Training in progress, step 10000, checkpoint

checkpoint-10000/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0e6c6c99cad3852e0f0054fae9887041ed563af6729b2bcf5931e225bcf0660c
+oid sha256:bda902a1ecab3327b631fa275c7ef42152ea06f97d3599b9a4e20f02e8da4890
 size 503128704

checkpoint-10000/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7dac54ccd7991b0f3ed8ee341ced9970de80dc1f8f3a513a99f7e93bdd9783af
+oid sha256:b095bf350dbdadc285de7e4c92935649c90b1d2a6d640e206360c99510343584
 size 1006351290

checkpoint-10000/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:56c6e19448573c7e2c1efb3c4876abd43251e67c2c88cba662ec83de112885c2
+oid sha256:53a47b99e43a46630effbfb45e7223c8a493d5b48b00aebac2470d03eafb7364
 size 14244

checkpoint-10000/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d8b563a93d2aa94339a2fe2f149c5bc78a6bdd0db192e77a5fe707f1d03f0d44
+oid sha256:baaddc0dc177f093cb7927eae851400a36fde6f45129bcaf2df443bdde5269f4
 size 1064

checkpoint-10000/trainer_state.json

@@ -1,7 +1,7 @@
 {
-  "best_metric": 3.7506744861602783,
-  "best_model_checkpoint": "/scratch/cl5625/exceptions/models/100M_1208/checkpoint-10000",
-  "epoch": 1.0763104079216446,
+  "best_metric": 3.7468345165252686,
+  "best_model_checkpoint": "/scratch/cl5625/exceptions/models/100M_high_2000_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": 1.6288660764694214,
-      "learning_rate": 0.000276,
-      "loss": 8.759,
+      "epoch": 0.005390835579514825,
+      "grad_norm": 1.6212843656539917,
+      "learning_rate": 0.0003,
+      "loss": 8.7944,
       "step": 50
     },
     {
-      "epoch": 0.010763104079216447,
-      "grad_norm": 8.2901611328125,
-      "learning_rate": 0.0005759999999999999,
-      "loss": 7.0168,
+      "epoch": 0.01078167115902965,
+      "grad_norm": 2.362013101577759,
+      "learning_rate": 0.0006,
+      "loss": 6.9994,
       "step": 100
     },
     {
-      "epoch": 0.01614465611882467,
-      "grad_norm": 1.3438377380371094,
-      "learning_rate": 0.0005997026182523434,
-      "loss": 6.6009,
+      "epoch": 0.016172506738544475,
+      "grad_norm": 1.731601357460022,
+      "learning_rate": 0.0005996762007555315,
+      "loss": 6.5225,
       "step": 150
     },
     {
-      "epoch": 0.021526208158432893,
-      "grad_norm": 1.8783726692199707,
-      "learning_rate": 0.000599379377222282,
-      "loss": 6.3134,
+      "epoch": 0.0215633423180593,
+      "grad_norm": 1.5711681842803955,
+      "learning_rate": 0.000599352401511063,
+      "loss": 6.2587,
       "step": 200
     },
     {
-      "epoch": 0.026907760198041114,
-      "grad_norm": 1.0426092147827148,
-      "learning_rate": 0.0005990561361922206,
-      "loss": 6.1356,
+      "epoch": 0.026954177897574125,
+      "grad_norm": 1.7903895378112793,
+      "learning_rate": 0.0005990286022665946,
+      "loss": 6.0733,
       "step": 250
     },
     {
-      "epoch": 0.03228931223764934,
-      "grad_norm": 1.021669864654541,
-      "learning_rate": 0.0005987328951621592,
-      "loss": 6.024,
+      "epoch": 0.03234501347708895,
+      "grad_norm": 0.9440882205963135,
+      "learning_rate": 0.0005987048030221263,
+      "loss": 5.9696,
       "step": 300
     },
     {
-      "epoch": 0.03767086427725756,
-      "grad_norm": 1.6005719900131226,
-      "learning_rate": 0.0005984096541320978,
-      "loss": 5.9093,
+      "epoch": 0.03773584905660377,
+      "grad_norm": 2.2796380519866943,
+      "learning_rate": 0.0005983810037776578,
+      "loss": 5.8928,
       "step": 350
     },
     {
-      "epoch": 0.04305241631686579,
-      "grad_norm": 1.0307564735412598,
-      "learning_rate": 0.0005980864131020364,
-      "loss": 5.8065,
+      "epoch": 0.0431266846361186,
+      "grad_norm": 1.00179123878479,
+      "learning_rate": 0.0005980572045331894,
+      "loss": 5.8188,
       "step": 400
     },
     {
-      "epoch": 0.048433968356474004,
-      "grad_norm": 1.7200894355773926,
-      "learning_rate": 0.0005977631720719749,
-      "loss": 5.7429,
+      "epoch": 0.04851752021563342,
+      "grad_norm": 1.4342464208602905,
+      "learning_rate": 0.0005977334052887209,
+      "loss": 5.713,
       "step": 450
     },
     {
-      "epoch": 0.05381552039608223,
-      "grad_norm": 1.1326223611831665,
-      "learning_rate": 0.0005974399310419136,
-      "loss": 5.6593,
+      "epoch": 0.05390835579514825,
+      "grad_norm": 1.159049153327942,
+      "learning_rate": 0.0005974096060442526,
+      "loss": 5.6576,
       "step": 500
     },
     {
-      "epoch": 0.05919707243569045,
-      "grad_norm": 1.3042094707489014,
-      "learning_rate": 0.0005971166900118521,
-      "loss": 5.587,
+      "epoch": 0.05929919137466307,
+      "grad_norm": 2.1126091480255127,
+      "learning_rate": 0.0005970858067997841,
+      "loss": 5.5631,
       "step": 550
     },
     {
-      "epoch": 0.06457862447529868,
-      "grad_norm": 0.9970183372497559,
-      "learning_rate": 0.0005967934489817907,
-      "loss": 5.5242,
+      "epoch": 0.0646900269541779,
+      "grad_norm": 0.9070990681648254,
+      "learning_rate": 0.0005967620075553157,
+      "loss": 5.5338,
       "step": 600
     },
     {
-      "epoch": 0.0699601765149069,
-      "grad_norm": 0.8722121119499207,
-      "learning_rate": 0.0005964702079517293,
-      "loss": 5.4456,
+      "epoch": 0.07008086253369272,
+      "grad_norm": 1.1499637365341187,
+      "learning_rate": 0.0005964382083108472,
+      "loss": 5.4333,
       "step": 650
     },
     {
-      "epoch": 0.07534172855451512,
-      "grad_norm": 1.1744099855422974,
-      "learning_rate": 0.0005961469669216679,
-      "loss": 5.409,
+      "epoch": 0.07547169811320754,
+      "grad_norm": 1.5855638980865479,
+      "learning_rate": 0.0005961144090663788,
+      "loss": 5.4047,
       "step": 700
     },
     {
-      "epoch": 0.08072328059412334,
-      "grad_norm": 1.2246752977371216,
-      "learning_rate": 0.0005958237258916065,
-      "loss": 5.3477,
+      "epoch": 0.08086253369272237,
+      "grad_norm": 0.9327629804611206,
+      "learning_rate": 0.0005957906098219104,
+      "loss": 5.3082,
       "step": 750
     },
     {
-      "epoch": 0.08610483263373157,
-      "grad_norm": 1.0104173421859741,
-      "learning_rate": 0.000595500484861545,
-      "loss": 5.2821,
+      "epoch": 0.0862533692722372,
+      "grad_norm": 1.2515809535980225,
+      "learning_rate": 0.0005954668105774419,
+      "loss": 5.2654,
       "step": 800
     },
     {
-      "epoch": 0.09148638467333979,
-      "grad_norm": 0.8681843876838684,
-      "learning_rate": 0.0005951772438314836,
-      "loss": 5.217,
+      "epoch": 0.09164420485175202,
+      "grad_norm": 1.3055354356765747,
+      "learning_rate": 0.0005951430113329735,
+      "loss": 5.2094,
       "step": 850
     },
     {
-      "epoch": 0.09686793671294801,
-      "grad_norm": 1.285335898399353,
-      "learning_rate": 0.0005948540028014222,
-      "loss": 5.164,
+      "epoch": 0.09703504043126684,
+      "grad_norm": 1.014483094215393,
+      "learning_rate": 0.0005948192120885051,
+      "loss": 5.183,
       "step": 900
     },
     {
-      "epoch": 0.10224948875255624,
-      "grad_norm": 1.4134410619735718,
-      "learning_rate": 0.0005945307617713608,
-      "loss": 5.1207,
+      "epoch": 0.10242587601078167,
+      "grad_norm": 1.166028618812561,
+      "learning_rate": 0.0005944954128440366,
+      "loss": 5.1241,
       "step": 950
     },
     {
-      "epoch": 0.10763104079216446,
-      "grad_norm": 1.1055577993392944,
-      "learning_rate": 0.0005942075207412994,
-      "loss": 5.081,
+      "epoch": 0.1078167115902965,
+      "grad_norm": 1.2375012636184692,
+      "learning_rate": 0.0005941716135995682,
+      "loss": 5.1101,
       "step": 1000
     },
     {
-      "epoch": 0.10763104079216446,
-      "eval_accuracy": 0.2257252879057058,
-      "eval_loss": 5.034178256988525,
-      "eval_runtime": 185.9242,
-      "eval_samples_per_second": 96.873,
-      "eval_steps_per_second": 6.056,
+      "epoch": 0.1078167115902965,
+      "eval_accuracy": 0.22754044304953452,
+      "eval_loss": 5.025298118591309,
+      "eval_runtime": 186.83,
+      "eval_samples_per_second": 96.403,
+      "eval_steps_per_second": 6.027,
       "step": 1000
     },
     {
-      "epoch": 0.11301259283177269,
-      "grad_norm": 1.1473057270050049,
-      "learning_rate": 0.0005938842797112379,
-      "loss": 5.051,
+      "epoch": 0.11320754716981132,
+      "grad_norm": 1.1835060119628906,
+      "learning_rate": 0.0005938478143550997,
+      "loss": 5.0739,
       "step": 1050
     },
     {
-      "epoch": 0.1183941448713809,
-      "grad_norm": 1.2688162326812744,
-      "learning_rate": 0.0005935610386811766,
-      "loss": 5.0342,
+      "epoch": 0.11859838274932614,
+      "grad_norm": 1.216452956199646,
+      "learning_rate": 0.0005935240151106314,
+      "loss": 4.9977,
       "step": 1100
     },
     {
-      "epoch": 0.12377569691098914,
-      "grad_norm": 1.471405029296875,
-      "learning_rate": 0.0005932377976511151,
-      "loss": 4.9927,
+      "epoch": 0.12398921832884097,
+      "grad_norm": 1.2591557502746582,
+      "learning_rate": 0.0005932002158661629,
+      "loss": 4.9973,
       "step": 1150
     },
     {
-      "epoch": 0.12915724895059735,
-      "grad_norm": 1.2830071449279785,
-      "learning_rate": 0.0005929145566210538,
-      "loss": 4.9706,
+      "epoch": 0.1293800539083558,
+      "grad_norm": 1.0126723051071167,
+      "learning_rate": 0.0005928764166216945,
+      "loss": 4.9635,
       "step": 1200
     },
     {
-      "epoch": 0.13453880099020557,
-      "grad_norm": 1.042227029800415,
-      "learning_rate": 0.0005925913155909923,
-      "loss": 4.9304,
+      "epoch": 0.1347708894878706,
+      "grad_norm": 1.0633987188339233,
+      "learning_rate": 0.000592552617377226,
+      "loss": 4.9085,
       "step": 1250
     },
     {
-      "epoch": 0.1399203530298138,
-      "grad_norm": 1.1886645555496216,
-      "learning_rate": 0.0005922680745609308,
-      "loss": 4.91,
+      "epoch": 0.14016172506738545,
+      "grad_norm": 1.1745322942733765,
+      "learning_rate": 0.0005922288181327577,
+      "loss": 4.9127,
       "step": 1300
     },
     {
-      "epoch": 0.14530190506942203,
-      "grad_norm": 1.1165701150894165,
-      "learning_rate": 0.0005919448335308695,
-      "loss": 4.8792,
+      "epoch": 0.14555256064690028,
+      "grad_norm": 0.8595957159996033,
+      "learning_rate": 0.0005919050188882893,
+      "loss": 4.8803,
       "step": 1350
     },
     {
-      "epoch": 0.15068345710903025,
-      "grad_norm": 1.2330048084259033,
-      "learning_rate": 0.000591621592500808,
-      "loss": 4.8518,
+      "epoch": 0.1509433962264151,
+      "grad_norm": 1.0559402704238892,
+      "learning_rate": 0.0005915812196438207,
+      "loss": 4.8456,
       "step": 1400
     },
     {
-      "epoch": 0.15606500914863847,
-      "grad_norm": 1.231316089630127,
-      "learning_rate": 0.0005912983514707467,
-      "loss": 4.8285,
+      "epoch": 0.15633423180592992,
+      "grad_norm": 1.1085131168365479,
+      "learning_rate": 0.0005912574203993524,
+      "loss": 4.8253,
       "step": 1450
     },
     {
-      "epoch": 0.16144656118824668,
-      "grad_norm": 1.064316749572754,
-      "learning_rate": 0.0005909751104406852,
-      "loss": 4.8077,
+      "epoch": 0.16172506738544473,
+      "grad_norm": 1.0638484954833984,
+      "learning_rate": 0.0005909336211548839,
+      "loss": 4.8124,
       "step": 1500
     },
     {
-      "epoch": 0.1668281132278549,
-      "grad_norm": 0.7813340425491333,
-      "learning_rate": 0.0005906518694106238,
-      "loss": 4.7872,
+      "epoch": 0.16711590296495957,
+      "grad_norm": 0.9378893971443176,
+      "learning_rate": 0.0005906098219104155,
+      "loss": 4.782,
       "step": 1550
     },
     {
-      "epoch": 0.17220966526746315,
-      "grad_norm": 0.8941433429718018,
-      "learning_rate": 0.0005903286283805624,
-      "loss": 4.7446,
+      "epoch": 0.1725067385444744,
+      "grad_norm": 1.1634156703948975,
+      "learning_rate": 0.000590286022665947,
+      "loss": 4.7891,
       "step": 1600
     },
     {
-      "epoch": 0.17759121730707136,
-      "grad_norm": 0.838981568813324,
-      "learning_rate": 0.000590005387350501,
-      "loss": 4.7486,
+      "epoch": 0.1778975741239892,
+      "grad_norm": 1.0891342163085938,
+      "learning_rate": 0.0005899622234214787,
+      "loss": 4.7388,
       "step": 1650
     },
     {
-      "epoch": 0.18297276934667958,
-      "grad_norm": 1.0024216175079346,
-      "learning_rate": 0.0005896821463204395,
-      "loss": 4.7169,
+      "epoch": 0.18328840970350405,
+      "grad_norm": 0.8270106315612793,
+      "learning_rate": 0.0005896384241770102,
+      "loss": 4.6976,
       "step": 1700
     },
     {
-      "epoch": 0.1883543213862878,
-      "grad_norm": 1.0789501667022705,
-      "learning_rate": 0.0005893589052903781,
-      "loss": 4.6844,
+      "epoch": 0.18867924528301888,
+      "grad_norm": 0.8764285445213318,
+      "learning_rate": 0.0005893146249325418,
+      "loss": 4.6808,
       "step": 1750
     },
     {
-      "epoch": 0.19373587342589602,
-      "grad_norm": 0.9427714347839355,
-      "learning_rate": 0.0005890356642603167,
-      "loss": 4.7106,
+      "epoch": 0.1940700808625337,
+      "grad_norm": 0.983856201171875,
+      "learning_rate": 0.0005889908256880733,
+      "loss": 4.6771,
       "step": 1800
     },
     {
-      "epoch": 0.19911742546550426,
-      "grad_norm": 1.179642677307129,
-      "learning_rate": 0.0005887124232302553,
-      "loss": 4.6463,
+      "epoch": 0.19946091644204852,
+      "grad_norm": 0.7953420877456665,
+      "learning_rate": 0.0005886670264436049,
+      "loss": 4.6796,
       "step": 1850
     },
     {
-      "epoch": 0.20449897750511248,
-      "grad_norm": 0.905066192150116,
-      "learning_rate": 0.0005883891822001939,
-      "loss": 4.6574,
+      "epoch": 0.20485175202156333,
+      "grad_norm": 0.8113346099853516,
+      "learning_rate": 0.0005883432271991365,
+      "loss": 4.6245,
       "step": 1900
     },
     {
-      "epoch": 0.2098805295447207,
-      "grad_norm": 1.0582997798919678,
-      "learning_rate": 0.0005880659411701324,
-      "loss": 4.5862,
+      "epoch": 0.21024258760107817,
+      "grad_norm": 0.9100364446640015,
+      "learning_rate": 0.0005880194279546681,
+      "loss": 4.6084,
       "step": 1950
     },
     {
-      "epoch": 0.2152620815843289,
-      "grad_norm": 0.9089963436126709,
-      "learning_rate": 0.0005877427001400711,
-      "loss": 4.5993,
+      "epoch": 0.215633423180593,
+      "grad_norm": 1.0331058502197266,
+      "learning_rate": 0.0005876956287101996,
+      "loss": 4.5823,
       "step": 2000
     },
     {
-      "epoch": 0.2152620815843289,
-      "eval_accuracy": 0.26972286233863824,
-      "eval_loss": 4.515867233276367,
-      "eval_runtime": 185.6561,
-      "eval_samples_per_second": 97.013,
-      "eval_steps_per_second": 6.065,
+      "epoch": 0.215633423180593,
+      "eval_accuracy": 0.27079287597783525,
+      "eval_loss": 4.509124279022217,
+      "eval_runtime": 188.1171,
+      "eval_samples_per_second": 95.744,
+      "eval_steps_per_second": 5.986,
       "step": 2000
     },
     {
-      "epoch": 0.22064363362393713,
-      "grad_norm": 0.8239651918411255,
-      "learning_rate": 0.0005874194591100097,
-      "loss": 4.5754,
+      "epoch": 0.2210242587601078,
+      "grad_norm": 1.0378620624542236,
+      "learning_rate": 0.0005873718294657312,
+      "loss": 4.5701,
       "step": 2050
     },
     {
-      "epoch": 0.22602518566354537,
-      "grad_norm": 0.9209245443344116,
-      "learning_rate": 0.0005870962180799483,
-      "loss": 4.527,
+      "epoch": 0.22641509433962265,
+      "grad_norm": 0.7657790780067444,
+      "learning_rate": 0.0005870480302212628,
+      "loss": 4.5394,
       "step": 2100
     },
     {
-      "epoch": 0.2314067377031536,
-      "grad_norm": 0.7858166694641113,
-      "learning_rate": 0.0005867729770498868,
-      "loss": 4.5283,
+      "epoch": 0.23180592991913745,
+      "grad_norm": 0.9886602759361267,
+      "learning_rate": 0.0005867242309767943,
+      "loss": 4.5494,
       "step": 2150
     },
     {
-      "epoch": 0.2367882897427618,
-      "grad_norm": 0.8506563901901245,
-      "learning_rate": 0.0005864497360198253,
-      "loss": 4.5017,
+      "epoch": 0.2371967654986523,
+      "grad_norm": 0.902481734752655,
+      "learning_rate": 0.0005864004317323259,
+      "loss": 4.5171,
       "step": 2200
     },
     {
-      "epoch": 0.24216984178237003,
-      "grad_norm": 0.903475284576416,
-      "learning_rate": 0.000586126494989764,
-      "loss": 4.5116,
+      "epoch": 0.24258760107816713,
+      "grad_norm": 0.8495748043060303,
+      "learning_rate": 0.0005860766324878575,
+      "loss": 4.502,
       "step": 2250
     },
     {
-      "epoch": 0.24755139382197827,
-      "grad_norm": 0.8945424556732178,
-      "learning_rate": 0.0005858032539597025,
-      "loss": 4.5007,
+      "epoch": 0.24797843665768193,
+      "grad_norm": 0.7360301613807678,
+      "learning_rate": 0.000585752833243389,
+      "loss": 4.4774,
       "step": 2300
     },
     {
-      "epoch": 0.2529329458615865,
-      "grad_norm": 0.6636348366737366,
-      "learning_rate": 0.0005854800129296412,
-      "loss": 4.4701,
+      "epoch": 0.25336927223719674,
+      "grad_norm": 0.8609463572502136,
+      "learning_rate": 0.0005854290339989206,
+      "loss": 4.4652,
       "step": 2350
     },
     {
-      "epoch": 0.2583144979011947,
-      "grad_norm": 1.0701271295547485,
-      "learning_rate": 0.0005851567718995797,
-      "loss": 4.4403,
+      "epoch": 0.2587601078167116,
+      "grad_norm": 0.9813892841339111,
+      "learning_rate": 0.0005851052347544521,
+      "loss": 4.454,
       "step": 2400
     },
     {
-      "epoch": 0.2636960499408029,
-      "grad_norm": 1.0419641733169556,
-      "learning_rate": 0.0005848335308695184,
-      "loss": 4.4431,
+      "epoch": 0.2641509433962264,
+      "grad_norm": 0.8736472129821777,
+      "learning_rate": 0.0005847814355099838,
+      "loss": 4.4437,
       "step": 2450
     },
     {
-      "epoch": 0.26907760198041114,
-      "grad_norm": 0.9337476491928101,
-      "learning_rate": 0.0005845102898394569,
-      "loss": 4.4188,
+      "epoch": 0.2695417789757412,
+      "grad_norm": 0.8053706884384155,
+      "learning_rate": 0.0005844576362655154,
+      "loss": 4.4595,
       "step": 2500
     },
     {
-      "epoch": 0.27445915402001936,
-      "grad_norm": 0.9284446835517883,
-      "learning_rate": 0.0005841870488093954,
-      "loss": 4.3951,
+      "epoch": 0.2749326145552561,
+      "grad_norm": 1.003642201423645,
+      "learning_rate": 0.0005841338370210469,
+      "loss": 4.4089,
       "step": 2550
     },
     {
-      "epoch": 0.2798407060596276,
-      "grad_norm": 1.0020089149475098,
-      "learning_rate": 0.0005838638077793341,
-      "loss": 4.3854,
+      "epoch": 0.2803234501347709,
+      "grad_norm": 0.9675222635269165,
+      "learning_rate": 0.0005838100377765785,
+      "loss": 4.3859,
       "step": 2600
     },
     {
-      "epoch": 0.2852222580992358,
-      "grad_norm": 0.8462651371955872,
-      "learning_rate": 0.0005835405667492726,
-      "loss": 4.379,
+      "epoch": 0.2857142857142857,
+      "grad_norm": 0.917807400226593,
+      "learning_rate": 0.0005834862385321101,
+      "loss": 4.3903,
       "step": 2650
     },
     {
-      "epoch": 0.29060381013884407,
-      "grad_norm": 0.8200508952140808,
-      "learning_rate": 0.0005832173257192113,
-      "loss": 4.3916,
+      "epoch": 0.29110512129380056,
+      "grad_norm": 1.174917221069336,
+      "learning_rate": 0.0005831624392876417,
+      "loss": 4.3836,
       "step": 2700
     },
     {
-      "epoch": 0.2959853621784523,
-      "grad_norm": 0.9552410840988159,
-      "learning_rate": 0.0005828940846891498,
-      "loss": 4.3722,
+      "epoch": 0.29649595687331537,
+      "grad_norm": 0.9486121535301208,
+      "learning_rate": 0.0005828386400431731,
+      "loss": 4.365,
       "step": 2750
     },
     {
-      "epoch": 0.3013669142180605,
-      "grad_norm": 1.0096497535705566,
-      "learning_rate": 0.0005825708436590884,
-      "loss": 4.3725,
+      "epoch": 0.3018867924528302,
+      "grad_norm": 0.8282912969589233,
+      "learning_rate": 0.0005825148407987048,
+      "loss": 4.3634,
       "step": 2800
     },
     {
-      "epoch": 0.3067484662576687,
-      "grad_norm": 0.9831181168556213,
-      "learning_rate": 0.000582247602629027,
-      "loss": 4.3666,
+      "epoch": 0.30727762803234504,
+      "grad_norm": 0.9368751645088196,
+      "learning_rate": 0.0005821910415542363,
+      "loss": 4.364,
       "step": 2850
     },
     {
-      "epoch": 0.31213001829727693,
-      "grad_norm": 1.0847501754760742,
-      "learning_rate": 0.0005819243615989656,
-      "loss": 4.3497,
+      "epoch": 0.31266846361185985,
+      "grad_norm": 1.105896234512329,
+      "learning_rate": 0.0005818672423097679,
+      "loss": 4.3451,
       "step": 2900
     },
     {
-      "epoch": 0.31751157033688515,
-      "grad_norm": 0.7545516490936279,
-      "learning_rate": 0.0005816011205689042,
-      "loss": 4.3086,
+      "epoch": 0.31805929919137466,
+      "grad_norm": 0.9352311491966248,
+      "learning_rate": 0.0005815434430652994,
+      "loss": 4.3512,
       "step": 2950
     },
     {
-      "epoch": 0.32289312237649337,
-      "grad_norm": 1.1041558980941772,
-      "learning_rate": 0.0005812778795388427,
-      "loss": 4.2969,
+      "epoch": 0.32345013477088946,
+      "grad_norm": 0.8877279758453369,
+      "learning_rate": 0.0005812196438208311,
+      "loss": 4.2972,
       "step": 3000
     },
     {
-      "epoch": 0.32289312237649337,
-      "eval_accuracy": 0.29862963718301744,
-      "eval_loss": 4.24181604385376,
-      "eval_runtime": 185.692,
-      "eval_samples_per_second": 96.994,
-      "eval_steps_per_second": 6.064,
+      "epoch": 0.32345013477088946,
+      "eval_accuracy": 0.298535000517731,
+      "eval_loss": 4.238403797149658,
+      "eval_runtime": 186.0259,
+      "eval_samples_per_second": 96.82,
+      "eval_steps_per_second": 6.053,
       "step": 3000
     },
     {
-      "epoch": 0.3282746744161016,
-      "grad_norm": 1.3289992809295654,
-      "learning_rate": 0.0005809546385087813,
-      "loss": 4.3089,
+      "epoch": 0.3288409703504043,
+      "grad_norm": 0.8135088086128235,
+      "learning_rate": 0.0005808958445763626,
+      "loss": 4.3073,
       "step": 3050
     },
     {
-      "epoch": 0.3336562264557098,
-      "grad_norm": 0.8491426110267639,
-      "learning_rate": 0.0005806313974787199,
-      "loss": 4.2836,
+      "epoch": 0.33423180592991913,
+      "grad_norm": 0.7028347849845886,
+      "learning_rate": 0.0005805720453318942,
+      "loss": 4.2885,
       "step": 3100
     },
     {
-      "epoch": 0.3390377784953181,
-      "grad_norm": 0.7015953660011292,
-      "learning_rate": 0.0005803081564486585,
-      "loss": 4.2754,
+      "epoch": 0.33962264150943394,
+      "grad_norm": 0.7956777811050415,
+      "learning_rate": 0.0005802482460874257,
+      "loss": 4.2783,
       "step": 3150
     },
     {
-      "epoch": 0.3444193305349263,
-      "grad_norm": 0.7845497131347656,
-      "learning_rate": 0.0005799849154185971,
-      "loss": 4.2675,
+      "epoch": 0.3450134770889488,
+      "grad_norm": 0.6486227512359619,
+      "learning_rate": 0.0005799244468429573,
+      "loss": 4.2986,
       "step": 3200
     },
     {
-      "epoch": 0.3498008825745345,
-      "grad_norm": 0.7819005250930786,
-      "learning_rate": 0.0005796616743885357,
-      "loss": 4.281,
+      "epoch": 0.3504043126684636,
+      "grad_norm": 0.773562490940094,
+      "learning_rate": 0.0005796006475984889,
+      "loss": 4.2811,
       "step": 3250
     },
     {
-      "epoch": 0.35518243461414273,
-      "grad_norm": 0.7934615015983582,
-      "learning_rate": 0.0005793384333584742,
-      "loss": 4.253,
+      "epoch": 0.3557951482479784,
+      "grad_norm": 0.8088301420211792,
+      "learning_rate": 0.0005792768483540205,
+      "loss": 4.2616,
       "step": 3300
     },
     {
-      "epoch": 0.36056398665375095,
-      "grad_norm": 0.8142829537391663,
-      "learning_rate": 0.0005790151923284129,
-      "loss": 4.2436,
+      "epoch": 0.3611859838274933,
+      "grad_norm": 0.7410290837287903,
+      "learning_rate": 0.000578953049109552,
+      "loss": 4.2709,
       "step": 3350
     },
     {
-      "epoch": 0.36594553869335916,
-      "grad_norm": 0.875823438167572,
-      "learning_rate": 0.0005786919512983514,
-      "loss": 4.2468,
+      "epoch": 0.3665768194070081,
+      "grad_norm": 0.7373089790344238,
+      "learning_rate": 0.0005786292498650836,
+      "loss": 4.2376,
       "step": 3400
     },
     {
-      "epoch": 0.3713270907329674,
-      "grad_norm": 0.816535472869873,
-      "learning_rate": 0.00057836871026829,
-      "loss": 4.2693,
+      "epoch": 0.3719676549865229,
+      "grad_norm": 0.7937477827072144,
+      "learning_rate": 0.0005783054506206152,
+      "loss": 4.25,
       "step": 3450
     },
     {
-      "epoch": 0.3767086427725756,
-      "grad_norm": 0.7482356429100037,
-      "learning_rate": 0.0005780454692382286,
-      "loss": 4.2367,
+      "epoch": 0.37735849056603776,
+      "grad_norm": 0.7315645813941956,
+      "learning_rate": 0.0005779816513761467,
+      "loss": 4.232,
       "step": 3500
     },
     {
-      "epoch": 0.3820901948121838,
-      "grad_norm": 0.8074214458465576,
-      "learning_rate": 0.0005777222282081672,
-      "loss": 4.2383,
+      "epoch": 0.38274932614555257,
+      "grad_norm": 1.0023199319839478,
+      "learning_rate": 0.0005776578521316782,
+      "loss": 4.2108,
       "step": 3550
     },
     {
-      "epoch": 0.38747174685179203,
-      "grad_norm": 1.1420109272003174,
-      "learning_rate": 0.0005773989871781058,
-      "loss": 4.2114,
+      "epoch": 0.3881401617250674,
+      "grad_norm": 0.8754860162734985,
+      "learning_rate": 0.0005773340528872099,
+      "loss": 4.2295,
       "step": 3600
     },
     {
-      "epoch": 0.3928532988914003,
-      "grad_norm": 0.9202913641929626,
-      "learning_rate": 0.0005770757461480443,
-      "loss": 4.2323,
+      "epoch": 0.3935309973045822,
+      "grad_norm": 0.775505542755127,
+      "learning_rate": 0.0005770102536427414,
+      "loss": 4.2242,
       "step": 3650
     },
     {
-      "epoch": 0.3982348509310085,
-      "grad_norm": 0.6997593641281128,
-      "learning_rate": 0.0005767525051179829,
+      "epoch": 0.39892183288409705,
+      "grad_norm": 0.6237397193908691,
+      "learning_rate": 0.000576686454398273,
       "loss": 4.2006,
       "step": 3700
     },
     {
-      "epoch": 0.40361640297061674,
-      "grad_norm": 0.6690722703933716,
-      "learning_rate": 0.0005764292640879215,
-      "loss": 4.1984,
+      "epoch": 0.40431266846361186,
+      "grad_norm": 0.673305094242096,
+      "learning_rate": 0.0005763626551538045,
+      "loss": 4.1807,
       "step": 3750
     },
     {
-      "epoch": 0.40899795501022496,
-      "grad_norm": 0.7391257882118225,
-      "learning_rate": 0.0005761060230578601,
-      "loss": 4.1973,
+      "epoch": 0.40970350404312667,
+      "grad_norm": 0.7659112215042114,
+      "learning_rate": 0.0005760388559093362,
+      "loss": 4.19,
       "step": 3800
     },
     {
-      "epoch": 0.4143795070498332,
-      "grad_norm": 0.7893011569976807,
-      "learning_rate": 0.0005757827820277987,
-      "loss": 4.1819,
+      "epoch": 0.41509433962264153,
+      "grad_norm": 0.7184351086616516,
+      "learning_rate": 0.0005757150566648678,
+      "loss": 4.1704,
       "step": 3850
     },
     {
-      "epoch": 0.4197610590894414,
-      "grad_norm": 0.792421817779541,
-      "learning_rate": 0.0005754595409977372,
-      "loss": 4.1899,
+      "epoch": 0.42048517520215634,
+      "grad_norm": 0.690081000328064,
+      "learning_rate": 0.0005753912574203993,
+      "loss": 4.1809,
       "step": 3900
     },
     {
-      "epoch": 0.4251426111290496,
-      "grad_norm": 0.7149876952171326,
-      "learning_rate": 0.0005751362999676759,
-      "loss": 4.1678,
+      "epoch": 0.42587601078167114,
+      "grad_norm": 0.9046959280967712,
+      "learning_rate": 0.0005750674581759309,
+      "loss": 4.1852,
       "step": 3950
     },
     {
-      "epoch": 0.4305241631686578,
-      "grad_norm": 0.6952950358390808,
-      "learning_rate": 0.0005748130589376144,
-      "loss": 4.1676,
+      "epoch": 0.431266846361186,
+      "grad_norm": 0.6259483098983765,
+      "learning_rate": 0.0005747436589314624,
+      "loss": 4.1641,
       "step": 4000
     },
     {
-      "epoch": 0.4305241631686578,
-      "eval_accuracy": 0.3120642408026145,
-      "eval_loss": 4.0956950187683105,
-      "eval_runtime": 185.4003,
-      "eval_samples_per_second": 97.147,
-      "eval_steps_per_second": 6.073,
+      "epoch": 0.431266846361186,
+      "eval_accuracy": 0.3115417290651147,
+      "eval_loss": 4.095163345336914,
+      "eval_runtime": 184.7258,
+      "eval_samples_per_second": 97.501,
+      "eval_steps_per_second": 6.096,
       "step": 4000
     },
     {
-      "epoch": 0.43590571520826604,
-      "grad_norm": 0.7889477014541626,
-      "learning_rate": 0.0005744898179075531,
-      "loss": 4.1673,
+      "epoch": 0.4366576819407008,
+      "grad_norm": 0.9814295768737793,
+      "learning_rate": 0.0005744198596869941,
+      "loss": 4.1534,
       "step": 4050
     },
     {
-      "epoch": 0.44128726724787426,
-      "grad_norm": 0.8199712038040161,
-      "learning_rate": 0.0005741665768774916,
-      "loss": 4.1462,
+      "epoch": 0.4420485175202156,
+      "grad_norm": 0.7232169508934021,
+      "learning_rate": 0.0005740960604425255,
+      "loss": 4.141,
       "step": 4100
     },
     {
-      "epoch": 0.44666881928748253,
-      "grad_norm": 0.6692017912864685,
-      "learning_rate": 0.0005738433358474301,
-      "loss": 4.1454,
+      "epoch": 0.4474393530997305,
+      "grad_norm": 0.6636750102043152,
+      "learning_rate": 0.0005737722611980572,
+      "loss": 4.1511,
       "step": 4150
     },
     {
-      "epoch": 0.45205037132709075,
-      "grad_norm": 0.7322275042533875,
-      "learning_rate": 0.0005735200948173688,
-      "loss": 4.1325,
+      "epoch": 0.4528301886792453,
+      "grad_norm": 0.9277781844139099,
+      "learning_rate": 0.0005734484619535887,
+      "loss": 4.137,
       "step": 4200
     },
     {
-      "epoch": 0.45743192336669897,
-      "grad_norm": 0.8394356369972229,
-      "learning_rate": 0.0005731968537873073,
-      "loss": 4.1269,
+      "epoch": 0.4582210242587601,
+      "grad_norm": 0.7925214767456055,
+      "learning_rate": 0.0005731246627091203,
+      "loss": 4.1349,
       "step": 4250
     },
     {
-      "epoch": 0.4628134754063072,
-      "grad_norm": 0.6714458465576172,
-      "learning_rate": 0.0005728736127572459,
-      "loss": 4.1348,
+      "epoch": 0.4636118598382749,
+      "grad_norm": 0.6720883846282959,
+      "learning_rate": 0.0005728008634646518,
+      "loss": 4.1219,
       "step": 4300
     },
     {
-      "epoch": 0.4681950274459154,
-      "grad_norm": 0.6486440300941467,
-      "learning_rate": 0.0005725503717271845,
-      "loss": 4.1298,
+      "epoch": 0.46900269541778977,
+      "grad_norm": 0.6587103009223938,
+      "learning_rate": 0.0005724770642201835,
+      "loss": 4.1188,
       "step": 4350
     },
     {
-      "epoch": 0.4735765794855236,
-      "grad_norm": 0.7148385643959045,
-      "learning_rate": 0.0005722271306971231,
-      "loss": 4.1185,
+      "epoch": 0.4743935309973046,
+      "grad_norm": 0.6791043877601624,
+      "learning_rate": 0.000572153264975715,
+      "loss": 4.1112,
       "step": 4400
     },
     {
-      "epoch": 0.47895813152513184,
-      "grad_norm": 0.7207582592964172,
-      "learning_rate": 0.0005719038896670617,
-      "loss": 4.1166,
+      "epoch": 0.4797843665768194,
+      "grad_norm": 0.7948880195617676,
+      "learning_rate": 0.0005718294657312466,
+      "loss": 4.1372,
       "step": 4450
     },
     {
-      "epoch": 0.48433968356474005,
-      "grad_norm": 0.6919761896133423,
-      "learning_rate": 0.0005715806486370003,
-      "loss": 4.1272,
+      "epoch": 0.48517520215633425,
+      "grad_norm": 0.7730107307434082,
+      "learning_rate": 0.0005715056664867781,
+      "loss": 4.0979,
       "step": 4500
     },
     {
-      "epoch": 0.48972123560434827,
-      "grad_norm": 0.6202664375305176,
-      "learning_rate": 0.0005712574076069388,
-      "loss": 4.1198,
+      "epoch": 0.49056603773584906,
+      "grad_norm": 0.6119707822799683,
+      "learning_rate": 0.0005711818672423097,
+      "loss": 4.1084,
       "step": 4550
     },
     {
-      "epoch": 0.49510278764395654,
-      "grad_norm": 0.6014512777328491,
-      "learning_rate": 0.0005709341665768774,
-      "loss": 4.1071,
+      "epoch": 0.49595687331536387,
+      "grad_norm": 0.7200744152069092,
+      "learning_rate": 0.0005708580679978413,
+      "loss": 4.1031,
       "step": 4600
     },
     {
-      "epoch": 0.5004843396835648,
-      "grad_norm": 0.6499958634376526,
-      "learning_rate": 0.0005706109255468161,
-      "loss": 4.0796,
+      "epoch": 0.5013477088948787,
+      "grad_norm": 0.6110667586326599,
+      "learning_rate": 0.0005705342687533729,
+      "loss": 4.0976,
       "step": 4650
     },
     {
-      "epoch": 0.505865891723173,
-      "grad_norm": 0.7313818335533142,
-      "learning_rate": 0.0005702876845167546,
-      "loss": 4.0998,
+      "epoch": 0.5067385444743935,
+      "grad_norm": 0.6757206916809082,
+      "learning_rate": 0.0005702104695089044,
+      "loss": 4.098,
       "step": 4700
     },
     {
-      "epoch": 0.5112474437627812,
-      "grad_norm": 0.6985046863555908,
-      "learning_rate": 0.0005699644434866932,
-      "loss": 4.0918,
+      "epoch": 0.5121293800539084,
+      "grad_norm": 0.6472781300544739,
+      "learning_rate": 0.000569886670264436,
+      "loss": 4.0992,
       "step": 4750
     },
     {
-      "epoch": 0.5166289958023894,
-      "grad_norm": 0.5906838178634644,
-      "learning_rate": 0.0005696412024566317,
-      "loss": 4.0921,
+      "epoch": 0.5175202156334232,
+      "grad_norm": 0.6886298060417175,
+      "learning_rate": 0.0005695628710199675,
+      "loss": 4.079,
       "step": 4800
     },
     {
-      "epoch": 0.5220105478419976,
-      "grad_norm": 0.8641616106033325,
-      "learning_rate": 0.0005693179614265704,
-      "loss": 4.0702,
+      "epoch": 0.522911051212938,
+      "grad_norm": 0.6832914352416992,
+      "learning_rate": 0.0005692390717754991,
+      "loss": 4.1078,
       "step": 4850
     },
     {
-      "epoch": 0.5273920998816058,
-      "grad_norm": 0.7014225721359253,
-      "learning_rate": 0.000568994720396509,
-      "loss": 4.0881,
+      "epoch": 0.5283018867924528,
+      "grad_norm": 0.6672677397727966,
+      "learning_rate": 0.0005689152725310306,
+      "loss": 4.0806,
       "step": 4900
     },
     {
-      "epoch": 0.5327736519212141,
-      "grad_norm": 0.6952133774757385,
-      "learning_rate": 0.0005686714793664476,
-      "loss": 4.0599,
+      "epoch": 0.5336927223719676,
+      "grad_norm": 0.7332984805107117,
+      "learning_rate": 0.0005685914732865623,
+      "loss": 4.0656,
       "step": 4950
     },
     {
-      "epoch": 0.5381552039608223,
-      "grad_norm": 0.75461745262146,
-      "learning_rate": 0.0005683482383363861,
-      "loss": 4.0642,
+      "epoch": 0.5390835579514824,
+      "grad_norm": 0.6176167130470276,
+      "learning_rate": 0.0005682676740420939,
+      "loss": 4.0686,
       "step": 5000
     },
     {
-      "epoch": 0.5381552039608223,
-      "eval_accuracy": 0.3205846916121383,
-      "eval_loss": 4.000431060791016,
-      "eval_runtime": 185.3844,
-      "eval_samples_per_second": 97.155,
-      "eval_steps_per_second": 6.074,
+      "epoch": 0.5390835579514824,
+      "eval_accuracy": 0.3214803173663931,
+      "eval_loss": 3.992448568344116,
+      "eval_runtime": 184.3526,
+      "eval_samples_per_second": 97.699,
+      "eval_steps_per_second": 6.108,
       "step": 5000
     },
     {
-      "epoch": 0.5435367560004305,
-      "grad_norm": 0.6787461042404175,
-      "learning_rate": 0.0005680249973063247,
-      "loss": 4.0693,
+      "epoch": 0.5444743935309974,
+      "grad_norm": 0.5917017459869385,
+      "learning_rate": 0.0005679438747976254,
+      "loss": 4.0621,
       "step": 5050
     },
     {
-      "epoch": 0.5489183080400387,
-      "grad_norm": 0.7117706537246704,
-      "learning_rate": 0.0005677017562762633,
-      "loss": 4.056,
+      "epoch": 0.5498652291105122,
+      "grad_norm": 0.6220250129699707,
+      "learning_rate": 0.000567620075553157,
+      "loss": 4.0531,
       "step": 5100
     },
     {
-      "epoch": 0.5542998600796469,
-      "grad_norm": 0.6522045135498047,
-      "learning_rate": 0.0005673785152462018,
-      "loss": 4.039,
+      "epoch": 0.555256064690027,
+      "grad_norm": 0.5982322096824646,
+      "learning_rate": 0.0005672962763086886,
+      "loss": 4.0624,
       "step": 5150
     },
     {
-      "epoch": 0.5596814121192552,
-      "grad_norm": 0.6263989806175232,
-      "learning_rate": 0.0005670552742161405,
-      "loss": 4.0767,
+      "epoch": 0.5606469002695418,
+      "grad_norm": 0.6485611796379089,
+      "learning_rate": 0.0005669724770642202,
+      "loss": 4.0528,
       "step": 5200
     },
     {
-      "epoch": 0.5650629641588634,
-      "grad_norm": 0.7219600081443787,
-      "learning_rate": 0.000566732033186079,
-      "loss": 4.0563,
+      "epoch": 0.5660377358490566,
+      "grad_norm": 0.6572372913360596,
+      "learning_rate": 0.0005666486778197517,
+      "loss": 4.0542,
       "step": 5250
     },
     {
-      "epoch": 0.5704445161984716,
-      "grad_norm": 0.6352048516273499,
-      "learning_rate": 0.0005664087921560177,
-      "loss": 4.0474,
+      "epoch": 0.5714285714285714,
+      "grad_norm": 0.6387195587158203,
+      "learning_rate": 0.0005663248785752833,
+      "loss": 4.0329,
       "step": 5300
     },
     {
-      "epoch": 0.5758260682380799,
-      "grad_norm": 0.6418537497520447,
-      "learning_rate": 0.0005660855511259562,
-      "loss": 4.0516,
+      "epoch": 0.5768194070080862,
+      "grad_norm": 0.6669923663139343,
+      "learning_rate": 0.0005660010793308148,
+      "loss": 4.0093,
       "step": 5350
     },
     {
-      "epoch": 0.5812076202776881,
-      "grad_norm": 0.7082635164260864,
-      "learning_rate": 0.0005657623100958947,
-      "loss": 4.0268,
+      "epoch": 0.5822102425876011,
+      "grad_norm": 0.7043260335922241,
+      "learning_rate": 0.0005656772800863465,
+      "loss": 4.0499,
       "step": 5400
     },
     {
-      "epoch": 0.5865891723172963,
-      "grad_norm": 0.6689324378967285,
-      "learning_rate": 0.0005654390690658334,
-      "loss": 4.024,
+      "epoch": 0.5876010781671159,
+      "grad_norm": 0.6300632953643799,
+      "learning_rate": 0.0005653534808418779,
+      "loss": 4.0284,
       "step": 5450
     },
     {
-      "epoch": 0.5919707243569046,
-      "grad_norm": 0.5856683850288391,
-      "learning_rate": 0.0005651158280357719,
-      "loss": 4.0459,
+      "epoch": 0.5929919137466307,
+      "grad_norm": 0.5939152240753174,
+      "learning_rate": 0.0005650296815974096,
+      "loss": 4.0374,
       "step": 5500
     },
     {
-      "epoch": 0.5973522763965128,
-      "grad_norm": 0.6519558429718018,
-      "learning_rate": 0.0005647925870057106,
-      "loss": 4.0324,
+      "epoch": 0.5983827493261455,
+      "grad_norm": 0.6550401449203491,
+      "learning_rate": 0.0005647058823529411,
+      "loss": 4.0211,
       "step": 5550
     },
     {
-      "epoch": 0.602733828436121,
-      "grad_norm": 0.6687760353088379,
-      "learning_rate": 0.0005644693459756491,
-      "loss": 4.035,
+      "epoch": 0.6037735849056604,
+      "grad_norm": 0.6555675864219666,
+      "learning_rate": 0.0005643820831084727,
+      "loss": 4.0216,
       "step": 5600
     },
     {
-      "epoch": 0.6081153804757292,
-      "grad_norm": 0.6159760355949402,
-      "learning_rate": 0.0005641461049455877,
-      "loss": 4.0104,
+      "epoch": 0.6091644204851752,
+      "grad_norm": 0.7580973505973816,
+      "learning_rate": 0.0005640582838640042,
+      "loss": 3.9976,
       "step": 5650
     },
     {
-      "epoch": 0.6134969325153374,
-      "grad_norm": 0.6416439414024353,
-      "learning_rate": 0.0005638228639155263,
-      "loss": 4.0014,
+      "epoch": 0.6145552560646901,
+      "grad_norm": 0.7556450963020325,
+      "learning_rate": 0.0005637344846195358,
+      "loss": 4.012,
       "step": 5700
     },
     {
-      "epoch": 0.6188784845549457,
-      "grad_norm": 0.7194835543632507,
-      "learning_rate": 0.000563499622885465,
-      "loss": 4.0274,
+      "epoch": 0.6199460916442049,
+      "grad_norm": 0.6850540041923523,
+      "learning_rate": 0.0005634106853750674,
+      "loss": 4.0064,
       "step": 5750
     },
     {
-      "epoch": 0.6242600365945539,
-      "grad_norm": 0.5478219389915466,
-      "learning_rate": 0.0005631763818554035,
-      "loss": 4.0012,
+      "epoch": 0.6253369272237197,
+      "grad_norm": 0.6250605583190918,
+      "learning_rate": 0.000563086886130599,
+      "loss": 4.0139,
       "step": 5800
     },
     {
-      "epoch": 0.6296415886341621,
-      "grad_norm": 1.0783581733703613,
-      "learning_rate": 0.000562853140825342,
-      "loss": 3.9946,
+      "epoch": 0.6307277628032345,
+      "grad_norm": 0.5788740515708923,
+      "learning_rate": 0.0005627630868861305,
+      "loss": 3.9968,
       "step": 5850
     },
     {
-      "epoch": 0.6350231406737703,
-      "grad_norm": 0.6307653188705444,
-      "learning_rate": 0.0005625298997952806,
-      "loss": 3.9929,
+      "epoch": 0.6361185983827493,
+      "grad_norm": 0.645414412021637,
+      "learning_rate": 0.0005624392876416621,
+      "loss": 3.9941,
       "step": 5900
     },
     {
-      "epoch": 0.6404046927133785,
-      "grad_norm": 0.5995813608169556,
-      "learning_rate": 0.0005622066587652192,
-      "loss": 3.994,
+      "epoch": 0.6415094339622641,
+      "grad_norm": 0.679947555065155,
+      "learning_rate": 0.0005621154883971937,
+      "loss": 3.9945,
       "step": 5950
     },
     {
-      "epoch": 0.6457862447529867,
-      "grad_norm": 0.6409510970115662,
-      "learning_rate": 0.0005618834177351578,
-      "loss": 4.0038,
+      "epoch": 0.6469002695417789,
+      "grad_norm": 0.637030303478241,
+      "learning_rate": 0.0005617916891527253,
+      "loss": 3.9931,
       "step": 6000
     },
     {
-      "epoch": 0.6457862447529867,
-      "eval_accuracy": 0.3277830540827355,
-      "eval_loss": 3.9248440265655518,
-      "eval_runtime": 185.5096,
-      "eval_samples_per_second": 97.089,
-      "eval_steps_per_second": 6.07,
+      "epoch": 0.6469002695417789,
+      "eval_accuracy": 0.32821647045222746,
+      "eval_loss": 3.9245009422302246,
+      "eval_runtime": 184.3541,
+      "eval_samples_per_second": 97.698,
+      "eval_steps_per_second": 6.108,
       "step": 6000
     },
     {
-      "epoch": 0.651167796792595,
-      "grad_norm": 0.6729882955551147,
-      "learning_rate": 0.0005615601767050964,
-      "loss": 3.9844,
+      "epoch": 0.6522911051212938,
+      "grad_norm": 0.5839635729789734,
+      "learning_rate": 0.0005614678899082568,
+      "loss": 3.9875,
       "step": 6050
     },
     {
-      "epoch": 0.6565493488322032,
-      "grad_norm": 0.6977255940437317,
-      "learning_rate": 0.000561236935675035,
-      "loss": 3.9991,
+      "epoch": 0.6576819407008087,
+      "grad_norm": 0.5757126808166504,
+      "learning_rate": 0.0005611440906637884,
+      "loss": 4.0056,
       "step": 6100
     },
     {
-      "epoch": 0.6619309008718114,
-      "grad_norm": 0.5876686573028564,
-      "learning_rate": 0.0005609136946449735,
-      "loss": 3.9916,
+      "epoch": 0.6630727762803235,
+      "grad_norm": 0.7211872339248657,
+      "learning_rate": 0.0005608267674042094,
+      "loss": 3.9899,
       "step": 6150
     },
     {
-      "epoch": 0.6673124529114196,
-      "grad_norm": 0.6658395528793335,
-      "learning_rate": 0.0005605904536149122,
-      "loss": 3.9722,
+      "epoch": 0.6684636118598383,
+      "grad_norm": 0.7232472896575928,
+      "learning_rate": 0.0005605029681597409,
+      "loss": 3.9868,
       "step": 6200
     },
     {
-      "epoch": 0.6726940049510278,
-      "grad_norm": 0.6371570825576782,
-      "learning_rate": 0.0005602672125848507,
-      "loss": 3.9791,
+      "epoch": 0.6738544474393531,
+      "grad_norm": 0.7061896920204163,
+      "learning_rate": 0.0005601791689152725,
+      "loss": 3.9796,
       "step": 6250
     },
     {
-      "epoch": 0.6780755569906362,
-      "grad_norm": 0.6571714282035828,
-      "learning_rate": 0.0005599439715547892,
-      "loss": 3.9892,
+      "epoch": 0.6792452830188679,
+      "grad_norm": 0.6948355436325073,
+      "learning_rate": 0.000559855369670804,
+      "loss": 3.983,
       "step": 6300
     },
     {
-      "epoch": 0.6834571090302444,
-      "grad_norm": 0.5972535610198975,
-      "learning_rate": 0.0005596207305247279,
-      "loss": 3.9656,
+      "epoch": 0.6846361185983828,
+      "grad_norm": 0.5587276220321655,
+      "learning_rate": 0.0005595315704263357,
+      "loss": 3.9645,
       "step": 6350
     },
     {
-      "epoch": 0.6888386610698526,
-      "grad_norm": 0.6958330273628235,
-      "learning_rate": 0.0005592974894946665,
-      "loss": 3.9722,
+      "epoch": 0.6900269541778976,
+      "grad_norm": 0.6232573986053467,
+      "learning_rate": 0.0005592077711818672,
+      "loss": 3.9776,
       "step": 6400
     },
     {
-      "epoch": 0.6942202131094608,
-      "grad_norm": 0.6522466540336609,
-      "learning_rate": 0.0005589742484646051,
-      "loss": 3.9688,
+      "epoch": 0.6954177897574124,
+      "grad_norm": 0.7714099884033203,
+      "learning_rate": 0.0005588839719373988,
+      "loss": 3.9541,
       "step": 6450
     },
     {
-      "epoch": 0.699601765149069,
-      "grad_norm": 0.6060879826545715,
-      "learning_rate": 0.0005586510074345436,
-      "loss": 3.9568,
+      "epoch": 0.7008086253369272,
+      "grad_norm": 0.6150910258293152,
+      "learning_rate": 0.0005585601726929303,
+      "loss": 3.9597,
       "step": 6500
     },
     {
-      "epoch": 0.7049833171886772,
-      "grad_norm": 0.5448448657989502,
-      "learning_rate": 0.0005583277664044822,
-      "loss": 3.9701,
+      "epoch": 0.706199460916442,
+      "grad_norm": 0.6611934304237366,
+      "learning_rate": 0.0005582363734484619,
+      "loss": 3.9441,
       "step": 6550
     },
     {
-      "epoch": 0.7103648692282855,
-      "grad_norm": 0.6579198837280273,
-      "learning_rate": 0.0005580045253744208,
-      "loss": 3.9474,
+      "epoch": 0.7115902964959568,
+      "grad_norm": 0.6238917112350464,
+      "learning_rate": 0.0005579125742039935,
+      "loss": 3.9504,
       "step": 6600
     },
     {
-      "epoch": 0.7157464212678937,
-      "grad_norm": 0.7126155495643616,
-      "learning_rate": 0.0005576812843443594,
-      "loss": 3.9606,
+      "epoch": 0.7169811320754716,
+      "grad_norm": 0.6062159538269043,
+      "learning_rate": 0.0005575887749595251,
+      "loss": 3.951,
       "step": 6650
     },
     {
-      "epoch": 0.7211279733075019,
-      "grad_norm": 0.5947892665863037,
-      "learning_rate": 0.000557358043314298,
-      "loss": 3.9326,
+      "epoch": 0.7223719676549866,
+      "grad_norm": 0.672397255897522,
+      "learning_rate": 0.0005572649757150566,
+      "loss": 3.9541,
       "step": 6700
     },
     {
-      "epoch": 0.7265095253471101,
-      "grad_norm": 0.5703105330467224,
-      "learning_rate": 0.0005570348022842365,
-      "loss": 3.9387,
+      "epoch": 0.7277628032345014,
+      "grad_norm": 0.6164179444313049,
+      "learning_rate": 0.0005569411764705882,
+      "loss": 3.9513,
       "step": 6750
     },
     {
-      "epoch": 0.7318910773867183,
-      "grad_norm": 0.6345952153205872,
-      "learning_rate": 0.0005567115612541752,
-      "loss": 3.9528,
+      "epoch": 0.7331536388140162,
+      "grad_norm": 0.6107237935066223,
+      "learning_rate": 0.0005566173772261198,
+      "loss": 3.9184,
       "step": 6800
     },
     {
-      "epoch": 0.7372726294263265,
-      "grad_norm": 0.6352825164794922,
-      "learning_rate": 0.0005563883202241137,
-      "loss": 3.9621,
+      "epoch": 0.738544474393531,
+      "grad_norm": 0.709151566028595,
+      "learning_rate": 0.0005562935779816513,
+      "loss": 3.9344,
       "step": 6850
     },
     {
-      "epoch": 0.7426541814659348,
-      "grad_norm": 0.605293333530426,
-      "learning_rate": 0.0005560650791940524,
-      "loss": 3.9476,
+      "epoch": 0.7439353099730458,
+      "grad_norm": 0.5938047766685486,
+      "learning_rate": 0.0005559697787371828,
+      "loss": 3.9487,
       "step": 6900
     },
     {
-      "epoch": 0.748035733505543,
-      "grad_norm": 0.629027247428894,
-      "learning_rate": 0.0005557418381639909,
-      "loss": 3.9496,
+      "epoch": 0.7493261455525606,
+      "grad_norm": 0.6733283400535583,
+      "learning_rate": 0.0005556459794927145,
+      "loss": 3.9294,
       "step": 6950
     },
     {
-      "epoch": 0.7534172855451512,
-      "grad_norm": 0.5393297672271729,
-      "learning_rate": 0.0005554185971339294,
-      "loss": 3.9256,
+      "epoch": 0.7547169811320755,
+      "grad_norm": 0.6545038819313049,
+      "learning_rate": 0.000555322180248246,
+      "loss": 3.9393,
       "step": 7000
     },
     {
-      "epoch": 0.7534172855451512,
-      "eval_accuracy": 0.3326138701278551,
-      "eval_loss": 3.8684041500091553,
-      "eval_runtime": 185.676,
-      "eval_samples_per_second": 97.002,
-      "eval_steps_per_second": 6.064,
+      "epoch": 0.7547169811320755,
+      "eval_accuracy": 0.3337426649793597,
+      "eval_loss": 3.8638250827789307,
+      "eval_runtime": 184.342,
+      "eval_samples_per_second": 97.704,
+      "eval_steps_per_second": 6.108,
       "step": 7000
     },
     {
-      "epoch": 0.7587988375847594,
-      "grad_norm": 0.6152430772781372,
-      "learning_rate": 0.0005550953561038681,
-      "loss": 3.9355,
+      "epoch": 0.7601078167115903,
+      "grad_norm": 0.7976385354995728,
+      "learning_rate": 0.0005549983810037776,
+      "loss": 3.9094,
       "step": 7050
     },
     {
-      "epoch": 0.7641803896243676,
-      "grad_norm": 0.5815154910087585,
-      "learning_rate": 0.0005547721150738066,
-      "loss": 3.9276,
+      "epoch": 0.7654986522911051,
+      "grad_norm": 0.6074730753898621,
+      "learning_rate": 0.0005546745817593091,
+      "loss": 3.9129,
       "step": 7100
     },
     {
-      "epoch": 0.7695619416639758,
-      "grad_norm": 0.5519508123397827,
-      "learning_rate": 0.0005544488740437452,
-      "loss": 3.9182,
+      "epoch": 0.77088948787062,
+      "grad_norm": 0.6270312070846558,
+      "learning_rate": 0.0005543507825148408,
+      "loss": 3.9175,
       "step": 7150
     },
     {
-      "epoch": 0.7749434937035841,
-      "grad_norm": 0.5207048654556274,
-      "learning_rate": 0.0005541256330136838,
-      "loss": 3.9456,
+      "epoch": 0.7762803234501348,
+      "grad_norm": 0.6784051060676575,
+      "learning_rate": 0.0005540269832703723,
+      "loss": 3.895,
       "step": 7200
     },
     {
-      "epoch": 0.7803250457431924,
-      "grad_norm": 0.670373797416687,
-      "learning_rate": 0.0005538023919836224,
-      "loss": 3.9291,
+      "epoch": 0.7816711590296496,
+      "grad_norm": 0.6523202657699585,
+      "learning_rate": 0.0005537031840259039,
+      "loss": 3.9237,
       "step": 7250
     },
     {
-      "epoch": 0.7857065977828006,
-      "grad_norm": 0.5639854073524475,
-      "learning_rate": 0.000553479150953561,
-      "loss": 3.9201,
+      "epoch": 0.7870619946091644,
+      "grad_norm": 0.6272968053817749,
+      "learning_rate": 0.0005533793847814354,
+      "loss": 3.8987,
       "step": 7300
     },
     {
-      "epoch": 0.7910881498224088,
-      "grad_norm": 0.6556904315948486,
-      "learning_rate": 0.0005531559099234996,
-      "loss": 3.9347,
+      "epoch": 0.7924528301886793,
+      "grad_norm": 0.5943434834480286,
+      "learning_rate": 0.000553055585536967,
+      "loss": 3.918,
       "step": 7350
     },
     {
-      "epoch": 0.796469701862017,
-      "grad_norm": 0.5908870100975037,
-      "learning_rate": 0.0005528326688934381,
-      "loss": 3.9179,
+      "epoch": 0.7978436657681941,
+      "grad_norm": 0.6551655530929565,
+      "learning_rate": 0.0005527317862924987,
+      "loss": 3.9191,
       "step": 7400
     },
     {
-      "epoch": 0.8018512539016253,
-      "grad_norm": 0.5798718333244324,
-      "learning_rate": 0.0005525094278633767,
-      "loss": 3.9148,
+      "epoch": 0.8032345013477089,
+      "grad_norm": 0.5754203200340271,
+      "learning_rate": 0.0005524079870480301,
+      "loss": 3.9101,
       "step": 7450
     },
     {
-      "epoch": 0.8072328059412335,
-      "grad_norm": 0.6623952388763428,
-      "learning_rate": 0.0005521861868333154,
-      "loss": 3.9189,
+      "epoch": 0.8086253369272237,
+      "grad_norm": 0.573434591293335,
+      "learning_rate": 0.0005520841878035618,
+      "loss": 3.929,
       "step": 7500
     },
     {
-      "epoch": 0.8126143579808417,
-      "grad_norm": 0.7111433744430542,
-      "learning_rate": 0.0005518629458032539,
-      "loss": 3.8909,
+      "epoch": 0.8140161725067385,
+      "grad_norm": 0.5679186582565308,
+      "learning_rate": 0.0005517603885590933,
+      "loss": 3.901,
       "step": 7550
     },
     {
-      "epoch": 0.8179959100204499,
-      "grad_norm": 0.6083288192749023,
-      "learning_rate": 0.0005515397047731925,
-      "loss": 3.9062,
+      "epoch": 0.8194070080862533,
+      "grad_norm": 0.6327888369560242,
+      "learning_rate": 0.0005514365893146249,
+      "loss": 3.8944,
       "step": 7600
     },
     {
-      "epoch": 0.8233774620600581,
-      "grad_norm": 0.6324785351753235,
-      "learning_rate": 0.000551216463743131,
-      "loss": 3.8885,
+      "epoch": 0.8247978436657682,
+      "grad_norm": 0.4915516674518585,
+      "learning_rate": 0.0005511127900701564,
+      "loss": 3.8896,
       "step": 7650
     },
     {
-      "epoch": 0.8287590140996663,
-      "grad_norm": 0.6778486371040344,
-      "learning_rate": 0.0005508932227130697,
-      "loss": 3.9184,
+      "epoch": 0.8301886792452831,
+      "grad_norm": 0.5711592435836792,
+      "learning_rate": 0.000550788990825688,
+      "loss": 3.917,
       "step": 7700
     },
     {
-      "epoch": 0.8341405661392746,
-      "grad_norm": 0.5886597037315369,
-      "learning_rate": 0.0005505699816830083,
-      "loss": 3.8933,
+      "epoch": 0.8355795148247979,
+      "grad_norm": 0.5974284410476685,
+      "learning_rate": 0.0005504651915812196,
+      "loss": 3.9154,
       "step": 7750
     },
     {
-      "epoch": 0.8395221181788828,
-      "grad_norm": 0.6386252641677856,
-      "learning_rate": 0.0005502467406529469,
-      "loss": 3.9005,
+      "epoch": 0.8409703504043127,
+      "grad_norm": 0.6452850699424744,
+      "learning_rate": 0.0005501413923367512,
+      "loss": 3.8774,
       "step": 7800
     },
     {
-      "epoch": 0.844903670218491,
-      "grad_norm": 0.581160306930542,
-      "learning_rate": 0.0005499234996228854,
-      "loss": 3.8939,
+      "epoch": 0.8463611859838275,
+      "grad_norm": 0.58039391040802,
+      "learning_rate": 0.0005498175930922827,
+      "loss": 3.8889,
       "step": 7850
     },
     {
-      "epoch": 0.8502852222580992,
-      "grad_norm": 0.6273126006126404,
-      "learning_rate": 0.000549600258592824,
-      "loss": 3.8867,
+      "epoch": 0.8517520215633423,
+      "grad_norm": 0.5420709252357483,
+      "learning_rate": 0.0005494937938478143,
+      "loss": 3.8947,
       "step": 7900
     },
     {
-      "epoch": 0.8556667742977074,
-      "grad_norm": 0.6495140790939331,
-      "learning_rate": 0.0005492770175627626,
-      "loss": 3.9011,
+      "epoch": 0.8571428571428571,
+      "grad_norm": 0.5335559248924255,
+      "learning_rate": 0.0005491699946033459,
+      "loss": 3.8837,
       "step": 7950
     },
     {
-      "epoch": 0.8610483263373157,
-      "grad_norm": 0.6411253213882446,
-      "learning_rate": 0.0005489537765327011,
-      "loss": 3.8889,
+      "epoch": 0.862533692722372,
+      "grad_norm": 0.564753532409668,
+      "learning_rate": 0.0005488461953588775,
+      "loss": 3.8848,
       "step": 8000
     },
     {
-      "epoch": 0.8610483263373157,
-      "eval_accuracy": 0.33802684834588476,
-      "eval_loss": 3.8203883171081543,
-      "eval_runtime": 185.6274,
-      "eval_samples_per_second": 97.028,
-      "eval_steps_per_second": 6.066,
+      "epoch": 0.862533692722372,
+      "eval_accuracy": 0.3382142745773647,
+      "eval_loss": 3.8157401084899902,
+      "eval_runtime": 184.0292,
+      "eval_samples_per_second": 97.87,
+      "eval_steps_per_second": 6.119,
       "step": 8000
     },
     {
-      "epoch": 0.8664298783769239,
-      "grad_norm": 0.5747044086456299,
-      "learning_rate": 0.0005486305355026398,
-      "loss": 3.8982,
+      "epoch": 0.8679245283018868,
+      "grad_norm": 0.5700918436050415,
+      "learning_rate": 0.000548522396114409,
+      "loss": 3.8844,
       "step": 8050
     },
     {
-      "epoch": 0.8718114304165321,
-      "grad_norm": 0.545897364616394,
-      "learning_rate": 0.0005483072944725783,
-      "loss": 3.8823,
+      "epoch": 0.8733153638814016,
+      "grad_norm": 0.6043035387992859,
+      "learning_rate": 0.0005481985968699406,
+      "loss": 3.8949,
       "step": 8100
     },
     {
-      "epoch": 0.8771929824561403,
-      "grad_norm": 0.6497473120689392,
-      "learning_rate": 0.000547984053442517,
-      "loss": 3.8708,
+      "epoch": 0.8787061994609164,
+      "grad_norm": 0.5310813188552856,
+      "learning_rate": 0.0005478747976254721,
+      "loss": 3.8731,
       "step": 8150
     },
     {
-      "epoch": 0.8825745344957485,
-      "grad_norm": 0.5869677066802979,
-      "learning_rate": 0.0005476608124124555,
-      "loss": 3.8915,
+      "epoch": 0.8840970350404312,
+      "grad_norm": 0.5834559202194214,
+      "learning_rate": 0.0005475509983810037,
+      "loss": 3.8765,
       "step": 8200
     },
     {
-      "epoch": 0.8879560865353568,
-      "grad_norm": 0.6175825595855713,
-      "learning_rate": 0.000547337571382394,
-      "loss": 3.8862,
+      "epoch": 0.889487870619946,
+      "grad_norm": 0.5974069237709045,
+      "learning_rate": 0.0005472271991365352,
+      "loss": 3.8714,
       "step": 8250
     },
     {
-      "epoch": 0.8933376385749651,
-      "grad_norm": 0.6012660264968872,
-      "learning_rate": 0.0005470143303523327,
-      "loss": 3.8825,
+      "epoch": 0.894878706199461,
+      "grad_norm": 0.6031675934791565,
+      "learning_rate": 0.0005469033998920669,
+      "loss": 3.8543,
       "step": 8300
     },
     {
-      "epoch": 0.8987191906145733,
-      "grad_norm": 0.5652008652687073,
-      "learning_rate": 0.0005466910893222712,
-      "loss": 3.8759,
+      "epoch": 0.9002695417789758,
+      "grad_norm": 0.5460870862007141,
+      "learning_rate": 0.0005465860766324878,
+      "loss": 3.8413,
       "step": 8350
     },
     {
-      "epoch": 0.9041007426541815,
-      "grad_norm": 0.5548220872879028,
-      "learning_rate": 0.0005463678482922099,
-      "loss": 3.8608,
+      "epoch": 0.9056603773584906,
+      "grad_norm": 0.5728166699409485,
+      "learning_rate": 0.0005462622773880194,
+      "loss": 3.888,
       "step": 8400
     },
     {
-      "epoch": 0.9094822946937897,
-      "grad_norm": 0.6385374069213867,
-      "learning_rate": 0.0005460446072621484,
-      "loss": 3.8615,
+      "epoch": 0.9110512129380054,
+      "grad_norm": 0.6531025171279907,
+      "learning_rate": 0.000545938478143551,
+      "loss": 3.8563,
       "step": 8450
     },
     {
-      "epoch": 0.9148638467333979,
-      "grad_norm": 0.660474419593811,
-      "learning_rate": 0.000545721366232087,
-      "loss": 3.8685,
+      "epoch": 0.9164420485175202,
+      "grad_norm": 0.5521366000175476,
+      "learning_rate": 0.0005456146788990825,
+      "loss": 3.8441,
       "step": 8500
     },
     {
-      "epoch": 0.9202453987730062,
-      "grad_norm": 0.6575385928153992,
-      "learning_rate": 0.0005453981252020256,
-      "loss": 3.8541,
+      "epoch": 0.921832884097035,
+      "grad_norm": 0.62563157081604,
+      "learning_rate": 0.000545290879654614,
+      "loss": 3.858,
       "step": 8550
     },
     {
-      "epoch": 0.9256269508126144,
-      "grad_norm": 0.5555447936058044,
-      "learning_rate": 0.0005450748841719643,
-      "loss": 3.877,
+      "epoch": 0.9272237196765498,
+      "grad_norm": 0.6563644409179688,
+      "learning_rate": 0.0005449670804101457,
+      "loss": 3.8569,
       "step": 8600
     },
     {
-      "epoch": 0.9310085028522226,
-      "grad_norm": 0.6523729562759399,
-      "learning_rate": 0.0005447516431419028,
-      "loss": 3.8495,
+      "epoch": 0.9326145552560647,
+      "grad_norm": 0.5994681119918823,
+      "learning_rate": 0.0005446432811656773,
+      "loss": 3.8598,
       "step": 8650
     },
     {
-      "epoch": 0.9363900548918308,
-      "grad_norm": 0.5873777270317078,
-      "learning_rate": 0.0005444284021118413,
-      "loss": 3.8549,
+      "epoch": 0.9380053908355795,
+      "grad_norm": 0.5892212986946106,
+      "learning_rate": 0.0005443194819212088,
+      "loss": 3.8369,
       "step": 8700
     },
     {
-      "epoch": 0.941771606931439,
-      "grad_norm": 0.5743074417114258,
-      "learning_rate": 0.0005441051610817799,
-      "loss": 3.8675,
+      "epoch": 0.9433962264150944,
+      "grad_norm": 0.6683197021484375,
+      "learning_rate": 0.0005439956826767404,
+      "loss": 3.855,
       "step": 8750
     },
     {
-      "epoch": 0.9471531589710472,
-      "grad_norm": 0.6072008013725281,
-      "learning_rate": 0.0005437819200517185,
-      "loss": 3.8561,
+      "epoch": 0.9487870619946092,
+      "grad_norm": 0.6023868918418884,
+      "learning_rate": 0.000543671883432272,
+      "loss": 3.8422,
       "step": 8800
     },
     {
-      "epoch": 0.9525347110106555,
-      "grad_norm": 0.5983343124389648,
-      "learning_rate": 0.0005434586790216571,
-      "loss": 3.8494,
+      "epoch": 0.954177897574124,
+      "grad_norm": 0.5741904973983765,
+      "learning_rate": 0.0005433480841878035,
+      "loss": 3.8376,
       "step": 8850
     },
     {
-      "epoch": 0.9579162630502637,
-      "grad_norm": 0.6041466593742371,
-      "learning_rate": 0.0005431354379915957,
-      "loss": 3.8476,
+      "epoch": 0.9595687331536388,
+      "grad_norm": 0.650631844997406,
+      "learning_rate": 0.000543024284943335,
+      "loss": 3.8293,
       "step": 8900
     },
     {
-      "epoch": 0.9632978150898719,
-      "grad_norm": 0.5659998059272766,
-      "learning_rate": 0.0005428121969615343,
-      "loss": 3.8525,
+      "epoch": 0.9649595687331537,
+      "grad_norm": 0.5889686346054077,
+      "learning_rate": 0.0005427004856988667,
+      "loss": 3.8449,
       "step": 8950
     },
     {
-      "epoch": 0.9686793671294801,
-      "grad_norm": 0.6054639220237732,
-      "learning_rate": 0.0005424889559314729,
-      "loss": 3.8508,
+      "epoch": 0.9703504043126685,
+      "grad_norm": 0.6525272130966187,
+      "learning_rate": 0.0005423766864543982,
+      "loss": 3.8455,
       "step": 9000
     },
     {
-      "epoch": 0.9686793671294801,
-      "eval_accuracy": 0.34104316116450256,
-      "eval_loss": 3.7828726768493652,
-      "eval_runtime": 185.41,
-      "eval_samples_per_second": 97.141,
-      "eval_steps_per_second": 6.073,
+      "epoch": 0.9703504043126685,
+      "eval_accuracy": 0.3415290568788089,
+      "eval_loss": 3.778981924057007,
+      "eval_runtime": 185.8166,
+      "eval_samples_per_second": 96.929,
+      "eval_steps_per_second": 6.06,
       "step": 9000
     },
     {
-      "epoch": 0.9740609191690883,
-      "grad_norm": 0.5490997433662415,
-      "learning_rate": 0.0005421657149014114,
-      "loss": 3.8243,
+      "epoch": 0.9757412398921833,
+      "grad_norm": 0.5841635465621948,
+      "learning_rate": 0.0005420528872099298,
+      "loss": 3.8287,
       "step": 9050
     },
     {
-      "epoch": 0.9794424712086965,
-      "grad_norm": 0.5536127090454102,
-      "learning_rate": 0.00054184247387135,
-      "loss": 3.8392,
+      "epoch": 0.9811320754716981,
+      "grad_norm": 0.5249017477035522,
+      "learning_rate": 0.0005417290879654613,
+      "loss": 3.828,
       "step": 9100
     },
     {
-      "epoch": 0.9848240232483048,
-      "grad_norm": 0.6085931658744812,
-      "learning_rate": 0.0005415192328412885,
-      "loss": 3.8462,
+      "epoch": 0.9865229110512129,
+      "grad_norm": 0.5723927617073059,
+      "learning_rate": 0.000541405288720993,
+      "loss": 3.8411,
       "step": 9150
     },
     {
-      "epoch": 0.9902055752879131,
-      "grad_norm": 0.6172447800636292,
-      "learning_rate": 0.0005411959918112272,
-      "loss": 3.8469,
+      "epoch": 0.9919137466307277,
+      "grad_norm": 0.541815459728241,
+      "learning_rate": 0.0005410814894765245,
+      "loss": 3.8225,
       "step": 9200
     },
     {
-      "epoch": 0.9955871273275213,
-      "grad_norm": 0.589678943157196,
-      "learning_rate": 0.0005408727507811658,
-      "loss": 3.8478,
+      "epoch": 0.9973045822102425,
+      "grad_norm": 0.6346942782402039,
+      "learning_rate": 0.0005407576902320561,
+      "loss": 3.8505,
       "step": 9250
     },
     {
-      "epoch": 1.0009686793671295,
-      "grad_norm": 0.5819297432899475,
-      "learning_rate": 0.0005405495097511044,
-      "loss": 3.8261,
+      "epoch": 1.0026954177897573,
+      "grad_norm": 0.6351709961891174,
+      "learning_rate": 0.0005404338909875876,
+      "loss": 3.7874,
       "step": 9300
     },
     {
-      "epoch": 1.0063502314067376,
-      "grad_norm": 0.6677573323249817,
-      "learning_rate": 0.0005402262687210429,
-      "loss": 3.7952,
+      "epoch": 1.0080862533692723,
+      "grad_norm": 0.5384916663169861,
+      "learning_rate": 0.0005401100917431192,
+      "loss": 3.7621,
       "step": 9350
     },
     {
-      "epoch": 1.011731783446346,
-      "grad_norm": 0.6139549016952515,
-      "learning_rate": 0.0005399030276909816,
-      "loss": 3.786,
+      "epoch": 1.013477088948787,
+      "grad_norm": 0.6582435965538025,
+      "learning_rate": 0.0005397862924986508,
+      "loss": 3.7765,
       "step": 9400
     },
     {
-      "epoch": 1.017113335485954,
-      "grad_norm": 0.5575100183486938,
-      "learning_rate": 0.0005395797866609201,
-      "loss": 3.7745,
+      "epoch": 1.0188679245283019,
+      "grad_norm": 0.5163936614990234,
+      "learning_rate": 0.0005394624932541824,
+      "loss": 3.7853,
       "step": 9450
     },
     {
-      "epoch": 1.0224948875255624,
-      "grad_norm": 0.5324420928955078,
-      "learning_rate": 0.0005392565456308587,
-      "loss": 3.7737,
+      "epoch": 1.0242587601078168,
+      "grad_norm": 0.5641939043998718,
+      "learning_rate": 0.0005391386940097139,
+      "loss": 3.7533,
       "step": 9500
     },
     {
-      "epoch": 1.0278764395651705,
-      "grad_norm": 0.5157521367073059,
-      "learning_rate": 0.0005389333046007973,
-      "loss": 3.767,
+      "epoch": 1.0296495956873315,
+      "grad_norm": 0.5747295618057251,
+      "learning_rate": 0.0005388148947652455,
+      "loss": 3.7649,
       "step": 9550
     },
     {
-      "epoch": 1.0332579916047788,
-      "grad_norm": 0.5273111462593079,
-      "learning_rate": 0.0005386100635707358,
-      "loss": 3.782,
+      "epoch": 1.0350404312668464,
+      "grad_norm": 0.5835955142974854,
+      "learning_rate": 0.000538491095520777,
+      "loss": 3.7931,
       "step": 9600
     },
     {
-      "epoch": 1.0386395436443872,
-      "grad_norm": 0.6705301403999329,
-      "learning_rate": 0.0005382868225406745,
-      "loss": 3.7519,
+      "epoch": 1.0404312668463611,
+      "grad_norm": 0.5599415302276611,
+      "learning_rate": 0.0005381672962763086,
+      "loss": 3.7511,
       "step": 9650
     },
     {
-      "epoch": 1.0440210956839953,
-      "grad_norm": 0.6219047904014587,
-      "learning_rate": 0.000537963581510613,
-      "loss": 3.762,
+      "epoch": 1.045822102425876,
+      "grad_norm": 0.5540547370910645,
+      "learning_rate": 0.0005378434970318403,
+      "loss": 3.7768,
       "step": 9700
     },
     {
-      "epoch": 1.0494026477236036,
-      "grad_norm": 0.644806981086731,
-      "learning_rate": 0.0005376403404805517,
-      "loss": 3.7834,
+      "epoch": 1.0512129380053907,
+      "grad_norm": 0.5856996178627014,
+      "learning_rate": 0.0005375196977873718,
+      "loss": 3.7522,
       "step": 9750
     },
     {
-      "epoch": 1.0547841997632117,
-      "grad_norm": 0.6287768483161926,
-      "learning_rate": 0.0005373170994504902,
-      "loss": 3.7877,
+      "epoch": 1.0566037735849056,
+      "grad_norm": 0.6114075183868408,
+      "learning_rate": 0.0005371958985429034,
+      "loss": 3.7652,
       "step": 9800
     },
     {
-      "epoch": 1.06016575180282,
-      "grad_norm": 0.5888068079948425,
-      "learning_rate": 0.0005369938584204287,
-      "loss": 3.7641,
+      "epoch": 1.0619946091644206,
+      "grad_norm": 0.5975747108459473,
+      "learning_rate": 0.0005368720992984349,
+      "loss": 3.7686,
       "step": 9850
     },
     {
-      "epoch": 1.0655473038424281,
-      "grad_norm": 0.5658965110778809,
-      "learning_rate": 0.0005366706173903674,
-      "loss": 3.7561,
+      "epoch": 1.0673854447439353,
+      "grad_norm": 0.6365913152694702,
+      "learning_rate": 0.0005365483000539665,
+      "loss": 3.7625,
       "step": 9900
     },
     {
-      "epoch": 1.0709288558820365,
-      "grad_norm": 0.5976064801216125,
-      "learning_rate": 0.0005363473763603059,
-      "loss": 3.7677,
+      "epoch": 1.0727762803234502,
+      "grad_norm": 0.6662463545799255,
+      "learning_rate": 0.0005362245008094981,
+      "loss": 3.78,
       "step": 9950
     },
     {
-      "epoch": 1.0763104079216446,
-      "grad_norm": 0.6668145060539246,
-      "learning_rate": 0.0005360241353302445,
-      "loss": 3.7758,
+      "epoch": 1.0781671159029649,
+      "grad_norm": 0.5857443809509277,
+      "learning_rate": 0.0005359007015650297,
+      "loss": 3.7676,
       "step": 10000
     },
     {
-      "epoch": 1.0763104079216446,
-      "eval_accuracy": 0.3436720177851739,
-      "eval_loss": 3.7506744861602783,
-      "eval_runtime": 185.8788,
-      "eval_samples_per_second": 96.896,
-      "eval_steps_per_second": 6.058,
+      "epoch": 1.0781671159029649,
+      "eval_accuracy": 0.34476778215878295,
+      "eval_loss": 3.7468345165252686,
+      "eval_runtime": 184.1281,
+      "eval_samples_per_second": 97.818,
+      "eval_steps_per_second": 6.115,
       "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": 8.3610314735616e+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 sha256:95b3b3fc5667e97ef3eac8d3857772a2ec3bc2a210796c6f18114614c7fa451d
+oid sha256:edc4e3e4516bbf1069b744f3697cef9eb8a2b5574a124f5bdcc4211e91e86f41
 size 5304