": 4050
+    },
+    {
+      "epoch": 0.3540594750152612,
+      "grad_norm": 2.048004150390625,
+      "learning_rate": 4.759449541782272e-06,
+      "loss": 2.3406,
+      "mean_token_accuracy": 0.4988502934575081,
+      "step": 4060
+    },
+    {
+      "epoch": 0.35493154268771254,
+      "grad_norm": 2.0867536067962646,
+      "learning_rate": 4.707693396682936e-06,
+      "loss": 2.3402,
+      "mean_token_accuracy": 0.49681996256113053,
+      "step": 4070
+    },
+    {
+      "epoch": 0.35580361036016395,
+      "grad_norm": 2.113605499267578,
+      "learning_rate": 4.656133426735315e-06,
+      "loss": 2.3566,
+      "mean_token_accuracy": 0.49332191944122317,
+      "step": 4080
+    },
+    {
+      "epoch": 0.35667567803261535,
+      "grad_norm": 2.0393636226654053,
+      "learning_rate": 4.604771543163706e-06,
+      "loss": 2.3758,
+      "mean_token_accuracy": 0.4844178080558777,
+      "step": 4090
+    },
+    {
+      "epoch": 0.3575477457050667,
+      "grad_norm": 2.128659248352051,
+      "learning_rate": 4.5536096498497295e-06,
+      "loss": 2.4016,
+      "mean_token_accuracy": 0.48309687077999114,
+      "step": 4100
+    },
+    {
+      "epoch": 0.3575477457050667,
+      "eval_runtime": 4.1322,
+      "eval_samples_per_second": 265.473,
+      "eval_steps_per_second": 66.55,
+      "step": 4100
+    },
+    {
+      "epoch": 0.3584198133775181,
+      "grad_norm": 1.9973739385604858,
+      "learning_rate": 4.502649643261779e-06,
+      "loss": 2.3148,
+      "mean_token_accuracy": 0.49669764935970306,
+      "step": 4110
+    },
+    {
+      "epoch": 0.35929188104996945,
+      "grad_norm": 2.3340232372283936,
+      "learning_rate": 4.451893412384707e-06,
+      "loss": 2.4094,
+      "mean_token_accuracy": 0.4848458930850029,
+      "step": 4120
+    },
+    {
+      "epoch": 0.36016394872242086,
+      "grad_norm": 2.3038649559020996,
+      "learning_rate": 4.401342838649818e-06,
+      "loss": 2.2703,
+      "mean_token_accuracy": 0.5089774921536445,
+      "step": 4130
+    },
+    {
+      "epoch": 0.36103601639487226,
+      "grad_norm": 2.1178901195526123,
+      "learning_rate": 4.350999795865109e-06,
+      "loss": 2.3234,
+      "mean_token_accuracy": 0.49991438537836075,
+      "step": 4140
+    },
+    {
+      "epoch": 0.3619080840673236,
+      "grad_norm": 2.2649543285369873,
+      "learning_rate": 4.300866150145837e-06,
+      "loss": 2.3312,
+      "mean_token_accuracy": 0.49960860460996626,
+      "step": 4150
+    },
+    {
+      "epoch": 0.362780151739775,
+      "grad_norm": 2.2314608097076416,
+      "learning_rate": 4.250943759845316e-06,
+      "loss": 2.3129,
+      "mean_token_accuracy": 0.49867905825376513,
+      "step": 4160
+    },
+    {
+      "epoch": 0.36365221941222636,
+      "grad_norm": 2.194746494293213,
+      "learning_rate": 4.201234475486063e-06,
+      "loss": 2.3188,
+      "mean_token_accuracy": 0.4971501961350441,
+      "step": 4170
+    },
+    {
+      "epoch": 0.36452428708467777,
+      "grad_norm": 2.028311014175415,
+      "learning_rate": 4.1517401396911725e-06,
+      "loss": 2.3961,
+      "mean_token_accuracy": 0.48145792335271836,
+      "step": 4180
+    },
+    {
+      "epoch": 0.36539635475712917,
+      "grad_norm": 1.9685331583023071,
+      "learning_rate": 4.1024625871160325e-06,
+      "loss": 2.3188,
+      "mean_token_accuracy": 0.4981164366006851,
+      "step": 4190
+    },
+    {
+      "epoch": 0.3662684224295805,
+      "grad_norm": 2.0334181785583496,
+      "learning_rate": 4.053403644380321e-06,
+      "loss": 2.3031,
+      "mean_token_accuracy": 0.4970034271478653,
+      "step": 4200
+    },
+    {
+      "epoch": 0.3662684224295805,
+      "eval_runtime": 4.1298,
+      "eval_samples_per_second": 265.631,
+      "eval_steps_per_second": 66.589,
+      "step": 4200
+    },
+    {
+      "epoch": 0.3671404901020319,
+      "grad_norm": 2.1037611961364746,
+      "learning_rate": 4.004565130000277e-06,
+      "loss": 2.3398,
+      "mean_token_accuracy": 0.4928326889872551,
+      "step": 4210
+    },
+    {
+      "epoch": 0.3680125577744833,
+      "grad_norm": 2.0131285190582275,
+      "learning_rate": 3.955948854321321e-06,
+      "loss": 2.3473,
+      "mean_token_accuracy": 0.4952543988823891,
+      "step": 4220
+    },
+    {
+      "epoch": 0.3688846254469347,
+      "grad_norm": 2.2105581760406494,
+      "learning_rate": 3.907556619450909e-06,
+      "loss": 2.3594,
+      "mean_token_accuracy": 0.49208659380674363,
+      "step": 4230
+    },
+    {
+      "epoch": 0.3697566931193861,
+      "grad_norm": 2.18249249458313,
+      "learning_rate": 3.859390219191775e-06,
+      "loss": 2.3383,
+      "mean_token_accuracy": 0.49487524628639223,
+      "step": 4240
+    },
+    {
+      "epoch": 0.37062876079183743,
+      "grad_norm": 2.298351526260376,
+      "learning_rate": 3.8114514389754098e-06,
+      "loss": 2.3234,
+      "mean_token_accuracy": 0.4956824779510498,
+      "step": 4250
+    },
+    {
+      "epoch": 0.37150082846428883,
+      "grad_norm": 1.9789235591888428,
+      "learning_rate": 3.7637420557958927e-06,
+      "loss": 2.3328,
+      "mean_token_accuracy": 0.5002323850989342,
+      "step": 4260
+    },
+    {
+      "epoch": 0.3723728961367402,
+      "grad_norm": 1.997238278388977,
+      "learning_rate": 3.7162638381440077e-06,
+      "loss": 2.3242,
+      "mean_token_accuracy": 0.4988869816064835,
+      "step": 4270
+    },
+    {
+      "epoch": 0.3732449638091916,
+      "grad_norm": 2.1409740447998047,
+      "learning_rate": 3.6690185459417107e-06,
+      "loss": 2.3516,
+      "mean_token_accuracy": 0.4943248450756073,
+      "step": 4280
+    },
+    {
+      "epoch": 0.374117031481643,
+      "grad_norm": 2.154865264892578,
+      "learning_rate": 3.622007930476865e-06,
+      "loss": 2.2715,
+      "mean_token_accuracy": 0.5112769082188606,
+      "step": 4290
+    },
+    {
+      "epoch": 0.37498909915409434,
+      "grad_norm": 1.9980154037475586,
+      "learning_rate": 3.575233734338356e-06,
+      "loss": 2.3711,
+      "mean_token_accuracy": 0.48721868693828585,
+      "step": 4300
+    },
+    {
+      "epoch": 0.37498909915409434,
+      "eval_runtime": 4.1412,
+      "eval_samples_per_second": 264.896,
+      "eval_steps_per_second": 66.405,
+      "step": 4300
+    },
+    {
+      "epoch": 0.37586116682654575,
+      "grad_norm": 2.0769875049591064,
+      "learning_rate": 3.528697691351465e-06,
+      "loss": 2.3777,
+      "mean_token_accuracy": 0.48463796377182006,
+      "step": 4310
+    },
+    {
+      "epoch": 0.37673323449899715,
+      "grad_norm": 2.0686168670654297,
+      "learning_rate": 3.4824015265136278e-06,
+      "loss": 2.3719,
+      "mean_token_accuracy": 0.49126712083816526,
+      "step": 4320
+    },
+    {
+      "epoch": 0.3776053021714485,
+      "grad_norm": 2.0578701496124268,
+      "learning_rate": 3.436346955930472e-06,
+      "loss": 2.3258,
+      "mean_token_accuracy": 0.4986301362514496,
+      "step": 4330
+    },
+    {
+      "epoch": 0.3784773698438999,
+      "grad_norm": 2.1568634510040283,
+      "learning_rate": 3.3905356867522187e-06,
+      "loss": 2.407,
+      "mean_token_accuracy": 0.4819593921303749,
+      "step": 4340
+    },
+    {
+      "epoch": 0.37934943751635125,
+      "grad_norm": 2.247126579284668,
+      "learning_rate": 3.344969417110391e-06,
+      "loss": 2.3586,
+      "mean_token_accuracy": 0.49480186104774476,
+      "step": 4350
+    },
+    {
+      "epoch": 0.38022150518880266,
+      "grad_norm": 2.140596628189087,
+      "learning_rate": 3.29964983605487e-06,
+      "loss": 2.4055,
+      "mean_token_accuracy": 0.48425880670547483,
+      "step": 4360
+    },
+    {
+      "epoch": 0.38109357286125406,
+      "grad_norm": 2.1944921016693115,
+      "learning_rate": 3.2545786234913e-06,
+      "loss": 2.2797,
+      "mean_token_accuracy": 0.504990217089653,
+      "step": 4370
+    },
+    {
+      "epoch": 0.3819656405337054,
+      "grad_norm": 2.0062012672424316,
+      "learning_rate": 3.2097574501187877e-06,
+      "loss": 2.3438,
+      "mean_token_accuracy": 0.49622064679861067,
+      "step": 4380
+    },
+    {
+      "epoch": 0.3828377082061568,
+      "grad_norm": 1.9321141242980957,
+      "learning_rate": 3.165187977368007e-06,
+      "loss": 2.3047,
+      "mean_token_accuracy": 0.4987035199999809,
+      "step": 4390
+    },
+    {
+      "epoch": 0.38370977587860816,
+      "grad_norm": 2.3334603309631348,
+      "learning_rate": 3.120871857339582e-06,
+      "loss": 2.4016,
+      "mean_token_accuracy": 0.48373287320137026,
+      "step": 4400
+    },
+    {
+      "epoch": 0.38370977587860816,
+      "eval_runtime": 4.1288,
+      "eval_samples_per_second": 265.693,
+      "eval_steps_per_second": 66.605,
+      "step": 4400
+    },
+    {
+      "epoch": 0.38458184355105957,
+      "grad_norm": 2.1358065605163574,
+      "learning_rate": 3.0768107327428766e-06,
+      "loss": 2.3242,
+      "mean_token_accuracy": 0.5046844437718392,
+      "step": 4410
+    },
+    {
+      "epoch": 0.38545391122351097,
+      "grad_norm": 2.107206344604492,
+      "learning_rate": 3.033006236835071e-06,
+      "loss": 2.3258,
+      "mean_token_accuracy": 0.4933341518044472,
+      "step": 4420
+    },
+    {
+      "epoch": 0.3863259788959623,
+      "grad_norm": 2.5342905521392822,
+      "learning_rate": 2.9894599933606518e-06,
+      "loss": 2.3164,
+      "mean_token_accuracy": 0.500330239534378,
+      "step": 4430
+    },
+    {
+      "epoch": 0.3871980465684137,
+      "grad_norm": 1.9564285278320312,
+      "learning_rate": 2.9461736164911934e-06,
+      "loss": 2.2578,
+      "mean_token_accuracy": 0.5108610525727272,
+      "step": 4440
+    },
+    {
+      "epoch": 0.38807011424086507,
+      "grad_norm": 2.2510104179382324,
+      "learning_rate": 2.903148710765552e-06,
+      "loss": 2.3363,
+      "mean_token_accuracy": 0.4905332654714584,
+      "step": 4450
+    },
+    {
+      "epoch": 0.3889421819133165,
+      "grad_norm": 2.2085652351379395,
+      "learning_rate": 2.8603868710303662e-06,
+      "loss": 2.2797,
+      "mean_token_accuracy": 0.5062377706170083,
+      "step": 4460
+    },
+    {
+      "epoch": 0.3898142495857679,
+      "grad_norm": 2.299708366394043,
+      "learning_rate": 2.8178896823809465e-06,
+      "loss": 2.3438,
+      "mean_token_accuracy": 0.48997064381837846,
+      "step": 4470
+    },
+    {
+      "epoch": 0.39068631725821923,
+      "grad_norm": 2.1641745567321777,
+      "learning_rate": 2.7756587201025297e-06,
+      "loss": 2.3352,
+      "mean_token_accuracy": 0.4941291570663452,
+      "step": 4480
+    },
+    {
+      "epoch": 0.39155838493067063,
+      "grad_norm": 2.290093183517456,
+      "learning_rate": 2.7336955496118666e-06,
+      "loss": 2.3477,
+      "mean_token_accuracy": 0.49297945350408556,
+      "step": 4490
+    },
+    {
+      "epoch": 0.392430452603122,
+      "grad_norm": 2.1219727993011475,
+      "learning_rate": 2.692001726399215e-06,
+      "loss": 2.3477,
+      "mean_token_accuracy": 0.49052103608846664,
+      "step": 4500
+    },
+    {
+      "epoch": 0.392430452603122,
+      "eval_runtime": 4.1268,
+      "eval_samples_per_second": 265.824,
+      "eval_steps_per_second": 66.638,
+      "step": 4500
+    },
+    {
+      "epoch": 0.3933025202755734,
+      "grad_norm": 2.1446876525878906,
+      "learning_rate": 2.6505787959706607e-06,
+      "loss": 2.2781,
+      "mean_token_accuracy": 0.5032656565308571,
+      "step": 4510
+    },
+    {
+      "epoch": 0.3941745879480248,
+      "grad_norm": 2.106083393096924,
+      "learning_rate": 2.609428293790852e-06,
+      "loss": 2.2598,
+      "mean_token_accuracy": 0.5072040125727654,
+      "step": 4520
+    },
+    {
+      "epoch": 0.39504665562047614,
+      "grad_norm": 2.3130080699920654,
+      "learning_rate": 2.5685517452260566e-06,
+      "loss": 2.3641,
+      "mean_token_accuracy": 0.49031311124563215,
+      "step": 4530
+    },
+    {
+      "epoch": 0.39591872329292754,
+      "grad_norm": 1.958322525024414,
+      "learning_rate": 2.5279506654876473e-06,
+      "loss": 2.2914,
+      "mean_token_accuracy": 0.49827544689178466,
+      "step": 4540
+    },
+    {
+      "epoch": 0.3967907909653789,
+      "grad_norm": 2.1239428520202637,
+      "learning_rate": 2.487626559575911e-06,
+      "loss": 2.3633,
+      "mean_token_accuracy": 0.4887964800000191,
+      "step": 4550
+    },
+    {
+      "epoch": 0.3976628586378303,
+      "grad_norm": 2.608161211013794,
+      "learning_rate": 2.4475809222242775e-06,
+      "loss": 2.3383,
+      "mean_token_accuracy": 0.4964652627706528,
+      "step": 4560
+    },
+    {
+      "epoch": 0.3985349263102817,
+      "grad_norm": 1.9930739402770996,
+      "learning_rate": 2.4078152378439033e-06,
+      "loss": 2.3453,
+      "mean_token_accuracy": 0.48918786346912385,
+      "step": 4570
+    },
+    {
+      "epoch": 0.39940699398273305,
+      "grad_norm": 2.3497517108917236,
+      "learning_rate": 2.3683309804686604e-06,
+      "loss": 2.3059,
+      "mean_token_accuracy": 0.4994129166007042,
+      "step": 4580
+    },
+    {
+      "epoch": 0.40027906165518445,
+      "grad_norm": 2.1560254096984863,
+      "learning_rate": 2.329129613700478e-06,
+      "loss": 2.3648,
+      "mean_token_accuracy": 0.4846257358789444,
+      "step": 4590
+    },
+    {
+      "epoch": 0.4011511293276358,
+      "grad_norm": 2.169948101043701,
+      "learning_rate": 2.29021259065511e-06,
+      "loss": 2.325,
+      "mean_token_accuracy": 0.49338307082653043,
+      "step": 4600
+    },
+    {
+      "epoch": 0.4011511293276358,
+      "eval_runtime": 4.1282,
+      "eval_samples_per_second": 265.73,
+      "eval_steps_per_second": 66.614,
+      "step": 4600
+    },
+    {
+      "epoch": 0.4020231970000872,
+      "grad_norm": 1.8934439420700073,
+      "learning_rate": 2.251581353908252e-06,
+      "loss": 2.3523,
+      "mean_token_accuracy": 0.4882583186030388,
+      "step": 4610
+    },
+    {
+      "epoch": 0.4028952646725386,
+      "grad_norm": 2.186927080154419,
+      "learning_rate": 2.2132373354420833e-06,
+      "loss": 2.3242,
+      "mean_token_accuracy": 0.4932974576950073,
+      "step": 4620
+    },
+    {
+      "epoch": 0.40376733234498996,
+      "grad_norm": 2.149789571762085,
+      "learning_rate": 2.1751819565921774e-06,
+      "loss": 2.3578,
+      "mean_token_accuracy": 0.4906678080558777,
+      "step": 4630
+    },
+    {
+      "epoch": 0.40463940001744136,
+      "grad_norm": 1.9236981868743896,
+      "learning_rate": 2.137416627994814e-06,
+      "loss": 2.3188,
+      "mean_token_accuracy": 0.49946184605360033,
+      "step": 4640
+    },
+    {
+      "epoch": 0.4055114676898927,
+      "grad_norm": 2.2763254642486572,
+      "learning_rate": 2.0999427495347035e-06,
+      "loss": 2.3609,
+      "mean_token_accuracy": 0.49170743972063063,
+      "step": 4650
+    },
+    {
+      "epoch": 0.4063835353623441,
+      "grad_norm": 2.1921656131744385,
+      "learning_rate": 2.0627617102930753e-06,
+      "loss": 2.3414,
+      "mean_token_accuracy": 0.4888331726193428,
+      "step": 4660
+    },
+    {
+      "epoch": 0.4072556030347955,
+      "grad_norm": 1.962241530418396,
+      "learning_rate": 2.02587488849621e-06,
+      "loss": 2.3109,
+      "mean_token_accuracy": 0.5015533298254014,
+      "step": 4670
+    },
+    {
+      "epoch": 0.40812767070724687,
+      "grad_norm": 2.1182403564453125,
+      "learning_rate": 1.989283651464329e-06,
+      "loss": 2.343,
+      "mean_token_accuracy": 0.4896526426076889,
+      "step": 4680
+    },
+    {
+      "epoch": 0.4089997383796983,
+      "grad_norm": 1.9797334671020508,
+      "learning_rate": 1.952989355560929e-06,
+      "loss": 2.3016,
+      "mean_token_accuracy": 0.5035225063562393,
+      "step": 4690
+    },
+    {
+      "epoch": 0.4098718060521496,
+      "grad_norm": 2.1399505138397217,
+      "learning_rate": 1.9169933461424928e-06,
+      "loss": 2.3422,
+      "mean_token_accuracy": 0.4882827788591385,
+      "step": 4700
+    },
+    {
+      "epoch": 0.4098718060521496,
+      "eval_runtime": 4.1337,
+      "eval_samples_per_second": 265.378,
+      "eval_steps_per_second": 66.526,
+      "step": 4700
+    },
+    {
+      "epoch": 0.410743873724601,
+      "grad_norm": 2.1622302532196045,
+      "learning_rate": 1.8812969575086272e-06,
+      "loss": 2.3266,
+      "mean_token_accuracy": 0.49628179967403413,
+      "step": 4710
+    },
+    {
+      "epoch": 0.41161594139705243,
+      "grad_norm": 2.1700828075408936,
+      "learning_rate": 1.8459015128525937e-06,
+      "loss": 2.2379,
+      "mean_token_accuracy": 0.5100905075669289,
+      "step": 4720
+    },
+    {
+      "epoch": 0.4124880090695038,
+      "grad_norm": 2.145601272583008,
+      "learning_rate": 1.8108083242122764e-06,
+      "loss": 2.2906,
+      "mean_token_accuracy": 0.498348830640316,
+      "step": 4730
+    },
+    {
+      "epoch": 0.4133600767419552,
+      "grad_norm": 2.2145471572875977,
+      "learning_rate": 1.7760186924215239e-06,
+      "loss": 2.2559,
+      "mean_token_accuracy": 0.5119618386030197,
+      "step": 4740
+    },
+    {
+      "epoch": 0.41423214441440653,
+      "grad_norm": 2.026996612548828,
+      "learning_rate": 1.7415339070619586e-06,
+      "loss": 2.3305,
+      "mean_token_accuracy": 0.4923801377415657,
+      "step": 4750
+    },
+    {
+      "epoch": 0.41510421208685794,
+      "grad_norm": 2.134308099746704,
+      "learning_rate": 1.7073552464151465e-06,
+      "loss": 2.3266,
+      "mean_token_accuracy": 0.4966487348079681,
+      "step": 4760
+    },
+    {
+      "epoch": 0.41597627975930934,
+      "grad_norm": 2.023016929626465,
+      "learning_rate": 1.6734839774152322e-06,
+      "loss": 2.3828,
+      "mean_token_accuracy": 0.49090019166469573,
+      "step": 4770
+    },
+    {
+      "epoch": 0.4168483474317607,
+      "grad_norm": 2.044248104095459,
+      "learning_rate": 1.6399213556019732e-06,
+      "loss": 2.2711,
+      "mean_token_accuracy": 0.5051981404423713,
+      "step": 4780
+    },
+    {
+      "epoch": 0.4177204151042121,
+      "grad_norm": 2.12876296043396,
+      "learning_rate": 1.6066686250741904e-06,
+      "loss": 2.3102,
+      "mean_token_accuracy": 0.5034368902444839,
+      "step": 4790
+    },
+    {
+      "epoch": 0.41859248277666344,
+      "grad_norm": 1.997663974761963,
+      "learning_rate": 1.573727018443667e-06,
+      "loss": 2.343,
+      "mean_token_accuracy": 0.4941780909895897,
+      "step": 4800
+    },
+    {
+      "epoch": 0.41859248277666344,
+      "eval_runtime": 4.1277,
+      "eval_samples_per_second": 265.768,
+      "eval_steps_per_second": 66.624,
+      "step": 4800
+    },
+    {
+      "epoch": 0.41946455044911485,
+      "grad_norm": 2.0558156967163086,
+      "learning_rate": 1.5410977567894403e-06,
+      "loss": 2.3367,
+      "mean_token_accuracy": 0.48852739930152894,
+      "step": 4810
+    },
+    {
+      "epoch": 0.42033661812156625,
+      "grad_norm": 2.0549166202545166,
+      "learning_rate": 1.5087820496125595e-06,
+      "loss": 2.3602,
+      "mean_token_accuracy": 0.49081457406282425,
+      "step": 4820
+    },
+    {
+      "epoch": 0.4212086857940176,
+      "grad_norm": 2.089341640472412,
+      "learning_rate": 1.4767810947912275e-06,
+      "loss": 2.3398,
+      "mean_token_accuracy": 0.48964040726423264,
+      "step": 4830
+    },
+    {
+      "epoch": 0.422080753466469,
+      "grad_norm": 1.991812825202942,
+      "learning_rate": 1.4450960785364244e-06,
+      "loss": 2.3359,
+      "mean_token_accuracy": 0.49184197932481766,
+      "step": 4840
+    },
+    {
+      "epoch": 0.42295282113892035,
+      "grad_norm": 2.083972454071045,
+      "learning_rate": 1.4137281753479092e-06,
+      "loss": 2.3641,
+      "mean_token_accuracy": 0.4926736786961555,
+      "step": 4850
+    },
+    {
+      "epoch": 0.42382488881137176,
+      "grad_norm": 2.251168966293335,
+      "learning_rate": 1.3826785479707128e-06,
+      "loss": 2.3477,
+      "mean_token_accuracy": 0.4906433433294296,
+      "step": 4860
+    },
+    {
+      "epoch": 0.42469695648382316,
+      "grad_norm": 2.016625165939331,
+      "learning_rate": 1.3519483473520124e-06,
+      "loss": 2.3766,
+      "mean_token_accuracy": 0.48362280130386354,
+      "step": 4870
+    },
+    {
+      "epoch": 0.4255690241562745,
+      "grad_norm": 2.0079569816589355,
+      "learning_rate": 1.3215387125984813e-06,
+      "loss": 2.3453,
+      "mean_token_accuracy": 0.49295499324798586,
+      "step": 4880
+    },
+    {
+      "epoch": 0.4264410918287259,
+      "grad_norm": 1.9797515869140625,
+      "learning_rate": 1.2914507709340596e-06,
+      "loss": 2.2984,
+      "mean_token_accuracy": 0.4988869860768318,
+      "step": 4890
+    },
+    {
+      "epoch": 0.42731315950117726,
+      "grad_norm": 2.122939348220825,
+      "learning_rate": 1.2616856376581766e-06,
+      "loss": 2.3273,
+      "mean_token_accuracy": 0.49677103757858276,
+      "step": 4900
+    },
+    {
+      "epoch": 0.42731315950117726,
+      "eval_runtime": 4.1347,
+      "eval_samples_per_second": 265.313,
+      "eval_steps_per_second": 66.51,
+      "step": 4900
+    },
+    {
+      "epoch": 0.42818522717362867,
+      "grad_norm": 1.929711103439331,
+      "learning_rate": 1.2322444161044e-06,
+      "loss": 2.3414,
+      "mean_token_accuracy": 0.4948874697089195,
+      "step": 4910
+    },
+    {
+      "epoch": 0.4290572948460801,
+      "grad_norm": 2.090467929840088,
+      "learning_rate": 1.2031281975995467e-06,
+      "loss": 2.3309,
+      "mean_token_accuracy": 0.4955479443073273,
+      "step": 4920
+    },
+    {
+      "epoch": 0.4299293625185314,
+      "grad_norm": 2.1671996116638184,
+      "learning_rate": 1.1743380614232213e-06,
+      "loss": 2.3312,
+      "mean_token_accuracy": 0.4941169276833534,
+      "step": 4930
+    },
+    {
+      "epoch": 0.4308014301909828,
+      "grad_norm": 2.2289273738861084,
+      "learning_rate": 1.1458750747678105e-06,
+      "loss": 2.3137,
+      "mean_token_accuracy": 0.5042074352502823,
+      "step": 4940
+    },
+    {
+      "epoch": 0.43167349786343423,
+      "grad_norm": 2.101097345352173,
+      "learning_rate": 1.1177402926989345e-06,
+      "loss": 2.3445,
+      "mean_token_accuracy": 0.49644080847501754,
+      "step": 4950
+    },
+    {
+      "epoch": 0.4325455655358856,
+      "grad_norm": 1.9031548500061035,
+      "learning_rate": 1.0899347581163222e-06,
+      "loss": 2.3484,
+      "mean_token_accuracy": 0.49497309029102327,
+      "step": 4960
+    },
+    {
+      "epoch": 0.433417633208337,
+      "grad_norm": 2.1439337730407715,
+      "learning_rate": 1.0624595017151685e-06,
+      "loss": 2.3438,
+      "mean_token_accuracy": 0.4897382572293282,
+      "step": 4970
+    },
+    {
+      "epoch": 0.43428970088078833,
+      "grad_norm": 2.309295892715454,
+      "learning_rate": 1.0353155419479122e-06,
+      "loss": 2.318,
+      "mean_token_accuracy": 0.4978595867753029,
+      "step": 4980
+    },
+    {
+      "epoch": 0.43516176855323974,
+      "grad_norm": 2.183516502380371,
+      "learning_rate": 1.0085038849865025e-06,
+      "loss": 2.3242,
+      "mean_token_accuracy": 0.49634295552968977,
+      "step": 4990
+    },
+    {
+      "epoch": 0.43603383622569114,
+      "grad_norm": 1.9440361261367798,
+      "learning_rate": 9.820255246850853e-07,
+      "loss": 2.3367,
+      "mean_token_accuracy": 0.4941291600465775,
+      "step": 5000
+    },
+    {
+      "epoch": 0.43603383622569114,
+      "eval_runtime": 4.133,
+      "eval_samples_per_second": 265.426,
+      "eval_steps_per_second": 66.538,
+      "step": 5000
+    },
+    {
+      "epoch": 0.4369059038981425,
+      "grad_norm": 2.001448154449463,
+      "learning_rate": 9.55881442543174e-07,
+      "loss": 2.3414,
+      "mean_token_accuracy": 0.4952054783701897,
+      "step": 5010
+    },
+    {
+      "epoch": 0.4377779715705939,
+      "grad_norm": 1.9931625127792358,
+      "learning_rate": 9.30072607669259e-07,
+      "loss": 2.3555,
+      "mean_token_accuracy": 0.4907534226775169,
+      "step": 5020
+    },
+    {
+      "epoch": 0.43865003924304524,
+      "grad_norm": 2.111868381500244,
+      "learning_rate": 9.045999767448988e-07,
+      "loss": 2.3062,
+      "mean_token_accuracy": 0.5013576313853264,
+      "step": 5030
+    },
+    {
+      "epoch": 0.43952210691549665,
+      "grad_norm": 2.0870048999786377,
+      "learning_rate": 8.794644939892361e-07,
+      "loss": 2.3742,
+      "mean_token_accuracy": 0.48999510556459425,
+      "step": 5040
+    },
+    {
+      "epoch": 0.44039417458794805,
+      "grad_norm": 2.1297481060028076,
+      "learning_rate": 8.546670911240196e-07,
+      "loss": 2.3508,
+      "mean_token_accuracy": 0.4961717247962952,
+      "step": 5050
+    },
+    {
+      "epoch": 0.4412662422603994,
+      "grad_norm": 1.9581613540649414,
+      "learning_rate": 8.302086873390536e-07,
+      "loss": 2.375,
+      "mean_token_accuracy": 0.48824608623981475,
+      "step": 5060
+    },
+    {
+      "epoch": 0.4421383099328508,
+      "grad_norm": 2.214728832244873,
+      "learning_rate": 8.060901892581241e-07,
+      "loss": 2.3484,
+      "mean_token_accuracy": 0.49313845038414,
+      "step": 5070
+    },
+    {
+      "epoch": 0.44301037760530215,
+      "grad_norm": 2.0651919841766357,
+      "learning_rate": 7.82312490905407e-07,
+      "loss": 2.3039,
+      "mean_token_accuracy": 0.5001223102211952,
+      "step": 5080
+    },
+    {
+      "epoch": 0.44388244527775356,
+      "grad_norm": 1.9561328887939453,
+      "learning_rate": 7.588764736723086e-07,
+      "loss": 2.3258,
+      "mean_token_accuracy": 0.4943737804889679,
+      "step": 5090
+    },
+    {
+      "epoch": 0.44475451295020496,
+      "grad_norm": 2.1118240356445312,
+      "learning_rate": 7.357830062848114e-07,
+      "loss": 2.2898,
+      "mean_token_accuracy": 0.5014432445168495,
+      "step": 5100
+    },
+    {
+      "epoch": 0.44475451295020496,
+      "eval_runtime": 4.1244,
+      "eval_samples_per_second": 265.977,
+      "eval_steps_per_second": 66.676,
+      "step": 5100
+    },
+    {
+      "epoch": 0.4456265806226563,
+      "grad_norm": 2.102750062942505,
+      "learning_rate": 7.130329447712581e-07,
+      "loss": 2.2426,
+      "mean_token_accuracy": 0.5095156550407409,
+      "step": 5110
+    },
+    {
+      "epoch": 0.4464986482951077,
+      "grad_norm": 1.938952088356018,
+      "learning_rate": 6.906271324306335e-07,
+      "loss": 2.3117,
+      "mean_token_accuracy": 0.4916340529918671,
+      "step": 5120
+    },
+    {
+      "epoch": 0.44737071596755906,
+      "grad_norm": 2.2127182483673096,
+      "learning_rate": 6.685663998012926e-07,
+      "loss": 2.3109,
+      "mean_token_accuracy": 0.49765166342258454,
+      "step": 5130
+    },
+    {
+      "epoch": 0.44824278364001047,
+      "grad_norm": 2.128307342529297,
+      "learning_rate": 6.468515646301865e-07,
+      "loss": 2.3414,
+      "mean_token_accuracy": 0.49337084144353865,
+      "step": 5140
+    },
+    {
+      "epoch": 0.44911485131246187,
+      "grad_norm": 2.370331287384033,
+      "learning_rate": 6.254834318425363e-07,
+      "loss": 2.3422,
+      "mean_token_accuracy": 0.49102250039577483,
+      "step": 5150
+    },
+    {
+      "epoch": 0.4499869189849132,
+      "grad_norm": 2.0760042667388916,
+      "learning_rate": 6.044627935120107e-07,
+      "loss": 2.325,
+      "mean_token_accuracy": 0.49480185359716417,
+      "step": 5160
+    },
+    {
+      "epoch": 0.4508589866573646,
+      "grad_norm": 1.944549322128296,
+      "learning_rate": 5.837904288313545e-07,
+      "loss": 2.3027,
+      "mean_token_accuracy": 0.49980430901050565,
+      "step": 5170
+    },
+    {
+      "epoch": 0.451731054329816,
+      "grad_norm": 2.2823469638824463,
+      "learning_rate": 5.634671040835104e-07,
+      "loss": 2.3875,
+      "mean_token_accuracy": 0.4848214253783226,
+      "step": 5180
+    },
+    {
+      "epoch": 0.4526031220022674,
+      "grad_norm": 2.3041868209838867,
+      "learning_rate": 5.43493572613214e-07,
+      "loss": 2.393,
+      "mean_token_accuracy": 0.485053813457489,
+      "step": 5190
+    },
+    {
+      "epoch": 0.4534751896747188,
+      "grad_norm": 2.0338685512542725,
+      "learning_rate": 5.238705747990669e-07,
+      "loss": 2.2703,
+      "mean_token_accuracy": 0.5066658541560173,
+      "step": 5200
+    },
+    {
+      "epoch": 0.4534751896747188,
+      "eval_runtime": 4.1386,
+      "eval_samples_per_second": 265.067,
+      "eval_steps_per_second": 66.448,
+      "step": 5200
+    },
+    {
+      "epoch": 0.45434725734717013,
+      "grad_norm": 2.0606861114501953,
+      "learning_rate": 5.045988380260935e-07,
+      "loss": 2.2828,
+      "mean_token_accuracy": 0.5026541098952293,
+      "step": 5210
+    },
+    {
+      "epoch": 0.45521932501962153,
+      "grad_norm": 2.008419990539551,
+      "learning_rate": 4.856790766587815e-07,
+      "loss": 2.3809,
+      "mean_token_accuracy": 0.49344422519207,
+      "step": 5220
+    },
+    {
+      "epoch": 0.4560913926920729,
+      "grad_norm": 1.9905202388763428,
+      "learning_rate": 4.6711199201459833e-07,
+      "loss": 2.3234,
+      "mean_token_accuracy": 0.49330968260765073,
+      "step": 5230
+    },
+    {
+      "epoch": 0.4569634603645243,
+      "grad_norm": 2.1715281009674072,
+      "learning_rate": 4.488982723379887e-07,
+      "loss": 2.2773,
+      "mean_token_accuracy": 0.5032044991850853,
+      "step": 5240
+    },
+    {
+      "epoch": 0.4578355280369757,
+      "grad_norm": 2.012258768081665,
+      "learning_rate": 4.3103859277488056e-07,
+      "loss": 2.2863,
+      "mean_token_accuracy": 0.5022015631198883,
+      "step": 5250
+    },
+    {
+      "epoch": 0.45870759570942704,
+      "grad_norm": 2.2576820850372314,
+      "learning_rate": 4.1353361534763657e-07,
+      "loss": 2.318,
+      "mean_token_accuracy": 0.49766389429569247,
+      "step": 5260
+    },
+    {
+      "epoch": 0.45957966338187844,
+      "grad_norm": 2.265260696411133,
+      "learning_rate": 3.963839889305343e-07,
+      "loss": 2.318,
+      "mean_token_accuracy": 0.496061646938324,
+      "step": 5270
+    },
+    {
+      "epoch": 0.4604517310543298,
+      "grad_norm": 2.1696579456329346,
+      "learning_rate": 3.7959034922569804e-07,
+      "loss": 2.3477,
+      "mean_token_accuracy": 0.4889799416065216,
+      "step": 5280
+    },
+    {
+      "epoch": 0.4613237987267812,
+      "grad_norm": 1.979793667793274,
+      "learning_rate": 3.631533187395453e-07,
+      "loss": 2.3445,
+      "mean_token_accuracy": 0.49362768679857255,
+      "step": 5290
+    },
+    {
+      "epoch": 0.4621958663992326,
+      "grad_norm": 2.143578052520752,
+      "learning_rate": 3.470735067597053e-07,
+      "loss": 2.3047,
+      "mean_token_accuracy": 0.4961839556694031,
+      "step": 5300
+    },
+    {
+      "epoch": 0.4621958663992326,
+      "eval_runtime": 4.1264,
+      "eval_samples_per_second": 265.847,
+      "eval_steps_per_second": 66.644,
+      "step": 5300
+    },
+    {
+      "epoch": 0.46306793407168395,
+      "grad_norm": 2.0299618244171143,
+      "learning_rate": 3.313515093324393e-07,
+      "loss": 2.2766,
+      "mean_token_accuracy": 0.5060543060302735,
+      "step": 5310
+    },
+    {
+      "epoch": 0.46394000174413536,
+      "grad_norm": 2.108973503112793,
+      "learning_rate": 3.1598790924053936e-07,
+      "loss": 2.3188,
+      "mean_token_accuracy": 0.5022504895925521,
+      "step": 5320
+    },
+    {
+      "epoch": 0.4648120694165867,
+      "grad_norm": 2.318573236465454,
+      "learning_rate": 3.009832759817344e-07,
+      "loss": 2.3641,
+      "mean_token_accuracy": 0.4872186854481697,
+      "step": 5330
+    },
+    {
+      "epoch": 0.4656841370890381,
+      "grad_norm": 1.9880313873291016,
+      "learning_rate": 2.8633816574757166e-07,
+      "loss": 2.2914,
+      "mean_token_accuracy": 0.5004770040512085,
+      "step": 5340
+    },
+    {
+      "epoch": 0.4665562047614895,
+      "grad_norm": 2.124990463256836,
+      "learning_rate": 2.720531214028055e-07,
+      "loss": 2.2773,
+      "mean_token_accuracy": 0.5056996047496796,
+      "step": 5350
+    },
+    {
+      "epoch": 0.46742827243394086,
+      "grad_norm": 2.1923046112060547,
+      "learning_rate": 2.5812867246527207e-07,
+      "loss": 2.2863,
+      "mean_token_accuracy": 0.5009295508265496,
+      "step": 5360
+    },
+    {
+      "epoch": 0.46830034010639227,
+      "grad_norm": 2.0684754848480225,
+      "learning_rate": 2.445653350862609e-07,
+      "loss": 2.3312,
+      "mean_token_accuracy": 0.49175636321306226,
+      "step": 5370
+    },
+    {
+      "epoch": 0.4691724077788436,
+      "grad_norm": 2.038372039794922,
+      "learning_rate": 2.3136361203138668e-07,
+      "loss": 2.343,
+      "mean_token_accuracy": 0.4960249558091164,
+      "step": 5380
+    },
+    {
+      "epoch": 0.470044475451295,
+      "grad_norm": 2.1624391078948975,
+      "learning_rate": 2.1852399266194312e-07,
+      "loss": 2.3234,
+      "mean_token_accuracy": 0.49827543944120406,
+      "step": 5390
+    },
+    {
+      "epoch": 0.4709165431237464,
+      "grad_norm": 1.905287504196167,
+      "learning_rate": 2.0604695291677523e-07,
+      "loss": 2.2687,
+      "mean_token_accuracy": 0.5026908054947853,
+      "step": 5400
+    },
+    {
+      "epoch": 0.4709165431237464,
+      "eval_runtime": 4.1281,
+      "eval_samples_per_second": 265.741,
+      "eval_steps_per_second": 66.617,
+      "step": 5400
+    },
+    {
+      "epoch": 0.47178861079619777,
+      "grad_norm": 2.169462203979492,
+      "learning_rate": 1.9393295529462674e-07,
+      "loss": 2.3,
+      "mean_token_accuracy": 0.4972235843539238,
+      "step": 5410
+    },
+    {
+      "epoch": 0.4726606784686492,
+      "grad_norm": 1.9786736965179443,
+      "learning_rate": 1.8218244883700386e-07,
+      "loss": 2.3375,
+      "mean_token_accuracy": 0.49372553527355195,
+      "step": 5420
+    },
+    {
+      "epoch": 0.4735327461411005,
+      "grad_norm": 1.8743739128112793,
+      "learning_rate": 1.7079586911152413e-07,
+      "loss": 2.2898,
+      "mean_token_accuracy": 0.5012842446565628,
+      "step": 5430
+    },
+    {
+      "epoch": 0.47440481381355193,
+      "grad_norm": 2.0569217205047607,
+      "learning_rate": 1.597736381957782e-07,
+      "loss": 2.2965,
+      "mean_token_accuracy": 0.5054549872875214,
+      "step": 5440
+    },
+    {
+      "epoch": 0.47527688148600333,
+      "grad_norm": 2.0441787242889404,
+      "learning_rate": 1.4911616466167345e-07,
+      "loss": 2.3328,
+      "mean_token_accuracy": 0.49340753704309465,
+      "step": 5450
+    },
+    {
+      "epoch": 0.4761489491584547,
+      "grad_norm": 2.0800154209136963,
+      "learning_rate": 1.3882384356030066e-07,
+      "loss": 2.3969,
+      "mean_token_accuracy": 0.4841976538300514,
+      "step": 5460
+    },
+    {
+      "epoch": 0.4770210168309061,
+      "grad_norm": 2.1264023780822754,
+      "learning_rate": 1.2889705640728445e-07,
+      "loss": 2.2801,
+      "mean_token_accuracy": 0.5074608579277993,
+      "step": 5470
+    },
+    {
+      "epoch": 0.47789308450335743,
+      "grad_norm": 2.0067479610443115,
+      "learning_rate": 1.1933617116863805e-07,
+      "loss": 2.3547,
+      "mean_token_accuracy": 0.4909368887543678,
+      "step": 5480
+    },
+    {
+      "epoch": 0.47876515217580884,
+      "grad_norm": 2.06038761138916,
+      "learning_rate": 1.1014154224713302e-07,
+      "loss": 2.3281,
+      "mean_token_accuracy": 0.5009662479162216,
+      "step": 5490
+    },
+    {
+      "epoch": 0.47963721984826024,
+      "grad_norm": 2.0036745071411133,
+      "learning_rate": 1.0131351046915094e-07,
+      "loss": 2.2766,
+      "mean_token_accuracy": 0.5009295508265496,
+      "step": 5500
+    },
+    {
+      "epoch": 0.47963721984826024,
+      "eval_runtime": 4.1322,
+      "eval_samples_per_second": 265.476,
+      "eval_steps_per_second": 66.551,
+      "step": 5500
+    },
+    {
+      "epoch": 0.4805092875207116,
+      "grad_norm": 2.004516363143921,
+      "learning_rate": 9.285240307206123e-08,
+      "loss": 2.3359,
+      "mean_token_accuracy": 0.493370845913887,
+      "step": 5510
+    },
+    {
+      "epoch": 0.481381355193163,
+      "grad_norm": 2.1120264530181885,
+      "learning_rate": 8.475853369207753e-08,
+      "loss": 2.3453,
+      "mean_token_accuracy": 0.49291829764842987,
+      "step": 5520
+    },
+    {
+      "epoch": 0.48225342286561435,
+      "grad_norm": 2.058835029602051,
+      "learning_rate": 7.703220235264708e-08,
+      "loss": 2.3211,
+      "mean_token_accuracy": 0.4960983365774155,
+      "step": 5530
+    },
+    {
+      "epoch": 0.48312549053806575,
+      "grad_norm": 1.964354395866394,
+      "learning_rate": 6.967369545331615e-08,
+      "loss": 2.3484,
+      "mean_token_accuracy": 0.49185420721769335,
+      "step": 5540
+    },
+    {
+      "epoch": 0.48399755821051715,
+      "grad_norm": 2.046884298324585,
+      "learning_rate": 6.26832857591242e-08,
+      "loss": 2.316,
+      "mean_token_accuracy": 0.4936399176716805,
+      "step": 5550
+    },
+    {
+      "epoch": 0.4848696258829685,
+      "grad_norm": 1.9410163164138794,
+      "learning_rate": 5.606123239048522e-08,
+      "loss": 2.3324,
+      "mean_token_accuracy": 0.49575415551662444,
+      "step": 5560
+    },
+    {
+      "epoch": 0.4857416935554199,
+      "grad_norm": 1.9944008588790894,
+      "learning_rate": 4.9807780813586615e-08,
+      "loss": 2.2844,
+      "mean_token_accuracy": 0.502140411734581,
+      "step": 5570
+    },
+    {
+      "epoch": 0.48661376122787126,
+      "grad_norm": 2.1313865184783936,
+      "learning_rate": 4.392316283128861e-08,
+      "loss": 2.2477,
+      "mean_token_accuracy": 0.5090386480093002,
+      "step": 5580
+    },
+    {
+      "epoch": 0.48748582890032266,
+      "grad_norm": 2.155336380004883,
+      "learning_rate": 3.840759657453452e-08,
+      "loss": 2.3332,
+      "mean_token_accuracy": 0.4918664380908012,
+      "step": 5590
+    },
+    {
+      "epoch": 0.48835789657277406,
+      "grad_norm": 2.1307590007781982,
+      "learning_rate": 3.326128649426053e-08,
+      "loss": 2.3508,
+      "mean_token_accuracy": 0.49245352298021317,
+      "step": 5600
+    },
+    {
+      "epoch": 0.48835789657277406,
+      "eval_runtime": 4.1308,
+      "eval_samples_per_second": 265.564,
+      "eval_steps_per_second": 66.573,
+      "step": 5600
+    },
+    {
+      "epoch": 0.4892299642452254,
+      "grad_norm": 2.0670790672302246,
+      "learning_rate": 2.8484423353822842e-08,
+      "loss": 2.2906,
+      "mean_token_accuracy": 0.5042441293597222,
+      "step": 5610
+    },
+    {
+      "epoch": 0.4901020319176768,
+      "grad_norm": 2.0412628650665283,
+      "learning_rate": 2.4077184221920068e-08,
+      "loss": 2.2426,
+      "mean_token_accuracy": 0.5120229914784431,
+      "step": 5620
+    },
+    {
+      "epoch": 0.4909740995901282,
+      "grad_norm": 2.019014596939087,
+      "learning_rate": 2.003973246603508e-08,
+      "loss": 2.2711,
+      "mean_token_accuracy": 0.5057607650756836,
+      "step": 5630
+    },
+    {
+      "epoch": 0.49184616726257957,
+      "grad_norm": 2.1194162368774414,
+      "learning_rate": 1.637221774637765e-08,
+      "loss": 2.3352,
+      "mean_token_accuracy": 0.4929427593946457,
+      "step": 5640
+    },
+    {
+      "epoch": 0.492718234935031,
+      "grad_norm": 2.091783285140991,
+      "learning_rate": 1.3074776010334466e-08,
+      "loss": 2.2695,
+      "mean_token_accuracy": 0.506029836833477,
+      "step": 5650
+    },
+    {
+      "epoch": 0.4935903026074823,
+      "grad_norm": 1.9325324296951294,
+      "learning_rate": 1.0147529487432028e-08,
+      "loss": 2.268,
+      "mean_token_accuracy": 0.5071061670780181,
+      "step": 5660
+    },
+    {
+      "epoch": 0.4944623702799337,
+      "grad_norm": 2.071957588195801,
+      "learning_rate": 7.590586684805834e-09,
+      "loss": 2.3746,
+      "mean_token_accuracy": 0.4878546938300133,
+      "step": 5670
+    },
+    {
+      "epoch": 0.49533443795238513,
+      "grad_norm": 2.0705039501190186,
+      "learning_rate": 5.4040423831802635e-09,
+      "loss": 2.307,
+      "mean_token_accuracy": 0.5019569426774979,
+      "step": 5680
+    },
+    {
+      "epoch": 0.4962065056248365,
+      "grad_norm": 2.1722216606140137,
+      "learning_rate": 3.587977633348061e-09,
+      "loss": 2.2875,
+      "mean_token_accuracy": 0.5036937370896339,
+      "step": 5690
+    },
+    {
+      "epoch": 0.4970785732972879,
+      "grad_norm": 2.023979902267456,
+      "learning_rate": 2.1424597531749524e-09,
+      "loss": 2.3387,
+      "mean_token_accuracy": 0.4925146758556366,
+      "step": 5700
+    },
+    {
+      "epoch": 0.4970785732972879,
+      "eval_runtime": 4.1318,
+      "eval_samples_per_second": 265.501,
+      "eval_steps_per_second": 66.557,
+      "step": 5700
+    },
+    {
+      "epoch": 0.49795064096973923,
+      "grad_norm": 2.0578551292419434,
+      "learning_rate": 1.0675423250994244e-09,
+      "loss": 2.3109,
+      "mean_token_accuracy": 0.5001345381140709,
+      "step": 5710
+    },
+    {
+      "epoch": 0.49882270864219064,
+      "grad_norm": 2.178114175796509,
+      "learning_rate": 3.6326519414431327e-10,
+      "loss": 2.3234,
+      "mean_token_accuracy": 0.49635518789291383,
+      "step": 5720
+    },
+    {
+      "epoch": 0.49969477631464204,
+      "grad_norm": 2.022113561630249,
+      "learning_rate": 2.965446644798142e-11,
+      "loss": 2.3328,
+      "mean_token_accuracy": 0.4927593007683754,
+      "step": 5730
+    },
+    {
+      "epoch": 0.5000436033836225,
+      "mean_token_accuracy": 0.49694227427244186,
+      "step": 5734,
+      "total_flos": 0.0,
+      "train_loss": 3.0764607222270666,
+      "train_runtime": 988.267,
+      "train_samples_per_second": 92.823,
+      "train_steps_per_second": 5.802
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 5734,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": false,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 0.0,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}