Training in progress, step 10000, checkpoint

checkpoint-10000/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da488ed6a1f1105a45388247997d2012d57e4fb51be4604a029fc353c16f08d2
+oid sha256:a41d45ba5400e4eb9ad33dc6eafaef29cc28b7e0b9c7d82db44a46da002a05cc
 size 503128704

checkpoint-10000/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c57d97a58b1568e9e013b75cb1ccf11dda6c0eb1f4d47811f67f565eefa5c8b6
+oid sha256:07195c78c2180f4569abc44ae5ed0d70b4449ca77f05e3d56b1ea610b75bdb6a
 size 1006351290

checkpoint-10000/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:87c5a3fa140d7fc2c169e29a28b60404fd6f51c9db380d7acd276a66def200df
+oid sha256:53a47b99e43a46630effbfb45e7223c8a493d5b48b00aebac2470d03eafb7364
 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": 3.7523419857025146,
-  "best_model_checkpoint": "/scratch/cl5625/exceptions/models/100M_1208/checkpoint-10000",
-  "epoch": 1.0763104079216446,
+  "best_metric": 3.749129056930542,
+  "best_model_checkpoint": "/scratch/cl5625/exceptions/models/100M_high_10_1208/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": 1.1599466800689697,
+      "epoch": 0.005390835579514825,
+      "grad_norm": 1.0746381282806396,
       "learning_rate": 0.0003,
-      "loss": 8.5114,
+      "loss": 8.7932,
       "step": 50
     },
     {
-      "epoch": 0.010763104079216447,
-      "grad_norm": 1.6068642139434814,
+      "epoch": 0.01078167115902965,
+      "grad_norm": 2.010450839996338,
       "learning_rate": 0.0006,
-      "loss": 6.963,
+      "loss": 6.983,
       "step": 100
     },
     {
-      "epoch": 0.01614465611882467,
-      "grad_norm": 3.1332640647888184,
-      "learning_rate": 0.0005996767589699385,
-      "loss": 6.4993,
+      "epoch": 0.016172506738544475,
+      "grad_norm": 2.0102932453155518,
+      "learning_rate": 0.0005996762007555315,
+      "loss": 6.5258,
       "step": 150
     },
     {
-      "epoch": 0.021526208158432893,
-      "grad_norm": 1.1792044639587402,
-      "learning_rate": 0.0005993535179398771,
-      "loss": 6.245,
+      "epoch": 0.0215633423180593,
+      "grad_norm": 1.1259429454803467,
+      "learning_rate": 0.000599352401511063,
+      "loss": 6.2429,
       "step": 200
     },
     {
-      "epoch": 0.026907760198041114,
-      "grad_norm": 1.4866029024124146,
-      "learning_rate": 0.0005990302769098158,
-      "loss": 6.0736,
+      "epoch": 0.026954177897574125,
+      "grad_norm": 1.3580305576324463,
+      "learning_rate": 0.0005990286022665946,
+      "loss": 6.0685,
       "step": 250
     },
     {
-      "epoch": 0.03228931223764934,
-      "grad_norm": 2.2453925609588623,
-      "learning_rate": 0.0005987070358797543,
-      "loss": 5.9807,
+      "epoch": 0.03234501347708895,
+      "grad_norm": 2.1491048336029053,
+      "learning_rate": 0.0005987048030221263,
+      "loss": 5.968,
       "step": 300
     },
     {
-      "epoch": 0.03767086427725756,
-      "grad_norm": 2.1176795959472656,
-      "learning_rate": 0.0005983837948496929,
-      "loss": 5.8913,
+      "epoch": 0.03773584905660377,
+      "grad_norm": 1.7612364292144775,
+      "learning_rate": 0.0005983810037776578,
+      "loss": 5.8845,
       "step": 350
     },
     {
-      "epoch": 0.04305241631686579,
-      "grad_norm": 1.5434505939483643,
-      "learning_rate": 0.0005980605538196314,
-      "loss": 5.8009,
+      "epoch": 0.0431266846361186,
+      "grad_norm": 1.019536018371582,
+      "learning_rate": 0.0005980572045331894,
+      "loss": 5.8118,
       "step": 400
     },
     {
-      "epoch": 0.048433968356474004,
-      "grad_norm": 0.9962345361709595,
-      "learning_rate": 0.0005977373127895701,
-      "loss": 5.7629,
+      "epoch": 0.04851752021563342,
+      "grad_norm": 1.70236074924469,
+      "learning_rate": 0.0005977334052887209,
+      "loss": 5.7138,
       "step": 450
     },
     {
-      "epoch": 0.05381552039608223,
-      "grad_norm": 2.0798397064208984,
-      "learning_rate": 0.0005974140717595086,
+      "epoch": 0.05390835579514825,
+      "grad_norm": 1.084849238395691,
+      "learning_rate": 0.0005974096060442526,
       "loss": 5.6589,
       "step": 500
     },
     {
-      "epoch": 0.05919707243569045,
-      "grad_norm": 1.4938079118728638,
-      "learning_rate": 0.0005970908307294472,
-      "loss": 5.58,
+      "epoch": 0.05929919137466307,
+      "grad_norm": 1.3619941473007202,
+      "learning_rate": 0.0005970858067997841,
+      "loss": 5.5683,
       "step": 550
     },
     {
-      "epoch": 0.06457862447529868,
-      "grad_norm": 1.5761438608169556,
-      "learning_rate": 0.0005967675896993858,
-      "loss": 5.5488,
+      "epoch": 0.0646900269541779,
+      "grad_norm": 1.2159621715545654,
+      "learning_rate": 0.0005967620075553157,
+      "loss": 5.5301,
       "step": 600
     },
     {
-      "epoch": 0.0699601765149069,
-      "grad_norm": 1.3473615646362305,
-      "learning_rate": 0.0005964443486693243,
-      "loss": 5.4381,
+      "epoch": 0.07008086253369272,
+      "grad_norm": 1.3903298377990723,
+      "learning_rate": 0.0005964382083108472,
+      "loss": 5.4299,
       "step": 650
     },
     {
-      "epoch": 0.07534172855451512,
-      "grad_norm": 1.1687018871307373,
-      "learning_rate": 0.000596121107639263,
-      "loss": 5.3875,
+      "epoch": 0.07547169811320754,
+      "grad_norm": 1.1805591583251953,
+      "learning_rate": 0.0005961144090663788,
+      "loss": 5.4018,
       "step": 700
     },
     {
-      "epoch": 0.08072328059412334,
-      "grad_norm": 1.4260847568511963,
-      "learning_rate": 0.0005957978666092015,
-      "loss": 5.3472,
+      "epoch": 0.08086253369272237,
+      "grad_norm": 1.206649899482727,
+      "learning_rate": 0.0005957906098219104,
+      "loss": 5.3101,
       "step": 750
     },
     {
-      "epoch": 0.08610483263373157,
-      "grad_norm": 0.8773511052131653,
-      "learning_rate": 0.0005954746255791401,
-      "loss": 5.2741,
+      "epoch": 0.0862533692722372,
+      "grad_norm": 1.4544278383255005,
+      "learning_rate": 0.0005954668105774419,
+      "loss": 5.2617,
       "step": 800
     },
     {
-      "epoch": 0.09148638467333979,
-      "grad_norm": 1.2318358421325684,
-      "learning_rate": 0.0005951513845490787,
-      "loss": 5.2467,
+      "epoch": 0.09164420485175202,
+      "grad_norm": 1.3878358602523804,
+      "learning_rate": 0.0005951430113329735,
+      "loss": 5.2069,
       "step": 850
     },
     {
-      "epoch": 0.09686793671294801,
-      "grad_norm": 1.2172234058380127,
-      "learning_rate": 0.0005948281435190174,
-      "loss": 5.2153,
+      "epoch": 0.09703504043126684,
+      "grad_norm": 1.1169546842575073,
+      "learning_rate": 0.0005948192120885051,
+      "loss": 5.1794,
       "step": 900
     },
     {
-      "epoch": 0.10224948875255624,
-      "grad_norm": 1.190338373184204,
-      "learning_rate": 0.0005945049024889559,
-      "loss": 5.1396,
+      "epoch": 0.10242587601078167,
+      "grad_norm": 0.9606800079345703,
+      "learning_rate": 0.0005944954128440366,
+      "loss": 5.1228,
       "step": 950
     },
     {
-      "epoch": 0.10763104079216446,
-      "grad_norm": 0.8306956887245178,
-      "learning_rate": 0.0005941816614588944,
-      "loss": 5.0999,
+      "epoch": 0.1078167115902965,
+      "grad_norm": 1.0928452014923096,
+      "learning_rate": 0.0005941716135995682,
+      "loss": 5.1074,
       "step": 1000
     },
     {
-      "epoch": 0.10763104079216446,
-      "eval_accuracy": 0.22666633056706703,
-      "eval_loss": 5.026983261108398,
-      "eval_runtime": 149.5498,
-      "eval_samples_per_second": 120.435,
-      "eval_steps_per_second": 7.529,
+      "epoch": 0.1078167115902965,
+      "eval_accuracy": 0.22748220510166597,
+      "eval_loss": 5.022502899169922,
+      "eval_runtime": 185.6777,
+      "eval_samples_per_second": 97.001,
+      "eval_steps_per_second": 6.064,
       "step": 1000
     },
     {
-      "epoch": 0.11301259283177269,
-      "grad_norm": 0.9132220149040222,
-      "learning_rate": 0.000593858420428833,
-      "loss": 5.0755,
+      "epoch": 0.11320754716981132,
+      "grad_norm": 1.1766444444656372,
+      "learning_rate": 0.0005938478143550997,
+      "loss": 5.0718,
       "step": 1050
     },
     {
-      "epoch": 0.1183941448713809,
-      "grad_norm": 0.9254953265190125,
-      "learning_rate": 0.0005935351793987716,
-      "loss": 5.0303,
+      "epoch": 0.11859838274932614,
+      "grad_norm": 1.2186388969421387,
+      "learning_rate": 0.0005935240151106314,
+      "loss": 4.9914,
       "step": 1100
     },
     {
-      "epoch": 0.12377569691098914,
-      "grad_norm": 1.0484514236450195,
-      "learning_rate": 0.0005932119383687103,
-      "loss": 4.9877,
+      "epoch": 0.12398921832884097,
+      "grad_norm": 1.047179937362671,
+      "learning_rate": 0.0005932002158661629,
+      "loss": 4.9909,
       "step": 1150
     },
     {
-      "epoch": 0.12915724895059735,
-      "grad_norm": 1.0418866872787476,
-      "learning_rate": 0.0005928886973386488,
-      "loss": 4.9893,
+      "epoch": 0.1293800539083558,
+      "grad_norm": 1.062658429145813,
+      "learning_rate": 0.0005928764166216945,
+      "loss": 4.9582,
       "step": 1200
     },
     {
-      "epoch": 0.13453880099020557,
-      "grad_norm": 1.0873647928237915,
-      "learning_rate": 0.0005925654563085874,
-      "loss": 4.958,
+      "epoch": 0.1347708894878706,
+      "grad_norm": 1.0171414613723755,
+      "learning_rate": 0.000592552617377226,
+      "loss": 4.9032,
       "step": 1250
     },
     {
-      "epoch": 0.1399203530298138,
-      "grad_norm": 1.3779596090316772,
-      "learning_rate": 0.000592242215278526,
-      "loss": 4.9229,
+      "epoch": 0.14016172506738545,
+      "grad_norm": 1.25285005569458,
+      "learning_rate": 0.0005922288181327577,
+      "loss": 4.9062,
       "step": 1300
     },
     {
-      "epoch": 0.14530190506942203,
-      "grad_norm": 0.8679800629615784,
-      "learning_rate": 0.0005919189742484645,
-      "loss": 4.8952,
+      "epoch": 0.14555256064690028,
+      "grad_norm": 0.8722177743911743,
+      "learning_rate": 0.0005919050188882893,
+      "loss": 4.8755,
       "step": 1350
     },
     {
-      "epoch": 0.15068345710903025,
-      "grad_norm": 1.008557915687561,
-      "learning_rate": 0.0005915957332184032,
-      "loss": 4.842,
+      "epoch": 0.1509433962264151,
+      "grad_norm": 0.9151164293289185,
+      "learning_rate": 0.0005915812196438207,
+      "loss": 4.8355,
       "step": 1400
     },
     {
-      "epoch": 0.15606500914863847,
-      "grad_norm": 0.9011673331260681,
-      "learning_rate": 0.0005912724921883417,
-      "loss": 4.8312,
+      "epoch": 0.15633423180592992,
+      "grad_norm": 0.9627906084060669,
+      "learning_rate": 0.0005912574203993524,
+      "loss": 4.8136,
       "step": 1450
     },
     {
-      "epoch": 0.16144656118824668,
-      "grad_norm": 1.0589803457260132,
-      "learning_rate": 0.0005909492511582803,
-      "loss": 4.8207,
+      "epoch": 0.16172506738544473,
+      "grad_norm": 0.9704756736755371,
+      "learning_rate": 0.0005909336211548839,
+      "loss": 4.8065,
       "step": 1500
     },
     {
-      "epoch": 0.1668281132278549,
-      "grad_norm": 1.0426925420761108,
-      "learning_rate": 0.0005906260101282189,
-      "loss": 4.7929,
+      "epoch": 0.16711590296495957,
+      "grad_norm": 0.8848446607589722,
+      "learning_rate": 0.0005906098219104155,
+      "loss": 4.7687,
       "step": 1550
     },
     {
-      "epoch": 0.17220966526746315,
-      "grad_norm": 1.0936744213104248,
-      "learning_rate": 0.0005903027690981575,
-      "loss": 4.7622,
+      "epoch": 0.1725067385444744,
+      "grad_norm": 1.641764521598816,
+      "learning_rate": 0.000590286022665947,
+      "loss": 4.7814,
       "step": 1600
     },
     {
-      "epoch": 0.17759121730707136,
-      "grad_norm": 0.9796872735023499,
-      "learning_rate": 0.000589979528068096,
-      "loss": 4.746,
+      "epoch": 0.1778975741239892,
+      "grad_norm": 1.0061463117599487,
+      "learning_rate": 0.0005899622234214787,
+      "loss": 4.7322,
       "step": 1650
     },
     {
-      "epoch": 0.18297276934667958,
-      "grad_norm": 1.108158826828003,
-      "learning_rate": 0.0005896562870380347,
-      "loss": 4.7278,
+      "epoch": 0.18328840970350405,
+      "grad_norm": 0.981390118598938,
+      "learning_rate": 0.0005896384241770102,
+      "loss": 4.6799,
       "step": 1700
     },
     {
-      "epoch": 0.1883543213862878,
-      "grad_norm": 0.9828539490699768,
-      "learning_rate": 0.0005893330460079732,
-      "loss": 4.7203,
+      "epoch": 0.18867924528301888,
+      "grad_norm": 0.9400816559791565,
+      "learning_rate": 0.0005893146249325418,
+      "loss": 4.6627,
       "step": 1750
     },
     {
-      "epoch": 0.19373587342589602,
-      "grad_norm": 0.8033989071846008,
-      "learning_rate": 0.0005890098049779118,
-      "loss": 4.6638,
+      "epoch": 0.1940700808625337,
+      "grad_norm": 0.7662665247917175,
+      "learning_rate": 0.0005889908256880733,
+      "loss": 4.6574,
       "step": 1800
     },
     {
-      "epoch": 0.19911742546550426,
-      "grad_norm": 0.9601740837097168,
-      "learning_rate": 0.0005886865639478504,
-      "loss": 4.663,
+      "epoch": 0.19946091644204852,
+      "grad_norm": 0.872972846031189,
+      "learning_rate": 0.0005886670264436049,
+      "loss": 4.6629,
       "step": 1850
     },
     {
-      "epoch": 0.20449897750511248,
-      "grad_norm": 0.8090372085571289,
-      "learning_rate": 0.0005883633229177889,
-      "loss": 4.6409,
+      "epoch": 0.20485175202156333,
+      "grad_norm": 1.0719066858291626,
+      "learning_rate": 0.0005883432271991365,
+      "loss": 4.6107,
       "step": 1900
     },
     {
-      "epoch": 0.2098805295447207,
-      "grad_norm": 0.8016367554664612,
-      "learning_rate": 0.0005880400818877276,
-      "loss": 4.631,
+      "epoch": 0.21024258760107817,
+      "grad_norm": 0.7163854241371155,
+      "learning_rate": 0.0005880194279546681,
+      "loss": 4.5934,
       "step": 1950
     },
     {
-      "epoch": 0.2152620815843289,
-      "grad_norm": 1.1094359159469604,
-      "learning_rate": 0.0005877168408576662,
-      "loss": 4.6114,
+      "epoch": 0.215633423180593,
+      "grad_norm": 0.863464891910553,
+      "learning_rate": 0.0005876956287101996,
+      "loss": 4.5689,
       "step": 2000
     },
     {
-      "epoch": 0.2152620815843289,
-      "eval_accuracy": 0.2693775634611638,
-      "eval_loss": 4.519698143005371,
-      "eval_runtime": 148.8899,
-      "eval_samples_per_second": 120.969,
-      "eval_steps_per_second": 7.563,
+      "epoch": 0.215633423180593,
+      "eval_accuracy": 0.2723961579904257,
+      "eval_loss": 4.493788242340088,
+      "eval_runtime": 183.9361,
+      "eval_samples_per_second": 97.92,
+      "eval_steps_per_second": 6.122,
       "step": 2000
     },
     {
-      "epoch": 0.22064363362393713,
-      "grad_norm": 1.1327083110809326,
-      "learning_rate": 0.0005873935998276048,
-      "loss": 4.5822,
+      "epoch": 0.2210242587601078,
+      "grad_norm": 1.1593316793441772,
+      "learning_rate": 0.0005873718294657312,
+      "loss": 4.559,
       "step": 2050
     },
     {
-      "epoch": 0.22602518566354537,
-      "grad_norm": 0.8756569623947144,
-      "learning_rate": 0.0005870703587975433,
-      "loss": 4.562,
+      "epoch": 0.22641509433962265,
+      "grad_norm": 1.1708295345306396,
+      "learning_rate": 0.0005870480302212628,
+      "loss": 4.5275,
       "step": 2100
     },
     {
-      "epoch": 0.2314067377031536,
-      "grad_norm": 1.4598721265792847,
-      "learning_rate": 0.0005867471177674818,
-      "loss": 4.5499,
+      "epoch": 0.23180592991913745,
+      "grad_norm": 0.8959186673164368,
+      "learning_rate": 0.0005867242309767943,
+      "loss": 4.5286,
       "step": 2150
     },
     {
-      "epoch": 0.2367882897427618,
-      "grad_norm": 0.9978944659233093,
-      "learning_rate": 0.0005864238767374205,
-      "loss": 4.5409,
+      "epoch": 0.2371967654986523,
+      "grad_norm": 1.2286344766616821,
+      "learning_rate": 0.0005864004317323259,
+      "loss": 4.4991,
       "step": 2200
     },
     {
-      "epoch": 0.24216984178237003,
-      "grad_norm": 1.0194833278656006,
-      "learning_rate": 0.0005861006357073591,
-      "loss": 4.5179,
+      "epoch": 0.24258760107816713,
+      "grad_norm": 0.8609565496444702,
+      "learning_rate": 0.0005860766324878575,
+      "loss": 4.4856,
       "step": 2250
     },
     {
-      "epoch": 0.24755139382197827,
-      "grad_norm": 0.8285874724388123,
-      "learning_rate": 0.0005857773946772977,
-      "loss": 4.4895,
+      "epoch": 0.24797843665768193,
+      "grad_norm": 0.8858295679092407,
+      "learning_rate": 0.000585752833243389,
+      "loss": 4.4644,
       "step": 2300
     },
     {
-      "epoch": 0.2529329458615865,
-      "grad_norm": 0.8295068740844727,
-      "learning_rate": 0.0005854541536472362,
-      "loss": 4.4668,
+      "epoch": 0.25336927223719674,
+      "grad_norm": 0.8810425400733948,
+      "learning_rate": 0.0005854290339989206,
+      "loss": 4.4562,
       "step": 2350
     },
     {
-      "epoch": 0.2583144979011947,
-      "grad_norm": 0.8978416323661804,
-      "learning_rate": 0.0005851309126171749,
-      "loss": 4.4467,
+      "epoch": 0.2587601078167116,
+      "grad_norm": 0.9185505509376526,
+      "learning_rate": 0.0005851052347544521,
+      "loss": 4.4452,
       "step": 2400
     },
     {
-      "epoch": 0.2636960499408029,
-      "grad_norm": 0.9676293730735779,
-      "learning_rate": 0.0005848076715871134,
-      "loss": 4.4305,
+      "epoch": 0.2641509433962264,
+      "grad_norm": 0.8982345461845398,
+      "learning_rate": 0.0005847814355099838,
+      "loss": 4.4337,
       "step": 2450
     },
     {
-      "epoch": 0.26907760198041114,
-      "grad_norm": 0.7877038717269897,
-      "learning_rate": 0.000584484430557052,
-      "loss": 4.4486,
+      "epoch": 0.2695417789757412,
+      "grad_norm": 0.7980067729949951,
+      "learning_rate": 0.0005844576362655154,
+      "loss": 4.452,
       "step": 2500
     },
     {
-      "epoch": 0.27445915402001936,
-      "grad_norm": 0.9541506767272949,
-      "learning_rate": 0.0005841611895269906,
-      "loss": 4.4264,
+      "epoch": 0.2749326145552561,
+      "grad_norm": 0.8528494834899902,
+      "learning_rate": 0.0005841338370210469,
+      "loss": 4.3994,
       "step": 2550
     },
     {
-      "epoch": 0.2798407060596276,
-      "grad_norm": 0.8499029874801636,
-      "learning_rate": 0.0005838379484969291,
-      "loss": 4.4156,
+      "epoch": 0.2803234501347709,
+      "grad_norm": 0.887252151966095,
+      "learning_rate": 0.0005838100377765785,
+      "loss": 4.3759,
       "step": 2600
     },
     {
-      "epoch": 0.2852222580992358,
-      "grad_norm": 0.7411248087882996,
-      "learning_rate": 0.0005835147074668678,
-      "loss": 4.4057,
+      "epoch": 0.2857142857142857,
+      "grad_norm": 0.8035648465156555,
+      "learning_rate": 0.0005834862385321101,
+      "loss": 4.3813,
       "step": 2650
     },
     {
-      "epoch": 0.29060381013884407,
-      "grad_norm": 0.8866831660270691,
-      "learning_rate": 0.0005831914664368063,
-      "loss": 4.3843,
+      "epoch": 0.29110512129380056,
+      "grad_norm": 1.654968023300171,
+      "learning_rate": 0.0005831624392876417,
+      "loss": 4.3762,
       "step": 2700
     },
     {
-      "epoch": 0.2959853621784523,
-      "grad_norm": 1.1369751691818237,
-      "learning_rate": 0.0005828682254067449,
-      "loss": 4.3582,
+      "epoch": 0.29649595687331537,
+      "grad_norm": 0.9251230955123901,
+      "learning_rate": 0.0005828386400431731,
+      "loss": 4.3589,
       "step": 2750
     },
     {
-      "epoch": 0.3013669142180605,
-      "grad_norm": 1.0001624822616577,
-      "learning_rate": 0.0005825449843766835,
-      "loss": 4.3353,
+      "epoch": 0.3018867924528302,
+      "grad_norm": 0.7364971041679382,
+      "learning_rate": 0.0005825148407987048,
+      "loss": 4.3577,
       "step": 2800
     },
     {
-      "epoch": 0.3067484662576687,
-      "grad_norm": 0.6730375289916992,
-      "learning_rate": 0.0005822217433466221,
-      "loss": 4.3423,
+      "epoch": 0.30727762803234504,
+      "grad_norm": 0.7727423906326294,
+      "learning_rate": 0.0005821910415542363,
+      "loss": 4.3556,
       "step": 2850
     },
     {
-      "epoch": 0.31213001829727693,
-      "grad_norm": 0.8990495204925537,
-      "learning_rate": 0.0005818985023165607,
-      "loss": 4.352,
+      "epoch": 0.31266846361185985,
+      "grad_norm": 0.8998560905456543,
+      "learning_rate": 0.0005818672423097679,
+      "loss": 4.3319,
       "step": 2900
     },
     {
-      "epoch": 0.31751157033688515,
-      "grad_norm": 0.7752662301063538,
-      "learning_rate": 0.0005815752612864992,
-      "loss": 4.3336,
+      "epoch": 0.31805929919137466,
+      "grad_norm": 0.9321883916854858,
+      "learning_rate": 0.0005815434430652994,
+      "loss": 4.3447,
       "step": 2950
     },
     {
-      "epoch": 0.32289312237649337,
-      "grad_norm": 0.8179628849029541,
-      "learning_rate": 0.0005812520202564378,
-      "loss": 4.315,
+      "epoch": 0.32345013477088946,
+      "grad_norm": 0.7349187731742859,
+      "learning_rate": 0.0005812196438208311,
+      "loss": 4.2912,
       "step": 3000
     },
     {
-      "epoch": 0.32289312237649337,
-      "eval_accuracy": 0.29780094160765636,
-      "eval_loss": 4.2412824630737305,
-      "eval_runtime": 149.0442,
-      "eval_samples_per_second": 120.843,
-      "eval_steps_per_second": 7.555,
+      "epoch": 0.32345013477088946,
+      "eval_accuracy": 0.2989078972287103,
+      "eval_loss": 4.2318949699401855,
+      "eval_runtime": 184.056,
+      "eval_samples_per_second": 97.856,
+      "eval_steps_per_second": 6.118,
       "step": 3000
     },
     {
-      "epoch": 0.3282746744161016,
-      "grad_norm": 0.8209251761436462,
-      "learning_rate": 0.0005809287792263764,
-      "loss": 4.3124,
+      "epoch": 0.3288409703504043,
+      "grad_norm": 0.7943829298019409,
+      "learning_rate": 0.0005808958445763626,
+      "loss": 4.299,
       "step": 3050
     },
     {
-      "epoch": 0.3336562264557098,
-      "grad_norm": 1.3617322444915771,
-      "learning_rate": 0.0005806055381963151,
-      "loss": 4.303,
+      "epoch": 0.33423180592991913,
+      "grad_norm": 0.8438714146614075,
+      "learning_rate": 0.0005805720453318942,
+      "loss": 4.2807,
       "step": 3100
     },
     {
-      "epoch": 0.3390377784953181,
-      "grad_norm": 0.8167195320129395,
-      "learning_rate": 0.0005802822971662536,
-      "loss": 4.3006,
+      "epoch": 0.33962264150943394,
+      "grad_norm": 0.7148887515068054,
+      "learning_rate": 0.0005802482460874257,
+      "loss": 4.2715,
       "step": 3150
     },
     {
-      "epoch": 0.3444193305349263,
-      "grad_norm": 0.8845825791358948,
-      "learning_rate": 0.0005799590561361922,
-      "loss": 4.2941,
+      "epoch": 0.3450134770889488,
+      "grad_norm": 0.6597511172294617,
+      "learning_rate": 0.0005799244468429573,
+      "loss": 4.2929,
       "step": 3200
     },
     {
-      "epoch": 0.3498008825745345,
-      "grad_norm": 0.900641143321991,
-      "learning_rate": 0.0005796358151061307,
-      "loss": 4.2486,
+      "epoch": 0.3504043126684636,
+      "grad_norm": 0.6622719764709473,
+      "learning_rate": 0.0005796006475984889,
+      "loss": 4.2765,
       "step": 3250
     },
     {
-      "epoch": 0.35518243461414273,
-      "grad_norm": 0.6514581441879272,
-      "learning_rate": 0.0005793125740760694,
-      "loss": 4.2632,
+      "epoch": 0.3557951482479784,
+      "grad_norm": 0.9331358075141907,
+      "learning_rate": 0.0005792768483540205,
+      "loss": 4.2567,
       "step": 3300
     },
     {
-      "epoch": 0.36056398665375095,
-      "grad_norm": 0.7339476346969604,
-      "learning_rate": 0.0005789893330460079,
-      "loss": 4.25,
+      "epoch": 0.3611859838274933,
+      "grad_norm": 0.6154801845550537,
+      "learning_rate": 0.000578953049109552,
+      "loss": 4.2658,
       "step": 3350
     },
     {
-      "epoch": 0.36594553869335916,
-      "grad_norm": 0.8381526470184326,
-      "learning_rate": 0.0005786660920159465,
-      "loss": 4.2415,
+      "epoch": 0.3665768194070081,
+      "grad_norm": 0.8815708756446838,
+      "learning_rate": 0.0005786292498650836,
+      "loss": 4.2332,
       "step": 3400
     },
     {
-      "epoch": 0.3713270907329674,
-      "grad_norm": 0.5960896611213684,
-      "learning_rate": 0.0005783428509858851,
-      "loss": 4.251,
+      "epoch": 0.3719676549865229,
+      "grad_norm": 0.7651433944702148,
+      "learning_rate": 0.0005783054506206152,
+      "loss": 4.2457,
       "step": 3450
     },
     {
-      "epoch": 0.3767086427725756,
-      "grad_norm": 0.7298816442489624,
-      "learning_rate": 0.0005780196099558237,
-      "loss": 4.2472,
+      "epoch": 0.37735849056603776,
+      "grad_norm": 0.7533497214317322,
+      "learning_rate": 0.0005779816513761467,
+      "loss": 4.225,
       "step": 3500
     },
     {
-      "epoch": 0.3820901948121838,
-      "grad_norm": 0.7578866481781006,
-      "learning_rate": 0.0005776963689257623,
-      "loss": 4.2265,
+      "epoch": 0.38274932614555257,
+      "grad_norm": 0.7970021367073059,
+      "learning_rate": 0.0005776578521316782,
+      "loss": 4.2038,
       "step": 3550
     },
     {
-      "epoch": 0.38747174685179203,
-      "grad_norm": 0.665746808052063,
-      "learning_rate": 0.0005773731278957008,
-      "loss": 4.2354,
+      "epoch": 0.3881401617250674,
+      "grad_norm": 0.7401692867279053,
+      "learning_rate": 0.0005773340528872099,
+      "loss": 4.2233,
       "step": 3600
     },
     {
-      "epoch": 0.3928532988914003,
-      "grad_norm": 0.6855995655059814,
-      "learning_rate": 0.0005770498868656394,
-      "loss": 4.2033,
+      "epoch": 0.3935309973045822,
+      "grad_norm": 0.6197974681854248,
+      "learning_rate": 0.0005770102536427414,
+      "loss": 4.218,
       "step": 3650
     },
     {
-      "epoch": 0.3982348509310085,
-      "grad_norm": 0.7479808926582336,
-      "learning_rate": 0.000576726645835578,
-      "loss": 4.2091,
+      "epoch": 0.39892183288409705,
+      "grad_norm": 0.5855727791786194,
+      "learning_rate": 0.000576686454398273,
+      "loss": 4.1972,
       "step": 3700
     },
     {
-      "epoch": 0.40361640297061674,
-      "grad_norm": 0.7702375054359436,
-      "learning_rate": 0.0005764034048055167,
-      "loss": 4.2211,
+      "epoch": 0.40431266846361186,
+      "grad_norm": 0.6316722631454468,
+      "learning_rate": 0.0005763626551538045,
+      "loss": 4.1756,
       "step": 3750
     },
     {
-      "epoch": 0.40899795501022496,
-      "grad_norm": 0.8518558144569397,
-      "learning_rate": 0.0005760801637754552,
-      "loss": 4.1696,
+      "epoch": 0.40970350404312667,
+      "grad_norm": 0.8590136170387268,
+      "learning_rate": 0.0005760388559093362,
+      "loss": 4.186,
       "step": 3800
     },
     {
-      "epoch": 0.4143795070498332,
-      "grad_norm": 0.7380365133285522,
-      "learning_rate": 0.0005757569227453937,
-      "loss": 4.1979,
+      "epoch": 0.41509433962264153,
+      "grad_norm": 0.7628465890884399,
+      "learning_rate": 0.0005757150566648678,
+      "loss": 4.1647,
       "step": 3850
     },
     {
-      "epoch": 0.4197610590894414,
-      "grad_norm": 0.6969083547592163,
-      "learning_rate": 0.0005754336817153324,
-      "loss": 4.1787,
+      "epoch": 0.42048517520215634,
+      "grad_norm": 0.7800719141960144,
+      "learning_rate": 0.0005753912574203993,
+      "loss": 4.1744,
       "step": 3900
     },
     {
-      "epoch": 0.4251426111290496,
-      "grad_norm": 1.0820659399032593,
-      "learning_rate": 0.0005751104406852709,
-      "loss": 4.1744,
+      "epoch": 0.42587601078167114,
+      "grad_norm": 0.7922886610031128,
+      "learning_rate": 0.0005750674581759309,
+      "loss": 4.1795,
       "step": 3950
     },
     {
-      "epoch": 0.4305241631686578,
-      "grad_norm": 0.6784261465072632,
-      "learning_rate": 0.0005747871996552096,
-      "loss": 4.1675,
+      "epoch": 0.431266846361186,
+      "grad_norm": 0.60633385181427,
+      "learning_rate": 0.0005747436589314624,
+      "loss": 4.1594,
       "step": 4000
     },
     {
-      "epoch": 0.4305241631686578,
-      "eval_accuracy": 0.3123320701710772,
-      "eval_loss": 4.096024036407471,
-      "eval_runtime": 149.6002,
-      "eval_samples_per_second": 120.394,
-      "eval_steps_per_second": 7.527,
+      "epoch": 0.431266846361186,
+      "eval_accuracy": 0.31194211495671104,
+      "eval_loss": 4.092201232910156,
+      "eval_runtime": 184.0379,
+      "eval_samples_per_second": 97.866,
+      "eval_steps_per_second": 6.118,
       "step": 4000
     },
     {
-      "epoch": 0.43590571520826604,
-      "grad_norm": 0.7904225587844849,
-      "learning_rate": 0.0005744639586251481,
-      "loss": 4.1546,
+      "epoch": 0.4366576819407008,
+      "grad_norm": 0.8343899846076965,
+      "learning_rate": 0.0005744198596869941,
+      "loss": 4.1458,
       "step": 4050
     },
     {
-      "epoch": 0.44128726724787426,
-      "grad_norm": 0.7725077867507935,
-      "learning_rate": 0.0005741407175950867,
-      "loss": 4.1414,
+      "epoch": 0.4420485175202156,
+      "grad_norm": 0.6843920946121216,
+      "learning_rate": 0.0005740960604425255,
+      "loss": 4.1337,
       "step": 4100
     },
     {
-      "epoch": 0.44666881928748253,
-      "grad_norm": 0.6624230742454529,
-      "learning_rate": 0.0005738174765650253,
-      "loss": 4.1319,
+      "epoch": 0.4474393530997305,
+      "grad_norm": 0.5905738472938538,
+      "learning_rate": 0.0005737722611980572,
+      "loss": 4.1462,
       "step": 4150
     },
     {
-      "epoch": 0.45205037132709075,
-      "grad_norm": 0.7300996780395508,
-      "learning_rate": 0.0005734942355349638,
-      "loss": 4.138,
+      "epoch": 0.4528301886792453,
+      "grad_norm": 0.7496142387390137,
+      "learning_rate": 0.0005734484619535887,
+      "loss": 4.1294,
       "step": 4200
     },
     {
-      "epoch": 0.45743192336669897,
-      "grad_norm": 0.7004075050354004,
-      "learning_rate": 0.0005731709945049025,
-      "loss": 4.1344,
+      "epoch": 0.4582210242587601,
+      "grad_norm": 0.6900115013122559,
+      "learning_rate": 0.0005731246627091203,
+      "loss": 4.1303,
       "step": 4250
     },
     {
-      "epoch": 0.4628134754063072,
-      "grad_norm": 0.7337325811386108,
-      "learning_rate": 0.000572847753474841,
-      "loss": 4.1462,
+      "epoch": 0.4636118598382749,
+      "grad_norm": 0.6268366575241089,
+      "learning_rate": 0.0005728008634646518,
+      "loss": 4.1138,
       "step": 4300
     },
     {
-      "epoch": 0.4681950274459154,
-      "grad_norm": 0.6394081115722656,
-      "learning_rate": 0.0005725245124447796,
-      "loss": 4.1319,
+      "epoch": 0.46900269541778977,
+      "grad_norm": 0.691112756729126,
+      "learning_rate": 0.0005724770642201835,
+      "loss": 4.1119,
       "step": 4350
     },
     {
-      "epoch": 0.4735765794855236,
-      "grad_norm": 0.5687220692634583,
-      "learning_rate": 0.0005722012714147182,
-      "loss": 4.129,
+      "epoch": 0.4743935309973046,
+      "grad_norm": 0.6356542706489563,
+      "learning_rate": 0.000572153264975715,
+      "loss": 4.1062,
       "step": 4400
     },
     {
-      "epoch": 0.47895813152513184,
-      "grad_norm": 0.6553306579589844,
-      "learning_rate": 0.0005718780303846568,
-      "loss": 4.1125,
+      "epoch": 0.4797843665768194,
+      "grad_norm": 0.753619372844696,
+      "learning_rate": 0.0005718294657312466,
+      "loss": 4.134,
       "step": 4450
     },
     {
-      "epoch": 0.48433968356474005,
-      "grad_norm": 0.7389906644821167,
-      "learning_rate": 0.0005715547893545953,
-      "loss": 4.1056,
+      "epoch": 0.48517520215633425,
+      "grad_norm": 0.7323446273803711,
+      "learning_rate": 0.0005715056664867781,
+      "loss": 4.097,
       "step": 4500
     },
     {
-      "epoch": 0.48972123560434827,
-      "grad_norm": 0.7992845177650452,
-      "learning_rate": 0.000571231548324534,
-      "loss": 4.1069,
+      "epoch": 0.49056603773584906,
+      "grad_norm": 0.656112790107727,
+      "learning_rate": 0.0005711818672423097,
+      "loss": 4.1037,
       "step": 4550
     },
     {
-      "epoch": 0.49510278764395654,
-      "grad_norm": 0.7084970474243164,
-      "learning_rate": 0.0005709083072944725,
-      "loss": 4.1046,
+      "epoch": 0.49595687331536387,
+      "grad_norm": 0.792602002620697,
+      "learning_rate": 0.0005708580679978413,
+      "loss": 4.0985,
       "step": 4600
     },
     {
-      "epoch": 0.5004843396835648,
-      "grad_norm": 0.7775879502296448,
-      "learning_rate": 0.0005705850662644111,
-      "loss": 4.1012,
+      "epoch": 0.5013477088948787,
+      "grad_norm": 0.6225011348724365,
+      "learning_rate": 0.0005705342687533729,
+      "loss": 4.0917,
       "step": 4650
     },
     {
-      "epoch": 0.505865891723173,
-      "grad_norm": 0.6137743592262268,
-      "learning_rate": 0.0005702618252343497,
-      "loss": 4.0934,
+      "epoch": 0.5067385444743935,
+      "grad_norm": 0.7103033065795898,
+      "learning_rate": 0.0005702104695089044,
+      "loss": 4.097,
       "step": 4700
     },
     {
-      "epoch": 0.5112474437627812,
-      "grad_norm": 0.7164639234542847,
-      "learning_rate": 0.0005699385842042882,
-      "loss": 4.1068,
+      "epoch": 0.5121293800539084,
+      "grad_norm": 0.7025964856147766,
+      "learning_rate": 0.000569886670264436,
+      "loss": 4.0961,
       "step": 4750
     },
     {
-      "epoch": 0.5166289958023894,
-      "grad_norm": 0.7044581770896912,
-      "learning_rate": 0.0005696153431742269,
-      "loss": 4.078,
+      "epoch": 0.5175202156334232,
+      "grad_norm": 0.6616626381874084,
+      "learning_rate": 0.0005695628710199675,
+      "loss": 4.0753,
       "step": 4800
     },
     {
-      "epoch": 0.5220105478419976,
-      "grad_norm": 0.635504961013794,
-      "learning_rate": 0.0005692921021441655,
-      "loss": 4.0693,
+      "epoch": 0.522911051212938,
+      "grad_norm": 0.6561509966850281,
+      "learning_rate": 0.0005692390717754991,
+      "loss": 4.0997,
       "step": 4850
     },
     {
-      "epoch": 0.5273920998816058,
-      "grad_norm": 0.607115626335144,
-      "learning_rate": 0.0005689688611141041,
-      "loss": 4.0796,
+      "epoch": 0.5283018867924528,
+      "grad_norm": 0.5918413400650024,
+      "learning_rate": 0.0005689152725310306,
+      "loss": 4.071,
       "step": 4900
     },
     {
-      "epoch": 0.5327736519212141,
-      "grad_norm": 0.6604019403457642,
-      "learning_rate": 0.0005686456200840426,
-      "loss": 4.0634,
+      "epoch": 0.5336927223719676,
+      "grad_norm": 0.7274987101554871,
+      "learning_rate": 0.0005685914732865623,
+      "loss": 4.0605,
       "step": 4950
     },
     {
-      "epoch": 0.5381552039608223,
-      "grad_norm": 0.7243701219558716,
-      "learning_rate": 0.0005683223790539811,
-      "loss": 4.0785,
+      "epoch": 0.5390835579514824,
+      "grad_norm": 0.6024713516235352,
+      "learning_rate": 0.0005682676740420939,
+      "loss": 4.0616,
       "step": 5000
     },
     {
-      "epoch": 0.5381552039608223,
-      "eval_accuracy": 0.3211391472986556,
-      "eval_loss": 3.996016263961792,
-      "eval_runtime": 148.8985,
-      "eval_samples_per_second": 120.962,
-      "eval_steps_per_second": 7.562,
+      "epoch": 0.5390835579514824,
+      "eval_accuracy": 0.3220235818054655,
+      "eval_loss": 3.988025665283203,
+      "eval_runtime": 183.7524,
+      "eval_samples_per_second": 98.018,
+      "eval_steps_per_second": 6.128,
       "step": 5000
     },
     {
-      "epoch": 0.5435367560004305,
-      "grad_norm": 0.6107810139656067,
-      "learning_rate": 0.0005679991380239198,
-      "loss": 4.0398,
+      "epoch": 0.5444743935309974,
+      "grad_norm": 0.614084780216217,
+      "learning_rate": 0.0005679438747976254,
+      "loss": 4.0595,
       "step": 5050
     },
     {
-      "epoch": 0.5489183080400387,
-      "grad_norm": 0.8288403749465942,
-      "learning_rate": 0.0005676758969938584,
-      "loss": 4.0597,
+      "epoch": 0.5498652291105122,
+      "grad_norm": 0.6400277018547058,
+      "learning_rate": 0.000567620075553157,
+      "loss": 4.0471,
       "step": 5100
     },
     {
-      "epoch": 0.5542998600796469,
-      "grad_norm": 0.6572490334510803,
-      "learning_rate": 0.000567352655963797,
-      "loss": 4.0519,
+      "epoch": 0.555256064690027,
+      "grad_norm": 0.6315747499465942,
+      "learning_rate": 0.0005672962763086886,
+      "loss": 4.0597,
       "step": 5150
     },
     {
-      "epoch": 0.5596814121192552,
-      "grad_norm": 0.6547346711158752,
-      "learning_rate": 0.0005670294149337355,
-      "loss": 4.029,
+      "epoch": 0.5606469002695418,
+      "grad_norm": 0.6158743500709534,
+      "learning_rate": 0.0005669724770642202,
+      "loss": 4.0514,
       "step": 5200
     },
     {
-      "epoch": 0.5650629641588634,
-      "grad_norm": 0.6182342171669006,
-      "learning_rate": 0.0005667061739036742,
-      "loss": 4.0407,
+      "epoch": 0.5660377358490566,
+      "grad_norm": 0.7305905222892761,
+      "learning_rate": 0.0005666486778197517,
+      "loss": 4.0467,
       "step": 5250
     },
     {
-      "epoch": 0.5704445161984716,
-      "grad_norm": 0.6051702499389648,
-      "learning_rate": 0.0005663829328736127,
-      "loss": 4.045,
+      "epoch": 0.5714285714285714,
+      "grad_norm": 0.6409223675727844,
+      "learning_rate": 0.0005663248785752833,
+      "loss": 4.03,
       "step": 5300
     },
     {
-      "epoch": 0.5758260682380799,
-      "grad_norm": 0.6755617260932922,
-      "learning_rate": 0.0005660596918435512,
-      "loss": 4.0245,
+      "epoch": 0.5768194070080862,
+      "grad_norm": 0.6823210120201111,
+      "learning_rate": 0.0005660010793308148,
+      "loss": 4.0071,
       "step": 5350
     },
     {
-      "epoch": 0.5812076202776881,
-      "grad_norm": 0.6777322888374329,
-      "learning_rate": 0.0005657364508134899,
-      "loss": 4.045,
+      "epoch": 0.5822102425876011,
+      "grad_norm": 0.634975790977478,
+      "learning_rate": 0.0005656772800863465,
+      "loss": 4.0469,
       "step": 5400
     },
     {
-      "epoch": 0.5865891723172963,
-      "grad_norm": 0.6305786371231079,
-      "learning_rate": 0.0005654132097834284,
-      "loss": 4.0334,
+      "epoch": 0.5876010781671159,
+      "grad_norm": 0.6946772336959839,
+      "learning_rate": 0.0005653534808418779,
+      "loss": 4.0274,
       "step": 5450
     },
     {
-      "epoch": 0.5919707243569046,
-      "grad_norm": 0.6374103426933289,
-      "learning_rate": 0.0005650899687533671,
-      "loss": 4.0208,
+      "epoch": 0.5929919137466307,
+      "grad_norm": 0.6479852795600891,
+      "learning_rate": 0.0005650296815974096,
+      "loss": 4.0358,
       "step": 5500
     },
     {
-      "epoch": 0.5973522763965128,
-      "grad_norm": 0.6658488512039185,
-      "learning_rate": 0.0005647667277233056,
-      "loss": 4.0439,
+      "epoch": 0.5983827493261455,
+      "grad_norm": 0.6506155133247375,
+      "learning_rate": 0.0005647058823529411,
+      "loss": 4.0165,
       "step": 5550
     },
     {
-      "epoch": 0.602733828436121,
-      "grad_norm": 0.8127887845039368,
-      "learning_rate": 0.0005644434866932442,
-      "loss": 4.0256,
+      "epoch": 0.6037735849056604,
+      "grad_norm": 0.626352846622467,
+      "learning_rate": 0.0005643820831084727,
+      "loss": 4.0194,
       "step": 5600
     },
     {
-      "epoch": 0.6081153804757292,
-      "grad_norm": 1.2196614742279053,
-      "learning_rate": 0.0005641202456631828,
-      "loss": 4.0248,
+      "epoch": 0.6091644204851752,
+      "grad_norm": 0.6344131827354431,
+      "learning_rate": 0.0005640582838640042,
+      "loss": 3.9937,
       "step": 5650
     },
     {
-      "epoch": 0.6134969325153374,
-      "grad_norm": 0.7336172461509705,
-      "learning_rate": 0.0005637970046331214,
-      "loss": 4.0063,
+      "epoch": 0.6145552560646901,
+      "grad_norm": 0.6841148734092712,
+      "learning_rate": 0.0005637344846195358,
+      "loss": 4.0074,
       "step": 5700
     },
     {
-      "epoch": 0.6188784845549457,
-      "grad_norm": 0.7298561334609985,
-      "learning_rate": 0.00056347376360306,
-      "loss": 4.0099,
+      "epoch": 0.6199460916442049,
+      "grad_norm": 0.6779347062110901,
+      "learning_rate": 0.0005634106853750674,
+      "loss": 4.0008,
       "step": 5750
     },
     {
-      "epoch": 0.6242600365945539,
-      "grad_norm": 0.7107455730438232,
-      "learning_rate": 0.0005631505225729985,
-      "loss": 3.985,
+      "epoch": 0.6253369272237197,
+      "grad_norm": 0.6471227407455444,
+      "learning_rate": 0.000563086886130599,
+      "loss": 4.0106,
       "step": 5800
     },
     {
-      "epoch": 0.6296415886341621,
-      "grad_norm": 0.6696971654891968,
-      "learning_rate": 0.0005628272815429371,
-      "loss": 4.0056,
+      "epoch": 0.6307277628032345,
+      "grad_norm": 0.6579660773277283,
+      "learning_rate": 0.0005627630868861305,
+      "loss": 3.9984,
       "step": 5850
     },
     {
-      "epoch": 0.6350231406737703,
-      "grad_norm": 0.5660277605056763,
-      "learning_rate": 0.0005625040405128757,
-      "loss": 3.9932,
+      "epoch": 0.6361185983827493,
+      "grad_norm": 0.7190978527069092,
+      "learning_rate": 0.0005624392876416621,
+      "loss": 3.993,
       "step": 5900
     },
     {
-      "epoch": 0.6404046927133785,
-      "grad_norm": 0.6211795210838318,
-      "learning_rate": 0.0005621807994828143,
-      "loss": 3.9809,
+      "epoch": 0.6415094339622641,
+      "grad_norm": 0.6463994979858398,
+      "learning_rate": 0.0005621154883971937,
+      "loss": 3.9932,
       "step": 5950
     },
     {
-      "epoch": 0.6457862447529867,
-      "grad_norm": 0.5881251096725464,
-      "learning_rate": 0.0005618575584527529,
-      "loss": 3.9737,
+      "epoch": 0.6469002695417789,
+      "grad_norm": 0.6113254427909851,
+      "learning_rate": 0.0005617916891527253,
+      "loss": 3.992,
       "step": 6000
     },
     {
-      "epoch": 0.6457862447529867,
-      "eval_accuracy": 0.3286207678477851,
-      "eval_loss": 3.9236061573028564,
-      "eval_runtime": 148.8618,
-      "eval_samples_per_second": 120.991,
-      "eval_steps_per_second": 7.564,
+      "epoch": 0.6469002695417789,
+      "eval_accuracy": 0.3278068490651668,
+      "eval_loss": 3.9221835136413574,
+      "eval_runtime": 183.734,
+      "eval_samples_per_second": 98.028,
+      "eval_steps_per_second": 6.128,
       "step": 6000
     },
     {
-      "epoch": 0.651167796792595,
-      "grad_norm": 0.5960618853569031,
-      "learning_rate": 0.0005615343174226915,
-      "loss": 4.001,
+      "epoch": 0.6522911051212938,
+      "grad_norm": 0.5573587417602539,
+      "learning_rate": 0.0005614678899082568,
+      "loss": 3.9855,
       "step": 6050
     },
     {
-      "epoch": 0.6565493488322032,
-      "grad_norm": 0.7855511903762817,
-      "learning_rate": 0.00056121107639263,
-      "loss": 3.9755,
+      "epoch": 0.6576819407008087,
+      "grad_norm": 0.618307888507843,
+      "learning_rate": 0.0005611440906637884,
+      "loss": 4.0032,
       "step": 6100
     },
     {
-      "epoch": 0.6619309008718114,
-      "grad_norm": 0.7068238854408264,
-      "learning_rate": 0.0005608878353625687,
-      "loss": 3.9568,
+      "epoch": 0.6630727762803235,
+      "grad_norm": 0.6326853036880493,
+      "learning_rate": 0.00056082029141932,
+      "loss": 3.986,
       "step": 6150
     },
     {
-      "epoch": 0.6673124529114196,
-      "grad_norm": 0.5632562041282654,
-      "learning_rate": 0.0005605645943325072,
-      "loss": 4.0027,
+      "epoch": 0.6684636118598383,
+      "grad_norm": 0.6867143511772156,
+      "learning_rate": 0.0005604964921748515,
+      "loss": 3.9849,
       "step": 6200
     },
     {
-      "epoch": 0.6726940049510278,
-      "grad_norm": 0.668786883354187,
-      "learning_rate": 0.0005602413533024458,
-      "loss": 3.9658,
+      "epoch": 0.6738544474393531,
+      "grad_norm": 0.7261978983879089,
+      "learning_rate": 0.000560172692930383,
+      "loss": 3.9775,
       "step": 6250
     },
     {
-      "epoch": 0.6780755569906362,
-      "grad_norm": 0.5909920334815979,
-      "learning_rate": 0.0005599181122723844,
-      "loss": 3.9732,
+      "epoch": 0.6792452830188679,
+      "grad_norm": 0.6933634877204895,
+      "learning_rate": 0.0005598488936859147,
+      "loss": 3.9771,
       "step": 6300
     },
     {
-      "epoch": 0.6834571090302444,
-      "grad_norm": 0.6160820722579956,
-      "learning_rate": 0.000559594871242323,
-      "loss": 3.9724,
+      "epoch": 0.6846361185983828,
+      "grad_norm": 0.6094374060630798,
+      "learning_rate": 0.0005595315704263357,
+      "loss": 3.9608,
       "step": 6350
     },
     {
-      "epoch": 0.6888386610698526,
-      "grad_norm": 0.6244766116142273,
-      "learning_rate": 0.0005592716302122616,
-      "loss": 3.9548,
+      "epoch": 0.6900269541778976,
+      "grad_norm": 0.5936299562454224,
+      "learning_rate": 0.0005592077711818672,
+      "loss": 3.9767,
       "step": 6400
     },
     {
-      "epoch": 0.6942202131094608,
-      "grad_norm": 0.7597425580024719,
-      "learning_rate": 0.0005589548540028014,
-      "loss": 3.9739,
+      "epoch": 0.6954177897574124,
+      "grad_norm": 0.7098467350006104,
+      "learning_rate": 0.0005588839719373988,
+      "loss": 3.9525,
       "step": 6450
     },
     {
-      "epoch": 0.699601765149069,
-      "grad_norm": 0.6434476375579834,
-      "learning_rate": 0.0005586316129727399,
-      "loss": 3.9689,
+      "epoch": 0.7008086253369272,
+      "grad_norm": 0.5945666432380676,
+      "learning_rate": 0.0005585601726929303,
+      "loss": 3.959,
       "step": 6500
     },
     {
-      "epoch": 0.7049833171886772,
-      "grad_norm": 0.6384700536727905,
-      "learning_rate": 0.0005583083719426786,
-      "loss": 3.9564,
+      "epoch": 0.706199460916442,
+      "grad_norm": 0.5693997144699097,
+      "learning_rate": 0.0005582363734484619,
+      "loss": 3.9383,
       "step": 6550
     },
     {
-      "epoch": 0.7103648692282855,
-      "grad_norm": 0.7494992017745972,
-      "learning_rate": 0.0005579851309126171,
-      "loss": 3.9539,
+      "epoch": 0.7115902964959568,
+      "grad_norm": 0.5391964316368103,
+      "learning_rate": 0.0005579125742039935,
+      "loss": 3.9467,
       "step": 6600
     },
     {
-      "epoch": 0.7157464212678937,
-      "grad_norm": 0.5634821653366089,
-      "learning_rate": 0.0005576618898825558,
-      "loss": 3.955,
+      "epoch": 0.7169811320754716,
+      "grad_norm": 0.6286468505859375,
+      "learning_rate": 0.0005575887749595251,
+      "loss": 3.9512,
       "step": 6650
     },
     {
-      "epoch": 0.7211279733075019,
-      "grad_norm": 0.5900028944015503,
-      "learning_rate": 0.0005573386488524943,
-      "loss": 3.9455,
+      "epoch": 0.7223719676549866,
+      "grad_norm": 0.6168409585952759,
+      "learning_rate": 0.0005572649757150566,
+      "loss": 3.9541,
       "step": 6700
     },
     {
-      "epoch": 0.7265095253471101,
-      "grad_norm": 0.6802782416343689,
-      "learning_rate": 0.0005570154078224328,
-      "loss": 3.9441,
+      "epoch": 0.7277628032345014,
+      "grad_norm": 0.6346463561058044,
+      "learning_rate": 0.0005569411764705882,
+      "loss": 3.9508,
       "step": 6750
     },
     {
-      "epoch": 0.7318910773867183,
-      "grad_norm": 0.5620344281196594,
-      "learning_rate": 0.0005566921667923715,
-      "loss": 3.9357,
+      "epoch": 0.7331536388140162,
+      "grad_norm": 0.5671019554138184,
+      "learning_rate": 0.0005566173772261198,
+      "loss": 3.9188,
       "step": 6800
     },
     {
-      "epoch": 0.7372726294263265,
-      "grad_norm": 0.5253839492797852,
-      "learning_rate": 0.00055636892576231,
-      "loss": 3.9395,
+      "epoch": 0.738544474393531,
+      "grad_norm": 0.6257506608963013,
+      "learning_rate": 0.0005562935779816513,
+      "loss": 3.9325,
       "step": 6850
     },
     {
-      "epoch": 0.7426541814659348,
-      "grad_norm": 0.5747552514076233,
-      "learning_rate": 0.0005560456847322487,
-      "loss": 3.9271,
+      "epoch": 0.7439353099730458,
+      "grad_norm": 0.7027499675750732,
+      "learning_rate": 0.0005559697787371828,
+      "loss": 3.9496,
       "step": 6900
     },
     {
-      "epoch": 0.748035733505543,
-      "grad_norm": 0.6038549542427063,
-      "learning_rate": 0.0005557224437021872,
-      "loss": 3.9523,
+      "epoch": 0.7493261455525606,
+      "grad_norm": 0.6026901006698608,
+      "learning_rate": 0.0005556459794927145,
+      "loss": 3.9281,
       "step": 6950
     },
     {
-      "epoch": 0.7534172855451512,
-      "grad_norm": 0.5900692343711853,
-      "learning_rate": 0.0005553992026721258,
-      "loss": 3.9263,
+      "epoch": 0.7547169811320755,
+      "grad_norm": 0.6102020740509033,
+      "learning_rate": 0.000555322180248246,
+      "loss": 3.9416,
       "step": 7000
     },
     {
-      "epoch": 0.7534172855451512,
-      "eval_accuracy": 0.333655851321996,
-      "eval_loss": 3.8643293380737305,
-      "eval_runtime": 149.0842,
-      "eval_samples_per_second": 120.811,
-      "eval_steps_per_second": 7.553,
+      "epoch": 0.7547169811320755,
+      "eval_accuracy": 0.33369692211358987,
+      "eval_loss": 3.8627331256866455,
+      "eval_runtime": 183.6876,
+      "eval_samples_per_second": 98.052,
+      "eval_steps_per_second": 6.13,
       "step": 7000
     },
     {
-      "epoch": 0.7587988375847594,
-      "grad_norm": 0.5576010346412659,
-      "learning_rate": 0.0005550759616420644,
-      "loss": 3.9293,
+      "epoch": 0.7601078167115903,
+      "grad_norm": 0.7307771444320679,
+      "learning_rate": 0.0005549983810037776,
+      "loss": 3.9079,
       "step": 7050
     },
     {
-      "epoch": 0.7641803896243676,
-      "grad_norm": 0.677987277507782,
-      "learning_rate": 0.000554752720612003,
-      "loss": 3.9408,
+      "epoch": 0.7654986522911051,
+      "grad_norm": 0.6346297860145569,
+      "learning_rate": 0.0005546745817593091,
+      "loss": 3.9094,
       "step": 7100
     },
     {
-      "epoch": 0.7695619416639758,
-      "grad_norm": 0.5040335059165955,
-      "learning_rate": 0.0005544294795819415,
-      "loss": 3.9226,
+      "epoch": 0.77088948787062,
+      "grad_norm": 0.6780383586883545,
+      "learning_rate": 0.0005543507825148408,
+      "loss": 3.914,
       "step": 7150
     },
     {
-      "epoch": 0.7749434937035841,
-      "grad_norm": 0.6092420816421509,
-      "learning_rate": 0.0005541062385518801,
-      "loss": 3.9272,
+      "epoch": 0.7762803234501348,
+      "grad_norm": 0.6258558630943298,
+      "learning_rate": 0.0005540269832703723,
+      "loss": 3.8931,
       "step": 7200
     },
     {
-      "epoch": 0.7803250457431924,
-      "grad_norm": 0.6148164868354797,
-      "learning_rate": 0.0005537829975218188,
-      "loss": 3.9033,
+      "epoch": 0.7816711590296496,
+      "grad_norm": 0.5334175825119019,
+      "learning_rate": 0.0005537031840259039,
+      "loss": 3.9205,
       "step": 7250
     },
     {
-      "epoch": 0.7857065977828006,
-      "grad_norm": 0.5672699213027954,
-      "learning_rate": 0.0005534597564917573,
-      "loss": 3.9245,
+      "epoch": 0.7870619946091644,
+      "grad_norm": 0.5545728802680969,
+      "learning_rate": 0.0005533793847814354,
+      "loss": 3.8958,
       "step": 7300
     },
     {
-      "epoch": 0.7910881498224088,
-      "grad_norm": 0.5728033781051636,
-      "learning_rate": 0.0005531365154616959,
-      "loss": 3.9277,
+      "epoch": 0.7924528301886793,
+      "grad_norm": 0.5616306662559509,
+      "learning_rate": 0.000553055585536967,
+      "loss": 3.9153,
       "step": 7350
     },
     {
-      "epoch": 0.796469701862017,
-      "grad_norm": 0.6223685145378113,
-      "learning_rate": 0.0005528132744316344,
-      "loss": 3.9036,
+      "epoch": 0.7978436657681941,
+      "grad_norm": 0.6454257965087891,
+      "learning_rate": 0.0005527317862924987,
+      "loss": 3.9148,
       "step": 7400
     },
     {
-      "epoch": 0.8018512539016253,
-      "grad_norm": 0.5439171195030212,
-      "learning_rate": 0.0005524900334015731,
-      "loss": 3.9139,
+      "epoch": 0.8032345013477089,
+      "grad_norm": 0.5397379398345947,
+      "learning_rate": 0.0005524079870480301,
+      "loss": 3.908,
       "step": 7450
     },
     {
-      "epoch": 0.8072328059412335,
-      "grad_norm": 0.6579694151878357,
-      "learning_rate": 0.0005521667923715117,
-      "loss": 3.9061,
+      "epoch": 0.8086253369272237,
+      "grad_norm": 0.5687892436981201,
+      "learning_rate": 0.0005520841878035618,
+      "loss": 3.9252,
       "step": 7500
     },
     {
-      "epoch": 0.8126143579808417,
-      "grad_norm": 0.6734451055526733,
-      "learning_rate": 0.0005518435513414502,
-      "loss": 3.9072,
+      "epoch": 0.8140161725067385,
+      "grad_norm": 0.6189422011375427,
+      "learning_rate": 0.0005517603885590933,
+      "loss": 3.8987,
       "step": 7550
     },
     {
-      "epoch": 0.8179959100204499,
-      "grad_norm": 0.5974473357200623,
-      "learning_rate": 0.0005515203103113888,
-      "loss": 3.9066,
+      "epoch": 0.8194070080862533,
+      "grad_norm": 0.5241925716400146,
+      "learning_rate": 0.0005514365893146249,
+      "loss": 3.8955,
       "step": 7600
     },
     {
-      "epoch": 0.8233774620600581,
-      "grad_norm": 0.6027451157569885,
-      "learning_rate": 0.0005511970692813274,
-      "loss": 3.9061,
+      "epoch": 0.8247978436657682,
+      "grad_norm": 0.5576037168502808,
+      "learning_rate": 0.0005511127900701564,
+      "loss": 3.8886,
       "step": 7650
     },
     {
-      "epoch": 0.8287590140996663,
-      "grad_norm": 0.6368857026100159,
-      "learning_rate": 0.000550873828251266,
-      "loss": 3.9254,
+      "epoch": 0.8301886792452831,
+      "grad_norm": 0.565155565738678,
+      "learning_rate": 0.000550788990825688,
+      "loss": 3.9167,
       "step": 7700
     },
     {
-      "epoch": 0.8341405661392746,
-      "grad_norm": 0.5675856471061707,
-      "learning_rate": 0.0005505505872212045,
-      "loss": 3.9026,
+      "epoch": 0.8355795148247979,
+      "grad_norm": 0.5731838345527649,
+      "learning_rate": 0.0005504651915812196,
+      "loss": 3.9103,
       "step": 7750
     },
     {
-      "epoch": 0.8395221181788828,
-      "grad_norm": 0.6578029990196228,
-      "learning_rate": 0.0005502273461911432,
-      "loss": 3.9009,
+      "epoch": 0.8409703504043127,
+      "grad_norm": 0.6731712818145752,
+      "learning_rate": 0.0005501413923367512,
+      "loss": 3.8767,
       "step": 7800
     },
     {
-      "epoch": 0.844903670218491,
-      "grad_norm": 0.7151255011558533,
-      "learning_rate": 0.0005499041051610817,
-      "loss": 3.8941,
+      "epoch": 0.8463611859838275,
+      "grad_norm": 0.5753887295722961,
+      "learning_rate": 0.0005498175930922827,
+      "loss": 3.8881,
       "step": 7850
     },
     {
-      "epoch": 0.8502852222580992,
-      "grad_norm": 0.5232037305831909,
-      "learning_rate": 0.0005495808641310204,
-      "loss": 3.8879,
+      "epoch": 0.8517520215633423,
+      "grad_norm": 0.520650327205658,
+      "learning_rate": 0.0005494937938478143,
+      "loss": 3.8946,
       "step": 7900
     },
     {
-      "epoch": 0.8556667742977074,
-      "grad_norm": 0.61030513048172,
-      "learning_rate": 0.0005492576231009589,
-      "loss": 3.8961,
+      "epoch": 0.8571428571428571,
+      "grad_norm": 0.555343747138977,
+      "learning_rate": 0.0005491699946033459,
+      "loss": 3.8822,
       "step": 7950
     },
     {
-      "epoch": 0.8610483263373157,
-      "grad_norm": 0.5574431419372559,
-      "learning_rate": 0.0005489343820708974,
-      "loss": 3.8761,
+      "epoch": 0.862533692722372,
+      "grad_norm": 0.6189568638801575,
+      "learning_rate": 0.0005488461953588775,
+      "loss": 3.8812,
       "step": 8000
     },
     {
-      "epoch": 0.8610483263373157,
-      "eval_accuracy": 0.3376209211570098,
-      "eval_loss": 3.8182339668273926,
-      "eval_runtime": 149.0778,
-      "eval_samples_per_second": 120.816,
-      "eval_steps_per_second": 7.553,
+      "epoch": 0.862533692722372,
+      "eval_accuracy": 0.3376978474015825,
+      "eval_loss": 3.8168375492095947,
+      "eval_runtime": 183.937,
+      "eval_samples_per_second": 97.919,
+      "eval_steps_per_second": 6.122,
       "step": 8000
     },
     {
-      "epoch": 0.8664298783769239,
-      "grad_norm": 0.611583411693573,
-      "learning_rate": 0.0005486111410408361,
-      "loss": 3.8816,
+      "epoch": 0.8679245283018868,
+      "grad_norm": 0.5607768297195435,
+      "learning_rate": 0.000548522396114409,
+      "loss": 3.884,
       "step": 8050
     },
     {
-      "epoch": 0.8718114304165321,
-      "grad_norm": 0.5672095417976379,
-      "learning_rate": 0.0005482879000107746,
-      "loss": 3.8771,
+      "epoch": 0.8733153638814016,
+      "grad_norm": 0.5763882398605347,
+      "learning_rate": 0.0005481985968699406,
+      "loss": 3.8958,
       "step": 8100
     },
     {
-      "epoch": 0.8771929824561403,
-      "grad_norm": 0.5900101065635681,
-      "learning_rate": 0.0005479646589807133,
-      "loss": 3.8695,
+      "epoch": 0.8787061994609164,
+      "grad_norm": 0.5309826135635376,
+      "learning_rate": 0.0005478747976254721,
+      "loss": 3.8734,
       "step": 8150
     },
     {
-      "epoch": 0.8825745344957485,
-      "grad_norm": 0.6257373690605164,
-      "learning_rate": 0.0005476414179506518,
-      "loss": 3.8712,
+      "epoch": 0.8840970350404312,
+      "grad_norm": 0.5828977227210999,
+      "learning_rate": 0.0005475509983810037,
+      "loss": 3.8736,
       "step": 8200
     },
     {
-      "epoch": 0.8879560865353568,
-      "grad_norm": 0.6412038207054138,
-      "learning_rate": 0.0005473181769205904,
-      "loss": 3.8893,
+      "epoch": 0.889487870619946,
+      "grad_norm": 0.6202263236045837,
+      "learning_rate": 0.0005472271991365352,
+      "loss": 3.8707,
       "step": 8250
     },
     {
-      "epoch": 0.8933376385749651,
-      "grad_norm": 0.5489782094955444,
-      "learning_rate": 0.000546994935890529,
-      "loss": 3.8712,
+      "epoch": 0.894878706199461,
+      "grad_norm": 0.6479662656784058,
+      "learning_rate": 0.0005469033998920669,
+      "loss": 3.8543,
       "step": 8300
     },
     {
-      "epoch": 0.8987191906145733,
-      "grad_norm": 0.5805892944335938,
-      "learning_rate": 0.0005466716948604677,
-      "loss": 3.8665,
+      "epoch": 0.9002695417789758,
+      "grad_norm": 0.5923611521720886,
+      "learning_rate": 0.0005465860766324878,
+      "loss": 3.8427,
       "step": 8350
     },
     {
-      "epoch": 0.9041007426541815,
-      "grad_norm": 0.5447978973388672,
-      "learning_rate": 0.0005463484538304062,
-      "loss": 3.8658,
+      "epoch": 0.9056603773584906,
+      "grad_norm": 0.5276738405227661,
+      "learning_rate": 0.0005462622773880194,
+      "loss": 3.8873,
       "step": 8400
     },
     {
-      "epoch": 0.9094822946937897,
-      "grad_norm": 0.6286603212356567,
-      "learning_rate": 0.0005460252128003447,
-      "loss": 3.8591,
+      "epoch": 0.9110512129380054,
+      "grad_norm": 0.7163664102554321,
+      "learning_rate": 0.000545938478143551,
+      "loss": 3.8539,
       "step": 8450
     },
     {
-      "epoch": 0.9148638467333979,
-      "grad_norm": 0.6845695376396179,
-      "learning_rate": 0.0005457019717702833,
-      "loss": 3.8676,
+      "epoch": 0.9164420485175202,
+      "grad_norm": 0.557249128818512,
+      "learning_rate": 0.0005456146788990825,
+      "loss": 3.8443,
       "step": 8500
     },
     {
-      "epoch": 0.9202453987730062,
-      "grad_norm": 0.6076292395591736,
-      "learning_rate": 0.0005453787307402219,
-      "loss": 3.8496,
+      "epoch": 0.921832884097035,
+      "grad_norm": 0.6361478567123413,
+      "learning_rate": 0.000545290879654614,
+      "loss": 3.8548,
       "step": 8550
     },
     {
-      "epoch": 0.9256269508126144,
-      "grad_norm": 0.5436860918998718,
-      "learning_rate": 0.0005450554897101605,
-      "loss": 3.8572,
+      "epoch": 0.9272237196765498,
+      "grad_norm": 0.6362162828445435,
+      "learning_rate": 0.0005449670804101457,
+      "loss": 3.857,
       "step": 8600
     },
     {
-      "epoch": 0.9310085028522226,
-      "grad_norm": 0.553849458694458,
-      "learning_rate": 0.0005447322486800991,
-      "loss": 3.8578,
+      "epoch": 0.9326145552560647,
+      "grad_norm": 0.6405161023139954,
+      "learning_rate": 0.0005446432811656773,
+      "loss": 3.8596,
       "step": 8650
     },
     {
-      "epoch": 0.9363900548918308,
-      "grad_norm": 0.6013088822364807,
-      "learning_rate": 0.0005444090076500377,
-      "loss": 3.853,
+      "epoch": 0.9380053908355795,
+      "grad_norm": 0.5607011914253235,
+      "learning_rate": 0.0005443194819212088,
+      "loss": 3.8363,
       "step": 8700
     },
     {
-      "epoch": 0.941771606931439,
-      "grad_norm": 0.5041967630386353,
-      "learning_rate": 0.0005440857666199763,
-      "loss": 3.8513,
+      "epoch": 0.9433962264150944,
+      "grad_norm": 0.6090148687362671,
+      "learning_rate": 0.0005439956826767404,
+      "loss": 3.8549,
       "step": 8750
     },
     {
-      "epoch": 0.9471531589710472,
-      "grad_norm": 0.5646461844444275,
-      "learning_rate": 0.0005437625255899148,
-      "loss": 3.8459,
+      "epoch": 0.9487870619946092,
+      "grad_norm": 0.6270147562026978,
+      "learning_rate": 0.000543671883432272,
+      "loss": 3.8424,
       "step": 8800
     },
     {
-      "epoch": 0.9525347110106555,
-      "grad_norm": 0.5224244594573975,
-      "learning_rate": 0.0005434392845598534,
-      "loss": 3.8614,
+      "epoch": 0.954177897574124,
+      "grad_norm": 0.562201738357544,
+      "learning_rate": 0.0005433480841878035,
+      "loss": 3.8382,
       "step": 8850
     },
     {
-      "epoch": 0.9579162630502637,
-      "grad_norm": 0.6048169732093811,
-      "learning_rate": 0.000543116043529792,
-      "loss": 3.8516,
+      "epoch": 0.9595687331536388,
+      "grad_norm": 0.6880311369895935,
+      "learning_rate": 0.000543024284943335,
+      "loss": 3.8296,
       "step": 8900
     },
     {
-      "epoch": 0.9632978150898719,
-      "grad_norm": 0.5460017919540405,
-      "learning_rate": 0.0005427928024997306,
-      "loss": 3.8384,
+      "epoch": 0.9649595687331537,
+      "grad_norm": 0.5794912576675415,
+      "learning_rate": 0.0005427004856988667,
+      "loss": 3.8442,
       "step": 8950
     },
     {
-      "epoch": 0.9686793671294801,
-      "grad_norm": 0.5394402146339417,
-      "learning_rate": 0.0005424695614696692,
-      "loss": 3.8471,
+      "epoch": 0.9703504043126685,
+      "grad_norm": 0.6744676232337952,
+      "learning_rate": 0.0005423766864543982,
+      "loss": 3.8445,
       "step": 9000
     },
     {
-      "epoch": 0.9686793671294801,
-      "eval_accuracy": 0.3415538297372306,
-      "eval_loss": 3.7804224491119385,
-      "eval_runtime": 148.5482,
-      "eval_samples_per_second": 121.247,
-      "eval_steps_per_second": 7.58,
+      "epoch": 0.9703504043126685,
+      "eval_accuracy": 0.3413188135408879,
+      "eval_loss": 3.77766489982605,
+      "eval_runtime": 184.153,
+      "eval_samples_per_second": 97.805,
+      "eval_steps_per_second": 6.114,
       "step": 9000
     },
     {
-      "epoch": 0.9740609191690883,
-      "grad_norm": 0.5256498456001282,
-      "learning_rate": 0.0005421463204396078,
-      "loss": 3.8335,
+      "epoch": 0.9757412398921833,
+      "grad_norm": 0.5510568022727966,
+      "learning_rate": 0.0005420528872099298,
+      "loss": 3.8276,
       "step": 9050
     },
     {
-      "epoch": 0.9794424712086965,
-      "grad_norm": 0.6665931940078735,
-      "learning_rate": 0.0005418230794095463,
-      "loss": 3.8403,
+      "epoch": 0.9811320754716981,
+      "grad_norm": 0.5297811627388,
+      "learning_rate": 0.0005417290879654613,
+      "loss": 3.8265,
       "step": 9100
     },
     {
-      "epoch": 0.9848240232483048,
-      "grad_norm": 0.5108867883682251,
-      "learning_rate": 0.000541499838379485,
-      "loss": 3.8502,
+      "epoch": 0.9865229110512129,
+      "grad_norm": 0.585360586643219,
+      "learning_rate": 0.000541405288720993,
+      "loss": 3.8399,
       "step": 9150
     },
     {
-      "epoch": 0.9902055752879131,
-      "grad_norm": 0.5940078496932983,
-      "learning_rate": 0.0005411830621700248,
-      "loss": 3.8331,
+      "epoch": 0.9919137466307277,
+      "grad_norm": 0.5567845702171326,
+      "learning_rate": 0.0005410814894765245,
+      "loss": 3.8217,
       "step": 9200
     },
     {
-      "epoch": 0.9955871273275213,
-      "grad_norm": 0.5534390807151794,
-      "learning_rate": 0.0005408598211399633,
-      "loss": 3.8375,
+      "epoch": 0.9973045822102425,
+      "grad_norm": 0.6347305774688721,
+      "learning_rate": 0.0005407576902320561,
+      "loss": 3.8527,
       "step": 9250
     },
     {
-      "epoch": 1.0009686793671295,
-      "grad_norm": 0.664844810962677,
-      "learning_rate": 0.0005405365801099019,
-      "loss": 3.8191,
+      "epoch": 1.0026954177897573,
+      "grad_norm": 0.5502357482910156,
+      "learning_rate": 0.0005404338909875876,
+      "loss": 3.7881,
       "step": 9300
     },
     {
-      "epoch": 1.0063502314067376,
-      "grad_norm": 0.5579114556312561,
-      "learning_rate": 0.0005402133390798405,
-      "loss": 3.7834,
+      "epoch": 1.0080862533692723,
+      "grad_norm": 0.5889478325843811,
+      "learning_rate": 0.0005401100917431192,
+      "loss": 3.7619,
       "step": 9350
     },
     {
-      "epoch": 1.011731783446346,
-      "grad_norm": 0.5882884860038757,
-      "learning_rate": 0.000539890098049779,
-      "loss": 3.7759,
+      "epoch": 1.013477088948787,
+      "grad_norm": 0.6611557006835938,
+      "learning_rate": 0.0005397862924986508,
+      "loss": 3.7757,
       "step": 9400
     },
     {
-      "epoch": 1.017113335485954,
-      "grad_norm": 0.6504033803939819,
-      "learning_rate": 0.0005395668570197177,
-      "loss": 3.7585,
+      "epoch": 1.0188679245283019,
+      "grad_norm": 0.5743335485458374,
+      "learning_rate": 0.0005394624932541824,
+      "loss": 3.7831,
       "step": 9450
     },
     {
-      "epoch": 1.0224948875255624,
-      "grad_norm": 0.520551860332489,
-      "learning_rate": 0.0005392436159896562,
-      "loss": 3.7759,
+      "epoch": 1.0242587601078168,
+      "grad_norm": 0.5449957847595215,
+      "learning_rate": 0.0005391386940097139,
+      "loss": 3.7516,
       "step": 9500
     },
     {
-      "epoch": 1.0278764395651705,
-      "grad_norm": 0.5106040239334106,
-      "learning_rate": 0.0005389203749595948,
-      "loss": 3.7818,
+      "epoch": 1.0296495956873315,
+      "grad_norm": 0.5865920186042786,
+      "learning_rate": 0.0005388148947652455,
+      "loss": 3.7644,
       "step": 9550
     },
     {
-      "epoch": 1.0332579916047788,
-      "grad_norm": 0.5361846089363098,
-      "learning_rate": 0.0005385971339295334,
-      "loss": 3.756,
+      "epoch": 1.0350404312668464,
+      "grad_norm": 0.564854621887207,
+      "learning_rate": 0.000538491095520777,
+      "loss": 3.7912,
       "step": 9600
     },
     {
-      "epoch": 1.0386395436443872,
-      "grad_norm": 0.5659493803977966,
-      "learning_rate": 0.000538273892899472,
-      "loss": 3.7683,
+      "epoch": 1.0404312668463611,
+      "grad_norm": 0.5819408297538757,
+      "learning_rate": 0.0005381672962763086,
+      "loss": 3.7516,
       "step": 9650
     },
     {
-      "epoch": 1.0440210956839953,
-      "grad_norm": 0.550243616104126,
-      "learning_rate": 0.0005379506518694106,
-      "loss": 3.7731,
+      "epoch": 1.045822102425876,
+      "grad_norm": 0.5492261052131653,
+      "learning_rate": 0.0005378434970318403,
+      "loss": 3.7754,
       "step": 9700
     },
     {
-      "epoch": 1.0494026477236036,
-      "grad_norm": 0.5463876724243164,
-      "learning_rate": 0.0005376274108393491,
-      "loss": 3.751,
+      "epoch": 1.0512129380053907,
+      "grad_norm": 0.670627236366272,
+      "learning_rate": 0.0005375196977873718,
+      "loss": 3.7525,
       "step": 9750
     },
     {
-      "epoch": 1.0547841997632117,
-      "grad_norm": 0.5370287299156189,
-      "learning_rate": 0.0005373041698092877,
-      "loss": 3.7825,
+      "epoch": 1.0566037735849056,
+      "grad_norm": 0.631523847579956,
+      "learning_rate": 0.0005371958985429034,
+      "loss": 3.7631,
       "step": 9800
     },
     {
-      "epoch": 1.06016575180282,
-      "grad_norm": 0.5432443022727966,
-      "learning_rate": 0.0005369809287792263,
-      "loss": 3.7691,
+      "epoch": 1.0619946091644206,
+      "grad_norm": 0.557306706905365,
+      "learning_rate": 0.0005368720992984349,
+      "loss": 3.767,
       "step": 9850
     },
     {
-      "epoch": 1.0655473038424281,
-      "grad_norm": 0.510743260383606,
-      "learning_rate": 0.000536657687749165,
-      "loss": 3.8028,
+      "epoch": 1.0673854447439353,
+      "grad_norm": 0.6992411613464355,
+      "learning_rate": 0.0005365483000539665,
+      "loss": 3.7646,
       "step": 9900
     },
     {
-      "epoch": 1.0709288558820365,
-      "grad_norm": 0.5883675813674927,
-      "learning_rate": 0.0005363344467191035,
-      "loss": 3.7823,
+      "epoch": 1.0727762803234502,
+      "grad_norm": 0.5577909350395203,
+      "learning_rate": 0.0005362245008094981,
+      "loss": 3.7781,
       "step": 9950
     },
     {
-      "epoch": 1.0763104079216446,
-      "grad_norm": 0.5291104316711426,
-      "learning_rate": 0.000536011205689042,
-      "loss": 3.7589,
+      "epoch": 1.0781671159029649,
+      "grad_norm": 0.5798797607421875,
+      "learning_rate": 0.0005359007015650297,
+      "loss": 3.7688,
       "step": 10000
     },
     {
-      "epoch": 1.0763104079216446,
-      "eval_accuracy": 0.34495249206806755,
-      "eval_loss": 3.7523419857025146,
-      "eval_runtime": 149.0472,
-      "eval_samples_per_second": 120.841,
-      "eval_steps_per_second": 7.555,
+      "epoch": 1.0781671159029649,
+      "eval_accuracy": 0.34479168579410213,
+      "eval_loss": 3.749129056930542,
+      "eval_runtime": 183.5981,
+      "eval_samples_per_second": 98.1,
+      "eval_steps_per_second": 6.133,
       "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": 8.3608485691392e+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 sha256:4222331dde522cb32c803259fd1784cc44dfa46cad6cba0d92642e134c2649bb
+oid sha256:46b6cf74c2912212e165b0372c0b6508024d7c992a638ccc24384ab9cb06eaca
 size 5304