Training in progress, step 10000, checkpoint

checkpoint-10000/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c7f4b66aff963b41d8ad24592337a5931f0e7c8e9a7ef03849882cb06b0959d3
+oid sha256:809ccb47a24ee60f65c7997a1d1589544f4b11b5307387cc901c7e656e29776e
 size 503128704

checkpoint-10000/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e6c79719de8dc628b11b4e0a640d95e3f26455ac74d27fc600a0d6cfc292fb8f
+oid sha256:3600ab41cdbd8a953a1f6d4de77eb7c8799b2237b9677c6e03575f0632bf9a0b
 size 1006351290

checkpoint-10000/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c4e6e63706b3e7c1569b5ffaa49431642792cc040955a53c8fc468d0452c9264
+oid sha256:d65e78f44114e21bbfe15593d61a8ef27409a026cbe91ed192a243a5a11f378b
 size 14244

checkpoint-10000/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:90c311f0069114271a5dba5d72dcffc82e7a33081dd5f150aa8c9b2278c1c497
+oid sha256:e4e7893396eb2644b36995e7c0917ee358383aadd367376b940738387a18a156
 size 1064

checkpoint-10000/trainer_state.json

@@ -1,7 +1,7 @@
 {
-  "best_metric": 3.7531447410583496,
-  "best_model_checkpoint": "/scratch/cl5625/exceptions/models/100M_495/checkpoint-10000",
-  "epoch": 1.0763104079216446,
+  "best_metric": 3.7646424770355225,
+  "best_model_checkpoint": "/scratch/cl5625/exceptions/models/100M_high_10_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": 3.403714418411255,
-      "learning_rate": 0.00028799999999999995,
-      "loss": 8.9829,
+      "epoch": 0.005390835579514825,
+      "grad_norm": 0.9467681646347046,
+      "learning_rate": 0.000276,
+      "loss": 8.7571,
       "step": 50
     },
     {
-      "epoch": 0.010763104079216447,
-      "grad_norm": 3.788738965988159,
-      "learning_rate": 0.000588,
-      "loss": 6.9704,
+      "epoch": 0.01078167115902965,
+      "grad_norm": 3.062793254852295,
+      "learning_rate": 0.0005759999999999999,
+      "loss": 6.9857,
       "step": 100
     },
     {
-      "epoch": 0.01614465611882467,
-      "grad_norm": 1.4487857818603516,
-      "learning_rate": 0.000599689688611141,
-      "loss": 6.4875,
+      "epoch": 0.016172506738544475,
+      "grad_norm": 2.382413148880005,
+      "learning_rate": 0.000599702104695089,
+      "loss": 6.597,
       "step": 150
     },
     {
-      "epoch": 0.021526208158432893,
-      "grad_norm": 2.067326068878174,
-      "learning_rate": 0.0005993664475810796,
-      "loss": 6.2749,
+      "epoch": 0.0215633423180593,
+      "grad_norm": 0.9547237753868103,
+      "learning_rate": 0.0005993783054506205,
+      "loss": 6.3206,
       "step": 200
     },
     {
-      "epoch": 0.026907760198041114,
-      "grad_norm": 1.0104901790618896,
-      "learning_rate": 0.0005990432065510182,
-      "loss": 6.0719,
+      "epoch": 0.026954177897574125,
+      "grad_norm": 1.4377750158309937,
+      "learning_rate": 0.0005990545062061521,
+      "loss": 6.1209,
       "step": 250
     },
     {
-      "epoch": 0.03228931223764934,
-      "grad_norm": 1.5355337858200073,
-      "learning_rate": 0.0005987199655209567,
-      "loss": 5.9689,
+      "epoch": 0.03234501347708895,
+      "grad_norm": 1.6764158010482788,
+      "learning_rate": 0.0005987307069616836,
+      "loss": 6.003,
       "step": 300
     },
     {
-      "epoch": 0.03767086427725756,
-      "grad_norm": 1.7764770984649658,
-      "learning_rate": 0.0005983967244908953,
-      "loss": 5.8883,
+      "epoch": 0.03773584905660377,
+      "grad_norm": 0.94362473487854,
+      "learning_rate": 0.0005984069077172153,
+      "loss": 5.9015,
       "step": 350
     },
     {
-      "epoch": 0.04305241631686579,
-      "grad_norm": 1.896673321723938,
-      "learning_rate": 0.0005980734834608338,
-      "loss": 5.7905,
+      "epoch": 0.0431266846361186,
+      "grad_norm": 1.4955400228500366,
+      "learning_rate": 0.0005980831084727469,
+      "loss": 5.8417,
       "step": 400
     },
     {
-      "epoch": 0.048433968356474004,
-      "grad_norm": 1.7309101819992065,
-      "learning_rate": 0.0005977502424307725,
-      "loss": 5.7165,
+      "epoch": 0.04851752021563342,
+      "grad_norm": 0.8737039566040039,
+      "learning_rate": 0.0005977593092282784,
+      "loss": 5.7544,
       "step": 450
     },
     {
-      "epoch": 0.05381552039608223,
-      "grad_norm": 1.5152183771133423,
-      "learning_rate": 0.0005974270014007111,
-      "loss": 5.6356,
+      "epoch": 0.05390835579514825,
+      "grad_norm": 1.6916964054107666,
+      "learning_rate": 0.00059743550998381,
+      "loss": 5.7004,
       "step": 500
     },
     {
-      "epoch": 0.05919707243569045,
-      "grad_norm": 1.2986925840377808,
-      "learning_rate": 0.0005971037603706497,
-      "loss": 5.5617,
+      "epoch": 0.05929919137466307,
+      "grad_norm": 1.3075671195983887,
+      "learning_rate": 0.0005971117107393416,
+      "loss": 5.5902,
       "step": 550
     },
     {
-      "epoch": 0.06457862447529868,
-      "grad_norm": 2.1971282958984375,
-      "learning_rate": 0.0005967805193405882,
-      "loss": 5.516,
+      "epoch": 0.0646900269541779,
+      "grad_norm": 1.5872267484664917,
+      "learning_rate": 0.0005967879114948732,
+      "loss": 5.5606,
       "step": 600
     },
     {
-      "epoch": 0.0699601765149069,
-      "grad_norm": 1.4911054372787476,
-      "learning_rate": 0.0005964572783105269,
-      "loss": 5.4241,
+      "epoch": 0.07008086253369272,
+      "grad_norm": 1.3364683389663696,
+      "learning_rate": 0.0005964641122504047,
+      "loss": 5.4624,
       "step": 650
     },
     {
-      "epoch": 0.07534172855451512,
-      "grad_norm": 1.4151604175567627,
-      "learning_rate": 0.0005961340372804654,
-      "loss": 5.362,
+      "epoch": 0.07547169811320754,
+      "grad_norm": 1.1322250366210938,
+      "learning_rate": 0.0005961403130059363,
+      "loss": 5.4412,
       "step": 700
     },
     {
-      "epoch": 0.08072328059412334,
-      "grad_norm": 1.035212755203247,
-      "learning_rate": 0.000595810796250404,
-      "loss": 5.3373,
+      "epoch": 0.08086253369272237,
+      "grad_norm": 1.526710867881775,
+      "learning_rate": 0.0005958165137614678,
+      "loss": 5.3674,
       "step": 750
     },
     {
-      "epoch": 0.08610483263373157,
-      "grad_norm": 1.0760645866394043,
-      "learning_rate": 0.0005954875552203426,
-      "loss": 5.2561,
+      "epoch": 0.0862533692722372,
+      "grad_norm": 1.0809352397918701,
+      "learning_rate": 0.0005954927145169995,
+      "loss": 5.3087,
       "step": 800
     },
     {
-      "epoch": 0.09148638467333979,
-      "grad_norm": 1.1470861434936523,
-      "learning_rate": 0.0005951643141902811,
-      "loss": 5.2337,
+      "epoch": 0.09164420485175202,
+      "grad_norm": 1.260084867477417,
+      "learning_rate": 0.0005951689152725309,
+      "loss": 5.2775,
       "step": 850
     },
     {
-      "epoch": 0.09686793671294801,
-      "grad_norm": 1.1675933599472046,
-      "learning_rate": 0.0005948410731602198,
-      "loss": 5.1697,
+      "epoch": 0.09703504043126684,
+      "grad_norm": 1.0708142518997192,
+      "learning_rate": 0.0005948451160280626,
+      "loss": 5.1994,
       "step": 900
     },
     {
-      "epoch": 0.10224948875255624,
-      "grad_norm": 0.6720972061157227,
-      "learning_rate": 0.0005945178321301583,
-      "loss": 5.1131,
+      "epoch": 0.10242587601078167,
+      "grad_norm": 0.828127920627594,
+      "learning_rate": 0.0005945213167835941,
+      "loss": 5.1905,
       "step": 950
     },
     {
-      "epoch": 0.10763104079216446,
-      "grad_norm": 0.8663607835769653,
-      "learning_rate": 0.000594194591100097,
-      "loss": 5.08,
+      "epoch": 0.1078167115902965,
+      "grad_norm": 1.1112550497055054,
+      "learning_rate": 0.0005941975175391257,
+      "loss": 5.134,
       "step": 1000
     },
     {
-      "epoch": 0.10763104079216446,
-      "eval_accuracy": 0.22722828330284353,
-      "eval_loss": 5.01658296585083,
-      "eval_runtime": 185.3477,
-      "eval_samples_per_second": 97.174,
-      "eval_steps_per_second": 6.075,
+      "epoch": 0.1078167115902965,
+      "eval_accuracy": 0.22364556298004884,
+      "eval_loss": 5.057981014251709,
+      "eval_runtime": 155.8033,
+      "eval_samples_per_second": 115.601,
+      "eval_steps_per_second": 7.227,
       "step": 1000
     },
     {
-      "epoch": 0.11301259283177269,
-      "grad_norm": 1.171659231185913,
-      "learning_rate": 0.0005938713500700355,
-      "loss": 5.0623,
+      "epoch": 0.11320754716981132,
+      "grad_norm": 1.0472087860107422,
+      "learning_rate": 0.0005938737182946572,
+      "loss": 5.0945,
       "step": 1050
     },
     {
-      "epoch": 0.1183941448713809,
-      "grad_norm": 1.2680262327194214,
-      "learning_rate": 0.000593548109039974,
-      "loss": 5.0281,
+      "epoch": 0.11859838274932614,
+      "grad_norm": 1.4178184270858765,
+      "learning_rate": 0.0005935499190501888,
+      "loss": 5.0532,
       "step": 1100
     },
     {
-      "epoch": 0.12377569691098914,
-      "grad_norm": 0.9914045929908752,
-      "learning_rate": 0.0005932248680099127,
-      "loss": 4.9721,
+      "epoch": 0.12398921832884097,
+      "grad_norm": 0.9259712100028992,
+      "learning_rate": 0.0005932261198057204,
+      "loss": 5.0331,
       "step": 1150
     },
     {
-      "epoch": 0.12915724895059735,
-      "grad_norm": 1.0515373945236206,
-      "learning_rate": 0.0005929016269798512,
-      "loss": 4.9805,
+      "epoch": 0.1293800539083558,
+      "grad_norm": 0.8550077080726624,
+      "learning_rate": 0.000592902320561252,
+      "loss": 5.0119,
       "step": 1200
     },
     {
-      "epoch": 0.13453880099020557,
-      "grad_norm": 1.0933362245559692,
-      "learning_rate": 0.0005925783859497898,
-      "loss": 4.8982,
+      "epoch": 0.1347708894878706,
+      "grad_norm": 0.8048805594444275,
+      "learning_rate": 0.0005925785213167835,
+      "loss": 4.9853,
       "step": 1250
     },
     {
-      "epoch": 0.1399203530298138,
-      "grad_norm": 1.5771769285202026,
-      "learning_rate": 0.0005922551449197284,
-      "loss": 4.9081,
+      "epoch": 0.14016172506738545,
+      "grad_norm": 1.0447702407836914,
+      "learning_rate": 0.0005922547220723151,
+      "loss": 4.9313,
       "step": 1300
     },
     {
-      "epoch": 0.14530190506942203,
-      "grad_norm": 1.2906471490859985,
-      "learning_rate": 0.0005919319038896671,
-      "loss": 4.8856,
+      "epoch": 0.14555256064690028,
+      "grad_norm": 1.220167636871338,
+      "learning_rate": 0.0005919309228278468,
+      "loss": 4.9171,
       "step": 1350
     },
     {
-      "epoch": 0.15068345710903025,
-      "grad_norm": 1.1471878290176392,
-      "learning_rate": 0.0005916086628596056,
-      "loss": 4.8496,
+      "epoch": 0.1509433962264151,
+      "grad_norm": 0.8236722350120544,
+      "learning_rate": 0.0005916071235833783,
+      "loss": 4.8912,
       "step": 1400
     },
     {
-      "epoch": 0.15606500914863847,
-      "grad_norm": 0.8490638732910156,
-      "learning_rate": 0.0005912854218295442,
-      "loss": 4.8223,
+      "epoch": 0.15633423180592992,
+      "grad_norm": 0.9686391949653625,
+      "learning_rate": 0.0005912833243389097,
+      "loss": 4.8686,
       "step": 1450
     },
     {
-      "epoch": 0.16144656118824668,
-      "grad_norm": 1.2815980911254883,
-      "learning_rate": 0.0005909621807994827,
-      "loss": 4.8234,
+      "epoch": 0.16172506738544473,
+      "grad_norm": 0.928552508354187,
+      "learning_rate": 0.0005909595250944414,
+      "loss": 4.8514,
       "step": 1500
     },
     {
-      "epoch": 0.1668281132278549,
-      "grad_norm": 1.162052035331726,
-      "learning_rate": 0.0005906389397694213,
-      "loss": 4.7904,
+      "epoch": 0.16711590296495957,
+      "grad_norm": 1.1130082607269287,
+      "learning_rate": 0.000590635725849973,
+      "loss": 4.8119,
       "step": 1550
     },
     {
-      "epoch": 0.17220966526746315,
-      "grad_norm": 1.0659350156784058,
-      "learning_rate": 0.00059031569873936,
-      "loss": 4.7667,
+      "epoch": 0.1725067385444744,
+      "grad_norm": 1.0583394765853882,
+      "learning_rate": 0.0005903119266055045,
+      "loss": 4.8314,
       "step": 1600
     },
     {
-      "epoch": 0.17759121730707136,
-      "grad_norm": 0.935391366481781,
-      "learning_rate": 0.0005899924577092985,
-      "loss": 4.734,
+      "epoch": 0.1778975741239892,
+      "grad_norm": 0.912996768951416,
+      "learning_rate": 0.0005899881273610361,
+      "loss": 4.809,
       "step": 1650
     },
     {
-      "epoch": 0.18297276934667958,
-      "grad_norm": 1.0256898403167725,
-      "learning_rate": 0.0005896692166792371,
-      "loss": 4.738,
+      "epoch": 0.18328840970350405,
+      "grad_norm": 0.7502346038818359,
+      "learning_rate": 0.0005896643281165677,
+      "loss": 4.7298,
       "step": 1700
     },
     {
-      "epoch": 0.1883543213862878,
-      "grad_norm": 0.8851535320281982,
-      "learning_rate": 0.0005893459756491757,
-      "loss": 4.6881,
+      "epoch": 0.18867924528301888,
+      "grad_norm": 1.1314946413040161,
+      "learning_rate": 0.0005893405288720993,
+      "loss": 4.7486,
       "step": 1750
     },
     {
-      "epoch": 0.19373587342589602,
-      "grad_norm": 1.0578432083129883,
-      "learning_rate": 0.0005890227346191143,
-      "loss": 4.6711,
+      "epoch": 0.1940700808625337,
+      "grad_norm": 1.0135972499847412,
+      "learning_rate": 0.0005890167296276308,
+      "loss": 4.7355,
       "step": 1800
     },
     {
-      "epoch": 0.19911742546550426,
-      "grad_norm": 1.0169049501419067,
-      "learning_rate": 0.0005886994935890529,
-      "loss": 4.6803,
+      "epoch": 0.19946091644204852,
+      "grad_norm": 0.8325548768043518,
+      "learning_rate": 0.0005886929303831624,
+      "loss": 4.6924,
       "step": 1850
     },
     {
-      "epoch": 0.20449897750511248,
-      "grad_norm": 0.8385932445526123,
-      "learning_rate": 0.0005883762525589915,
-      "loss": 4.6329,
+      "epoch": 0.20485175202156333,
+      "grad_norm": 0.8735172152519226,
+      "learning_rate": 0.0005883691311386939,
+      "loss": 4.6884,
       "step": 1900
     },
     {
-      "epoch": 0.2098805295447207,
-      "grad_norm": 1.224091649055481,
-      "learning_rate": 0.00058805301152893,
-      "loss": 4.6126,
+      "epoch": 0.21024258760107817,
+      "grad_norm": 1.009144902229309,
+      "learning_rate": 0.0005880453318942256,
+      "loss": 4.6563,
       "step": 1950
     },
     {
-      "epoch": 0.2152620815843289,
-      "grad_norm": 1.068516492843628,
-      "learning_rate": 0.0005877297704988686,
-      "loss": 4.6063,
+      "epoch": 0.215633423180593,
+      "grad_norm": 1.0438272953033447,
+      "learning_rate": 0.0005877215326497571,
+      "loss": 4.6368,
       "step": 2000
     },
     {
-      "epoch": 0.2152620815843289,
-      "eval_accuracy": 0.26832732464754905,
-      "eval_loss": 4.5254340171813965,
-      "eval_runtime": 184.072,
-      "eval_samples_per_second": 97.848,
-      "eval_steps_per_second": 6.117,
+      "epoch": 0.215633423180593,
+      "eval_accuracy": 0.26113689383776234,
+      "eval_loss": 4.590283393859863,
+      "eval_runtime": 153.3564,
+      "eval_samples_per_second": 117.445,
+      "eval_steps_per_second": 7.342,
       "step": 2000
     },
     {
-      "epoch": 0.22064363362393713,
-      "grad_norm": 1.1607660055160522,
-      "learning_rate": 0.0005874065294688072,
-      "loss": 4.5794,
+      "epoch": 0.2210242587601078,
+      "grad_norm": 0.8841946721076965,
+      "learning_rate": 0.0005873977334052887,
+      "loss": 4.6585,
       "step": 2050
     },
     {
-      "epoch": 0.22602518566354537,
-      "grad_norm": 0.8364855051040649,
-      "learning_rate": 0.0005870832884387457,
-      "loss": 4.5403,
+      "epoch": 0.22641509433962265,
+      "grad_norm": 0.9768223762512207,
+      "learning_rate": 0.0005870739341608202,
+      "loss": 4.6248,
       "step": 2100
     },
     {
-      "epoch": 0.2314067377031536,
-      "grad_norm": 1.0584214925765991,
-      "learning_rate": 0.0005867600474086844,
-      "loss": 4.5344,
+      "epoch": 0.23180592991913745,
+      "grad_norm": 0.88718181848526,
+      "learning_rate": 0.0005867501349163519,
+      "loss": 4.5645,
       "step": 2150
     },
     {
-      "epoch": 0.2367882897427618,
-      "grad_norm": 0.8201826214790344,
-      "learning_rate": 0.0005864368063786229,
-      "loss": 4.5198,
+      "epoch": 0.2371967654986523,
+      "grad_norm": 0.928945779800415,
+      "learning_rate": 0.0005864263356718833,
+      "loss": 4.6031,
       "step": 2200
     },
     {
-      "epoch": 0.24216984178237003,
-      "grad_norm": 0.8647080063819885,
-      "learning_rate": 0.0005861135653485616,
-      "loss": 4.5024,
+      "epoch": 0.24258760107816713,
+      "grad_norm": 0.9557152986526489,
+      "learning_rate": 0.000586102536427415,
+      "loss": 4.5394,
       "step": 2250
     },
     {
-      "epoch": 0.24755139382197827,
-      "grad_norm": 1.2521132230758667,
-      "learning_rate": 0.0005857903243185001,
-      "loss": 4.4975,
+      "epoch": 0.24797843665768193,
+      "grad_norm": 0.938102126121521,
+      "learning_rate": 0.0005857787371829465,
+      "loss": 4.5208,
       "step": 2300
     },
     {
-      "epoch": 0.2529329458615865,
-      "grad_norm": 0.8756551742553711,
-      "learning_rate": 0.0005854670832884386,
-      "loss": 4.4975,
+      "epoch": 0.25336927223719674,
+      "grad_norm": 0.8805590271949768,
+      "learning_rate": 0.0005854549379384781,
+      "loss": 4.5136,
       "step": 2350
     },
     {
-      "epoch": 0.2583144979011947,
-      "grad_norm": 0.9284951090812683,
-      "learning_rate": 0.0005851438422583773,
-      "loss": 4.4694,
+      "epoch": 0.2587601078167116,
+      "grad_norm": 1.2024765014648438,
+      "learning_rate": 0.0005851311386940096,
+      "loss": 4.5026,
       "step": 2400
     },
     {
-      "epoch": 0.2636960499408029,
-      "grad_norm": 0.8526713252067566,
-      "learning_rate": 0.0005848206012283159,
-      "loss": 4.4355,
+      "epoch": 0.2641509433962264,
+      "grad_norm": 0.9804381132125854,
+      "learning_rate": 0.0005848073394495412,
+      "loss": 4.5039,
       "step": 2450
     },
     {
-      "epoch": 0.26907760198041114,
-      "grad_norm": 0.8831415176391602,
-      "learning_rate": 0.0005844973601982545,
-      "loss": 4.4085,
+      "epoch": 0.2695417789757412,
+      "grad_norm": 0.9215933084487915,
+      "learning_rate": 0.0005844835402050728,
+      "loss": 4.4967,
       "step": 2500
     },
     {
-      "epoch": 0.27445915402001936,
-      "grad_norm": 0.9462632536888123,
-      "learning_rate": 0.000584174119168193,
-      "loss": 4.4267,
+      "epoch": 0.2749326145552561,
+      "grad_norm": 0.9548609852790833,
+      "learning_rate": 0.0005841597409606044,
+      "loss": 4.4551,
       "step": 2550
     },
     {
-      "epoch": 0.2798407060596276,
-      "grad_norm": 0.9993759393692017,
-      "learning_rate": 0.0005838508781381316,
-      "loss": 4.4065,
+      "epoch": 0.2803234501347709,
+      "grad_norm": 0.8881437182426453,
+      "learning_rate": 0.000583835941716136,
+      "loss": 4.4451,
       "step": 2600
     },
     {
-      "epoch": 0.2852222580992358,
-      "grad_norm": 0.8348021507263184,
-      "learning_rate": 0.0005835276371080702,
-      "loss": 4.3843,
+      "epoch": 0.2857142857142857,
+      "grad_norm": 1.0776073932647705,
+      "learning_rate": 0.0005835121424716675,
+      "loss": 4.4423,
       "step": 2650
     },
     {
-      "epoch": 0.29060381013884407,
-      "grad_norm": 0.779861330986023,
-      "learning_rate": 0.0005832043960780088,
-      "loss": 4.389,
+      "epoch": 0.29110512129380056,
+      "grad_norm": 0.9139967560768127,
+      "learning_rate": 0.0005831883432271992,
+      "loss": 4.4323,
       "step": 2700
     },
     {
-      "epoch": 0.2959853621784523,
-      "grad_norm": 0.863554835319519,
-      "learning_rate": 0.0005828811550479474,
-      "loss": 4.3843,
+      "epoch": 0.29649595687331537,
+      "grad_norm": 0.8609856367111206,
+      "learning_rate": 0.0005828645439827307,
+      "loss": 4.3929,
       "step": 2750
     },
     {
-      "epoch": 0.3013669142180605,
-      "grad_norm": 0.8065552115440369,
-      "learning_rate": 0.0005825579140178859,
-      "loss": 4.3862,
+      "epoch": 0.3018867924528302,
+      "grad_norm": 0.8271277546882629,
+      "learning_rate": 0.0005825407447382622,
+      "loss": 4.4317,
       "step": 2800
     },
     {
-      "epoch": 0.3067484662576687,
-      "grad_norm": 0.7346043586730957,
-      "learning_rate": 0.0005822346729878246,
-      "loss": 4.3553,
+      "epoch": 0.30727762803234504,
+      "grad_norm": 0.7900182604789734,
+      "learning_rate": 0.0005822169454937938,
+      "loss": 4.3984,
       "step": 2850
     },
     {
-      "epoch": 0.31213001829727693,
-      "grad_norm": 0.9188678860664368,
-      "learning_rate": 0.0005819114319577631,
-      "loss": 4.3256,
+      "epoch": 0.31266846361185985,
+      "grad_norm": 0.7303901314735413,
+      "learning_rate": 0.0005818931462493254,
+      "loss": 4.3783,
       "step": 2900
     },
     {
-      "epoch": 0.31751157033688515,
-      "grad_norm": 0.705977201461792,
-      "learning_rate": 0.0005815881909277017,
-      "loss": 4.3318,
+      "epoch": 0.31805929919137466,
+      "grad_norm": 0.6963273286819458,
+      "learning_rate": 0.0005815693470048569,
+      "loss": 4.3785,
       "step": 2950
     },
     {
-      "epoch": 0.32289312237649337,
-      "grad_norm": 0.9044803380966187,
-      "learning_rate": 0.0005812649498976403,
-      "loss": 4.3318,
+      "epoch": 0.32345013477088946,
+      "grad_norm": 0.7128702998161316,
+      "learning_rate": 0.0005812455477603885,
+      "loss": 4.3722,
       "step": 3000
     },
     {
-      "epoch": 0.32289312237649337,
-      "eval_accuracy": 0.2972615886725453,
-      "eval_loss": 4.248094081878662,
-      "eval_runtime": 184.1083,
-      "eval_samples_per_second": 97.828,
-      "eval_steps_per_second": 6.116,
+      "epoch": 0.32345013477088946,
+      "eval_accuracy": 0.2934933978702513,
+      "eval_loss": 4.272831916809082,
+      "eval_runtime": 153.1756,
+      "eval_samples_per_second": 117.584,
+      "eval_steps_per_second": 7.351,
       "step": 3000
     },
     {
-      "epoch": 0.3282746744161016,
-      "grad_norm": 0.7348528504371643,
-      "learning_rate": 0.0005809417088675789,
-      "loss": 4.3203,
+      "epoch": 0.3288409703504043,
+      "grad_norm": 0.8196587562561035,
+      "learning_rate": 0.0005809217485159201,
+      "loss": 4.3483,
       "step": 3050
     },
     {
-      "epoch": 0.3336562264557098,
-      "grad_norm": 0.8613377213478088,
-      "learning_rate": 0.0005806184678375175,
-      "loss": 4.3203,
+      "epoch": 0.33423180592991913,
+      "grad_norm": 0.730968177318573,
+      "learning_rate": 0.0005805979492714517,
+      "loss": 4.3439,
       "step": 3100
     },
     {
-      "epoch": 0.3390377784953181,
-      "grad_norm": 0.6706194281578064,
-      "learning_rate": 0.000580295226807456,
-      "loss": 4.2918,
+      "epoch": 0.33962264150943394,
+      "grad_norm": 0.7348554134368896,
+      "learning_rate": 0.0005802741500269832,
+      "loss": 4.307,
       "step": 3150
     },
     {
-      "epoch": 0.3444193305349263,
-      "grad_norm": 0.8804624080657959,
-      "learning_rate": 0.0005799719857773946,
-      "loss": 4.2862,
+      "epoch": 0.3450134770889488,
+      "grad_norm": 0.7481687068939209,
+      "learning_rate": 0.0005799503507825148,
+      "loss": 4.3136,
       "step": 3200
     },
     {
-      "epoch": 0.3498008825745345,
-      "grad_norm": 0.6948880553245544,
-      "learning_rate": 0.0005796487447473331,
-      "loss": 4.2839,
+      "epoch": 0.3504043126684636,
+      "grad_norm": 0.7450636029243469,
+      "learning_rate": 0.0005796265515380463,
+      "loss": 4.2863,
       "step": 3250
     },
     {
-      "epoch": 0.35518243461414273,
-      "grad_norm": 0.768258810043335,
-      "learning_rate": 0.0005793255037172718,
-      "loss": 4.2639,
+      "epoch": 0.3557951482479784,
+      "grad_norm": 0.86126309633255,
+      "learning_rate": 0.000579302752293578,
+      "loss": 4.3108,
       "step": 3300
     },
     {
-      "epoch": 0.36056398665375095,
-      "grad_norm": 0.7073678970336914,
-      "learning_rate": 0.0005790022626872104,
-      "loss": 4.2469,
+      "epoch": 0.3611859838274933,
+      "grad_norm": 0.7103971838951111,
+      "learning_rate": 0.0005789789530491095,
+      "loss": 4.287,
       "step": 3350
     },
     {
-      "epoch": 0.36594553869335916,
-      "grad_norm": 0.8866555690765381,
-      "learning_rate": 0.000578679021657149,
-      "loss": 4.2393,
+      "epoch": 0.3665768194070081,
+      "grad_norm": 0.8686874508857727,
+      "learning_rate": 0.0005786551538046411,
+      "loss": 4.2802,
       "step": 3400
     },
     {
-      "epoch": 0.3713270907329674,
-      "grad_norm": 0.629542887210846,
-      "learning_rate": 0.0005783557806270875,
-      "loss": 4.2315,
+      "epoch": 0.3719676549865229,
+      "grad_norm": 0.6263065934181213,
+      "learning_rate": 0.0005783313545601726,
+      "loss": 4.2675,
       "step": 3450
     },
     {
-      "epoch": 0.3767086427725756,
-      "grad_norm": 0.8919417858123779,
-      "learning_rate": 0.0005780325395970262,
-      "loss": 4.2348,
+      "epoch": 0.37735849056603776,
+      "grad_norm": 0.7077925205230713,
+      "learning_rate": 0.0005780075553157043,
+      "loss": 4.2435,
       "step": 3500
     },
     {
-      "epoch": 0.3820901948121838,
-      "grad_norm": 0.7498871088027954,
-      "learning_rate": 0.0005777092985669647,
-      "loss": 4.2246,
+      "epoch": 0.38274932614555257,
+      "grad_norm": 0.8216277360916138,
+      "learning_rate": 0.0005776837560712357,
+      "loss": 4.2474,
       "step": 3550
     },
     {
-      "epoch": 0.38747174685179203,
-      "grad_norm": 0.7134843468666077,
-      "learning_rate": 0.0005773860575369033,
-      "loss": 4.2223,
+      "epoch": 0.3881401617250674,
+      "grad_norm": 0.8052435517311096,
+      "learning_rate": 0.0005773599568267673,
+      "loss": 4.2508,
       "step": 3600
     },
     {
-      "epoch": 0.3928532988914003,
-      "grad_norm": 0.64930260181427,
-      "learning_rate": 0.0005770628165068419,
-      "loss": 4.1985,
+      "epoch": 0.3935309973045822,
+      "grad_norm": 1.003618836402893,
+      "learning_rate": 0.0005770361575822989,
+      "loss": 4.2544,
       "step": 3650
     },
     {
-      "epoch": 0.3982348509310085,
-      "grad_norm": 0.6372562646865845,
-      "learning_rate": 0.0005767395754767804,
-      "loss": 4.2213,
+      "epoch": 0.39892183288409705,
+      "grad_norm": 0.8883004784584045,
+      "learning_rate": 0.0005767123583378305,
+      "loss": 4.228,
       "step": 3700
     },
     {
-      "epoch": 0.40361640297061674,
-      "grad_norm": 0.8313030004501343,
-      "learning_rate": 0.0005764163344467191,
-      "loss": 4.1942,
+      "epoch": 0.40431266846361186,
+      "grad_norm": 0.8125618696212769,
+      "learning_rate": 0.000576388559093362,
+      "loss": 4.2213,
       "step": 3750
     },
     {
-      "epoch": 0.40899795501022496,
-      "grad_norm": 0.7237107753753662,
-      "learning_rate": 0.0005760930934166576,
-      "loss": 4.1753,
+      "epoch": 0.40970350404312667,
+      "grad_norm": 0.7294287085533142,
+      "learning_rate": 0.0005760647598488936,
+      "loss": 4.2398,
       "step": 3800
     },
     {
-      "epoch": 0.4143795070498332,
-      "grad_norm": 0.602759063243866,
-      "learning_rate": 0.0005757698523865963,
-      "loss": 4.1893,
+      "epoch": 0.41509433962264153,
+      "grad_norm": 0.7409259080886841,
+      "learning_rate": 0.0005757409606044253,
+      "loss": 4.2228,
       "step": 3850
     },
     {
-      "epoch": 0.4197610590894414,
-      "grad_norm": 0.6982997059822083,
-      "learning_rate": 0.0005754466113565348,
-      "loss": 4.1906,
+      "epoch": 0.42048517520215634,
+      "grad_norm": 0.9595204591751099,
+      "learning_rate": 0.0005754171613599568,
+      "loss": 4.189,
       "step": 3900
     },
     {
-      "epoch": 0.4251426111290496,
-      "grad_norm": 0.8566523790359497,
-      "learning_rate": 0.0005751233703264733,
-      "loss": 4.1769,
+      "epoch": 0.42587601078167114,
+      "grad_norm": 0.6629497408866882,
+      "learning_rate": 0.0005750933621154884,
+      "loss": 4.1884,
       "step": 3950
     },
     {
-      "epoch": 0.4305241631686578,
-      "grad_norm": 0.7572330236434937,
-      "learning_rate": 0.000574800129296412,
-      "loss": 4.1659,
+      "epoch": 0.431266846361186,
+      "grad_norm": 0.8883629441261292,
+      "learning_rate": 0.0005747695628710199,
+      "loss": 4.1956,
       "step": 4000
     },
     {
-      "epoch": 0.4305241631686578,
-      "eval_accuracy": 0.31151891195867365,
-      "eval_loss": 4.096553802490234,
-      "eval_runtime": 183.6747,
-      "eval_samples_per_second": 98.059,
-      "eval_steps_per_second": 6.13,
+      "epoch": 0.431266846361186,
+      "eval_accuracy": 0.30888592652826535,
+      "eval_loss": 4.123743534088135,
+      "eval_runtime": 153.2705,
+      "eval_samples_per_second": 117.511,
+      "eval_steps_per_second": 7.346,
       "step": 4000
     },
     {
-      "epoch": 0.43590571520826604,
-      "grad_norm": 0.7648617029190063,
-      "learning_rate": 0.0005744768882663505,
-      "loss": 4.1546,
+      "epoch": 0.4366576819407008,
+      "grad_norm": 0.6576101779937744,
+      "learning_rate": 0.0005744457636265515,
+      "loss": 4.1943,
       "step": 4050
     },
     {
-      "epoch": 0.44128726724787426,
-      "grad_norm": 0.8174023032188416,
-      "learning_rate": 0.0005741536472362891,
-      "loss": 4.1598,
+      "epoch": 0.4420485175202156,
+      "grad_norm": 0.6346908211708069,
+      "learning_rate": 0.0005741219643820831,
+      "loss": 4.1902,
       "step": 4100
     },
     {
-      "epoch": 0.44666881928748253,
-      "grad_norm": 0.7902569770812988,
-      "learning_rate": 0.0005738304062062277,
-      "loss": 4.1417,
+      "epoch": 0.4474393530997305,
+      "grad_norm": 0.6723673343658447,
+      "learning_rate": 0.0005737981651376146,
+      "loss": 4.1837,
       "step": 4150
     },
     {
-      "epoch": 0.45205037132709075,
-      "grad_norm": 0.775563657283783,
-      "learning_rate": 0.0005735071651761664,
-      "loss": 4.1588,
+      "epoch": 0.4528301886792453,
+      "grad_norm": 0.8088533282279968,
+      "learning_rate": 0.0005734743658931462,
+      "loss": 4.1711,
       "step": 4200
     },
     {
-      "epoch": 0.45743192336669897,
-      "grad_norm": 0.7469772100448608,
-      "learning_rate": 0.0005731839241461049,
-      "loss": 4.129,
+      "epoch": 0.4582210242587601,
+      "grad_norm": 0.5922421813011169,
+      "learning_rate": 0.0005731505666486778,
+      "loss": 4.1633,
       "step": 4250
     },
     {
-      "epoch": 0.4628134754063072,
-      "grad_norm": 0.7498493194580078,
-      "learning_rate": 0.0005728606831160435,
-      "loss": 4.1349,
+      "epoch": 0.4636118598382749,
+      "grad_norm": 0.7532811164855957,
+      "learning_rate": 0.0005728267674042093,
+      "loss": 4.1571,
       "step": 4300
     },
     {
-      "epoch": 0.4681950274459154,
-      "grad_norm": 0.777646541595459,
-      "learning_rate": 0.000572537442085982,
-      "loss": 4.1185,
+      "epoch": 0.46900269541778977,
+      "grad_norm": 0.7632968425750732,
+      "learning_rate": 0.0005725029681597409,
+      "loss": 4.1624,
       "step": 4350
     },
     {
-      "epoch": 0.4735765794855236,
-      "grad_norm": 0.7824798822402954,
-      "learning_rate": 0.0005722142010559206,
-      "loss": 4.1398,
+      "epoch": 0.4743935309973046,
+      "grad_norm": 0.683935284614563,
+      "learning_rate": 0.0005721791689152725,
+      "loss": 4.1474,
       "step": 4400
     },
     {
-      "epoch": 0.47895813152513184,
-      "grad_norm": 0.7924827337265015,
-      "learning_rate": 0.0005718909600258593,
-      "loss": 4.12,
+      "epoch": 0.4797843665768194,
+      "grad_norm": 0.828084409236908,
+      "learning_rate": 0.0005718553696708041,
+      "loss": 4.1485,
       "step": 4450
     },
     {
-      "epoch": 0.48433968356474005,
-      "grad_norm": 0.6577082872390747,
-      "learning_rate": 0.0005715677189957978,
-      "loss": 4.1047,
+      "epoch": 0.48517520215633425,
+      "grad_norm": 0.6277663707733154,
+      "learning_rate": 0.0005715315704263356,
+      "loss": 4.1318,
       "step": 4500
     },
     {
-      "epoch": 0.48972123560434827,
-      "grad_norm": 0.7151188254356384,
-      "learning_rate": 0.0005712444779657364,
-      "loss": 4.1206,
+      "epoch": 0.49056603773584906,
+      "grad_norm": 0.7237159013748169,
+      "learning_rate": 0.0005712077711818672,
+      "loss": 4.1216,
       "step": 4550
     },
     {
-      "epoch": 0.49510278764395654,
-      "grad_norm": 0.64276123046875,
-      "learning_rate": 0.000570921236935675,
-      "loss": 4.1035,
+      "epoch": 0.49595687331536387,
+      "grad_norm": 0.7076147794723511,
+      "learning_rate": 0.0005708839719373987,
+      "loss": 4.1303,
       "step": 4600
     },
     {
-      "epoch": 0.5004843396835648,
-      "grad_norm": 0.6700813174247742,
-      "learning_rate": 0.0005705979959056136,
-      "loss": 4.0922,
+      "epoch": 0.5013477088948787,
+      "grad_norm": 0.6837750673294067,
+      "learning_rate": 0.0005705601726929304,
+      "loss": 4.1089,
       "step": 4650
     },
     {
-      "epoch": 0.505865891723173,
-      "grad_norm": 0.6062188744544983,
-      "learning_rate": 0.0005702747548755522,
-      "loss": 4.0885,
+      "epoch": 0.5067385444743935,
+      "grad_norm": 0.8128517270088196,
+      "learning_rate": 0.0005702363734484619,
+      "loss": 4.1158,
       "step": 4700
     },
     {
-      "epoch": 0.5112474437627812,
-      "grad_norm": 0.6247645616531372,
-      "learning_rate": 0.0005699515138454908,
-      "loss": 4.0825,
+      "epoch": 0.5121293800539084,
+      "grad_norm": 0.5411143898963928,
+      "learning_rate": 0.0005699125742039935,
+      "loss": 4.0971,
       "step": 4750
     },
     {
-      "epoch": 0.5166289958023894,
-      "grad_norm": 0.6468704342842102,
-      "learning_rate": 0.0005696282728154293,
-      "loss": 4.1031,
+      "epoch": 0.5175202156334232,
+      "grad_norm": 0.6508839130401611,
+      "learning_rate": 0.000569588774959525,
+      "loss": 4.1118,
       "step": 4800
     },
     {
-      "epoch": 0.5220105478419976,
-      "grad_norm": 0.5434349179267883,
-      "learning_rate": 0.0005693050317853679,
-      "loss": 4.0938,
+      "epoch": 0.522911051212938,
+      "grad_norm": 0.6948415040969849,
+      "learning_rate": 0.0005692649757150567,
+      "loss": 4.1042,
       "step": 4850
     },
     {
-      "epoch": 0.5273920998816058,
-      "grad_norm": 0.7494503855705261,
-      "learning_rate": 0.0005689817907553065,
-      "loss": 4.0747,
+      "epoch": 0.5283018867924528,
+      "grad_norm": 0.6959540247917175,
+      "learning_rate": 0.0005689411764705881,
+      "loss": 4.1124,
       "step": 4900
     },
     {
-      "epoch": 0.5327736519212141,
-      "grad_norm": 0.5512547492980957,
-      "learning_rate": 0.000568658549725245,
-      "loss": 4.0845,
+      "epoch": 0.5336927223719676,
+      "grad_norm": 0.7044671177864075,
+      "learning_rate": 0.0005686173772261197,
+      "loss": 4.0879,
       "step": 4950
     },
     {
-      "epoch": 0.5381552039608223,
-      "grad_norm": 0.6949486136436462,
-      "learning_rate": 0.0005683353086951837,
-      "loss": 4.0724,
+      "epoch": 0.5390835579514824,
+      "grad_norm": 0.633337676525116,
+      "learning_rate": 0.0005682935779816514,
+      "loss": 4.076,
       "step": 5000
     },
     {
-      "epoch": 0.5381552039608223,
-      "eval_accuracy": 0.3202588416015827,
-      "eval_loss": 3.9990289211273193,
-      "eval_runtime": 184.0076,
-      "eval_samples_per_second": 97.882,
-      "eval_steps_per_second": 6.119,
+      "epoch": 0.5390835579514824,
+      "eval_accuracy": 0.31857580836933636,
+      "eval_loss": 4.020145893096924,
+      "eval_runtime": 153.5029,
+      "eval_samples_per_second": 117.333,
+      "eval_steps_per_second": 7.335,
       "step": 5000
     },
     {
-      "epoch": 0.5435367560004305,
-      "grad_norm": 0.6374182105064392,
-      "learning_rate": 0.0005680120676651222,
-      "loss": 4.0595,
+      "epoch": 0.5444743935309974,
+      "grad_norm": 0.5795180201530457,
+      "learning_rate": 0.0005679697787371829,
+      "loss": 4.0695,
       "step": 5050
     },
     {
-      "epoch": 0.5489183080400387,
-      "grad_norm": 0.6074301600456238,
-      "learning_rate": 0.0005676888266350609,
-      "loss": 4.0512,
+      "epoch": 0.5498652291105122,
+      "grad_norm": 0.7057806849479675,
+      "learning_rate": 0.0005676459794927145,
+      "loss": 4.0772,
       "step": 5100
     },
     {
-      "epoch": 0.5542998600796469,
-      "grad_norm": 0.5869015455245972,
-      "learning_rate": 0.0005673655856049994,
-      "loss": 4.0554,
+      "epoch": 0.555256064690027,
+      "grad_norm": 0.6123654246330261,
+      "learning_rate": 0.000567322180248246,
+      "loss": 4.0789,
       "step": 5150
     },
     {
-      "epoch": 0.5596814121192552,
-      "grad_norm": 0.636626660823822,
-      "learning_rate": 0.0005670423445749379,
-      "loss": 4.0444,
+      "epoch": 0.5606469002695418,
+      "grad_norm": 0.6368526220321655,
+      "learning_rate": 0.0005669983810037777,
+      "loss": 4.0569,
       "step": 5200
     },
     {
-      "epoch": 0.5650629641588634,
-      "grad_norm": 0.6491479277610779,
-      "learning_rate": 0.0005667191035448766,
-      "loss": 4.0441,
+      "epoch": 0.5660377358490566,
+      "grad_norm": 0.7483193278312683,
+      "learning_rate": 0.0005666745817593092,
+      "loss": 4.0595,
       "step": 5250
     },
     {
-      "epoch": 0.5704445161984716,
-      "grad_norm": 0.6757907271385193,
-      "learning_rate": 0.0005663958625148152,
-      "loss": 4.0463,
+      "epoch": 0.5714285714285714,
+      "grad_norm": 0.6111021041870117,
+      "learning_rate": 0.0005663507825148408,
+      "loss": 4.042,
       "step": 5300
     },
     {
-      "epoch": 0.5758260682380799,
-      "grad_norm": 0.6085935831069946,
-      "learning_rate": 0.0005660726214847538,
-      "loss": 4.0338,
+      "epoch": 0.5768194070080862,
+      "grad_norm": 0.7408255934715271,
+      "learning_rate": 0.0005660269832703723,
+      "loss": 4.0586,
       "step": 5350
     },
     {
-      "epoch": 0.5812076202776881,
-      "grad_norm": 0.5898264050483704,
-      "learning_rate": 0.0005657493804546923,
-      "loss": 4.045,
+      "epoch": 0.5822102425876011,
+      "grad_norm": 0.717922031879425,
+      "learning_rate": 0.0005657031840259039,
+      "loss": 4.0458,
       "step": 5400
     },
     {
-      "epoch": 0.5865891723172963,
-      "grad_norm": 0.7703672051429749,
-      "learning_rate": 0.0005654261394246309,
-      "loss": 4.0247,
+      "epoch": 0.5876010781671159,
+      "grad_norm": 0.6948340535163879,
+      "learning_rate": 0.0005653793847814355,
+      "loss": 4.0578,
       "step": 5450
     },
     {
-      "epoch": 0.5919707243569046,
-      "grad_norm": 0.6306314468383789,
-      "learning_rate": 0.0005651028983945695,
-      "loss": 4.0191,
+      "epoch": 0.5929919137466307,
+      "grad_norm": 0.6431124210357666,
+      "learning_rate": 0.000565055585536967,
+      "loss": 4.0547,
       "step": 5500
     },
     {
-      "epoch": 0.5973522763965128,
-      "grad_norm": 0.6411070227622986,
-      "learning_rate": 0.000564779657364508,
-      "loss": 4.0399,
+      "epoch": 0.5983827493261455,
+      "grad_norm": 0.6256029605865479,
+      "learning_rate": 0.0005647317862924986,
+      "loss": 4.0613,
       "step": 5550
     },
     {
-      "epoch": 0.602733828436121,
-      "grad_norm": 0.7495064735412598,
-      "learning_rate": 0.0005644564163344467,
-      "loss": 4.0159,
+      "epoch": 0.6037735849056604,
+      "grad_norm": 0.6387584209442139,
+      "learning_rate": 0.0005644079870480302,
+      "loss": 4.0393,
       "step": 5600
     },
     {
-      "epoch": 0.6081153804757292,
-      "grad_norm": 0.6548343896865845,
-      "learning_rate": 0.0005641331753043852,
-      "loss": 4.0313,
+      "epoch": 0.6091644204851752,
+      "grad_norm": 0.5745953321456909,
+      "learning_rate": 0.0005640841878035617,
+      "loss": 4.0271,
       "step": 5650
     },
     {
-      "epoch": 0.6134969325153374,
-      "grad_norm": 0.6325730085372925,
-      "learning_rate": 0.0005638099342743239,
-      "loss": 4.0135,
+      "epoch": 0.6145552560646901,
+      "grad_norm": 0.6845305562019348,
+      "learning_rate": 0.0005637603885590933,
+      "loss": 4.0351,
       "step": 5700
     },
     {
-      "epoch": 0.6188784845549457,
-      "grad_norm": 0.5813512802124023,
-      "learning_rate": 0.0005634866932442624,
-      "loss": 4.0193,
+      "epoch": 0.6199460916442049,
+      "grad_norm": 0.6254762411117554,
+      "learning_rate": 0.0005634365893146248,
+      "loss": 4.0458,
       "step": 5750
     },
     {
-      "epoch": 0.6242600365945539,
-      "grad_norm": 0.7371019124984741,
-      "learning_rate": 0.000563163452214201,
-      "loss": 3.9911,
+      "epoch": 0.6253369272237197,
+      "grad_norm": 0.5808873772621155,
+      "learning_rate": 0.0005631127900701565,
+      "loss": 4.0354,
       "step": 5800
     },
     {
-      "epoch": 0.6296415886341621,
-      "grad_norm": 0.5817407369613647,
-      "learning_rate": 0.0005628402111841396,
-      "loss": 4.0193,
+      "epoch": 0.6307277628032345,
+      "grad_norm": 0.650372326374054,
+      "learning_rate": 0.000562788990825688,
+      "loss": 4.0449,
       "step": 5850
     },
     {
-      "epoch": 0.6350231406737703,
-      "grad_norm": 0.6328310966491699,
-      "learning_rate": 0.0005625169701540782,
-      "loss": 3.9958,
+      "epoch": 0.6361185983827493,
+      "grad_norm": 0.6067761182785034,
+      "learning_rate": 0.0005624651915812196,
+      "loss": 4.014,
       "step": 5900
     },
     {
-      "epoch": 0.6404046927133785,
-      "grad_norm": 0.7258861064910889,
-      "learning_rate": 0.0005621937291240168,
-      "loss": 3.9997,
+      "epoch": 0.6415094339622641,
+      "grad_norm": 0.6532148718833923,
+      "learning_rate": 0.0005621413923367511,
+      "loss": 4.0074,
       "step": 5950
     },
     {
-      "epoch": 0.6457862447529867,
-      "grad_norm": 0.6138096451759338,
-      "learning_rate": 0.0005618704880939553,
-      "loss": 3.9907,
+      "epoch": 0.6469002695417789,
+      "grad_norm": 0.6020557284355164,
+      "learning_rate": 0.0005618175930922828,
+      "loss": 4.0178,
       "step": 6000
     },
     {
-      "epoch": 0.6457862447529867,
-      "eval_accuracy": 0.32727434126198807,
-      "eval_loss": 3.926227331161499,
-      "eval_runtime": 183.661,
-      "eval_samples_per_second": 98.067,
-      "eval_steps_per_second": 6.131,
+      "epoch": 0.6469002695417789,
+      "eval_accuracy": 0.3249853508744004,
+      "eval_loss": 3.9412128925323486,
+      "eval_runtime": 153.3779,
+      "eval_samples_per_second": 117.429,
+      "eval_steps_per_second": 7.341,
       "step": 6000
     },
     {
-      "epoch": 0.651167796792595,
-      "grad_norm": 0.6955623626708984,
-      "learning_rate": 0.0005615472470638939,
-      "loss": 3.9931,
+      "epoch": 0.6522911051212938,
+      "grad_norm": 0.6567849516868591,
+      "learning_rate": 0.0005614937938478143,
+      "loss": 3.9889,
       "step": 6050
     },
     {
-      "epoch": 0.6565493488322032,
-      "grad_norm": 0.5675811767578125,
-      "learning_rate": 0.0005612240060338325,
-      "loss": 3.9823,
+      "epoch": 0.6576819407008087,
+      "grad_norm": 0.7067267894744873,
+      "learning_rate": 0.0005611699946033459,
+      "loss": 4.0251,
       "step": 6100
     },
     {
-      "epoch": 0.6619309008718114,
-      "grad_norm": 0.6101326942443848,
-      "learning_rate": 0.0005609007650037711,
-      "loss": 3.9581,
+      "epoch": 0.6630727762803235,
+      "grad_norm": 0.5447336435317993,
+      "learning_rate": 0.0005608461953588774,
+      "loss": 4.0062,
       "step": 6150
     },
     {
-      "epoch": 0.6673124529114196,
-      "grad_norm": 0.5811055302619934,
-      "learning_rate": 0.0005605775239737097,
-      "loss": 3.9611,
+      "epoch": 0.6684636118598383,
+      "grad_norm": 0.6051011681556702,
+      "learning_rate": 0.000560522396114409,
+      "loss": 3.9816,
       "step": 6200
     },
     {
-      "epoch": 0.6726940049510278,
-      "grad_norm": 0.6250592470169067,
-      "learning_rate": 0.0005602542829436483,
-      "loss": 3.9918,
+      "epoch": 0.6738544474393531,
+      "grad_norm": 0.6470229625701904,
+      "learning_rate": 0.0005601985968699405,
+      "loss": 3.9996,
       "step": 6250
     },
     {
-      "epoch": 0.6780755569906362,
-      "grad_norm": 0.6698481440544128,
-      "learning_rate": 0.0005599310419135868,
-      "loss": 3.9753,
+      "epoch": 0.6792452830188679,
+      "grad_norm": 0.699021577835083,
+      "learning_rate": 0.0005598747976254721,
+      "loss": 4.0176,
       "step": 6300
     },
     {
-      "epoch": 0.6834571090302444,
-      "grad_norm": 0.682300865650177,
-      "learning_rate": 0.0005596078008835255,
-      "loss": 3.9756,
+      "epoch": 0.6846361185983828,
+      "grad_norm": 0.6702553629875183,
+      "learning_rate": 0.0005595509983810038,
+      "loss": 3.9787,
       "step": 6350
     },
     {
-      "epoch": 0.6888386610698526,
-      "grad_norm": 0.5564926266670227,
-      "learning_rate": 0.000559284559853464,
-      "loss": 3.9725,
+      "epoch": 0.6900269541778976,
+      "grad_norm": 0.5683832764625549,
+      "learning_rate": 0.0005592271991365353,
+      "loss": 3.9837,
       "step": 6400
     },
     {
-      "epoch": 0.6942202131094608,
-      "grad_norm": 0.5958801507949829,
-      "learning_rate": 0.0005589613188234026,
-      "loss": 3.9805,
+      "epoch": 0.6954177897574124,
+      "grad_norm": 0.6393706202507019,
+      "learning_rate": 0.0005589033998920669,
+      "loss": 3.9689,
       "step": 6450
     },
     {
-      "epoch": 0.699601765149069,
-      "grad_norm": 0.6558675169944763,
-      "learning_rate": 0.0005586380777933412,
-      "loss": 3.9501,
+      "epoch": 0.7008086253369272,
+      "grad_norm": 0.6676239967346191,
+      "learning_rate": 0.0005585796006475984,
+      "loss": 3.983,
       "step": 6500
     },
     {
-      "epoch": 0.7049833171886772,
-      "grad_norm": 0.5374082922935486,
-      "learning_rate": 0.0005583148367632797,
-      "loss": 3.9658,
+      "epoch": 0.706199460916442,
+      "grad_norm": 0.5636298060417175,
+      "learning_rate": 0.0005582558014031301,
+      "loss": 4.0066,
       "step": 6550
     },
     {
-      "epoch": 0.7103648692282855,
-      "grad_norm": 0.5923078656196594,
-      "learning_rate": 0.0005579915957332184,
-      "loss": 3.9615,
+      "epoch": 0.7115902964959568,
+      "grad_norm": 0.618071973323822,
+      "learning_rate": 0.0005579320021586616,
+      "loss": 3.9622,
       "step": 6600
     },
     {
-      "epoch": 0.7157464212678937,
-      "grad_norm": 0.635322093963623,
-      "learning_rate": 0.0005576683547031569,
-      "loss": 3.964,
+      "epoch": 0.7169811320754716,
+      "grad_norm": 0.5887283086776733,
+      "learning_rate": 0.0005576082029141932,
+      "loss": 3.9926,
       "step": 6650
     },
     {
-      "epoch": 0.7211279733075019,
-      "grad_norm": 0.6079649329185486,
-      "learning_rate": 0.0005573451136730956,
-      "loss": 3.9573,
+      "epoch": 0.7223719676549866,
+      "grad_norm": 0.6095805168151855,
+      "learning_rate": 0.0005572844036697247,
+      "loss": 3.9656,
       "step": 6700
     },
     {
-      "epoch": 0.7265095253471101,
-      "grad_norm": 0.6069414615631104,
-      "learning_rate": 0.0005570218726430341,
-      "loss": 3.918,
+      "epoch": 0.7277628032345014,
+      "grad_norm": 0.5963723063468933,
+      "learning_rate": 0.0005569606044252563,
+      "loss": 3.9725,
       "step": 6750
     },
     {
-      "epoch": 0.7318910773867183,
-      "grad_norm": 0.6097965240478516,
-      "learning_rate": 0.0005566986316129728,
-      "loss": 3.9562,
+      "epoch": 0.7331536388140162,
+      "grad_norm": 0.6220936179161072,
+      "learning_rate": 0.0005566368051807879,
+      "loss": 3.9604,
       "step": 6800
     },
     {
-      "epoch": 0.7372726294263265,
-      "grad_norm": 0.6329127550125122,
-      "learning_rate": 0.0005563753905829113,
-      "loss": 3.9635,
+      "epoch": 0.738544474393531,
+      "grad_norm": 0.7677363753318787,
+      "learning_rate": 0.0005563130059363194,
+      "loss": 3.9399,
       "step": 6850
     },
     {
-      "epoch": 0.7426541814659348,
-      "grad_norm": 0.5949454307556152,
-      "learning_rate": 0.0005560521495528498,
-      "loss": 3.9418,
+      "epoch": 0.7439353099730458,
+      "grad_norm": 0.7177647352218628,
+      "learning_rate": 0.000555989206691851,
+      "loss": 3.9652,
       "step": 6900
     },
     {
-      "epoch": 0.748035733505543,
-      "grad_norm": 0.6039018630981445,
-      "learning_rate": 0.0005557289085227884,
-      "loss": 3.9432,
+      "epoch": 0.7493261455525606,
+      "grad_norm": 0.5666942596435547,
+      "learning_rate": 0.0005556654074473826,
+      "loss": 3.9621,
       "step": 6950
     },
     {
-      "epoch": 0.7534172855451512,
-      "grad_norm": 0.6565973162651062,
-      "learning_rate": 0.000555405667492727,
-      "loss": 3.9245,
+      "epoch": 0.7547169811320755,
+      "grad_norm": 0.6877643465995789,
+      "learning_rate": 0.0005553416082029141,
+      "loss": 3.9497,
       "step": 7000
     },
     {
-      "epoch": 0.7534172855451512,
-      "eval_accuracy": 0.33284975554730034,
-      "eval_loss": 3.869795799255371,
-      "eval_runtime": 183.8335,
-      "eval_samples_per_second": 97.975,
-      "eval_steps_per_second": 6.125,
+      "epoch": 0.7547169811320755,
+      "eval_accuracy": 0.33131622868868676,
+      "eval_loss": 3.8834235668182373,
+      "eval_runtime": 153.3456,
+      "eval_samples_per_second": 117.454,
+      "eval_steps_per_second": 7.343,
       "step": 7000
     },
     {
-      "epoch": 0.7587988375847594,
-      "grad_norm": 0.5873769521713257,
-      "learning_rate": 0.0005550824264626657,
-      "loss": 3.9419,
+      "epoch": 0.7601078167115903,
+      "grad_norm": 0.6757393479347229,
+      "learning_rate": 0.0005550178089584457,
+      "loss": 3.9406,
       "step": 7050
     },
     {
-      "epoch": 0.7641803896243676,
-      "grad_norm": 0.5975030660629272,
-      "learning_rate": 0.0005547591854326042,
-      "loss": 3.9205,
+      "epoch": 0.7654986522911051,
+      "grad_norm": 0.6207107901573181,
+      "learning_rate": 0.0005546940097139772,
+      "loss": 3.9529,
       "step": 7100
     },
     {
-      "epoch": 0.7695619416639758,
-      "grad_norm": 0.7010316252708435,
-      "learning_rate": 0.0005544359444025428,
-      "loss": 3.921,
+      "epoch": 0.77088948787062,
+      "grad_norm": 0.5236271619796753,
+      "learning_rate": 0.0005543702104695089,
+      "loss": 3.9499,
       "step": 7150
     },
     {
-      "epoch": 0.7749434937035841,
-      "grad_norm": 0.6028709411621094,
-      "learning_rate": 0.0005541127033724813,
-      "loss": 3.9054,
+      "epoch": 0.7762803234501348,
+      "grad_norm": 0.6513909101486206,
+      "learning_rate": 0.0005540464112250404,
+      "loss": 3.9531,
       "step": 7200
     },
     {
-      "epoch": 0.7803250457431924,
-      "grad_norm": 0.6167606711387634,
-      "learning_rate": 0.0005537894623424199,
-      "loss": 3.9206,
+      "epoch": 0.7816711590296496,
+      "grad_norm": 0.5805194973945618,
+      "learning_rate": 0.000553722611980572,
+      "loss": 3.91,
       "step": 7250
     },
     {
-      "epoch": 0.7857065977828006,
-      "grad_norm": 0.6583150029182434,
-      "learning_rate": 0.0005534662213123586,
-      "loss": 3.9263,
+      "epoch": 0.7870619946091644,
+      "grad_norm": 0.6637202501296997,
+      "learning_rate": 0.0005533988127361035,
+      "loss": 3.948,
       "step": 7300
     },
     {
-      "epoch": 0.7910881498224088,
-      "grad_norm": 0.6138869524002075,
-      "learning_rate": 0.0005531429802822971,
-      "loss": 3.9296,
+      "epoch": 0.7924528301886793,
+      "grad_norm": 0.558056116104126,
+      "learning_rate": 0.0005530750134916352,
+      "loss": 3.9333,
       "step": 7350
     },
     {
-      "epoch": 0.796469701862017,
-      "grad_norm": 0.7079772353172302,
-      "learning_rate": 0.0005528197392522357,
-      "loss": 3.9301,
+      "epoch": 0.7978436657681941,
+      "grad_norm": 0.7991946339607239,
+      "learning_rate": 0.0005527512142471668,
+      "loss": 3.9324,
       "step": 7400
     },
     {
-      "epoch": 0.8018512539016253,
-      "grad_norm": 0.6088504195213318,
-      "learning_rate": 0.0005524964982221743,
-      "loss": 3.9284,
+      "epoch": 0.8032345013477089,
+      "grad_norm": 0.5782480835914612,
+      "learning_rate": 0.0005524274150026982,
+      "loss": 3.9235,
       "step": 7450
     },
     {
-      "epoch": 0.8072328059412335,
-      "grad_norm": 0.5483489632606506,
-      "learning_rate": 0.0005521732571921129,
-      "loss": 3.8974,
+      "epoch": 0.8086253369272237,
+      "grad_norm": 0.661090612411499,
+      "learning_rate": 0.0005521036157582299,
+      "loss": 3.9098,
       "step": 7500
     },
     {
-      "epoch": 0.8126143579808417,
-      "grad_norm": 0.6508141160011292,
-      "learning_rate": 0.0005518500161620514,
-      "loss": 3.916,
+      "epoch": 0.8140161725067385,
+      "grad_norm": 0.6254606246948242,
+      "learning_rate": 0.0005517798165137614,
+      "loss": 3.936,
       "step": 7550
     },
     {
-      "epoch": 0.8179959100204499,
-      "grad_norm": 0.5982174277305603,
-      "learning_rate": 0.00055152677513199,
-      "loss": 3.9016,
+      "epoch": 0.8194070080862533,
+      "grad_norm": 0.5371659398078918,
+      "learning_rate": 0.000551456017269293,
+      "loss": 3.938,
       "step": 7600
     },
     {
-      "epoch": 0.8233774620600581,
-      "grad_norm": 0.5615628361701965,
-      "learning_rate": 0.0005512035341019286,
-      "loss": 3.9112,
+      "epoch": 0.8247978436657682,
+      "grad_norm": 0.644661009311676,
+      "learning_rate": 0.0005511322180248245,
+      "loss": 3.9149,
       "step": 7650
     },
     {
-      "epoch": 0.8287590140996663,
-      "grad_norm": 0.6235767006874084,
-      "learning_rate": 0.0005508802930718672,
-      "loss": 3.9002,
+      "epoch": 0.8301886792452831,
+      "grad_norm": 0.6013936996459961,
+      "learning_rate": 0.0005508084187803562,
+      "loss": 3.9182,
       "step": 7700
     },
     {
-      "epoch": 0.8341405661392746,
-      "grad_norm": 0.6372842788696289,
-      "learning_rate": 0.0005505570520418058,
-      "loss": 3.9002,
+      "epoch": 0.8355795148247979,
+      "grad_norm": 0.5374178886413574,
+      "learning_rate": 0.0005504846195358877,
+      "loss": 3.9007,
       "step": 7750
     },
     {
-      "epoch": 0.8395221181788828,
-      "grad_norm": 0.5565422773361206,
-      "learning_rate": 0.0005502338110117443,
-      "loss": 3.886,
+      "epoch": 0.8409703504043127,
+      "grad_norm": 0.6176519393920898,
+      "learning_rate": 0.0005501608202914193,
+      "loss": 3.9278,
       "step": 7800
     },
     {
-      "epoch": 0.844903670218491,
-      "grad_norm": 0.5980644822120667,
-      "learning_rate": 0.000549910569981683,
-      "loss": 3.8933,
+      "epoch": 0.8463611859838275,
+      "grad_norm": 0.6597527861595154,
+      "learning_rate": 0.0005498370210469508,
+      "loss": 3.9062,
       "step": 7850
     },
     {
-      "epoch": 0.8502852222580992,
-      "grad_norm": 0.5744678378105164,
-      "learning_rate": 0.0005495873289516215,
-      "loss": 3.8747,
+      "epoch": 0.8517520215633423,
+      "grad_norm": 0.5706176161766052,
+      "learning_rate": 0.0005495132218024824,
+      "loss": 3.9084,
       "step": 7900
     },
     {
-      "epoch": 0.8556667742977074,
-      "grad_norm": 0.5711344480514526,
-      "learning_rate": 0.0005492640879215602,
-      "loss": 3.8882,
+      "epoch": 0.8571428571428571,
+      "grad_norm": 0.6241531372070312,
+      "learning_rate": 0.000549189422558014,
+      "loss": 3.9047,
       "step": 7950
     },
     {
-      "epoch": 0.8610483263373157,
-      "grad_norm": 0.559182345867157,
-      "learning_rate": 0.0005489408468914987,
-      "loss": 3.8694,
+      "epoch": 0.862533692722372,
+      "grad_norm": 0.7237799763679504,
+      "learning_rate": 0.0005488656233135456,
+      "loss": 3.8934,
       "step": 8000
     },
     {
-      "epoch": 0.8610483263373157,
-      "eval_accuracy": 0.3374283882398026,
-      "eval_loss": 3.820354461669922,
-      "eval_runtime": 183.958,
-      "eval_samples_per_second": 97.908,
-      "eval_steps_per_second": 6.121,
+      "epoch": 0.862533692722372,
+      "eval_accuracy": 0.33549306300205106,
+      "eval_loss": 3.8358263969421387,
+      "eval_runtime": 153.0976,
+      "eval_samples_per_second": 117.644,
+      "eval_steps_per_second": 7.355,
       "step": 8000
     },
     {
-      "epoch": 0.8664298783769239,
-      "grad_norm": 0.5815455317497253,
-      "learning_rate": 0.0005486176058614372,
-      "loss": 3.8719,
+      "epoch": 0.8679245283018868,
+      "grad_norm": 0.5674304366111755,
+      "learning_rate": 0.0005485418240690771,
+      "loss": 3.9022,
       "step": 8050
     },
     {
-      "epoch": 0.8718114304165321,
-      "grad_norm": 0.556144654750824,
-      "learning_rate": 0.0005482943648313759,
-      "loss": 3.8804,
+      "epoch": 0.8733153638814016,
+      "grad_norm": 0.6205889582633972,
+      "learning_rate": 0.0005482180248246087,
+      "loss": 3.8871,
       "step": 8100
     },
     {
-      "epoch": 0.8771929824561403,
-      "grad_norm": 0.5465092658996582,
-      "learning_rate": 0.0005479711238013145,
-      "loss": 3.8732,
+      "epoch": 0.8787061994609164,
+      "grad_norm": 0.5285912156105042,
+      "learning_rate": 0.0005478942255801403,
+      "loss": 3.8994,
       "step": 8150
     },
     {
-      "epoch": 0.8825745344957485,
-      "grad_norm": 0.6253562569618225,
-      "learning_rate": 0.0005476478827712531,
-      "loss": 3.8984,
+      "epoch": 0.8840970350404312,
+      "grad_norm": 0.5494291186332703,
+      "learning_rate": 0.0005475704263356718,
+      "loss": 3.8885,
       "step": 8200
     },
     {
-      "epoch": 0.8879560865353568,
-      "grad_norm": 0.5278693437576294,
-      "learning_rate": 0.0005473246417411916,
-      "loss": 3.8743,
+      "epoch": 0.889487870619946,
+      "grad_norm": 0.6058967709541321,
+      "learning_rate": 0.0005472466270912034,
+      "loss": 3.8736,
       "step": 8250
     },
     {
-      "epoch": 0.8933376385749651,
-      "grad_norm": 0.6014917492866516,
-      "learning_rate": 0.0005470014007111302,
-      "loss": 3.8721,
+      "epoch": 0.894878706199461,
+      "grad_norm": 0.6890183687210083,
+      "learning_rate": 0.000546922827846735,
+      "loss": 3.8986,
       "step": 8300
     },
     {
-      "epoch": 0.8987191906145733,
-      "grad_norm": 0.5801762342453003,
-      "learning_rate": 0.0005466781596810688,
-      "loss": 3.891,
+      "epoch": 0.9002695417789758,
+      "grad_norm": 0.625860333442688,
+      "learning_rate": 0.0005465990286022665,
+      "loss": 3.8816,
       "step": 8350
     },
     {
-      "epoch": 0.9041007426541815,
-      "grad_norm": 0.559445858001709,
-      "learning_rate": 0.0005463549186510073,
-      "loss": 3.8685,
+      "epoch": 0.9056603773584906,
+      "grad_norm": 0.6296173334121704,
+      "learning_rate": 0.0005462752293577981,
+      "loss": 3.8932,
       "step": 8400
     },
     {
-      "epoch": 0.9094822946937897,
-      "grad_norm": 0.5420745015144348,
-      "learning_rate": 0.000546031677620946,
-      "loss": 3.8733,
+      "epoch": 0.9110512129380054,
+      "grad_norm": 0.5746651291847229,
+      "learning_rate": 0.0005459514301133296,
+      "loss": 3.9057,
       "step": 8450
     },
     {
-      "epoch": 0.9148638467333979,
-      "grad_norm": 0.5471014380455017,
-      "learning_rate": 0.0005457084365908845,
-      "loss": 3.8714,
+      "epoch": 0.9164420485175202,
+      "grad_norm": 0.6228104829788208,
+      "learning_rate": 0.0005456276308688613,
+      "loss": 3.8985,
       "step": 8500
     },
     {
-      "epoch": 0.9202453987730062,
-      "grad_norm": 0.5171864032745361,
-      "learning_rate": 0.0005453851955608232,
-      "loss": 3.8572,
+      "epoch": 0.921832884097035,
+      "grad_norm": 0.5756477117538452,
+      "learning_rate": 0.0005453038316243929,
+      "loss": 3.8781,
       "step": 8550
     },
     {
-      "epoch": 0.9256269508126144,
-      "grad_norm": 0.5859182476997375,
-      "learning_rate": 0.0005450619545307617,
-      "loss": 3.8806,
+      "epoch": 0.9272237196765498,
+      "grad_norm": 0.5988864302635193,
+      "learning_rate": 0.0005449800323799244,
+      "loss": 3.8724,
       "step": 8600
     },
     {
-      "epoch": 0.9310085028522226,
-      "grad_norm": 0.5607949495315552,
-      "learning_rate": 0.0005447387135007003,
-      "loss": 3.8568,
+      "epoch": 0.9326145552560647,
+      "grad_norm": 0.5490165948867798,
+      "learning_rate": 0.000544656233135456,
+      "loss": 3.8638,
       "step": 8650
     },
     {
-      "epoch": 0.9363900548918308,
-      "grad_norm": 0.5764897465705872,
-      "learning_rate": 0.0005444154724706389,
-      "loss": 3.8628,
+      "epoch": 0.9380053908355795,
+      "grad_norm": 0.5722247362136841,
+      "learning_rate": 0.0005443324338909875,
+      "loss": 3.8698,
       "step": 8700
     },
     {
-      "epoch": 0.941771606931439,
-      "grad_norm": 0.5329459309577942,
-      "learning_rate": 0.0005440922314405775,
-      "loss": 3.8604,
+      "epoch": 0.9433962264150944,
+      "grad_norm": 0.592191755771637,
+      "learning_rate": 0.0005440086346465192,
+      "loss": 3.8706,
       "step": 8750
     },
     {
-      "epoch": 0.9471531589710472,
-      "grad_norm": 0.712491512298584,
-      "learning_rate": 0.0005437689904105161,
-      "loss": 3.8376,
+      "epoch": 0.9487870619946092,
+      "grad_norm": 0.5509844422340393,
+      "learning_rate": 0.0005436848354020506,
+      "loss": 3.8615,
       "step": 8800
     },
     {
-      "epoch": 0.9525347110106555,
-      "grad_norm": 0.6513589024543762,
-      "learning_rate": 0.0005434457493804546,
-      "loss": 3.8705,
+      "epoch": 0.954177897574124,
+      "grad_norm": 0.6250285506248474,
+      "learning_rate": 0.0005433610361575823,
+      "loss": 3.8504,
       "step": 8850
     },
     {
-      "epoch": 0.9579162630502637,
-      "grad_norm": 0.5159483551979065,
-      "learning_rate": 0.0005431225083503932,
-      "loss": 3.8302,
+      "epoch": 0.9595687331536388,
+      "grad_norm": 0.5853751301765442,
+      "learning_rate": 0.0005430372369131138,
+      "loss": 3.8685,
       "step": 8900
     },
     {
-      "epoch": 0.9632978150898719,
-      "grad_norm": 0.619147539138794,
-      "learning_rate": 0.0005427992673203318,
-      "loss": 3.8394,
+      "epoch": 0.9649595687331537,
+      "grad_norm": 0.6114296913146973,
+      "learning_rate": 0.0005427134376686454,
+      "loss": 3.8746,
       "step": 8950
     },
     {
-      "epoch": 0.9686793671294801,
-      "grad_norm": 0.5892804861068726,
-      "learning_rate": 0.0005424760262902704,
-      "loss": 3.8479,
+      "epoch": 0.9703504043126685,
+      "grad_norm": 0.5549982190132141,
+      "learning_rate": 0.0005423896384241769,
+      "loss": 3.8623,
       "step": 9000
     },
     {
-      "epoch": 0.9686793671294801,
-      "eval_accuracy": 0.34035245475666587,
-      "eval_loss": 3.7828423976898193,
-      "eval_runtime": 183.4819,
-      "eval_samples_per_second": 98.162,
-      "eval_steps_per_second": 6.137,
+      "epoch": 0.9703504043126685,
+      "eval_accuracy": 0.33976007921230134,
+      "eval_loss": 3.796833038330078,
+      "eval_runtime": 153.5005,
+      "eval_samples_per_second": 117.335,
+      "eval_steps_per_second": 7.335,
       "step": 9000
     },
     {
-      "epoch": 0.9740609191690883,
-      "grad_norm": 0.5918485522270203,
-      "learning_rate": 0.000542152785260209,
-      "loss": 3.8558,
+      "epoch": 0.9757412398921833,
+      "grad_norm": 0.5819941759109497,
+      "learning_rate": 0.0005420658391797086,
+      "loss": 3.8561,
       "step": 9050
     },
     {
-      "epoch": 0.9794424712086965,
-      "grad_norm": 0.48340779542922974,
-      "learning_rate": 0.0005418295442301476,
-      "loss": 3.8433,
+      "epoch": 0.9811320754716981,
+      "grad_norm": 0.6158163547515869,
+      "learning_rate": 0.0005417420399352401,
+      "loss": 3.8681,
       "step": 9100
     },
     {
-      "epoch": 0.9848240232483048,
-      "grad_norm": 0.559395968914032,
-      "learning_rate": 0.0005415063032000861,
-      "loss": 3.847,
+      "epoch": 0.9865229110512129,
+      "grad_norm": 0.5715709924697876,
+      "learning_rate": 0.0005414182406907717,
+      "loss": 3.8422,
       "step": 9150
     },
     {
-      "epoch": 0.9902055752879131,
-      "grad_norm": 0.5653789043426514,
-      "learning_rate": 0.0005411830621700248,
-      "loss": 3.8355,
+      "epoch": 0.9919137466307277,
+      "grad_norm": 0.5587881207466125,
+      "learning_rate": 0.0005410944414463032,
+      "loss": 3.8401,
       "step": 9200
     },
     {
-      "epoch": 0.9955871273275213,
-      "grad_norm": 0.5224032402038574,
-      "learning_rate": 0.0005408598211399633,
-      "loss": 3.8249,
+      "epoch": 0.9973045822102425,
+      "grad_norm": 0.5460543036460876,
+      "learning_rate": 0.0005407706422018348,
+      "loss": 3.8508,
       "step": 9250
     },
     {
-      "epoch": 1.0009686793671295,
-      "grad_norm": 0.572032630443573,
-      "learning_rate": 0.0005405365801099019,
-      "loss": 3.828,
+      "epoch": 1.0026954177897573,
+      "grad_norm": 0.6066132187843323,
+      "learning_rate": 0.0005404468429573664,
+      "loss": 3.8255,
       "step": 9300
     },
     {
-      "epoch": 1.0063502314067376,
-      "grad_norm": 0.5785325169563293,
-      "learning_rate": 0.0005402133390798405,
-      "loss": 3.7596,
+      "epoch": 1.0080862533692723,
+      "grad_norm": 0.5967060327529907,
+      "learning_rate": 0.000540123043712898,
+      "loss": 3.7934,
       "step": 9350
     },
     {
-      "epoch": 1.011731783446346,
-      "grad_norm": 0.5417508482933044,
-      "learning_rate": 0.000539890098049779,
-      "loss": 3.7565,
+      "epoch": 1.013477088948787,
+      "grad_norm": 0.5728771686553955,
+      "learning_rate": 0.0005397992444684295,
+      "loss": 3.7864,
       "step": 9400
     },
     {
-      "epoch": 1.017113335485954,
-      "grad_norm": 0.6021242141723633,
-      "learning_rate": 0.0005395668570197177,
-      "loss": 3.7638,
+      "epoch": 1.0188679245283019,
+      "grad_norm": 0.6757526993751526,
+      "learning_rate": 0.0005394754452239611,
+      "loss": 3.7864,
       "step": 9450
     },
     {
-      "epoch": 1.0224948875255624,
-      "grad_norm": 0.5668533444404602,
-      "learning_rate": 0.0005392436159896562,
-      "loss": 3.7673,
+      "epoch": 1.0242587601078168,
+      "grad_norm": 0.6057505011558533,
+      "learning_rate": 0.0005391516459794927,
+      "loss": 3.7886,
       "step": 9500
     },
     {
-      "epoch": 1.0278764395651705,
-      "grad_norm": 0.5705947279930115,
-      "learning_rate": 0.0005389203749595948,
-      "loss": 3.7738,
+      "epoch": 1.0296495956873315,
+      "grad_norm": 0.5644155740737915,
+      "learning_rate": 0.0005388278467350242,
+      "loss": 3.787,
       "step": 9550
     },
     {
-      "epoch": 1.0332579916047788,
-      "grad_norm": 0.5880433320999146,
-      "learning_rate": 0.0005385971339295334,
-      "loss": 3.7634,
+      "epoch": 1.0350404312668464,
+      "grad_norm": 0.6167004108428955,
+      "learning_rate": 0.0005385040474905557,
+      "loss": 3.7731,
       "step": 9600
     },
     {
-      "epoch": 1.0386395436443872,
-      "grad_norm": 0.629132866859436,
-      "learning_rate": 0.000538273892899472,
-      "loss": 3.7783,
+      "epoch": 1.0404312668463611,
+      "grad_norm": 0.6160714626312256,
+      "learning_rate": 0.0005381802482460874,
+      "loss": 3.8133,
       "step": 9650
     },
     {
-      "epoch": 1.0440210956839953,
-      "grad_norm": 0.6227521300315857,
-      "learning_rate": 0.0005379506518694106,
-      "loss": 3.7852,
+      "epoch": 1.045822102425876,
+      "grad_norm": 0.6024379134178162,
+      "learning_rate": 0.000537856449001619,
+      "loss": 3.8009,
       "step": 9700
     },
     {
-      "epoch": 1.0494026477236036,
-      "grad_norm": 0.5902397632598877,
-      "learning_rate": 0.0005376274108393491,
-      "loss": 3.7773,
+      "epoch": 1.0512129380053907,
+      "grad_norm": 0.5305362343788147,
+      "learning_rate": 0.0005375326497571505,
+      "loss": 3.7983,
       "step": 9750
     },
     {
-      "epoch": 1.0547841997632117,
-      "grad_norm": 0.5385565161705017,
-      "learning_rate": 0.0005373041698092877,
-      "loss": 3.7692,
+      "epoch": 1.0566037735849056,
+      "grad_norm": 0.6367751955986023,
+      "learning_rate": 0.000537208850512682,
+      "loss": 3.8077,
       "step": 9800
     },
     {
-      "epoch": 1.06016575180282,
-      "grad_norm": 0.5663689374923706,
-      "learning_rate": 0.0005369809287792263,
-      "loss": 3.7817,
+      "epoch": 1.0619946091644206,
+      "grad_norm": 0.53994220495224,
+      "learning_rate": 0.0005368850512682137,
+      "loss": 3.8023,
       "step": 9850
     },
     {
-      "epoch": 1.0655473038424281,
-      "grad_norm": 0.5260084867477417,
-      "learning_rate": 0.000536657687749165,
-      "loss": 3.7742,
+      "epoch": 1.0673854447439353,
+      "grad_norm": 0.5434533953666687,
+      "learning_rate": 0.0005365612520237453,
+      "loss": 3.7914,
       "step": 9900
     },
     {
-      "epoch": 1.0709288558820365,
-      "grad_norm": 0.5762213468551636,
-      "learning_rate": 0.0005363344467191035,
-      "loss": 3.7833,
+      "epoch": 1.0727762803234502,
+      "grad_norm": 0.6101065278053284,
+      "learning_rate": 0.0005362374527792768,
+      "loss": 3.788,
       "step": 9950
     },
     {
-      "epoch": 1.0763104079216446,
-      "grad_norm": 0.6788592338562012,
-      "learning_rate": 0.000536011205689042,
-      "loss": 3.7598,
+      "epoch": 1.0781671159029649,
+      "grad_norm": 0.6409618258476257,
+      "learning_rate": 0.0005359136535348084,
+      "loss": 3.7815,
       "step": 10000
     },
     {
-      "epoch": 1.0763104079216446,
-      "eval_accuracy": 0.34371200204788965,
-      "eval_loss": 3.7531447410583496,
-      "eval_runtime": 183.7635,
-      "eval_samples_per_second": 98.012,
-      "eval_steps_per_second": 6.127,
+      "epoch": 1.0781671159029649,
+      "eval_accuracy": 0.3427394500490622,
+      "eval_loss": 3.7646424770355225,
+      "eval_runtime": 153.0954,
+      "eval_samples_per_second": 117.646,
+      "eval_steps_per_second": 7.355,
       "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:f2b29136d15578405af07a641ddcf73a5754a95ec72bcf6dd9a64313eeb6b854
+oid sha256:03acfd6b89028e5050aa86a8794f9f5dee8ad92e507888c3f06d8005b87b81ce
 size 5304