Training in progress, step 10000, checkpoint

checkpoint-10000/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0f573415a31a6e5e51de63c04347402fa6cc4b78a86595a2c6839881eece7e5a
+oid sha256:137fe9dcc15d3345d074f44c433f9fecc3988737c5e320320ed99ccd75cfb52e
 size 503128704

checkpoint-10000/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:45b835c2609a57b61f9b9dc224466501c6f9fcef5931a37c129bb1c842b091cc
+oid sha256:bf4950616eaa7d17ab89514c942a0cebb59d222366e4cd9a2ade93d042bb1a50
 size 1006351290

checkpoint-10000/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7eaa6575e8a76da450c71dba12e5221e4cd19663f53f5a5eccef335047c9f788
+oid sha256:1abdd4c7c045dcb1f8a6404bddc83e4806c8d3e3173eb1017f9b082b3c161e85
 size 14244

checkpoint-10000/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:90c311f0069114271a5dba5d72dcffc82e7a33081dd5f150aa8c9b2278c1c497
+oid sha256:baaddc0dc177f093cb7927eae851400a36fde6f45129bcaf2df443bdde5269f4
 size 1064

checkpoint-10000/trainer_state.json

@@ -1,7 +1,7 @@
 {
-  "best_metric": 3.7458531856536865,
-  "best_model_checkpoint": "/scratch/cl5625/exceptions/models/100M_495/checkpoint-10000",
-  "epoch": 1.0763104079216446,
+  "best_metric": 3.7537264823913574,
+  "best_model_checkpoint": "/scratch/cl5625/exceptions/models/100M_high_100_495/checkpoint-10000",
+  "epoch": 1.0781671159029649,
   "eval_steps": 1000,
   "global_step": 10000,
   "is_hyper_param_search": false,
@@ -9,1498 +9,1498 @@
   "is_world_process_zero": true,
   "log_history": [
     {
-      "epoch": 0.005381552039608223,
-      "grad_norm": 1.075736165046692,
+      "epoch": 0.005390835579514825,
+      "grad_norm": 1.41281259059906,
       "learning_rate": 0.0003,
-      "loss": 8.6392,
+      "loss": 8.517,
       "step": 50
     },
     {
-      "epoch": 0.010763104079216447,
-      "grad_norm": 1.940716028213501,
+      "epoch": 0.01078167115902965,
+      "grad_norm": 10.04233169555664,
       "learning_rate": 0.0006,
-      "loss": 6.9097,
+      "loss": 6.9303,
       "step": 100
     },
     {
-      "epoch": 0.01614465611882467,
-      "grad_norm": 2.641298294067383,
-      "learning_rate": 0.0005996767589699385,
-      "loss": 6.4736,
+      "epoch": 0.016172506738544475,
+      "grad_norm": 1.715460181236267,
+      "learning_rate": 0.0005996762007555315,
+      "loss": 6.5396,
       "step": 150
     },
     {
-      "epoch": 0.021526208158432893,
-      "grad_norm": 1.0494722127914429,
-      "learning_rate": 0.0005993535179398771,
-      "loss": 6.2312,
+      "epoch": 0.0215633423180593,
+      "grad_norm": 1.74041748046875,
+      "learning_rate": 0.000599352401511063,
+      "loss": 6.3013,
       "step": 200
     },
     {
-      "epoch": 0.026907760198041114,
-      "grad_norm": 2.403583526611328,
-      "learning_rate": 0.0005990302769098158,
-      "loss": 6.0645,
+      "epoch": 0.026954177897574125,
+      "grad_norm": 1.5424238443374634,
+      "learning_rate": 0.0005990286022665946,
+      "loss": 6.1131,
       "step": 250
     },
     {
-      "epoch": 0.03228931223764934,
-      "grad_norm": 1.0529396533966064,
-      "learning_rate": 0.0005987070358797543,
-      "loss": 5.9584,
+      "epoch": 0.03234501347708895,
+      "grad_norm": 1.5338594913482666,
+      "learning_rate": 0.0005987048030221263,
+      "loss": 5.999,
       "step": 300
     },
     {
-      "epoch": 0.03767086427725756,
-      "grad_norm": 2.3611762523651123,
-      "learning_rate": 0.0005983837948496929,
-      "loss": 5.8904,
+      "epoch": 0.03773584905660377,
+      "grad_norm": 1.283864974975586,
+      "learning_rate": 0.0005983810037776578,
+      "loss": 5.9014,
       "step": 350
     },
     {
-      "epoch": 0.04305241631686579,
-      "grad_norm": 1.3422589302062988,
-      "learning_rate": 0.0005980605538196314,
-      "loss": 5.7955,
+      "epoch": 0.0431266846361186,
+      "grad_norm": 1.1902074813842773,
+      "learning_rate": 0.0005980572045331894,
+      "loss": 5.8383,
       "step": 400
     },
     {
-      "epoch": 0.048433968356474004,
-      "grad_norm": 1.2432260513305664,
-      "learning_rate": 0.0005977373127895701,
-      "loss": 5.7188,
+      "epoch": 0.04851752021563342,
+      "grad_norm": 0.7562912702560425,
+      "learning_rate": 0.0005977334052887209,
+      "loss": 5.7487,
       "step": 450
     },
     {
-      "epoch": 0.05381552039608223,
-      "grad_norm": 1.03367280960083,
-      "learning_rate": 0.0005974140717595086,
-      "loss": 5.6017,
+      "epoch": 0.05390835579514825,
+      "grad_norm": 1.742124080657959,
+      "learning_rate": 0.0005974096060442526,
+      "loss": 5.6927,
       "step": 500
     },
     {
-      "epoch": 0.05919707243569045,
-      "grad_norm": 1.6495734453201294,
-      "learning_rate": 0.0005970908307294472,
-      "loss": 5.5867,
+      "epoch": 0.05929919137466307,
+      "grad_norm": 1.421120524406433,
+      "learning_rate": 0.0005970858067997841,
+      "loss": 5.5825,
       "step": 550
     },
     {
-      "epoch": 0.06457862447529868,
-      "grad_norm": 1.1987271308898926,
-      "learning_rate": 0.0005967675896993858,
-      "loss": 5.4715,
+      "epoch": 0.0646900269541779,
+      "grad_norm": 1.1182944774627686,
+      "learning_rate": 0.0005967620075553157,
+      "loss": 5.5529,
       "step": 600
     },
     {
-      "epoch": 0.0699601765149069,
-      "grad_norm": 1.3340237140655518,
-      "learning_rate": 0.0005964443486693243,
-      "loss": 5.4042,
+      "epoch": 0.07008086253369272,
+      "grad_norm": 1.370255708694458,
+      "learning_rate": 0.0005964382083108472,
+      "loss": 5.4469,
       "step": 650
     },
     {
-      "epoch": 0.07534172855451512,
-      "grad_norm": 1.4124592542648315,
-      "learning_rate": 0.000596121107639263,
-      "loss": 5.3681,
+      "epoch": 0.07547169811320754,
+      "grad_norm": 1.5170753002166748,
+      "learning_rate": 0.0005961144090663788,
+      "loss": 5.4229,
       "step": 700
     },
     {
-      "epoch": 0.08072328059412334,
-      "grad_norm": 1.2840919494628906,
-      "learning_rate": 0.0005957978666092015,
-      "loss": 5.3004,
+      "epoch": 0.08086253369272237,
+      "grad_norm": 1.429090976715088,
+      "learning_rate": 0.0005957906098219104,
+      "loss": 5.3479,
       "step": 750
     },
     {
-      "epoch": 0.08610483263373157,
-      "grad_norm": 1.2653297185897827,
-      "learning_rate": 0.0005954746255791401,
-      "loss": 5.2782,
+      "epoch": 0.0862533692722372,
+      "grad_norm": 1.4216498136520386,
+      "learning_rate": 0.0005954668105774419,
+      "loss": 5.2936,
       "step": 800
     },
     {
-      "epoch": 0.09148638467333979,
-      "grad_norm": 1.1787012815475464,
-      "learning_rate": 0.0005951513845490787,
-      "loss": 5.2122,
+      "epoch": 0.09164420485175202,
+      "grad_norm": 1.17503023147583,
+      "learning_rate": 0.0005951430113329735,
+      "loss": 5.2588,
       "step": 850
     },
     {
-      "epoch": 0.09686793671294801,
-      "grad_norm": 1.4027178287506104,
-      "learning_rate": 0.0005948281435190174,
-      "loss": 5.1643,
+      "epoch": 0.09703504043126684,
+      "grad_norm": 1.144158124923706,
+      "learning_rate": 0.0005948192120885051,
+      "loss": 5.1691,
       "step": 900
     },
     {
-      "epoch": 0.10224948875255624,
-      "grad_norm": 1.2877130508422852,
-      "learning_rate": 0.0005945049024889559,
-      "loss": 5.1141,
+      "epoch": 0.10242587601078167,
+      "grad_norm": 0.9053890109062195,
+      "learning_rate": 0.0005944954128440366,
+      "loss": 5.1664,
       "step": 950
     },
     {
-      "epoch": 0.10763104079216446,
-      "grad_norm": 1.1237081289291382,
-      "learning_rate": 0.0005941816614588944,
-      "loss": 5.0786,
+      "epoch": 0.1078167115902965,
+      "grad_norm": 1.2537935972213745,
+      "learning_rate": 0.0005941716135995682,
+      "loss": 5.1099,
       "step": 1000
     },
     {
-      "epoch": 0.10763104079216446,
-      "eval_accuracy": 0.22782348382229126,
-      "eval_loss": 5.011038780212402,
-      "eval_runtime": 182.7004,
-      "eval_samples_per_second": 98.582,
-      "eval_steps_per_second": 6.163,
+      "epoch": 0.1078167115902965,
+      "eval_accuracy": 0.22626724851012445,
+      "eval_loss": 5.036267280578613,
+      "eval_runtime": 183.603,
+      "eval_samples_per_second": 98.098,
+      "eval_steps_per_second": 6.133,
       "step": 1000
     },
     {
-      "epoch": 0.11301259283177269,
-      "grad_norm": 1.0486116409301758,
-      "learning_rate": 0.000593858420428833,
-      "loss": 5.0249,
+      "epoch": 0.11320754716981132,
+      "grad_norm": 1.1290090084075928,
+      "learning_rate": 0.0005938478143550997,
+      "loss": 5.065,
       "step": 1050
     },
     {
-      "epoch": 0.1183941448713809,
-      "grad_norm": 1.4592251777648926,
-      "learning_rate": 0.0005935351793987716,
-      "loss": 5.0193,
+      "epoch": 0.11859838274932614,
+      "grad_norm": 1.6579389572143555,
+      "learning_rate": 0.0005935240151106314,
+      "loss": 5.0244,
       "step": 1100
     },
     {
-      "epoch": 0.12377569691098914,
-      "grad_norm": 0.8498998880386353,
-      "learning_rate": 0.0005932119383687103,
-      "loss": 4.9749,
+      "epoch": 0.12398921832884097,
+      "grad_norm": 1.2606121301651,
+      "learning_rate": 0.0005932002158661629,
+      "loss": 5.0081,
       "step": 1150
     },
     {
-      "epoch": 0.12915724895059735,
-      "grad_norm": 1.0922001600265503,
-      "learning_rate": 0.0005928886973386488,
-      "loss": 4.9523,
+      "epoch": 0.1293800539083558,
+      "grad_norm": 1.0182017087936401,
+      "learning_rate": 0.0005928764166216945,
+      "loss": 4.991,
       "step": 1200
     },
     {
-      "epoch": 0.13453880099020557,
-      "grad_norm": 1.251257061958313,
-      "learning_rate": 0.0005925654563085874,
-      "loss": 4.9168,
+      "epoch": 0.1347708894878706,
+      "grad_norm": 0.9977413415908813,
+      "learning_rate": 0.000592552617377226,
+      "loss": 4.9628,
       "step": 1250
     },
     {
-      "epoch": 0.1399203530298138,
-      "grad_norm": 0.999203085899353,
-      "learning_rate": 0.000592242215278526,
-      "loss": 4.896,
+      "epoch": 0.14016172506738545,
+      "grad_norm": 0.8428726196289062,
+      "learning_rate": 0.0005922288181327577,
+      "loss": 4.9069,
       "step": 1300
     },
     {
-      "epoch": 0.14530190506942203,
-      "grad_norm": 1.1269627809524536,
-      "learning_rate": 0.0005919189742484645,
-      "loss": 4.8705,
+      "epoch": 0.14555256064690028,
+      "grad_norm": 1.242508888244629,
+      "learning_rate": 0.0005919050188882893,
+      "loss": 4.8888,
       "step": 1350
     },
     {
-      "epoch": 0.15068345710903025,
-      "grad_norm": 0.9638675451278687,
-      "learning_rate": 0.0005915957332184032,
-      "loss": 4.8564,
+      "epoch": 0.1509433962264151,
+      "grad_norm": 0.9166982173919678,
+      "learning_rate": 0.0005915812196438207,
+      "loss": 4.8623,
       "step": 1400
     },
     {
-      "epoch": 0.15606500914863847,
-      "grad_norm": 1.0301928520202637,
-      "learning_rate": 0.0005912724921883417,
-      "loss": 4.8593,
+      "epoch": 0.15633423180592992,
+      "grad_norm": 1.2252544164657593,
+      "learning_rate": 0.0005912574203993524,
+      "loss": 4.8387,
       "step": 1450
     },
     {
-      "epoch": 0.16144656118824668,
-      "grad_norm": 1.0656464099884033,
-      "learning_rate": 0.0005909492511582803,
-      "loss": 4.7941,
+      "epoch": 0.16172506738544473,
+      "grad_norm": 1.0784257650375366,
+      "learning_rate": 0.0005909336211548839,
+      "loss": 4.832,
       "step": 1500
     },
     {
-      "epoch": 0.1668281132278549,
-      "grad_norm": 1.2099454402923584,
-      "learning_rate": 0.0005906260101282189,
-      "loss": 4.79,
+      "epoch": 0.16711590296495957,
+      "grad_norm": 0.8483706712722778,
+      "learning_rate": 0.0005906098219104155,
+      "loss": 4.7853,
       "step": 1550
     },
     {
-      "epoch": 0.17220966526746315,
-      "grad_norm": 0.8705623745918274,
-      "learning_rate": 0.0005903027690981575,
-      "loss": 4.7639,
+      "epoch": 0.1725067385444744,
+      "grad_norm": 0.8520029783248901,
+      "learning_rate": 0.000590286022665947,
+      "loss": 4.8012,
       "step": 1600
     },
     {
-      "epoch": 0.17759121730707136,
-      "grad_norm": 1.1025254726409912,
-      "learning_rate": 0.000589979528068096,
-      "loss": 4.7225,
+      "epoch": 0.1778975741239892,
+      "grad_norm": 0.9755344390869141,
+      "learning_rate": 0.0005899622234214787,
+      "loss": 4.78,
       "step": 1650
     },
     {
-      "epoch": 0.18297276934667958,
-      "grad_norm": 0.9097400307655334,
-      "learning_rate": 0.0005896562870380347,
-      "loss": 4.7106,
+      "epoch": 0.18328840970350405,
+      "grad_norm": 0.9586187601089478,
+      "learning_rate": 0.0005896384241770102,
+      "loss": 4.6985,
       "step": 1700
     },
     {
-      "epoch": 0.1883543213862878,
-      "grad_norm": 0.8317919969558716,
-      "learning_rate": 0.0005893330460079732,
-      "loss": 4.7051,
+      "epoch": 0.18867924528301888,
+      "grad_norm": 1.0500203371047974,
+      "learning_rate": 0.0005893146249325418,
+      "loss": 4.7112,
       "step": 1750
     },
     {
-      "epoch": 0.19373587342589602,
-      "grad_norm": 0.9143258333206177,
-      "learning_rate": 0.0005890098049779118,
-      "loss": 4.6752,
+      "epoch": 0.1940700808625337,
+      "grad_norm": 0.9213830232620239,
+      "learning_rate": 0.0005889908256880733,
+      "loss": 4.6982,
       "step": 1800
     },
     {
-      "epoch": 0.19911742546550426,
-      "grad_norm": 1.1971888542175293,
-      "learning_rate": 0.0005886865639478504,
-      "loss": 4.6509,
+      "epoch": 0.19946091644204852,
+      "grad_norm": 1.0179482698440552,
+      "learning_rate": 0.0005886670264436049,
+      "loss": 4.656,
       "step": 1850
     },
     {
-      "epoch": 0.20449897750511248,
-      "grad_norm": 1.1611573696136475,
-      "learning_rate": 0.0005883633229177889,
-      "loss": 4.6188,
+      "epoch": 0.20485175202156333,
+      "grad_norm": 1.0454351902008057,
+      "learning_rate": 0.0005883432271991365,
+      "loss": 4.6528,
       "step": 1900
     },
     {
-      "epoch": 0.2098805295447207,
-      "grad_norm": 1.1406153440475464,
-      "learning_rate": 0.0005880400818877276,
-      "loss": 4.5752,
+      "epoch": 0.21024258760107817,
+      "grad_norm": 1.1703290939331055,
+      "learning_rate": 0.0005880194279546681,
+      "loss": 4.6166,
       "step": 1950
     },
     {
-      "epoch": 0.2152620815843289,
-      "grad_norm": 0.9002125263214111,
-      "learning_rate": 0.0005877168408576662,
-      "loss": 4.5645,
+      "epoch": 0.215633423180593,
+      "grad_norm": 1.1421180963516235,
+      "learning_rate": 0.0005876956287101996,
+      "loss": 4.5976,
       "step": 2000
     },
     {
-      "epoch": 0.2152620815843289,
-      "eval_accuracy": 0.2710199605133675,
-      "eval_loss": 4.505298614501953,
-      "eval_runtime": 183.0944,
-      "eval_samples_per_second": 98.37,
-      "eval_steps_per_second": 6.15,
+      "epoch": 0.215633423180593,
+      "eval_accuracy": 0.26813989841606906,
+      "eval_loss": 4.5355610847473145,
+      "eval_runtime": 183.4606,
+      "eval_samples_per_second": 98.174,
+      "eval_steps_per_second": 6.138,
       "step": 2000
     },
     {
-      "epoch": 0.22064363362393713,
-      "grad_norm": 1.0004241466522217,
-      "learning_rate": 0.0005873935998276048,
-      "loss": 4.5533,
+      "epoch": 0.2210242587601078,
+      "grad_norm": 1.0642319917678833,
+      "learning_rate": 0.0005873718294657312,
+      "loss": 4.6073,
       "step": 2050
     },
     {
-      "epoch": 0.22602518566354537,
-      "grad_norm": 0.9126260280609131,
-      "learning_rate": 0.0005870703587975433,
-      "loss": 4.5523,
+      "epoch": 0.22641509433962265,
+      "grad_norm": 0.8122314810752869,
+      "learning_rate": 0.0005870480302212628,
+      "loss": 4.5755,
       "step": 2100
     },
     {
-      "epoch": 0.2314067377031536,
-      "grad_norm": 1.2047978639602661,
-      "learning_rate": 0.0005867471177674818,
-      "loss": 4.533,
+      "epoch": 0.23180592991913745,
+      "grad_norm": 1.088830590248108,
+      "learning_rate": 0.0005867242309767943,
+      "loss": 4.5186,
       "step": 2150
     },
     {
-      "epoch": 0.2367882897427618,
-      "grad_norm": 1.082648754119873,
-      "learning_rate": 0.0005864238767374205,
-      "loss": 4.5228,
+      "epoch": 0.2371967654986523,
+      "grad_norm": 0.9628719091415405,
+      "learning_rate": 0.0005864004317323259,
+      "loss": 4.5535,
       "step": 2200
     },
     {
-      "epoch": 0.24216984178237003,
-      "grad_norm": 0.9497154355049133,
-      "learning_rate": 0.0005861006357073591,
-      "loss": 4.4829,
+      "epoch": 0.24258760107816713,
+      "grad_norm": 0.679783284664154,
+      "learning_rate": 0.0005860766324878575,
+      "loss": 4.4903,
       "step": 2250
     },
     {
-      "epoch": 0.24755139382197827,
-      "grad_norm": 1.155327558517456,
-      "learning_rate": 0.0005857773946772977,
-      "loss": 4.4893,
+      "epoch": 0.24797843665768193,
+      "grad_norm": 0.8863517045974731,
+      "learning_rate": 0.000585752833243389,
+      "loss": 4.4702,
       "step": 2300
     },
     {
-      "epoch": 0.2529329458615865,
-      "grad_norm": 0.8410912156105042,
-      "learning_rate": 0.0005854541536472362,
-      "loss": 4.4655,
+      "epoch": 0.25336927223719674,
+      "grad_norm": 0.7841287851333618,
+      "learning_rate": 0.0005854290339989206,
+      "loss": 4.4647,
       "step": 2350
     },
     {
-      "epoch": 0.2583144979011947,
-      "grad_norm": 1.152588963508606,
-      "learning_rate": 0.0005851309126171749,
-      "loss": 4.4482,
+      "epoch": 0.2587601078167116,
+      "grad_norm": 1.5073508024215698,
+      "learning_rate": 0.0005851052347544521,
+      "loss": 4.4585,
       "step": 2400
     },
     {
-      "epoch": 0.2636960499408029,
-      "grad_norm": 0.9096517562866211,
-      "learning_rate": 0.0005848076715871134,
-      "loss": 4.4409,
+      "epoch": 0.2641509433962264,
+      "grad_norm": 1.0127826929092407,
+      "learning_rate": 0.0005847814355099838,
+      "loss": 4.4628,
       "step": 2450
     },
     {
-      "epoch": 0.26907760198041114,
-      "grad_norm": 0.7436021566390991,
-      "learning_rate": 0.000584484430557052,
-      "loss": 4.4293,
+      "epoch": 0.2695417789757412,
+      "grad_norm": 0.8225167393684387,
+      "learning_rate": 0.0005844576362655154,
+      "loss": 4.4588,
       "step": 2500
     },
     {
-      "epoch": 0.27445915402001936,
-      "grad_norm": 0.9664739370346069,
-      "learning_rate": 0.0005841611895269906,
-      "loss": 4.4294,
+      "epoch": 0.2749326145552561,
+      "grad_norm": 0.9775927662849426,
+      "learning_rate": 0.0005841338370210469,
+      "loss": 4.4167,
       "step": 2550
     },
     {
-      "epoch": 0.2798407060596276,
-      "grad_norm": 0.8059970140457153,
-      "learning_rate": 0.0005838379484969291,
-      "loss": 4.415,
+      "epoch": 0.2803234501347709,
+      "grad_norm": 0.8481923937797546,
+      "learning_rate": 0.0005838100377765785,
+      "loss": 4.4041,
       "step": 2600
     },
     {
-      "epoch": 0.2852222580992358,
-      "grad_norm": 0.8837782144546509,
-      "learning_rate": 0.0005835147074668678,
-      "loss": 4.3814,
+      "epoch": 0.2857142857142857,
+      "grad_norm": 1.06608247756958,
+      "learning_rate": 0.0005834862385321101,
+      "loss": 4.4038,
       "step": 2650
     },
     {
-      "epoch": 0.29060381013884407,
-      "grad_norm": 0.8020942807197571,
-      "learning_rate": 0.0005831914664368063,
-      "loss": 4.3716,
+      "epoch": 0.29110512129380056,
+      "grad_norm": 0.995085597038269,
+      "learning_rate": 0.0005831624392876417,
+      "loss": 4.3995,
       "step": 2700
     },
     {
-      "epoch": 0.2959853621784523,
-      "grad_norm": 0.9676242470741272,
-      "learning_rate": 0.0005828682254067449,
-      "loss": 4.3542,
+      "epoch": 0.29649595687331537,
+      "grad_norm": 1.057483434677124,
+      "learning_rate": 0.0005828386400431731,
+      "loss": 4.3603,
       "step": 2750
     },
     {
-      "epoch": 0.3013669142180605,
-      "grad_norm": 0.6955452561378479,
-      "learning_rate": 0.0005825449843766835,
-      "loss": 4.3551,
+      "epoch": 0.3018867924528302,
+      "grad_norm": 0.6513780951499939,
+      "learning_rate": 0.0005825148407987048,
+      "loss": 4.3982,
       "step": 2800
     },
     {
-      "epoch": 0.3067484662576687,
-      "grad_norm": 0.7632643580436707,
-      "learning_rate": 0.0005822217433466221,
-      "loss": 4.3336,
+      "epoch": 0.30727762803234504,
+      "grad_norm": 0.9269214272499084,
+      "learning_rate": 0.0005821910415542363,
+      "loss": 4.368,
       "step": 2850
     },
     {
-      "epoch": 0.31213001829727693,
-      "grad_norm": 0.7833709716796875,
-      "learning_rate": 0.0005818985023165607,
-      "loss": 4.3283,
+      "epoch": 0.31266846361185985,
+      "grad_norm": 0.8944932818412781,
+      "learning_rate": 0.0005818672423097679,
+      "loss": 4.3447,
       "step": 2900
     },
     {
-      "epoch": 0.31751157033688515,
-      "grad_norm": 0.7635897397994995,
-      "learning_rate": 0.0005815752612864992,
-      "loss": 4.3279,
+      "epoch": 0.31805929919137466,
+      "grad_norm": 0.7875288128852844,
+      "learning_rate": 0.0005815434430652994,
+      "loss": 4.3437,
       "step": 2950
     },
     {
-      "epoch": 0.32289312237649337,
-      "grad_norm": 0.7228420376777649,
-      "learning_rate": 0.0005812520202564378,
-      "loss": 4.3021,
+      "epoch": 0.32345013477088946,
+      "grad_norm": 0.7112038731575012,
+      "learning_rate": 0.0005812196438208311,
+      "loss": 4.3426,
       "step": 3000
     },
     {
-      "epoch": 0.32289312237649337,
-      "eval_accuracy": 0.29860888448144485,
-      "eval_loss": 4.235284328460693,
-      "eval_runtime": 182.6772,
-      "eval_samples_per_second": 98.595,
-      "eval_steps_per_second": 6.164,
+      "epoch": 0.32345013477088946,
+      "eval_accuracy": 0.2970420011862722,
+      "eval_loss": 4.25105094909668,
+      "eval_runtime": 183.3424,
+      "eval_samples_per_second": 98.237,
+      "eval_steps_per_second": 6.142,
       "step": 3000
     },
     {
-      "epoch": 0.3282746744161016,
-      "grad_norm": 0.6172655820846558,
-      "learning_rate": 0.0005809287792263764,
-      "loss": 4.2975,
+      "epoch": 0.3288409703504043,
+      "grad_norm": 0.7651824951171875,
+      "learning_rate": 0.0005808958445763626,
+      "loss": 4.3231,
       "step": 3050
     },
     {
-      "epoch": 0.3336562264557098,
-      "grad_norm": 0.7698323130607605,
-      "learning_rate": 0.0005806055381963151,
-      "loss": 4.2838,
+      "epoch": 0.33423180592991913,
+      "grad_norm": 0.7703312039375305,
+      "learning_rate": 0.0005805720453318942,
+      "loss": 4.3135,
       "step": 3100
     },
     {
-      "epoch": 0.3390377784953181,
-      "grad_norm": 1.0566166639328003,
-      "learning_rate": 0.0005802822971662536,
-      "loss": 4.2915,
+      "epoch": 0.33962264150943394,
+      "grad_norm": 0.7293694615364075,
+      "learning_rate": 0.0005802482460874257,
+      "loss": 4.2798,
       "step": 3150
     },
     {
-      "epoch": 0.3444193305349263,
-      "grad_norm": 0.6601986289024353,
-      "learning_rate": 0.0005799590561361922,
-      "loss": 4.273,
+      "epoch": 0.3450134770889488,
+      "grad_norm": 0.8173927068710327,
+      "learning_rate": 0.0005799244468429573,
+      "loss": 4.2867,
       "step": 3200
     },
     {
-      "epoch": 0.3498008825745345,
-      "grad_norm": 0.6495400667190552,
-      "learning_rate": 0.0005796358151061307,
-      "loss": 4.285,
+      "epoch": 0.3504043126684636,
+      "grad_norm": 0.7781753540039062,
+      "learning_rate": 0.0005796006475984889,
+      "loss": 4.2612,
       "step": 3250
     },
     {
-      "epoch": 0.35518243461414273,
-      "grad_norm": 0.7450563311576843,
-      "learning_rate": 0.0005793125740760694,
-      "loss": 4.2673,
+      "epoch": 0.3557951482479784,
+      "grad_norm": 0.8917109966278076,
+      "learning_rate": 0.0005792768483540205,
+      "loss": 4.2869,
       "step": 3300
     },
     {
-      "epoch": 0.36056398665375095,
-      "grad_norm": 1.018758773803711,
-      "learning_rate": 0.0005789893330460079,
-      "loss": 4.2647,
+      "epoch": 0.3611859838274933,
+      "grad_norm": 0.6178274154663086,
+      "learning_rate": 0.000578953049109552,
+      "loss": 4.2661,
       "step": 3350
     },
     {
-      "epoch": 0.36594553869335916,
-      "grad_norm": 0.7435822486877441,
-      "learning_rate": 0.0005786660920159465,
-      "loss": 4.2384,
+      "epoch": 0.3665768194070081,
+      "grad_norm": 0.9008888006210327,
+      "learning_rate": 0.0005786292498650836,
+      "loss": 4.2523,
       "step": 3400
     },
     {
-      "epoch": 0.3713270907329674,
-      "grad_norm": 0.8073232769966125,
-      "learning_rate": 0.0005783428509858851,
-      "loss": 4.2306,
+      "epoch": 0.3719676549865229,
+      "grad_norm": 0.7638339400291443,
+      "learning_rate": 0.0005783054506206152,
+      "loss": 4.2423,
       "step": 3450
     },
     {
-      "epoch": 0.3767086427725756,
-      "grad_norm": 0.65106201171875,
-      "learning_rate": 0.0005780196099558237,
-      "loss": 4.2243,
+      "epoch": 0.37735849056603776,
+      "grad_norm": 0.6660911440849304,
+      "learning_rate": 0.0005779816513761467,
+      "loss": 4.2227,
       "step": 3500
     },
     {
-      "epoch": 0.3820901948121838,
-      "grad_norm": 0.7388604283332825,
-      "learning_rate": 0.0005776963689257623,
-      "loss": 4.2108,
+      "epoch": 0.38274932614555257,
+      "grad_norm": 0.808186411857605,
+      "learning_rate": 0.0005776578521316782,
+      "loss": 4.2255,
       "step": 3550
     },
     {
-      "epoch": 0.38747174685179203,
-      "grad_norm": 0.9689790606498718,
-      "learning_rate": 0.0005773731278957008,
-      "loss": 4.2189,
+      "epoch": 0.3881401617250674,
+      "grad_norm": 0.8280903100967407,
+      "learning_rate": 0.0005773340528872099,
+      "loss": 4.227,
       "step": 3600
     },
     {
-      "epoch": 0.3928532988914003,
-      "grad_norm": 0.5655866265296936,
-      "learning_rate": 0.0005770498868656394,
-      "loss": 4.1889,
+      "epoch": 0.3935309973045822,
+      "grad_norm": 0.903641402721405,
+      "learning_rate": 0.0005770102536427414,
+      "loss": 4.2364,
       "step": 3650
     },
     {
-      "epoch": 0.3982348509310085,
-      "grad_norm": 0.7019178867340088,
-      "learning_rate": 0.000576726645835578,
-      "loss": 4.1981,
+      "epoch": 0.39892183288409705,
+      "grad_norm": 0.7360288500785828,
+      "learning_rate": 0.000576686454398273,
+      "loss": 4.2028,
       "step": 3700
     },
     {
-      "epoch": 0.40361640297061674,
-      "grad_norm": 0.6695815324783325,
-      "learning_rate": 0.0005764034048055167,
-      "loss": 4.1944,
+      "epoch": 0.40431266846361186,
+      "grad_norm": 0.8260166049003601,
+      "learning_rate": 0.0005763626551538045,
+      "loss": 4.2001,
       "step": 3750
     },
     {
-      "epoch": 0.40899795501022496,
-      "grad_norm": 0.7157506346702576,
-      "learning_rate": 0.0005760801637754552,
-      "loss": 4.1936,
+      "epoch": 0.40970350404312667,
+      "grad_norm": 0.6548323631286621,
+      "learning_rate": 0.0005760388559093362,
+      "loss": 4.2199,
       "step": 3800
     },
     {
-      "epoch": 0.4143795070498332,
-      "grad_norm": 0.8047358989715576,
-      "learning_rate": 0.0005757569227453937,
-      "loss": 4.1804,
+      "epoch": 0.41509433962264153,
+      "grad_norm": 0.7552820444107056,
+      "learning_rate": 0.0005757150566648678,
+      "loss": 4.2009,
       "step": 3850
     },
     {
-      "epoch": 0.4197610590894414,
-      "grad_norm": 0.6181033253669739,
-      "learning_rate": 0.0005754336817153324,
-      "loss": 4.1882,
+      "epoch": 0.42048517520215634,
+      "grad_norm": 0.8771376013755798,
+      "learning_rate": 0.0005753912574203993,
+      "loss": 4.1679,
       "step": 3900
     },
     {
-      "epoch": 0.4251426111290496,
-      "grad_norm": 0.6974236369132996,
-      "learning_rate": 0.0005751104406852709,
-      "loss": 4.1637,
+      "epoch": 0.42587601078167114,
+      "grad_norm": 0.6558768153190613,
+      "learning_rate": 0.0005750674581759309,
+      "loss": 4.1696,
       "step": 3950
     },
     {
-      "epoch": 0.4305241631686578,
-      "grad_norm": 0.6980713605880737,
-      "learning_rate": 0.0005747871996552096,
-      "loss": 4.1463,
+      "epoch": 0.431266846361186,
+      "grad_norm": 0.7329567670822144,
+      "learning_rate": 0.0005747436589314624,
+      "loss": 4.174,
       "step": 4000
     },
     {
-      "epoch": 0.4305241631686578,
-      "eval_accuracy": 0.31158182198071827,
-      "eval_loss": 4.100734710693359,
-      "eval_runtime": 182.4679,
-      "eval_samples_per_second": 98.708,
-      "eval_steps_per_second": 6.171,
+      "epoch": 0.431266846361186,
+      "eval_accuracy": 0.3116716779189408,
+      "eval_loss": 4.103002548217773,
+      "eval_runtime": 183.3675,
+      "eval_samples_per_second": 98.224,
+      "eval_steps_per_second": 6.141,
       "step": 4000
     },
     {
-      "epoch": 0.43590571520826604,
-      "grad_norm": 0.8112263679504395,
-      "learning_rate": 0.0005744639586251481,
-      "loss": 4.1637,
+      "epoch": 0.4366576819407008,
+      "grad_norm": 0.666535496711731,
+      "learning_rate": 0.0005744198596869941,
+      "loss": 4.1692,
       "step": 4050
     },
     {
-      "epoch": 0.44128726724787426,
-      "grad_norm": 0.6137505173683167,
-      "learning_rate": 0.0005741407175950867,
-      "loss": 4.1281,
+      "epoch": 0.4420485175202156,
+      "grad_norm": 0.6010123491287231,
+      "learning_rate": 0.0005740960604425255,
+      "loss": 4.1791,
       "step": 4100
     },
     {
-      "epoch": 0.44666881928748253,
-      "grad_norm": 0.7151933908462524,
-      "learning_rate": 0.0005738174765650253,
-      "loss": 4.1454,
+      "epoch": 0.4474393530997305,
+      "grad_norm": 0.6479814052581787,
+      "learning_rate": 0.0005737722611980572,
+      "loss": 4.1603,
       "step": 4150
     },
     {
-      "epoch": 0.45205037132709075,
-      "grad_norm": 0.6336910724639893,
-      "learning_rate": 0.0005734942355349638,
-      "loss": 4.1361,
+      "epoch": 0.4528301886792453,
+      "grad_norm": 0.64080810546875,
+      "learning_rate": 0.0005734484619535887,
+      "loss": 4.1515,
       "step": 4200
     },
     {
-      "epoch": 0.45743192336669897,
-      "grad_norm": 0.7914460301399231,
-      "learning_rate": 0.0005731709945049025,
-      "loss": 4.134,
+      "epoch": 0.4582210242587601,
+      "grad_norm": 0.5678718090057373,
+      "learning_rate": 0.0005731246627091203,
+      "loss": 4.1433,
       "step": 4250
     },
     {
-      "epoch": 0.4628134754063072,
-      "grad_norm": 1.2934105396270752,
-      "learning_rate": 0.000572847753474841,
-      "loss": 4.1562,
+      "epoch": 0.4636118598382749,
+      "grad_norm": 0.8291149735450745,
+      "learning_rate": 0.0005728008634646518,
+      "loss": 4.1365,
       "step": 4300
     },
     {
-      "epoch": 0.4681950274459154,
-      "grad_norm": 0.7198614478111267,
-      "learning_rate": 0.0005725245124447796,
-      "loss": 4.1097,
+      "epoch": 0.46900269541778977,
+      "grad_norm": 0.6569222211837769,
+      "learning_rate": 0.0005724770642201835,
+      "loss": 4.1442,
       "step": 4350
     },
     {
-      "epoch": 0.4735765794855236,
-      "grad_norm": 0.7265746593475342,
-      "learning_rate": 0.0005722012714147182,
-      "loss": 4.1082,
+      "epoch": 0.4743935309973046,
+      "grad_norm": 0.7683055996894836,
+      "learning_rate": 0.000572153264975715,
+      "loss": 4.1277,
       "step": 4400
     },
     {
-      "epoch": 0.47895813152513184,
-      "grad_norm": 0.6873067617416382,
-      "learning_rate": 0.0005718780303846568,
-      "loss": 4.1105,
+      "epoch": 0.4797843665768194,
+      "grad_norm": 0.7972410321235657,
+      "learning_rate": 0.0005718294657312466,
+      "loss": 4.1351,
       "step": 4450
     },
     {
-      "epoch": 0.48433968356474005,
-      "grad_norm": 0.6153125762939453,
-      "learning_rate": 0.0005715547893545953,
-      "loss": 4.1073,
+      "epoch": 0.48517520215633425,
+      "grad_norm": 0.6762375235557556,
+      "learning_rate": 0.0005715056664867781,
+      "loss": 4.1153,
       "step": 4500
     },
     {
-      "epoch": 0.48972123560434827,
-      "grad_norm": 0.6878327131271362,
-      "learning_rate": 0.000571231548324534,
-      "loss": 4.0956,
+      "epoch": 0.49056603773584906,
+      "grad_norm": 0.553313672542572,
+      "learning_rate": 0.0005711818672423097,
+      "loss": 4.1043,
       "step": 4550
     },
     {
-      "epoch": 0.49510278764395654,
-      "grad_norm": 0.8140902519226074,
-      "learning_rate": 0.0005709083072944725,
-      "loss": 4.0975,
+      "epoch": 0.49595687331536387,
+      "grad_norm": 0.6392484903335571,
+      "learning_rate": 0.0005708580679978413,
+      "loss": 4.114,
       "step": 4600
     },
     {
-      "epoch": 0.5004843396835648,
-      "grad_norm": 0.571739137172699,
-      "learning_rate": 0.0005705850662644111,
-      "loss": 4.09,
+      "epoch": 0.5013477088948787,
+      "grad_norm": 0.6732353568077087,
+      "learning_rate": 0.0005705342687533729,
+      "loss": 4.0907,
       "step": 4650
     },
     {
-      "epoch": 0.505865891723173,
-      "grad_norm": 0.7585338950157166,
-      "learning_rate": 0.0005702618252343497,
-      "loss": 4.1061,
+      "epoch": 0.5067385444743935,
+      "grad_norm": 0.689285397529602,
+      "learning_rate": 0.0005702104695089044,
+      "loss": 4.0959,
       "step": 4700
     },
     {
-      "epoch": 0.5112474437627812,
-      "grad_norm": 0.8002457618713379,
-      "learning_rate": 0.0005699385842042882,
-      "loss": 4.0763,
+      "epoch": 0.5121293800539084,
+      "grad_norm": 0.6244467496871948,
+      "learning_rate": 0.000569886670264436,
+      "loss": 4.0792,
       "step": 4750
     },
     {
-      "epoch": 0.5166289958023894,
-      "grad_norm": 0.6828979253768921,
-      "learning_rate": 0.0005696153431742269,
-      "loss": 4.0878,
+      "epoch": 0.5175202156334232,
+      "grad_norm": 0.5999540686607361,
+      "learning_rate": 0.0005695628710199675,
+      "loss": 4.0924,
       "step": 4800
     },
     {
-      "epoch": 0.5220105478419976,
-      "grad_norm": 0.6259872913360596,
-      "learning_rate": 0.0005692921021441655,
-      "loss": 4.0798,
+      "epoch": 0.522911051212938,
+      "grad_norm": 0.6409953236579895,
+      "learning_rate": 0.0005692390717754991,
+      "loss": 4.0864,
       "step": 4850
     },
     {
-      "epoch": 0.5273920998816058,
-      "grad_norm": 0.7003547549247742,
-      "learning_rate": 0.0005689688611141041,
-      "loss": 4.0822,
+      "epoch": 0.5283018867924528,
+      "grad_norm": 0.690284788608551,
+      "learning_rate": 0.0005689152725310306,
+      "loss": 4.097,
       "step": 4900
     },
     {
-      "epoch": 0.5327736519212141,
-      "grad_norm": 0.6551280617713928,
-      "learning_rate": 0.0005686456200840426,
-      "loss": 4.0599,
+      "epoch": 0.5336927223719676,
+      "grad_norm": 0.6556663513183594,
+      "learning_rate": 0.0005685914732865623,
+      "loss": 4.0683,
       "step": 4950
     },
     {
-      "epoch": 0.5381552039608223,
-      "grad_norm": 0.7073853611946106,
-      "learning_rate": 0.0005683223790539811,
-      "loss": 4.0502,
+      "epoch": 0.5390835579514824,
+      "grad_norm": 0.7610403299331665,
+      "learning_rate": 0.0005682676740420939,
+      "loss": 4.0588,
       "step": 5000
     },
     {
-      "epoch": 0.5381552039608223,
-      "eval_accuracy": 0.3215845154858072,
-      "eval_loss": 3.9914491176605225,
-      "eval_runtime": 182.6307,
-      "eval_samples_per_second": 98.62,
-      "eval_steps_per_second": 6.165,
+      "epoch": 0.5390835579514824,
+      "eval_accuracy": 0.32074832286118693,
+      "eval_loss": 4.004209518432617,
+      "eval_runtime": 183.2771,
+      "eval_samples_per_second": 98.272,
+      "eval_steps_per_second": 6.144,
       "step": 5000
     },
     {
-      "epoch": 0.5435367560004305,
-      "grad_norm": 0.5926269888877869,
-      "learning_rate": 0.0005679991380239198,
-      "loss": 4.0245,
+      "epoch": 0.5444743935309974,
+      "grad_norm": 0.6256934404373169,
+      "learning_rate": 0.0005679438747976254,
+      "loss": 4.0548,
       "step": 5050
     },
     {
-      "epoch": 0.5489183080400387,
-      "grad_norm": 0.6609042882919312,
-      "learning_rate": 0.0005676758969938584,
-      "loss": 4.0562,
+      "epoch": 0.5498652291105122,
+      "grad_norm": 0.7420689463615417,
+      "learning_rate": 0.000567620075553157,
+      "loss": 4.0608,
       "step": 5100
     },
     {
-      "epoch": 0.5542998600796469,
-      "grad_norm": 0.6182550191879272,
-      "learning_rate": 0.000567352655963797,
-      "loss": 4.0355,
+      "epoch": 0.555256064690027,
+      "grad_norm": 0.6498412489891052,
+      "learning_rate": 0.0005672962763086886,
+      "loss": 4.0613,
       "step": 5150
     },
     {
-      "epoch": 0.5596814121192552,
-      "grad_norm": 0.538492739200592,
-      "learning_rate": 0.0005670294149337355,
-      "loss": 4.045,
+      "epoch": 0.5606469002695418,
+      "grad_norm": 0.6685329079627991,
+      "learning_rate": 0.0005669724770642202,
+      "loss": 4.0443,
       "step": 5200
     },
     {
-      "epoch": 0.5650629641588634,
-      "grad_norm": 0.6184486746788025,
-      "learning_rate": 0.0005667061739036742,
-      "loss": 4.0402,
+      "epoch": 0.5660377358490566,
+      "grad_norm": 0.6504703164100647,
+      "learning_rate": 0.0005666486778197517,
+      "loss": 4.0435,
       "step": 5250
     },
     {
-      "epoch": 0.5704445161984716,
-      "grad_norm": 0.6973541378974915,
-      "learning_rate": 0.0005663829328736127,
-      "loss": 4.0271,
+      "epoch": 0.5714285714285714,
+      "grad_norm": 0.6796240210533142,
+      "learning_rate": 0.0005663248785752833,
+      "loss": 4.026,
       "step": 5300
     },
     {
-      "epoch": 0.5758260682380799,
-      "grad_norm": 0.6938584446907043,
-      "learning_rate": 0.0005660596918435512,
-      "loss": 4.0294,
+      "epoch": 0.5768194070080862,
+      "grad_norm": 0.5926557183265686,
+      "learning_rate": 0.0005660010793308148,
+      "loss": 4.0387,
       "step": 5350
     },
     {
-      "epoch": 0.5812076202776881,
-      "grad_norm": 0.734360933303833,
-      "learning_rate": 0.0005657364508134899,
-      "loss": 4.0206,
+      "epoch": 0.5822102425876011,
+      "grad_norm": 0.6693084836006165,
+      "learning_rate": 0.0005656772800863465,
+      "loss": 4.0303,
       "step": 5400
     },
     {
-      "epoch": 0.5865891723172963,
-      "grad_norm": 0.6802305579185486,
-      "learning_rate": 0.0005654132097834284,
-      "loss": 4.0411,
+      "epoch": 0.5876010781671159,
+      "grad_norm": 0.6638355255126953,
+      "learning_rate": 0.0005653534808418779,
+      "loss": 4.0393,
       "step": 5450
     },
     {
-      "epoch": 0.5919707243569046,
-      "grad_norm": 0.6002328991889954,
-      "learning_rate": 0.0005650899687533671,
-      "loss": 4.0047,
+      "epoch": 0.5929919137466307,
+      "grad_norm": 0.5713601112365723,
+      "learning_rate": 0.0005650296815974096,
+      "loss": 4.0419,
       "step": 5500
     },
     {
-      "epoch": 0.5973522763965128,
-      "grad_norm": 0.5953026413917542,
-      "learning_rate": 0.0005647667277233056,
-      "loss": 4.02,
+      "epoch": 0.5983827493261455,
+      "grad_norm": 0.6896920204162598,
+      "learning_rate": 0.0005647058823529411,
+      "loss": 4.0443,
       "step": 5550
     },
     {
-      "epoch": 0.602733828436121,
-      "grad_norm": 0.6416895985603333,
-      "learning_rate": 0.0005644434866932442,
-      "loss": 4.022,
+      "epoch": 0.6037735849056604,
+      "grad_norm": 0.6261423826217651,
+      "learning_rate": 0.0005643820831084727,
+      "loss": 4.0231,
       "step": 5600
     },
     {
-      "epoch": 0.6081153804757292,
-      "grad_norm": 0.6067703366279602,
-      "learning_rate": 0.0005641202456631828,
-      "loss": 4.0064,
+      "epoch": 0.6091644204851752,
+      "grad_norm": 0.6080160140991211,
+      "learning_rate": 0.0005640582838640042,
+      "loss": 4.0137,
       "step": 5650
     },
     {
-      "epoch": 0.6134969325153374,
-      "grad_norm": 0.6449361443519592,
-      "learning_rate": 0.0005637970046331214,
-      "loss": 3.9941,
+      "epoch": 0.6145552560646901,
+      "grad_norm": 0.5479706525802612,
+      "learning_rate": 0.0005637344846195358,
+      "loss": 4.0175,
       "step": 5700
     },
     {
-      "epoch": 0.6188784845549457,
-      "grad_norm": 0.5998024940490723,
-      "learning_rate": 0.00056347376360306,
-      "loss": 4.008,
+      "epoch": 0.6199460916442049,
+      "grad_norm": 0.6059190630912781,
+      "learning_rate": 0.0005634106853750674,
+      "loss": 4.0272,
       "step": 5750
     },
     {
-      "epoch": 0.6242600365945539,
-      "grad_norm": 0.6512130498886108,
-      "learning_rate": 0.0005631505225729985,
-      "loss": 3.9929,
+      "epoch": 0.6253369272237197,
+      "grad_norm": 0.5997337698936462,
+      "learning_rate": 0.000563086886130599,
+      "loss": 4.0172,
       "step": 5800
     },
     {
-      "epoch": 0.6296415886341621,
-      "grad_norm": 0.6279934048652649,
-      "learning_rate": 0.0005628272815429371,
-      "loss": 4.0128,
+      "epoch": 0.6307277628032345,
+      "grad_norm": 0.648623526096344,
+      "learning_rate": 0.0005627630868861305,
+      "loss": 4.027,
       "step": 5850
     },
     {
-      "epoch": 0.6350231406737703,
-      "grad_norm": 0.5534334778785706,
-      "learning_rate": 0.0005625040405128757,
-      "loss": 3.979,
+      "epoch": 0.6361185983827493,
+      "grad_norm": 0.5959683656692505,
+      "learning_rate": 0.0005624392876416621,
+      "loss": 3.9989,
       "step": 5900
     },
     {
-      "epoch": 0.6404046927133785,
-      "grad_norm": 0.600577175617218,
-      "learning_rate": 0.0005621807994828143,
-      "loss": 3.9642,
+      "epoch": 0.6415094339622641,
+      "grad_norm": 0.6386790871620178,
+      "learning_rate": 0.0005621154883971937,
+      "loss": 3.9924,
       "step": 5950
     },
     {
-      "epoch": 0.6457862447529867,
-      "grad_norm": 0.6984274387359619,
-      "learning_rate": 0.0005618575584527529,
-      "loss": 3.9897,
+      "epoch": 0.6469002695417789,
+      "grad_norm": 0.5958532691001892,
+      "learning_rate": 0.0005617916891527253,
+      "loss": 4.0027,
       "step": 6000
     },
     {
-      "epoch": 0.6457862447529867,
-      "eval_accuracy": 0.3287275736365068,
-      "eval_loss": 3.9194459915161133,
-      "eval_runtime": 182.389,
-      "eval_samples_per_second": 98.75,
-      "eval_steps_per_second": 6.174,
+      "epoch": 0.6469002695417789,
+      "eval_accuracy": 0.32683625281832007,
+      "eval_loss": 3.926405668258667,
+      "eval_runtime": 183.1995,
+      "eval_samples_per_second": 98.314,
+      "eval_steps_per_second": 6.146,
       "step": 6000
     },
     {
-      "epoch": 0.651167796792595,
-      "grad_norm": 0.6016660332679749,
-      "learning_rate": 0.0005615343174226915,
-      "loss": 3.9889,
+      "epoch": 0.6522911051212938,
+      "grad_norm": 0.6504858732223511,
+      "learning_rate": 0.0005614678899082568,
+      "loss": 3.976,
       "step": 6050
     },
     {
-      "epoch": 0.6565493488322032,
-      "grad_norm": 0.90537428855896,
-      "learning_rate": 0.00056121107639263,
-      "loss": 3.9729,
+      "epoch": 0.6576819407008087,
+      "grad_norm": 0.75847989320755,
+      "learning_rate": 0.0005611440906637884,
+      "loss": 4.0059,
       "step": 6100
     },
     {
-      "epoch": 0.6619309008718114,
-      "grad_norm": 0.5522285103797913,
-      "learning_rate": 0.0005608878353625687,
-      "loss": 3.9668,
+      "epoch": 0.6630727762803235,
+      "grad_norm": 0.6028177738189697,
+      "learning_rate": 0.00056082029141932,
+      "loss": 3.9891,
       "step": 6150
     },
     {
-      "epoch": 0.6673124529114196,
-      "grad_norm": 0.6631731390953064,
-      "learning_rate": 0.0005605645943325072,
-      "loss": 3.9806,
+      "epoch": 0.6684636118598383,
+      "grad_norm": 0.6849642992019653,
+      "learning_rate": 0.0005604964921748515,
+      "loss": 3.9681,
       "step": 6200
     },
     {
-      "epoch": 0.6726940049510278,
-      "grad_norm": 0.6557983756065369,
-      "learning_rate": 0.0005602413533024458,
-      "loss": 3.9728,
+      "epoch": 0.6738544474393531,
+      "grad_norm": 0.6571520566940308,
+      "learning_rate": 0.000560172692930383,
+      "loss": 3.9844,
       "step": 6250
     },
     {
-      "epoch": 0.6780755569906362,
-      "grad_norm": 0.7173463702201843,
-      "learning_rate": 0.0005599181122723844,
-      "loss": 3.9589,
+      "epoch": 0.6792452830188679,
+      "grad_norm": 0.6362521648406982,
+      "learning_rate": 0.0005598488936859147,
+      "loss": 4.0014,
       "step": 6300
     },
     {
-      "epoch": 0.6834571090302444,
-      "grad_norm": 0.5700642466545105,
-      "learning_rate": 0.000559594871242323,
-      "loss": 3.9823,
+      "epoch": 0.6846361185983828,
+      "grad_norm": 0.6521901488304138,
+      "learning_rate": 0.0005595250944414463,
+      "loss": 3.9638,
       "step": 6350
     },
     {
-      "epoch": 0.6888386610698526,
-      "grad_norm": 0.6668356657028198,
-      "learning_rate": 0.0005592716302122616,
-      "loss": 3.9757,
+      "epoch": 0.6900269541778976,
+      "grad_norm": 0.6911128163337708,
+      "learning_rate": 0.0005592012951969778,
+      "loss": 3.9705,
       "step": 6400
     },
     {
-      "epoch": 0.6942202131094608,
-      "grad_norm": 0.6102961301803589,
-      "learning_rate": 0.0005589483891822001,
-      "loss": 3.9594,
+      "epoch": 0.6954177897574124,
+      "grad_norm": 0.627589762210846,
+      "learning_rate": 0.0005588774959525094,
+      "loss": 3.9556,
       "step": 6450
     },
     {
-      "epoch": 0.699601765149069,
-      "grad_norm": 0.904893696308136,
-      "learning_rate": 0.0005586251481521387,
-      "loss": 3.9342,
+      "epoch": 0.7008086253369272,
+      "grad_norm": 0.6308057904243469,
+      "learning_rate": 0.000558553696708041,
+      "loss": 3.9665,
       "step": 6500
     },
     {
-      "epoch": 0.7049833171886772,
-      "grad_norm": 0.5942201614379883,
-      "learning_rate": 0.0005583019071220773,
-      "loss": 3.9506,
+      "epoch": 0.706199460916442,
+      "grad_norm": 0.6295865774154663,
+      "learning_rate": 0.0005582298974635726,
+      "loss": 3.9923,
       "step": 6550
     },
     {
-      "epoch": 0.7103648692282855,
-      "grad_norm": 0.6090822815895081,
-      "learning_rate": 0.000557978666092016,
-      "loss": 3.943,
+      "epoch": 0.7115902964959568,
+      "grad_norm": 0.5598770976066589,
+      "learning_rate": 0.0005579060982191041,
+      "loss": 3.945,
       "step": 6600
     },
     {
-      "epoch": 0.7157464212678937,
-      "grad_norm": 0.6859561800956726,
-      "learning_rate": 0.0005576554250619545,
-      "loss": 3.9675,
+      "epoch": 0.7169811320754716,
+      "grad_norm": 0.5894781351089478,
+      "learning_rate": 0.0005575822989746357,
+      "loss": 3.9769,
       "step": 6650
     },
     {
-      "epoch": 0.7211279733075019,
-      "grad_norm": 0.610731840133667,
-      "learning_rate": 0.0005573386488524943,
-      "loss": 3.9524,
+      "epoch": 0.7223719676549866,
+      "grad_norm": 0.6093037128448486,
+      "learning_rate": 0.0005572584997301672,
+      "loss": 3.9523,
       "step": 6700
     },
     {
-      "epoch": 0.7265095253471101,
-      "grad_norm": 0.6064528226852417,
-      "learning_rate": 0.0005570154078224328,
-      "loss": 3.9637,
+      "epoch": 0.7277628032345014,
+      "grad_norm": 0.5324745774269104,
+      "learning_rate": 0.0005569347004856989,
+      "loss": 3.958,
       "step": 6750
     },
     {
-      "epoch": 0.7318910773867183,
-      "grad_norm": 0.645495593547821,
-      "learning_rate": 0.0005566921667923715,
-      "loss": 3.9507,
+      "epoch": 0.7331536388140162,
+      "grad_norm": 0.5724645256996155,
+      "learning_rate": 0.0005566109012412303,
+      "loss": 3.9453,
       "step": 6800
     },
     {
-      "epoch": 0.7372726294263265,
-      "grad_norm": 0.6026754975318909,
-      "learning_rate": 0.00055636892576231,
-      "loss": 3.9365,
+      "epoch": 0.738544474393531,
+      "grad_norm": 0.6083112359046936,
+      "learning_rate": 0.000556287101996762,
+      "loss": 3.9246,
       "step": 6850
     },
     {
-      "epoch": 0.7426541814659348,
-      "grad_norm": 0.5844349265098572,
-      "learning_rate": 0.0005560456847322487,
-      "loss": 3.921,
+      "epoch": 0.7439353099730458,
+      "grad_norm": 0.6993039846420288,
+      "learning_rate": 0.0005559633027522935,
+      "loss": 3.9524,
       "step": 6900
     },
     {
-      "epoch": 0.748035733505543,
-      "grad_norm": 0.5787811875343323,
-      "learning_rate": 0.0005557224437021872,
-      "loss": 3.959,
+      "epoch": 0.7493261455525606,
+      "grad_norm": 0.5825726985931396,
+      "learning_rate": 0.0005556395035078251,
+      "loss": 3.949,
       "step": 6950
     },
     {
-      "epoch": 0.7534172855451512,
-      "grad_norm": 0.6540557146072388,
-      "learning_rate": 0.0005553992026721258,
-      "loss": 3.9287,
+      "epoch": 0.7547169811320755,
+      "grad_norm": 0.6970276832580566,
+      "learning_rate": 0.0005553157042633566,
+      "loss": 3.9344,
       "step": 7000
     },
     {
-      "epoch": 0.7534172855451512,
-      "eval_accuracy": 0.33348081151972686,
-      "eval_loss": 3.8603153228759766,
-      "eval_runtime": 182.9316,
-      "eval_samples_per_second": 98.458,
-      "eval_steps_per_second": 6.155,
+      "epoch": 0.7547169811320755,
+      "eval_accuracy": 0.3327122009913272,
+      "eval_loss": 3.8720343112945557,
+      "eval_runtime": 183.416,
+      "eval_samples_per_second": 98.198,
+      "eval_steps_per_second": 6.139,
       "step": 7000
     },
     {
-      "epoch": 0.7587988375847594,
-      "grad_norm": 0.5272918343544006,
-      "learning_rate": 0.0005550759616420644,
-      "loss": 3.9179,
+      "epoch": 0.7601078167115903,
+      "grad_norm": 0.6229957342147827,
+      "learning_rate": 0.0005549919050188882,
+      "loss": 3.9287,
       "step": 7050
     },
     {
-      "epoch": 0.7641803896243676,
-      "grad_norm": 0.5821722745895386,
-      "learning_rate": 0.000554752720612003,
-      "loss": 3.9573,
+      "epoch": 0.7654986522911051,
+      "grad_norm": 0.6007510423660278,
+      "learning_rate": 0.0005546681057744198,
+      "loss": 3.9394,
       "step": 7100
     },
     {
-      "epoch": 0.7695619416639758,
-      "grad_norm": 0.5356701612472534,
-      "learning_rate": 0.0005544294795819415,
-      "loss": 3.9199,
+      "epoch": 0.77088948787062,
+      "grad_norm": 0.5614594221115112,
+      "learning_rate": 0.0005543443065299514,
+      "loss": 3.9401,
       "step": 7150
     },
     {
-      "epoch": 0.7749434937035841,
-      "grad_norm": 0.6949185132980347,
-      "learning_rate": 0.0005541062385518801,
-      "loss": 3.9261,
+      "epoch": 0.7762803234501348,
+      "grad_norm": 0.6459982395172119,
+      "learning_rate": 0.000554020507285483,
+      "loss": 3.9381,
       "step": 7200
     },
     {
-      "epoch": 0.7803250457431924,
-      "grad_norm": 0.6216970682144165,
-      "learning_rate": 0.0005537829975218188,
-      "loss": 3.9127,
+      "epoch": 0.7816711590296496,
+      "grad_norm": 0.5772397518157959,
+      "learning_rate": 0.0005536967080410145,
+      "loss": 3.8955,
       "step": 7250
     },
     {
-      "epoch": 0.7857065977828006,
-      "grad_norm": 0.6515355110168457,
-      "learning_rate": 0.0005534597564917573,
-      "loss": 3.8967,
+      "epoch": 0.7870619946091644,
+      "grad_norm": 0.6932154297828674,
+      "learning_rate": 0.0005533729087965462,
+      "loss": 3.9357,
       "step": 7300
     },
     {
-      "epoch": 0.7910881498224088,
-      "grad_norm": 0.600446879863739,
-      "learning_rate": 0.0005531365154616959,
-      "loss": 3.9118,
+      "epoch": 0.7924528301886793,
+      "grad_norm": 0.5726342797279358,
+      "learning_rate": 0.0005530491095520777,
+      "loss": 3.9201,
       "step": 7350
     },
     {
-      "epoch": 0.796469701862017,
-      "grad_norm": 0.5874130129814148,
-      "learning_rate": 0.0005528132744316344,
-      "loss": 3.9172,
+      "epoch": 0.7978436657681941,
+      "grad_norm": 0.642338752746582,
+      "learning_rate": 0.0005527253103076093,
+      "loss": 3.9198,
       "step": 7400
     },
     {
-      "epoch": 0.8018512539016253,
-      "grad_norm": 0.5926812887191772,
-      "learning_rate": 0.0005524900334015731,
-      "loss": 3.8874,
+      "epoch": 0.8032345013477089,
+      "grad_norm": 0.5481148958206177,
+      "learning_rate": 0.0005524015110631408,
+      "loss": 3.9085,
       "step": 7450
     },
     {
-      "epoch": 0.8072328059412335,
-      "grad_norm": 0.5605654120445251,
-      "learning_rate": 0.0005521667923715117,
-      "loss": 3.9155,
+      "epoch": 0.8086253369272237,
+      "grad_norm": 0.6552069187164307,
+      "learning_rate": 0.0005520777118186724,
+      "loss": 3.8957,
       "step": 7500
     },
     {
-      "epoch": 0.8126143579808417,
-      "grad_norm": 0.5662755966186523,
-      "learning_rate": 0.0005518435513414502,
-      "loss": 3.8875,
+      "epoch": 0.8140161725067385,
+      "grad_norm": 0.5861802101135254,
+      "learning_rate": 0.0005517539125742039,
+      "loss": 3.9231,
       "step": 7550
     },
     {
-      "epoch": 0.8179959100204499,
-      "grad_norm": 0.596157968044281,
-      "learning_rate": 0.0005515203103113888,
-      "loss": 3.9132,
+      "epoch": 0.8194070080862533,
+      "grad_norm": 0.5593369007110596,
+      "learning_rate": 0.0005514301133297355,
+      "loss": 3.9242,
       "step": 7600
     },
     {
-      "epoch": 0.8233774620600581,
-      "grad_norm": 0.5835180878639221,
-      "learning_rate": 0.0005511970692813274,
-      "loss": 3.9026,
+      "epoch": 0.8247978436657682,
+      "grad_norm": 0.5746191143989563,
+      "learning_rate": 0.0005511063140852671,
+      "loss": 3.8999,
       "step": 7650
     },
     {
-      "epoch": 0.8287590140996663,
-      "grad_norm": 0.6493765711784363,
-      "learning_rate": 0.000550873828251266,
-      "loss": 3.9017,
+      "epoch": 0.8301886792452831,
+      "grad_norm": 0.5479608178138733,
+      "learning_rate": 0.0005507825148407987,
+      "loss": 3.903,
       "step": 7700
     },
     {
-      "epoch": 0.8341405661392746,
-      "grad_norm": 0.6501802206039429,
-      "learning_rate": 0.0005505505872212045,
-      "loss": 3.8935,
+      "epoch": 0.8355795148247979,
+      "grad_norm": 0.5137019753456116,
+      "learning_rate": 0.0005504587155963302,
+      "loss": 3.885,
       "step": 7750
     },
     {
-      "epoch": 0.8395221181788828,
-      "grad_norm": 0.6308815479278564,
-      "learning_rate": 0.0005502273461911432,
-      "loss": 3.8995,
+      "epoch": 0.8409703504043127,
+      "grad_norm": 0.5970565676689148,
+      "learning_rate": 0.0005501349163518618,
+      "loss": 3.9148,
       "step": 7800
     },
     {
-      "epoch": 0.844903670218491,
-      "grad_norm": 0.5835139751434326,
-      "learning_rate": 0.0005499041051610817,
-      "loss": 3.8875,
+      "epoch": 0.8463611859838275,
+      "grad_norm": 0.632587730884552,
+      "learning_rate": 0.0005498111171073933,
+      "loss": 3.8933,
       "step": 7850
     },
     {
-      "epoch": 0.8502852222580992,
-      "grad_norm": 0.5937869548797607,
-      "learning_rate": 0.0005495808641310204,
-      "loss": 3.8825,
+      "epoch": 0.8517520215633423,
+      "grad_norm": 0.6267861127853394,
+      "learning_rate": 0.000549487317862925,
+      "loss": 3.8961,
       "step": 7900
     },
     {
-      "epoch": 0.8556667742977074,
-      "grad_norm": 0.6203649640083313,
-      "learning_rate": 0.0005492576231009589,
-      "loss": 3.8721,
+      "epoch": 0.8571428571428571,
+      "grad_norm": 0.5923589468002319,
+      "learning_rate": 0.0005491635186184565,
+      "loss": 3.8911,
       "step": 7950
     },
     {
-      "epoch": 0.8610483263373157,
-      "grad_norm": 0.5419115424156189,
-      "learning_rate": 0.0005489343820708974,
-      "loss": 3.8684,
+      "epoch": 0.862533692722372,
+      "grad_norm": 0.7637770175933838,
+      "learning_rate": 0.0005488397193739881,
+      "loss": 3.8805,
       "step": 8000
     },
     {
-      "epoch": 0.8610483263373157,
-      "eval_accuracy": 0.33826784045105723,
-      "eval_loss": 3.812558174133301,
-      "eval_runtime": 182.9459,
-      "eval_samples_per_second": 98.45,
-      "eval_steps_per_second": 6.155,
+      "epoch": 0.862533692722372,
+      "eval_accuracy": 0.337080264387245,
+      "eval_loss": 3.8229458332061768,
+      "eval_runtime": 183.1908,
+      "eval_samples_per_second": 98.318,
+      "eval_steps_per_second": 6.147,
       "step": 8000
     },
     {
-      "epoch": 0.8664298783769239,
-      "grad_norm": 0.57059645652771,
-      "learning_rate": 0.0005486111410408361,
-      "loss": 3.8651,
+      "epoch": 0.8679245283018868,
+      "grad_norm": 0.5714002847671509,
+      "learning_rate": 0.000548522396114409,
+      "loss": 3.8866,
       "step": 8050
     },
     {
-      "epoch": 0.8718114304165321,
-      "grad_norm": 0.5633136630058289,
-      "learning_rate": 0.0005482879000107746,
-      "loss": 3.8726,
+      "epoch": 0.8733153638814016,
+      "grad_norm": 0.586208164691925,
+      "learning_rate": 0.0005481985968699406,
+      "loss": 3.872,
       "step": 8100
     },
     {
-      "epoch": 0.8771929824561403,
-      "grad_norm": 0.48724743723869324,
-      "learning_rate": 0.0005479646589807133,
-      "loss": 3.8727,
+      "epoch": 0.8787061994609164,
+      "grad_norm": 0.5501817464828491,
+      "learning_rate": 0.0005478747976254721,
+      "loss": 3.8843,
       "step": 8150
     },
     {
-      "epoch": 0.8825745344957485,
-      "grad_norm": 0.6151993274688721,
-      "learning_rate": 0.0005476414179506518,
-      "loss": 3.8839,
+      "epoch": 0.8840970350404312,
+      "grad_norm": 0.524149477481842,
+      "learning_rate": 0.0005475509983810037,
+      "loss": 3.8754,
       "step": 8200
     },
     {
-      "epoch": 0.8879560865353568,
-      "grad_norm": 0.5960584282875061,
-      "learning_rate": 0.0005473181769205904,
-      "loss": 3.8591,
+      "epoch": 0.889487870619946,
+      "grad_norm": 0.6137134432792664,
+      "learning_rate": 0.0005472271991365352,
+      "loss": 3.8619,
       "step": 8250
     },
     {
-      "epoch": 0.8933376385749651,
-      "grad_norm": 0.5807567834854126,
-      "learning_rate": 0.000546994935890529,
-      "loss": 3.8761,
+      "epoch": 0.894878706199461,
+      "grad_norm": 0.6286455392837524,
+      "learning_rate": 0.0005469033998920669,
+      "loss": 3.8837,
       "step": 8300
     },
     {
-      "epoch": 0.8987191906145733,
-      "grad_norm": 0.5733466744422913,
-      "learning_rate": 0.0005466716948604677,
-      "loss": 3.8595,
+      "epoch": 0.9002695417789758,
+      "grad_norm": 0.62054443359375,
+      "learning_rate": 0.0005465796006475984,
+      "loss": 3.8658,
       "step": 8350
     },
     {
-      "epoch": 0.9041007426541815,
-      "grad_norm": 0.5487543344497681,
-      "learning_rate": 0.0005463484538304062,
-      "loss": 3.8807,
+      "epoch": 0.9056603773584906,
+      "grad_norm": 0.5985867977142334,
+      "learning_rate": 0.00054625580140313,
+      "loss": 3.8821,
       "step": 8400
     },
     {
-      "epoch": 0.9094822946937897,
-      "grad_norm": 0.6290964484214783,
-      "learning_rate": 0.0005460252128003447,
-      "loss": 3.8613,
+      "epoch": 0.9110512129380054,
+      "grad_norm": 0.5693569183349609,
+      "learning_rate": 0.0005459320021586615,
+      "loss": 3.8917,
       "step": 8450
     },
     {
-      "epoch": 0.9148638467333979,
-      "grad_norm": 0.5757058262825012,
-      "learning_rate": 0.0005457019717702833,
-      "loss": 3.8531,
+      "epoch": 0.9164420485175202,
+      "grad_norm": 0.6075363755226135,
+      "learning_rate": 0.0005456146788990825,
+      "loss": 3.8836,
       "step": 8500
     },
     {
-      "epoch": 0.9202453987730062,
-      "grad_norm": 0.5177820324897766,
-      "learning_rate": 0.0005453787307402219,
-      "loss": 3.8668,
+      "epoch": 0.921832884097035,
+      "grad_norm": 0.5467968583106995,
+      "learning_rate": 0.000545290879654614,
+      "loss": 3.8649,
       "step": 8550
     },
     {
-      "epoch": 0.9256269508126144,
-      "grad_norm": 0.5315573811531067,
-      "learning_rate": 0.0005450554897101605,
-      "loss": 3.8399,
+      "epoch": 0.9272237196765498,
+      "grad_norm": 0.5427431464195251,
+      "learning_rate": 0.0005449670804101457,
+      "loss": 3.8583,
       "step": 8600
     },
     {
-      "epoch": 0.9310085028522226,
-      "grad_norm": 0.5855271816253662,
-      "learning_rate": 0.0005447322486800991,
-      "loss": 3.8501,
+      "epoch": 0.9326145552560647,
+      "grad_norm": 0.5329350233078003,
+      "learning_rate": 0.0005446432811656773,
+      "loss": 3.8498,
       "step": 8650
     },
     {
-      "epoch": 0.9363900548918308,
-      "grad_norm": 0.532733678817749,
-      "learning_rate": 0.0005444090076500377,
-      "loss": 3.8375,
+      "epoch": 0.9380053908355795,
+      "grad_norm": 0.5499143004417419,
+      "learning_rate": 0.0005443194819212088,
+      "loss": 3.8575,
       "step": 8700
     },
     {
-      "epoch": 0.941771606931439,
-      "grad_norm": 0.5375049710273743,
-      "learning_rate": 0.0005440857666199763,
-      "loss": 3.8297,
+      "epoch": 0.9433962264150944,
+      "grad_norm": 0.5377353429794312,
+      "learning_rate": 0.0005439956826767404,
+      "loss": 3.854,
       "step": 8750
     },
     {
-      "epoch": 0.9471531589710472,
-      "grad_norm": 0.5517140030860901,
-      "learning_rate": 0.0005437625255899148,
-      "loss": 3.8428,
+      "epoch": 0.9487870619946092,
+      "grad_norm": 0.5699849724769592,
+      "learning_rate": 0.000543671883432272,
+      "loss": 3.8472,
       "step": 8800
     },
     {
-      "epoch": 0.9525347110106555,
-      "grad_norm": 0.596404492855072,
-      "learning_rate": 0.0005434392845598534,
-      "loss": 3.8426,
+      "epoch": 0.954177897574124,
+      "grad_norm": 0.6077035069465637,
+      "learning_rate": 0.0005433480841878035,
+      "loss": 3.838,
       "step": 8850
     },
     {
-      "epoch": 0.9579162630502637,
-      "grad_norm": Infinity,
-      "learning_rate": 0.0005431225083503932,
-      "loss": 3.8442,
+      "epoch": 0.9595687331536388,
+      "grad_norm": 0.5633051991462708,
+      "learning_rate": 0.000543024284943335,
+      "loss": 3.8574,
       "step": 8900
     },
     {
-      "epoch": 0.9632978150898719,
-      "grad_norm": 0.6046130061149597,
-      "learning_rate": 0.0005427992673203318,
-      "loss": 3.856,
+      "epoch": 0.9649595687331537,
+      "grad_norm": 0.6209304928779602,
+      "learning_rate": 0.0005427004856988667,
+      "loss": 3.8596,
       "step": 8950
     },
     {
-      "epoch": 0.9686793671294801,
-      "grad_norm": 0.5675839185714722,
-      "learning_rate": 0.0005424760262902704,
-      "loss": 3.8242,
+      "epoch": 0.9703504043126685,
+      "grad_norm": 0.5542908906936646,
+      "learning_rate": 0.0005423766864543982,
+      "loss": 3.8495,
       "step": 9000
     },
     {
-      "epoch": 0.9686793671294801,
-      "eval_accuracy": 0.3418150312795366,
-      "eval_loss": 3.774980068206787,
-      "eval_runtime": 182.6819,
-      "eval_samples_per_second": 98.592,
-      "eval_steps_per_second": 6.164,
+      "epoch": 0.9703504043126685,
+      "eval_accuracy": 0.34108444926187204,
+      "eval_loss": 3.7841978073120117,
+      "eval_runtime": 183.3669,
+      "eval_samples_per_second": 98.224,
+      "eval_steps_per_second": 6.141,
       "step": 9000
     },
     {
-      "epoch": 0.9740609191690883,
-      "grad_norm": 0.5818553566932678,
-      "learning_rate": 0.000542152785260209,
+      "epoch": 0.9757412398921833,
+      "grad_norm": 0.5503915548324585,
+      "learning_rate": 0.0005420528872099298,
       "loss": 3.8423,
       "step": 9050
     },
     {
-      "epoch": 0.9794424712086965,
-      "grad_norm": 0.5600106716156006,
-      "learning_rate": 0.0005418295442301476,
-      "loss": 3.8477,
+      "epoch": 0.9811320754716981,
+      "grad_norm": 0.5703542232513428,
+      "learning_rate": 0.0005417290879654613,
+      "loss": 3.8572,
       "step": 9100
     },
     {
-      "epoch": 0.9848240232483048,
-      "grad_norm": 0.5427900552749634,
-      "learning_rate": 0.0005415063032000861,
-      "loss": 3.8259,
+      "epoch": 0.9865229110512129,
+      "grad_norm": 0.5498520135879517,
+      "learning_rate": 0.000541405288720993,
+      "loss": 3.8299,
       "step": 9150
     },
     {
-      "epoch": 0.9902055752879131,
-      "grad_norm": 0.6167811155319214,
-      "learning_rate": 0.0005411830621700248,
-      "loss": 3.8275,
+      "epoch": 0.9919137466307277,
+      "grad_norm": 0.5754765868186951,
+      "learning_rate": 0.0005410814894765245,
+      "loss": 3.8266,
       "step": 9200
     },
     {
-      "epoch": 0.9955871273275213,
-      "grad_norm": 0.6433165669441223,
-      "learning_rate": 0.0005408598211399633,
-      "loss": 3.8424,
+      "epoch": 0.9973045822102425,
+      "grad_norm": 0.5759992599487305,
+      "learning_rate": 0.0005407576902320561,
+      "loss": 3.8378,
       "step": 9250
     },
     {
-      "epoch": 1.0009686793671295,
-      "grad_norm": 0.6788349747657776,
-      "learning_rate": 0.0005405365801099019,
-      "loss": 3.827,
+      "epoch": 1.0026954177897573,
+      "grad_norm": 0.5715844035148621,
+      "learning_rate": 0.0005404338909875876,
+      "loss": 3.813,
       "step": 9300
     },
     {
-      "epoch": 1.0063502314067376,
-      "grad_norm": 0.5517417788505554,
-      "learning_rate": 0.0005402133390798405,
-      "loss": 3.7509,
+      "epoch": 1.0080862533692723,
+      "grad_norm": 0.5600173473358154,
+      "learning_rate": 0.0005401100917431192,
+      "loss": 3.7802,
       "step": 9350
     },
     {
-      "epoch": 1.011731783446346,
-      "grad_norm": 0.5560055375099182,
-      "learning_rate": 0.000539890098049779,
-      "loss": 3.752,
+      "epoch": 1.013477088948787,
+      "grad_norm": 0.5711151957511902,
+      "learning_rate": 0.0005397862924986508,
+      "loss": 3.7743,
       "step": 9400
     },
     {
-      "epoch": 1.017113335485954,
-      "grad_norm": 0.5539925694465637,
-      "learning_rate": 0.0005395668570197177,
-      "loss": 3.7523,
+      "epoch": 1.0188679245283019,
+      "grad_norm": 0.5806585550308228,
+      "learning_rate": 0.0005394624932541824,
+      "loss": 3.7737,
       "step": 9450
     },
     {
-      "epoch": 1.0224948875255624,
-      "grad_norm": 0.6510425806045532,
-      "learning_rate": 0.0005392436159896562,
-      "loss": 3.7867,
+      "epoch": 1.0242587601078168,
+      "grad_norm": 0.5773636102676392,
+      "learning_rate": 0.0005391386940097139,
+      "loss": 3.7727,
       "step": 9500
     },
     {
-      "epoch": 1.0278764395651705,
-      "grad_norm": 0.6388372778892517,
-      "learning_rate": 0.0005389203749595948,
-      "loss": 3.7772,
+      "epoch": 1.0296495956873315,
+      "grad_norm": 0.5350525975227356,
+      "learning_rate": 0.0005388148947652455,
+      "loss": 3.7729,
       "step": 9550
     },
     {
-      "epoch": 1.0332579916047788,
-      "grad_norm": 0.6440207362174988,
-      "learning_rate": 0.0005385971339295334,
-      "loss": 3.742,
+      "epoch": 1.0350404312668464,
+      "grad_norm": 0.6753097176551819,
+      "learning_rate": 0.000538491095520777,
+      "loss": 3.7603,
       "step": 9600
     },
     {
-      "epoch": 1.0386395436443872,
-      "grad_norm": 0.5718036890029907,
-      "learning_rate": 0.000538273892899472,
-      "loss": 3.7687,
+      "epoch": 1.0404312668463611,
+      "grad_norm": 0.6020064949989319,
+      "learning_rate": 0.0005381672962763086,
+      "loss": 3.8016,
       "step": 9650
     },
     {
-      "epoch": 1.0440210956839953,
-      "grad_norm": 0.5422919392585754,
-      "learning_rate": 0.0005379506518694106,
-      "loss": 3.759,
+      "epoch": 1.045822102425876,
+      "grad_norm": 0.5449894666671753,
+      "learning_rate": 0.0005378434970318403,
+      "loss": 3.7855,
       "step": 9700
     },
     {
-      "epoch": 1.0494026477236036,
-      "grad_norm": 0.5563080310821533,
-      "learning_rate": 0.0005376274108393491,
-      "loss": 3.7722,
+      "epoch": 1.0512129380053907,
+      "grad_norm": 0.49985501170158386,
+      "learning_rate": 0.0005375196977873718,
+      "loss": 3.784,
       "step": 9750
     },
     {
-      "epoch": 1.0547841997632117,
-      "grad_norm": 0.5674169063568115,
-      "learning_rate": 0.0005373041698092877,
-      "loss": 3.7597,
+      "epoch": 1.0566037735849056,
+      "grad_norm": 0.5777371525764465,
+      "learning_rate": 0.0005371958985429034,
+      "loss": 3.7929,
       "step": 9800
     },
     {
-      "epoch": 1.06016575180282,
-      "grad_norm": 0.5650156140327454,
-      "learning_rate": 0.0005369809287792263,
-      "loss": 3.7682,
+      "epoch": 1.0619946091644206,
+      "grad_norm": 0.5492783188819885,
+      "learning_rate": 0.0005368720992984349,
+      "loss": 3.7887,
       "step": 9850
     },
     {
-      "epoch": 1.0655473038424281,
-      "grad_norm": 0.625595211982727,
-      "learning_rate": 0.000536657687749165,
-      "loss": 3.7546,
+      "epoch": 1.0673854447439353,
+      "grad_norm": 0.5796025991439819,
+      "learning_rate": 0.0005365483000539665,
+      "loss": 3.7788,
       "step": 9900
     },
     {
-      "epoch": 1.0709288558820365,
-      "grad_norm": 0.5709121227264404,
-      "learning_rate": 0.0005363344467191035,
-      "loss": 3.7682,
+      "epoch": 1.0727762803234502,
+      "grad_norm": 0.6108049154281616,
+      "learning_rate": 0.0005362245008094981,
+      "loss": 3.772,
       "step": 9950
     },
     {
-      "epoch": 1.0763104079216446,
-      "grad_norm": 0.5910200476646423,
-      "learning_rate": 0.000536011205689042,
-      "loss": 3.7605,
+      "epoch": 1.0781671159029649,
+      "grad_norm": 0.5750243067741394,
+      "learning_rate": 0.0005359007015650297,
+      "loss": 3.7678,
       "step": 10000
     },
     {
-      "epoch": 1.0763104079216446,
-      "eval_accuracy": 0.34491804910263035,
-      "eval_loss": 3.7458531856536865,
-      "eval_runtime": 182.5556,
-      "eval_samples_per_second": 98.66,
-      "eval_steps_per_second": 6.168,
+      "epoch": 1.0781671159029649,
+      "eval_accuracy": 0.34407403347008747,
+      "eval_loss": 3.7537264823913574,
+      "eval_runtime": 183.1293,
+      "eval_samples_per_second": 98.351,
+      "eval_steps_per_second": 6.149,
       "step": 10000
     }
   ],
   "logging_steps": 50,
-  "max_steps": 92910,
+  "max_steps": 92750,
   "num_input_tokens_seen": 0,
   "num_train_epochs": 10,
   "save_steps": 10000,
@@ -1516,7 +1516,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 8.360953085952e+16,
+  "total_flos": 8.36069179392e+16,
   "train_batch_size": 32,
   "trial_name": null,
   "trial_params": null

checkpoint-10000/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:df0910feec2c6b0f91a8c136538b32a570b9c17dd79f334444184194ebab2b39
+oid sha256:c99fc110a216bd7f9f70a3a44c7200551bf47aad0c84df11cd047859a8caff8b
 size 5304