Training in progress, step 10000, checkpoint

checkpoint-10000/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:25deb513368145b7e944c5045f37b0e6f12625ba91774faa9180764f7471dfeb
+oid sha256:070e2b2d14b1bd87e86c5976ba07f8004c9eb9841a428baaffb52e6864ab76c4
 size 503128704

checkpoint-10000/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f999256541f86dc6d7445a7e94b1941a07d8318b6279b27f101bd0fdebf226d3
+oid sha256:b3b9a7c435f4ba35e5f399d3055c77b80e15715bc331fc845581a6f2f06adc79
 size 1006351290

checkpoint-10000/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8c85560428bbc40e9dac5c7f840c356a4d2ac2d603c9375614da04a65f4bfd8a
+oid sha256:53a47b99e43a46630effbfb45e7223c8a493d5b48b00aebac2470d03eafb7364
 size 14244

checkpoint-10000/scheduler.pt

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

checkpoint-10000/trainer_state.json

@@ -1,7 +1,7 @@
 {
-  "best_metric": 3.7408206462860107,
-  "best_model_checkpoint": "/scratch/cl5625/exceptions/models/100M_1208/checkpoint-10000",
-  "epoch": 1.0763104079216446,
+  "best_metric": 3.7435693740844727,
+  "best_model_checkpoint": "/scratch/cl5625/exceptions/models/100M_high_1000_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": 4.5097432136535645,
+      "epoch": 0.005390835579514825,
+      "grad_norm": 1.6313424110412598,
       "learning_rate": 0.0003,
-      "loss": 8.6268,
+      "loss": 8.7945,
       "step": 50
     },
     {
-      "epoch": 0.010763104079216447,
-      "grad_norm": 2.590397834777832,
+      "epoch": 0.01078167115902965,
+      "grad_norm": 2.126284599304199,
       "learning_rate": 0.0006,
-      "loss": 6.9687,
+      "loss": 6.9919,
       "step": 100
     },
     {
-      "epoch": 0.01614465611882467,
-      "grad_norm": 2.468424081802368,
-      "learning_rate": 0.0005996767589699385,
-      "loss": 6.5499,
+      "epoch": 0.016172506738544475,
+      "grad_norm": 1.1851738691329956,
+      "learning_rate": 0.0005996762007555315,
+      "loss": 6.5278,
       "step": 150
     },
     {
-      "epoch": 0.021526208158432893,
-      "grad_norm": 1.429585576057434,
-      "learning_rate": 0.0005993535179398771,
-      "loss": 6.2827,
+      "epoch": 0.0215633423180593,
+      "grad_norm": 1.667454719543457,
+      "learning_rate": 0.000599352401511063,
+      "loss": 6.2628,
       "step": 200
     },
     {
-      "epoch": 0.026907760198041114,
-      "grad_norm": 1.5892139673233032,
-      "learning_rate": 0.0005990302769098158,
-      "loss": 6.1392,
+      "epoch": 0.026954177897574125,
+      "grad_norm": 0.8496881723403931,
+      "learning_rate": 0.0005990286022665946,
+      "loss": 6.0804,
       "step": 250
     },
     {
-      "epoch": 0.03228931223764934,
-      "grad_norm": 2.1433768272399902,
-      "learning_rate": 0.0005987070358797543,
-      "loss": 5.9916,
+      "epoch": 0.03234501347708895,
+      "grad_norm": 2.0049641132354736,
+      "learning_rate": 0.0005987048030221263,
+      "loss": 5.9684,
       "step": 300
     },
     {
-      "epoch": 0.03767086427725756,
-      "grad_norm": 1.4292688369750977,
-      "learning_rate": 0.0005983837948496929,
-      "loss": 5.9183,
+      "epoch": 0.03773584905660377,
+      "grad_norm": 1.6133452653884888,
+      "learning_rate": 0.0005983810037776578,
+      "loss": 5.8836,
       "step": 350
     },
     {
-      "epoch": 0.04305241631686579,
-      "grad_norm": 1.8690763711929321,
-      "learning_rate": 0.0005980605538196314,
-      "loss": 5.8218,
+      "epoch": 0.0431266846361186,
+      "grad_norm": 1.9989739656448364,
+      "learning_rate": 0.0005980572045331894,
+      "loss": 5.8135,
       "step": 400
     },
     {
-      "epoch": 0.048433968356474004,
-      "grad_norm": 1.1355857849121094,
-      "learning_rate": 0.0005977373127895701,
-      "loss": 5.7251,
+      "epoch": 0.04851752021563342,
+      "grad_norm": 1.4867981672286987,
+      "learning_rate": 0.0005977334052887209,
+      "loss": 5.7163,
       "step": 450
     },
     {
-      "epoch": 0.05381552039608223,
-      "grad_norm": 1.7394969463348389,
-      "learning_rate": 0.0005974140717595086,
-      "loss": 5.6661,
+      "epoch": 0.05390835579514825,
+      "grad_norm": 1.8946263790130615,
+      "learning_rate": 0.0005974096060442526,
+      "loss": 5.6654,
       "step": 500
     },
     {
-      "epoch": 0.05919707243569045,
-      "grad_norm": 1.3288583755493164,
-      "learning_rate": 0.0005970908307294472,
-      "loss": 5.6096,
+      "epoch": 0.05929919137466307,
+      "grad_norm": 1.4091145992279053,
+      "learning_rate": 0.0005970858067997841,
+      "loss": 5.5712,
       "step": 550
     },
     {
-      "epoch": 0.06457862447529868,
-      "grad_norm": 1.4666411876678467,
-      "learning_rate": 0.0005967675896993858,
-      "loss": 5.5193,
+      "epoch": 0.0646900269541779,
+      "grad_norm": 1.4549801349639893,
+      "learning_rate": 0.0005967620075553157,
+      "loss": 5.5428,
       "step": 600
     },
     {
-      "epoch": 0.0699601765149069,
-      "grad_norm": 1.0586670637130737,
-      "learning_rate": 0.0005964443486693243,
-      "loss": 5.4711,
+      "epoch": 0.07008086253369272,
+      "grad_norm": 1.4586806297302246,
+      "learning_rate": 0.0005964382083108472,
+      "loss": 5.4383,
       "step": 650
     },
     {
-      "epoch": 0.07534172855451512,
-      "grad_norm": 1.1426615715026855,
-      "learning_rate": 0.000596121107639263,
-      "loss": 5.3864,
+      "epoch": 0.07547169811320754,
+      "grad_norm": 1.593138337135315,
+      "learning_rate": 0.0005961144090663788,
+      "loss": 5.4101,
       "step": 700
     },
     {
-      "epoch": 0.08072328059412334,
-      "grad_norm": 0.9819518327713013,
-      "learning_rate": 0.0005957978666092015,
-      "loss": 5.3386,
+      "epoch": 0.08086253369272237,
+      "grad_norm": 1.3180700540542603,
+      "learning_rate": 0.0005957906098219104,
+      "loss": 5.3208,
       "step": 750
     },
     {
-      "epoch": 0.08610483263373157,
-      "grad_norm": 1.3782721757888794,
-      "learning_rate": 0.0005954746255791401,
-      "loss": 5.2565,
+      "epoch": 0.0862533692722372,
+      "grad_norm": 1.2883632183074951,
+      "learning_rate": 0.0005954668105774419,
+      "loss": 5.2751,
       "step": 800
     },
     {
-      "epoch": 0.09148638467333979,
-      "grad_norm": 1.3616726398468018,
-      "learning_rate": 0.0005951513845490787,
-      "loss": 5.2438,
+      "epoch": 0.09164420485175202,
+      "grad_norm": 1.293715476989746,
+      "learning_rate": 0.0005951430113329735,
+      "loss": 5.2205,
       "step": 850
     },
     {
-      "epoch": 0.09686793671294801,
-      "grad_norm": 1.4289978742599487,
-      "learning_rate": 0.0005948281435190174,
-      "loss": 5.1724,
+      "epoch": 0.09703504043126684,
+      "grad_norm": 1.1325150728225708,
+      "learning_rate": 0.0005948192120885051,
+      "loss": 5.1856,
       "step": 900
     },
     {
-      "epoch": 0.10224948875255624,
-      "grad_norm": 0.9795863628387451,
-      "learning_rate": 0.0005945049024889559,
-      "loss": 5.1255,
+      "epoch": 0.10242587601078167,
+      "grad_norm": 1.1576133966445923,
+      "learning_rate": 0.0005944954128440366,
+      "loss": 5.1321,
       "step": 950
     },
     {
-      "epoch": 0.10763104079216446,
-      "grad_norm": 1.8023662567138672,
-      "learning_rate": 0.0005941816614588944,
-      "loss": 5.093,
+      "epoch": 0.1078167115902965,
+      "grad_norm": 1.04938542842865,
+      "learning_rate": 0.0005941716135995682,
+      "loss": 5.1164,
       "step": 1000
     },
     {
-      "epoch": 0.10763104079216446,
-      "eval_accuracy": 0.22650661082197973,
-      "eval_loss": 5.036528587341309,
-      "eval_runtime": 180.4694,
-      "eval_samples_per_second": 99.801,
-      "eval_steps_per_second": 6.239,
+      "epoch": 0.1078167115902965,
+      "eval_accuracy": 0.22693144361333437,
+      "eval_loss": 5.0317254066467285,
+      "eval_runtime": 185.7135,
+      "eval_samples_per_second": 96.983,
+      "eval_steps_per_second": 6.063,
       "step": 1000
     },
     {
-      "epoch": 0.11301259283177269,
-      "grad_norm": 1.4686137437820435,
-      "learning_rate": 0.000593858420428833,
-      "loss": 5.0711,
+      "epoch": 0.11320754716981132,
+      "grad_norm": 1.0911813974380493,
+      "learning_rate": 0.0005938478143550997,
+      "loss": 5.0764,
       "step": 1050
     },
     {
-      "epoch": 0.1183941448713809,
-      "grad_norm": 0.9618597626686096,
-      "learning_rate": 0.0005935351793987716,
-      "loss": 5.0164,
+      "epoch": 0.11859838274932614,
+      "grad_norm": 0.8553237915039062,
+      "learning_rate": 0.0005935240151106314,
+      "loss": 5.0024,
       "step": 1100
     },
     {
-      "epoch": 0.12377569691098914,
-      "grad_norm": 0.8527742028236389,
-      "learning_rate": 0.0005932119383687103,
-      "loss": 4.9661,
+      "epoch": 0.12398921832884097,
+      "grad_norm": 1.2252320051193237,
+      "learning_rate": 0.0005932002158661629,
+      "loss": 5.0013,
       "step": 1150
     },
     {
-      "epoch": 0.12915724895059735,
-      "grad_norm": 1.0223184823989868,
-      "learning_rate": 0.0005928886973386488,
-      "loss": 4.9618,
+      "epoch": 0.1293800539083558,
+      "grad_norm": 1.0430017709732056,
+      "learning_rate": 0.0005928764166216945,
+      "loss": 4.9657,
       "step": 1200
     },
     {
-      "epoch": 0.13453880099020557,
-      "grad_norm": 1.365889072418213,
-      "learning_rate": 0.0005925654563085874,
-      "loss": 4.9245,
+      "epoch": 0.1347708894878706,
+      "grad_norm": 1.000243902206421,
+      "learning_rate": 0.000592552617377226,
+      "loss": 4.9126,
       "step": 1250
     },
     {
-      "epoch": 0.1399203530298138,
-      "grad_norm": 0.9643410444259644,
-      "learning_rate": 0.000592242215278526,
-      "loss": 4.9062,
+      "epoch": 0.14016172506738545,
+      "grad_norm": 1.2672815322875977,
+      "learning_rate": 0.0005922288181327577,
+      "loss": 4.9131,
       "step": 1300
     },
     {
-      "epoch": 0.14530190506942203,
-      "grad_norm": 1.0850372314453125,
-      "learning_rate": 0.0005919189742484645,
-      "loss": 4.9208,
+      "epoch": 0.14555256064690028,
+      "grad_norm": 0.8482131361961365,
+      "learning_rate": 0.0005919050188882893,
+      "loss": 4.8801,
       "step": 1350
     },
     {
-      "epoch": 0.15068345710903025,
-      "grad_norm": 1.0376044511795044,
-      "learning_rate": 0.0005915957332184032,
-      "loss": 4.8545,
+      "epoch": 0.1509433962264151,
+      "grad_norm": 1.2260438203811646,
+      "learning_rate": 0.0005915812196438207,
+      "loss": 4.8478,
       "step": 1400
     },
     {
-      "epoch": 0.15606500914863847,
-      "grad_norm": 0.8379460573196411,
-      "learning_rate": 0.0005912724921883417,
-      "loss": 4.8182,
+      "epoch": 0.15633423180592992,
+      "grad_norm": 0.9943752884864807,
+      "learning_rate": 0.0005912574203993524,
+      "loss": 4.8204,
       "step": 1450
     },
     {
-      "epoch": 0.16144656118824668,
-      "grad_norm": 1.0445117950439453,
-      "learning_rate": 0.0005909492511582803,
-      "loss": 4.8267,
+      "epoch": 0.16172506738544473,
+      "grad_norm": 0.9812339544296265,
+      "learning_rate": 0.0005909336211548839,
+      "loss": 4.8138,
       "step": 1500
     },
     {
-      "epoch": 0.1668281132278549,
-      "grad_norm": 1.0523771047592163,
-      "learning_rate": 0.0005906260101282189,
-      "loss": 4.7692,
+      "epoch": 0.16711590296495957,
+      "grad_norm": 0.9307915568351746,
+      "learning_rate": 0.0005906098219104155,
+      "loss": 4.7793,
       "step": 1550
     },
     {
-      "epoch": 0.17220966526746315,
-      "grad_norm": 0.8830639719963074,
-      "learning_rate": 0.0005903027690981575,
-      "loss": 4.7444,
+      "epoch": 0.1725067385444744,
+      "grad_norm": 1.3052858114242554,
+      "learning_rate": 0.000590286022665947,
+      "loss": 4.7939,
       "step": 1600
     },
     {
-      "epoch": 0.17759121730707136,
-      "grad_norm": 0.9524279236793518,
-      "learning_rate": 0.000589979528068096,
-      "loss": 4.7351,
+      "epoch": 0.1778975741239892,
+      "grad_norm": 1.1240953207015991,
+      "learning_rate": 0.0005899622234214787,
+      "loss": 4.7417,
       "step": 1650
     },
     {
-      "epoch": 0.18297276934667958,
-      "grad_norm": 0.8538389801979065,
-      "learning_rate": 0.0005896562870380347,
-      "loss": 4.715,
+      "epoch": 0.18328840970350405,
+      "grad_norm": 1.0071319341659546,
+      "learning_rate": 0.0005896384241770102,
+      "loss": 4.6972,
       "step": 1700
     },
     {
-      "epoch": 0.1883543213862878,
-      "grad_norm": 1.3721179962158203,
-      "learning_rate": 0.0005893330460079732,
-      "loss": 4.6854,
+      "epoch": 0.18867924528301888,
+      "grad_norm": 0.891978919506073,
+      "learning_rate": 0.0005893146249325418,
+      "loss": 4.688,
       "step": 1750
     },
     {
-      "epoch": 0.19373587342589602,
-      "grad_norm": 1.0248409509658813,
-      "learning_rate": 0.0005890098049779118,
-      "loss": 4.6675,
+      "epoch": 0.1940700808625337,
+      "grad_norm": 1.0216976404190063,
+      "learning_rate": 0.0005889908256880733,
+      "loss": 4.6766,
       "step": 1800
     },
     {
-      "epoch": 0.19911742546550426,
-      "grad_norm": 0.8168042898178101,
-      "learning_rate": 0.0005886865639478504,
-      "loss": 4.6301,
+      "epoch": 0.19946091644204852,
+      "grad_norm": 0.738538920879364,
+      "learning_rate": 0.0005886670264436049,
+      "loss": 4.6883,
       "step": 1850
     },
     {
-      "epoch": 0.20449897750511248,
-      "grad_norm": 0.9217840433120728,
-      "learning_rate": 0.0005883633229177889,
-      "loss": 4.6206,
+      "epoch": 0.20485175202156333,
+      "grad_norm": 0.8313882350921631,
+      "learning_rate": 0.0005883432271991365,
+      "loss": 4.6281,
       "step": 1900
     },
     {
-      "epoch": 0.2098805295447207,
-      "grad_norm": 1.1758180856704712,
-      "learning_rate": 0.0005880400818877276,
-      "loss": 4.592,
+      "epoch": 0.21024258760107817,
+      "grad_norm": 0.7849212884902954,
+      "learning_rate": 0.0005880194279546681,
+      "loss": 4.6127,
       "step": 1950
     },
     {
-      "epoch": 0.2152620815843289,
-      "grad_norm": 1.1105222702026367,
-      "learning_rate": 0.0005877168408576662,
-      "loss": 4.5667,
+      "epoch": 0.215633423180593,
+      "grad_norm": 0.9173558354377747,
+      "learning_rate": 0.0005876956287101996,
+      "loss": 4.5858,
       "step": 2000
     },
     {
-      "epoch": 0.2152620815843289,
-      "eval_accuracy": 0.2720056595116205,
-      "eval_loss": 4.506523609161377,
-      "eval_runtime": 181.174,
-      "eval_samples_per_second": 99.413,
-      "eval_steps_per_second": 6.215,
+      "epoch": 0.215633423180593,
+      "eval_accuracy": 0.26942873897132447,
+      "eval_loss": 4.5194478034973145,
+      "eval_runtime": 185.3555,
+      "eval_samples_per_second": 97.17,
+      "eval_steps_per_second": 6.075,
       "step": 2000
     },
     {
-      "epoch": 0.22064363362393713,
-      "grad_norm": 1.0530309677124023,
-      "learning_rate": 0.0005873935998276048,
-      "loss": 4.5455,
+      "epoch": 0.2210242587601078,
+      "grad_norm": 1.065422534942627,
+      "learning_rate": 0.0005873718294657312,
+      "loss": 4.5794,
       "step": 2050
     },
     {
-      "epoch": 0.22602518566354537,
-      "grad_norm": 0.7890729308128357,
-      "learning_rate": 0.0005870703587975433,
-      "loss": 4.5395,
+      "epoch": 0.22641509433962265,
+      "grad_norm": 0.9498239755630493,
+      "learning_rate": 0.0005870480302212628,
+      "loss": 4.5481,
       "step": 2100
     },
     {
-      "epoch": 0.2314067377031536,
-      "grad_norm": 0.9836071729660034,
-      "learning_rate": 0.0005867471177674818,
-      "loss": 4.5303,
+      "epoch": 0.23180592991913745,
+      "grad_norm": 0.9778485298156738,
+      "learning_rate": 0.0005867242309767943,
+      "loss": 4.5447,
       "step": 2150
     },
     {
-      "epoch": 0.2367882897427618,
-      "grad_norm": 1.1541528701782227,
-      "learning_rate": 0.0005864238767374205,
-      "loss": 4.5087,
+      "epoch": 0.2371967654986523,
+      "grad_norm": 0.9899240732192993,
+      "learning_rate": 0.0005864004317323259,
+      "loss": 4.5173,
       "step": 2200
     },
     {
-      "epoch": 0.24216984178237003,
-      "grad_norm": 0.8824282288551331,
-      "learning_rate": 0.0005861006357073591,
-      "loss": 4.5137,
+      "epoch": 0.24258760107816713,
+      "grad_norm": 0.874481201171875,
+      "learning_rate": 0.0005860766324878575,
+      "loss": 4.4996,
       "step": 2250
     },
     {
-      "epoch": 0.24755139382197827,
-      "grad_norm": 0.8979189395904541,
-      "learning_rate": 0.0005857773946772977,
-      "loss": 4.4942,
+      "epoch": 0.24797843665768193,
+      "grad_norm": 0.8830015063285828,
+      "learning_rate": 0.000585752833243389,
+      "loss": 4.4761,
       "step": 2300
     },
     {
-      "epoch": 0.2529329458615865,
-      "grad_norm": 0.7571122646331787,
-      "learning_rate": 0.0005854541536472362,
-      "loss": 4.4605,
+      "epoch": 0.25336927223719674,
+      "grad_norm": 0.7012396454811096,
+      "learning_rate": 0.0005854290339989206,
+      "loss": 4.4658,
       "step": 2350
     },
     {
-      "epoch": 0.2583144979011947,
-      "grad_norm": 0.7347502708435059,
-      "learning_rate": 0.0005851309126171749,
-      "loss": 4.4495,
+      "epoch": 0.2587601078167116,
+      "grad_norm": 0.885820746421814,
+      "learning_rate": 0.0005851052347544521,
+      "loss": 4.4561,
       "step": 2400
     },
     {
-      "epoch": 0.2636960499408029,
-      "grad_norm": 1.111667275428772,
-      "learning_rate": 0.0005848076715871134,
-      "loss": 4.4425,
+      "epoch": 0.2641509433962264,
+      "grad_norm": 0.8533319234848022,
+      "learning_rate": 0.0005847814355099838,
+      "loss": 4.4457,
       "step": 2450
     },
     {
-      "epoch": 0.26907760198041114,
-      "grad_norm": 0.7959837913513184,
-      "learning_rate": 0.000584484430557052,
-      "loss": 4.4107,
+      "epoch": 0.2695417789757412,
+      "grad_norm": 0.7030578851699829,
+      "learning_rate": 0.0005844576362655154,
+      "loss": 4.4615,
       "step": 2500
     },
     {
-      "epoch": 0.27445915402001936,
-      "grad_norm": 1.070115566253662,
-      "learning_rate": 0.0005841611895269906,
-      "loss": 4.4069,
+      "epoch": 0.2749326145552561,
+      "grad_norm": 0.9385006427764893,
+      "learning_rate": 0.0005841338370210469,
+      "loss": 4.4116,
       "step": 2550
     },
     {
-      "epoch": 0.2798407060596276,
-      "grad_norm": 0.8109025359153748,
-      "learning_rate": 0.0005838379484969291,
-      "loss": 4.3963,
+      "epoch": 0.2803234501347709,
+      "grad_norm": 0.9079705476760864,
+      "learning_rate": 0.0005838100377765785,
+      "loss": 4.3873,
       "step": 2600
     },
     {
-      "epoch": 0.2852222580992358,
-      "grad_norm": 0.8632609248161316,
-      "learning_rate": 0.0005835147074668678,
-      "loss": 4.3746,
+      "epoch": 0.2857142857142857,
+      "grad_norm": 0.886477530002594,
+      "learning_rate": 0.0005834862385321101,
+      "loss": 4.3929,
       "step": 2650
     },
     {
-      "epoch": 0.29060381013884407,
-      "grad_norm": 0.9841556549072266,
-      "learning_rate": 0.0005831914664368063,
-      "loss": 4.3615,
+      "epoch": 0.29110512129380056,
+      "grad_norm": 1.088681936264038,
+      "learning_rate": 0.0005831624392876417,
+      "loss": 4.383,
       "step": 2700
     },
     {
-      "epoch": 0.2959853621784523,
-      "grad_norm": 0.9777256846427917,
-      "learning_rate": 0.0005828682254067449,
-      "loss": 4.3684,
+      "epoch": 0.29649595687331537,
+      "grad_norm": 0.964737594127655,
+      "learning_rate": 0.0005828386400431731,
+      "loss": 4.3635,
       "step": 2750
     },
     {
-      "epoch": 0.3013669142180605,
-      "grad_norm": 0.820572555065155,
-      "learning_rate": 0.0005825449843766835,
-      "loss": 4.3447,
+      "epoch": 0.3018867924528302,
+      "grad_norm": 0.8155286312103271,
+      "learning_rate": 0.0005825148407987048,
+      "loss": 4.3646,
       "step": 2800
     },
     {
-      "epoch": 0.3067484662576687,
-      "grad_norm": 1.0357141494750977,
-      "learning_rate": 0.0005822217433466221,
-      "loss": 4.3187,
+      "epoch": 0.30727762803234504,
+      "grad_norm": 0.858011782169342,
+      "learning_rate": 0.0005821910415542363,
+      "loss": 4.3625,
       "step": 2850
     },
     {
-      "epoch": 0.31213001829727693,
-      "grad_norm": 0.9645561575889587,
-      "learning_rate": 0.0005818985023165607,
-      "loss": 4.3365,
+      "epoch": 0.31266846361185985,
+      "grad_norm": 0.9244931936264038,
+      "learning_rate": 0.0005818672423097679,
+      "loss": 4.3421,
       "step": 2900
     },
     {
-      "epoch": 0.31751157033688515,
-      "grad_norm": 0.8094056844711304,
-      "learning_rate": 0.0005815752612864992,
-      "loss": 4.3157,
+      "epoch": 0.31805929919137466,
+      "grad_norm": 0.8701003789901733,
+      "learning_rate": 0.0005815434430652994,
+      "loss": 4.3505,
       "step": 2950
     },
     {
-      "epoch": 0.32289312237649337,
-      "grad_norm": 0.8351282477378845,
-      "learning_rate": 0.0005812520202564378,
-      "loss": 4.3217,
+      "epoch": 0.32345013477088946,
+      "grad_norm": 0.7689580917358398,
+      "learning_rate": 0.0005812196438208311,
+      "loss": 4.2979,
       "step": 3000
     },
     {
-      "epoch": 0.32289312237649337,
-      "eval_accuracy": 0.2993755392578638,
-      "eval_loss": 4.2360100746154785,
-      "eval_runtime": 180.1462,
-      "eval_samples_per_second": 99.98,
-      "eval_steps_per_second": 6.25,
+      "epoch": 0.32345013477088946,
+      "eval_accuracy": 0.2978724352078383,
+      "eval_loss": 4.2380828857421875,
+      "eval_runtime": 189.3634,
+      "eval_samples_per_second": 95.113,
+      "eval_steps_per_second": 5.946,
       "step": 3000
     },
     {
-      "epoch": 0.3282746744161016,
-      "grad_norm": 0.8822458386421204,
-      "learning_rate": 0.0005809287792263764,
-      "loss": 4.2922,
+      "epoch": 0.3288409703504043,
+      "grad_norm": 0.7187278866767883,
+      "learning_rate": 0.0005808958445763626,
+      "loss": 4.3058,
       "step": 3050
     },
     {
-      "epoch": 0.3336562264557098,
-      "grad_norm": 0.670447587966919,
-      "learning_rate": 0.0005806055381963151,
-      "loss": 4.2889,
+      "epoch": 0.33423180592991913,
+      "grad_norm": 0.8372971415519714,
+      "learning_rate": 0.0005805720453318942,
+      "loss": 4.2845,
       "step": 3100
     },
     {
-      "epoch": 0.3390377784953181,
-      "grad_norm": 0.7028505802154541,
-      "learning_rate": 0.0005802822971662536,
-      "loss": 4.2853,
+      "epoch": 0.33962264150943394,
+      "grad_norm": 0.7437776923179626,
+      "learning_rate": 0.0005802482460874257,
+      "loss": 4.2759,
       "step": 3150
     },
     {
-      "epoch": 0.3444193305349263,
-      "grad_norm": 0.659954845905304,
-      "learning_rate": 0.0005799590561361922,
-      "loss": 4.2604,
+      "epoch": 0.3450134770889488,
+      "grad_norm": 0.6668998599052429,
+      "learning_rate": 0.0005799244468429573,
+      "loss": 4.2978,
       "step": 3200
     },
     {
-      "epoch": 0.3498008825745345,
-      "grad_norm": 0.6329878568649292,
-      "learning_rate": 0.0005796358151061307,
-      "loss": 4.2712,
+      "epoch": 0.3504043126684636,
+      "grad_norm": 0.7630629539489746,
+      "learning_rate": 0.0005796006475984889,
+      "loss": 4.2804,
       "step": 3250
     },
     {
-      "epoch": 0.35518243461414273,
-      "grad_norm": 0.751659631729126,
-      "learning_rate": 0.0005793125740760694,
-      "loss": 4.2444,
+      "epoch": 0.3557951482479784,
+      "grad_norm": 0.9011856913566589,
+      "learning_rate": 0.0005792768483540205,
+      "loss": 4.259,
       "step": 3300
     },
     {
-      "epoch": 0.36056398665375095,
-      "grad_norm": 0.6843229532241821,
-      "learning_rate": 0.0005789893330460079,
-      "loss": 4.2497,
+      "epoch": 0.3611859838274933,
+      "grad_norm": 0.6641933917999268,
+      "learning_rate": 0.000578953049109552,
+      "loss": 4.2697,
       "step": 3350
     },
     {
-      "epoch": 0.36594553869335916,
-      "grad_norm": 0.683152437210083,
-      "learning_rate": 0.0005786660920159465,
-      "loss": 4.2446,
+      "epoch": 0.3665768194070081,
+      "grad_norm": 0.753118097782135,
+      "learning_rate": 0.0005786292498650836,
+      "loss": 4.2391,
       "step": 3400
     },
     {
-      "epoch": 0.3713270907329674,
-      "grad_norm": 0.8310645818710327,
-      "learning_rate": 0.0005783428509858851,
-      "loss": 4.2479,
+      "epoch": 0.3719676549865229,
+      "grad_norm": 0.7131187319755554,
+      "learning_rate": 0.0005783054506206152,
+      "loss": 4.2466,
       "step": 3450
     },
     {
-      "epoch": 0.3767086427725756,
-      "grad_norm": 0.8316429853439331,
-      "learning_rate": 0.0005780196099558237,
-      "loss": 4.2324,
+      "epoch": 0.37735849056603776,
+      "grad_norm": 0.7759909629821777,
+      "learning_rate": 0.0005779816513761467,
+      "loss": 4.2255,
       "step": 3500
     },
     {
-      "epoch": 0.3820901948121838,
-      "grad_norm": 0.7475887537002563,
-      "learning_rate": 0.0005776963689257623,
-      "loss": 4.2272,
+      "epoch": 0.38274932614555257,
+      "grad_norm": 0.7672677040100098,
+      "learning_rate": 0.0005776578521316782,
+      "loss": 4.2113,
       "step": 3550
     },
     {
-      "epoch": 0.38747174685179203,
-      "grad_norm": 0.7887517809867859,
-      "learning_rate": 0.0005773731278957008,
-      "loss": 4.2269,
+      "epoch": 0.3881401617250674,
+      "grad_norm": 0.7773898839950562,
+      "learning_rate": 0.0005773340528872099,
+      "loss": 4.2286,
       "step": 3600
     },
     {
-      "epoch": 0.3928532988914003,
-      "grad_norm": 0.8017256259918213,
-      "learning_rate": 0.0005770498868656394,
-      "loss": 4.1896,
+      "epoch": 0.3935309973045822,
+      "grad_norm": 0.6527792811393738,
+      "learning_rate": 0.0005770102536427414,
+      "loss": 4.2187,
       "step": 3650
     },
     {
-      "epoch": 0.3982348509310085,
-      "grad_norm": 0.7304349541664124,
-      "learning_rate": 0.000576726645835578,
-      "loss": 4.1759,
+      "epoch": 0.39892183288409705,
+      "grad_norm": 0.662846028804779,
+      "learning_rate": 0.000576686454398273,
+      "loss": 4.2004,
       "step": 3700
     },
     {
-      "epoch": 0.40361640297061674,
-      "grad_norm": 0.7166491746902466,
-      "learning_rate": 0.0005764034048055167,
-      "loss": 4.1959,
+      "epoch": 0.40431266846361186,
+      "grad_norm": 0.6254228353500366,
+      "learning_rate": 0.0005763626551538045,
+      "loss": 4.1774,
       "step": 3750
     },
     {
-      "epoch": 0.40899795501022496,
-      "grad_norm": 0.6879552602767944,
-      "learning_rate": 0.0005760801637754552,
-      "loss": 4.1706,
+      "epoch": 0.40970350404312667,
+      "grad_norm": 0.6896920204162598,
+      "learning_rate": 0.0005760388559093362,
+      "loss": 4.1907,
       "step": 3800
     },
     {
-      "epoch": 0.4143795070498332,
-      "grad_norm": 0.6761829853057861,
-      "learning_rate": 0.0005757569227453937,
-      "loss": 4.1944,
+      "epoch": 0.41509433962264153,
+      "grad_norm": 0.8081727623939514,
+      "learning_rate": 0.0005757150566648678,
+      "loss": 4.1674,
       "step": 3850
     },
     {
-      "epoch": 0.4197610590894414,
-      "grad_norm": 0.7459233403205872,
-      "learning_rate": 0.0005754336817153324,
-      "loss": 4.1711,
+      "epoch": 0.42048517520215634,
+      "grad_norm": 0.7900207042694092,
+      "learning_rate": 0.0005753912574203993,
+      "loss": 4.1772,
       "step": 3900
     },
     {
-      "epoch": 0.4251426111290496,
-      "grad_norm": 0.7789084315299988,
-      "learning_rate": 0.0005751104406852709,
-      "loss": 4.1653,
+      "epoch": 0.42587601078167114,
+      "grad_norm": 0.8366458415985107,
+      "learning_rate": 0.0005750674581759309,
+      "loss": 4.1825,
       "step": 3950
     },
     {
-      "epoch": 0.4305241631686578,
-      "grad_norm": 0.8571879267692566,
-      "learning_rate": 0.0005747871996552096,
-      "loss": 4.1519,
+      "epoch": 0.431266846361186,
+      "grad_norm": 0.7199099063873291,
+      "learning_rate": 0.0005747436589314624,
+      "loss": 4.1628,
       "step": 4000
     },
     {
-      "epoch": 0.4305241631686578,
-      "eval_accuracy": 0.31302614481843616,
-      "eval_loss": 4.0840888023376465,
-      "eval_runtime": 180.3251,
-      "eval_samples_per_second": 99.881,
-      "eval_steps_per_second": 6.244,
+      "epoch": 0.431266846361186,
+      "eval_accuracy": 0.311244020152503,
+      "eval_loss": 4.09364128112793,
+      "eval_runtime": 189.5468,
+      "eval_samples_per_second": 95.021,
+      "eval_steps_per_second": 5.94,
       "step": 4000
     },
     {
-      "epoch": 0.43590571520826604,
-      "grad_norm": 0.653843879699707,
-      "learning_rate": 0.0005744639586251481,
-      "loss": 4.1578,
+      "epoch": 0.4366576819407008,
+      "grad_norm": 0.8956255316734314,
+      "learning_rate": 0.0005744198596869941,
+      "loss": 4.1492,
       "step": 4050
     },
     {
-      "epoch": 0.44128726724787426,
-      "grad_norm": 0.756898045539856,
-      "learning_rate": 0.0005741407175950867,
-      "loss": 4.147,
+      "epoch": 0.4420485175202156,
+      "grad_norm": 0.6956295967102051,
+      "learning_rate": 0.0005740960604425255,
+      "loss": 4.1354,
       "step": 4100
     },
     {
-      "epoch": 0.44666881928748253,
-      "grad_norm": 0.6926664710044861,
-      "learning_rate": 0.0005738174765650253,
-      "loss": 4.1192,
+      "epoch": 0.4474393530997305,
+      "grad_norm": 0.6194539070129395,
+      "learning_rate": 0.0005737722611980572,
+      "loss": 4.1458,
       "step": 4150
     },
     {
-      "epoch": 0.45205037132709075,
-      "grad_norm": 0.7730582356452942,
-      "learning_rate": 0.0005734942355349638,
-      "loss": 4.1537,
+      "epoch": 0.4528301886792453,
+      "grad_norm": 0.8528458476066589,
+      "learning_rate": 0.0005734484619535887,
+      "loss": 4.131,
       "step": 4200
     },
     {
-      "epoch": 0.45743192336669897,
-      "grad_norm": 0.6861958503723145,
-      "learning_rate": 0.0005731709945049025,
-      "loss": 4.1408,
+      "epoch": 0.4582210242587601,
+      "grad_norm": 0.8363102078437805,
+      "learning_rate": 0.0005731246627091203,
+      "loss": 4.1347,
       "step": 4250
     },
     {
-      "epoch": 0.4628134754063072,
-      "grad_norm": 0.7236369252204895,
-      "learning_rate": 0.000572847753474841,
-      "loss": 4.1348,
+      "epoch": 0.4636118598382749,
+      "grad_norm": 0.7403669357299805,
+      "learning_rate": 0.0005728008634646518,
+      "loss": 4.1184,
       "step": 4300
     },
     {
-      "epoch": 0.4681950274459154,
-      "grad_norm": 0.6467661261558533,
-      "learning_rate": 0.0005725245124447796,
-      "loss": 4.1233,
+      "epoch": 0.46900269541778977,
+      "grad_norm": 0.6248896718025208,
+      "learning_rate": 0.0005724770642201835,
+      "loss": 4.117,
       "step": 4350
     },
     {
-      "epoch": 0.4735765794855236,
-      "grad_norm": 0.6053397059440613,
-      "learning_rate": 0.0005722012714147182,
-      "loss": 4.1101,
+      "epoch": 0.4743935309973046,
+      "grad_norm": 0.6075791716575623,
+      "learning_rate": 0.000572153264975715,
+      "loss": 4.1078,
       "step": 4400
     },
     {
-      "epoch": 0.47895813152513184,
-      "grad_norm": 0.690331757068634,
-      "learning_rate": 0.0005718780303846568,
-      "loss": 4.0914,
+      "epoch": 0.4797843665768194,
+      "grad_norm": 0.712907612323761,
+      "learning_rate": 0.0005718294657312466,
+      "loss": 4.1338,
       "step": 4450
     },
     {
-      "epoch": 0.48433968356474005,
-      "grad_norm": 0.6854518055915833,
-      "learning_rate": 0.0005715547893545953,
-      "loss": 4.0947,
+      "epoch": 0.48517520215633425,
+      "grad_norm": 0.5888732075691223,
+      "learning_rate": 0.0005715056664867781,
+      "loss": 4.0942,
       "step": 4500
     },
     {
-      "epoch": 0.48972123560434827,
-      "grad_norm": 0.8158701062202454,
-      "learning_rate": 0.000571231548324534,
-      "loss": 4.1069,
+      "epoch": 0.49056603773584906,
+      "grad_norm": 0.7003511190414429,
+      "learning_rate": 0.0005711818672423097,
+      "loss": 4.1055,
       "step": 4550
     },
     {
-      "epoch": 0.49510278764395654,
-      "grad_norm": 0.6854028701782227,
-      "learning_rate": 0.0005709083072944725,
-      "loss": 4.0801,
+      "epoch": 0.49595687331536387,
+      "grad_norm": 0.7173230648040771,
+      "learning_rate": 0.0005708580679978413,
+      "loss": 4.1009,
       "step": 4600
     },
     {
-      "epoch": 0.5004843396835648,
-      "grad_norm": 0.5569154024124146,
-      "learning_rate": 0.0005705850662644111,
-      "loss": 4.0691,
+      "epoch": 0.5013477088948787,
+      "grad_norm": 0.6557320356369019,
+      "learning_rate": 0.0005705342687533729,
+      "loss": 4.0945,
       "step": 4650
     },
     {
-      "epoch": 0.505865891723173,
-      "grad_norm": 0.6735650300979614,
-      "learning_rate": 0.0005702618252343497,
-      "loss": 4.0966,
+      "epoch": 0.5067385444743935,
+      "grad_norm": 0.6504918336868286,
+      "learning_rate": 0.0005702104695089044,
+      "loss": 4.0943,
       "step": 4700
     },
     {
-      "epoch": 0.5112474437627812,
-      "grad_norm": 0.6227396726608276,
-      "learning_rate": 0.0005699385842042882,
-      "loss": 4.0652,
+      "epoch": 0.5121293800539084,
+      "grad_norm": 0.5261765122413635,
+      "learning_rate": 0.000569886670264436,
+      "loss": 4.0974,
       "step": 4750
     },
     {
-      "epoch": 0.5166289958023894,
-      "grad_norm": 0.7985734939575195,
-      "learning_rate": 0.0005696153431742269,
-      "loss": 4.0739,
+      "epoch": 0.5175202156334232,
+      "grad_norm": 0.6606865525245667,
+      "learning_rate": 0.0005695628710199675,
+      "loss": 4.0726,
       "step": 4800
     },
     {
-      "epoch": 0.5220105478419976,
-      "grad_norm": 0.5978109240531921,
-      "learning_rate": 0.0005692921021441655,
-      "loss": 4.084,
+      "epoch": 0.522911051212938,
+      "grad_norm": 0.7215351462364197,
+      "learning_rate": 0.0005692390717754991,
+      "loss": 4.1032,
       "step": 4850
     },
     {
-      "epoch": 0.5273920998816058,
-      "grad_norm": 0.6848176121711731,
-      "learning_rate": 0.0005689688611141041,
-      "loss": 4.053,
+      "epoch": 0.5283018867924528,
+      "grad_norm": 0.5598901510238647,
+      "learning_rate": 0.00056892174851592,
+      "loss": 4.0747,
       "step": 4900
     },
     {
-      "epoch": 0.5327736519212141,
-      "grad_norm": 0.8177867531776428,
-      "learning_rate": 0.0005686456200840426,
-      "loss": 4.0574,
+      "epoch": 0.5336927223719676,
+      "grad_norm": 0.7433397769927979,
+      "learning_rate": 0.0005685979492714516,
+      "loss": 4.061,
       "step": 4950
     },
     {
-      "epoch": 0.5381552039608223,
-      "grad_norm": 0.6719267964363098,
-      "learning_rate": 0.0005683223790539811,
-      "loss": 4.0713,
+      "epoch": 0.5390835579514824,
+      "grad_norm": 0.5735206007957458,
+      "learning_rate": 0.0005682741500269833,
+      "loss": 4.0616,
       "step": 5000
     },
     {
-      "epoch": 0.5381552039608223,
-      "eval_accuracy": 0.3224673201992998,
-      "eval_loss": 3.9863152503967285,
-      "eval_runtime": 180.4165,
-      "eval_samples_per_second": 99.83,
-      "eval_steps_per_second": 6.241,
+      "epoch": 0.5390835579514824,
+      "eval_accuracy": 0.3216426447807879,
+      "eval_loss": 3.9913065433502197,
+      "eval_runtime": 185.4453,
+      "eval_samples_per_second": 97.123,
+      "eval_steps_per_second": 6.072,
       "step": 5000
     },
     {
-      "epoch": 0.5435367560004305,
-      "grad_norm": 0.6600003838539124,
-      "learning_rate": 0.0005679991380239198,
-      "loss": 4.0633,
+      "epoch": 0.5444743935309974,
+      "grad_norm": 0.6067236065864563,
+      "learning_rate": 0.0005679503507825148,
+      "loss": 4.0595,
       "step": 5050
     },
     {
-      "epoch": 0.5489183080400387,
-      "grad_norm": 0.7393937110900879,
-      "learning_rate": 0.0005676758969938584,
-      "loss": 4.0429,
+      "epoch": 0.5498652291105122,
+      "grad_norm": 0.6634421944618225,
+      "learning_rate": 0.0005676265515380464,
+      "loss": 4.048,
       "step": 5100
     },
     {
-      "epoch": 0.5542998600796469,
-      "grad_norm": 0.6118986010551453,
-      "learning_rate": 0.000567352655963797,
-      "loss": 4.048,
+      "epoch": 0.555256064690027,
+      "grad_norm": 0.5526515245437622,
+      "learning_rate": 0.0005673027522935779,
+      "loss": 4.0608,
       "step": 5150
     },
     {
-      "epoch": 0.5596814121192552,
-      "grad_norm": 0.6689847111701965,
-      "learning_rate": 0.0005670294149337355,
-      "loss": 4.0392,
+      "epoch": 0.5606469002695418,
+      "grad_norm": 0.6474317312240601,
+      "learning_rate": 0.0005669789530491095,
+      "loss": 4.0511,
       "step": 5200
     },
     {
-      "epoch": 0.5650629641588634,
-      "grad_norm": 0.6301764249801636,
-      "learning_rate": 0.0005667061739036742,
-      "loss": 4.0404,
+      "epoch": 0.5660377358490566,
+      "grad_norm": 0.7668002843856812,
+      "learning_rate": 0.0005666551538046411,
+      "loss": 4.0463,
       "step": 5250
     },
     {
-      "epoch": 0.5704445161984716,
-      "grad_norm": 0.6697719097137451,
-      "learning_rate": 0.0005663829328736127,
-      "loss": 4.0368,
+      "epoch": 0.5714285714285714,
+      "grad_norm": 0.6317028403282166,
+      "learning_rate": 0.0005663313545601727,
+      "loss": 4.0299,
       "step": 5300
     },
     {
-      "epoch": 0.5758260682380799,
-      "grad_norm": 0.5661592483520508,
-      "learning_rate": 0.0005660596918435512,
-      "loss": 4.0047,
+      "epoch": 0.5768194070080862,
+      "grad_norm": 0.6677391529083252,
+      "learning_rate": 0.0005660075553157042,
+      "loss": 4.004,
       "step": 5350
     },
     {
-      "epoch": 0.5812076202776881,
-      "grad_norm": 0.6363985538482666,
-      "learning_rate": 0.0005657364508134899,
-      "loss": 4.0396,
+      "epoch": 0.5822102425876011,
+      "grad_norm": 0.5952298641204834,
+      "learning_rate": 0.0005656837560712358,
+      "loss": 4.0466,
       "step": 5400
     },
     {
-      "epoch": 0.5865891723172963,
-      "grad_norm": 0.6611218452453613,
-      "learning_rate": 0.0005654132097834284,
-      "loss": 4.0312,
+      "epoch": 0.5876010781671159,
+      "grad_norm": 0.6425842642784119,
+      "learning_rate": 0.0005653599568267674,
+      "loss": 4.026,
       "step": 5450
     },
     {
-      "epoch": 0.5919707243569046,
-      "grad_norm": 0.6354134678840637,
-      "learning_rate": 0.0005650899687533671,
-      "loss": 4.0169,
+      "epoch": 0.5929919137466307,
+      "grad_norm": 0.657423198223114,
+      "learning_rate": 0.0005650361575822989,
+      "loss": 4.0337,
       "step": 5500
     },
     {
-      "epoch": 0.5973522763965128,
-      "grad_norm": 0.6671438813209534,
-      "learning_rate": 0.0005647667277233056,
-      "loss": 4.0032,
+      "epoch": 0.5983827493261455,
+      "grad_norm": 0.6006332635879517,
+      "learning_rate": 0.0005647123583378305,
+      "loss": 4.0168,
       "step": 5550
     },
     {
-      "epoch": 0.602733828436121,
-      "grad_norm": 0.6798700094223022,
-      "learning_rate": 0.0005644434866932442,
-      "loss": 4.0315,
+      "epoch": 0.6037735849056604,
+      "grad_norm": 0.605254590511322,
+      "learning_rate": 0.0005643885590933621,
+      "loss": 4.0186,
       "step": 5600
     },
     {
-      "epoch": 0.6081153804757292,
-      "grad_norm": 0.6767042279243469,
-      "learning_rate": 0.0005641202456631828,
-      "loss": 3.9929,
+      "epoch": 0.6091644204851752,
+      "grad_norm": 0.5642939805984497,
+      "learning_rate": 0.0005640647598488936,
+      "loss": 3.9927,
       "step": 5650
     },
     {
-      "epoch": 0.6134969325153374,
-      "grad_norm": 1.2276337146759033,
-      "learning_rate": 0.0005637970046331214,
-      "loss": 4.0043,
+      "epoch": 0.6145552560646901,
+      "grad_norm": 0.641412079334259,
+      "learning_rate": 0.0005637409606044252,
+      "loss": 4.006,
       "step": 5700
     },
     {
-      "epoch": 0.6188784845549457,
-      "grad_norm": 0.6766164302825928,
-      "learning_rate": 0.00056347376360306,
-      "loss": 3.9801,
+      "epoch": 0.6199460916442049,
+      "grad_norm": 0.6121648550033569,
+      "learning_rate": 0.0005634171613599567,
+      "loss": 3.9975,
       "step": 5750
     },
     {
-      "epoch": 0.6242600365945539,
-      "grad_norm": 0.7068859934806824,
-      "learning_rate": 0.0005631505225729985,
-      "loss": 3.9791,
+      "epoch": 0.6253369272237197,
+      "grad_norm": 0.5894919633865356,
+      "learning_rate": 0.0005630933621154884,
+      "loss": 4.0098,
       "step": 5800
     },
     {
-      "epoch": 0.6296415886341621,
-      "grad_norm": 0.6442428231239319,
-      "learning_rate": 0.0005628272815429371,
-      "loss": 3.9831,
+      "epoch": 0.6307277628032345,
+      "grad_norm": 0.5354415774345398,
+      "learning_rate": 0.0005627695628710199,
+      "loss": 3.9951,
       "step": 5850
     },
     {
-      "epoch": 0.6350231406737703,
-      "grad_norm": 0.5629743933677673,
-      "learning_rate": 0.0005625040405128757,
-      "loss": 3.9629,
+      "epoch": 0.6361185983827493,
+      "grad_norm": 0.6370319724082947,
+      "learning_rate": 0.0005624457636265515,
+      "loss": 3.9929,
       "step": 5900
     },
     {
-      "epoch": 0.6404046927133785,
-      "grad_norm": 0.6905534863471985,
-      "learning_rate": 0.0005621807994828143,
-      "loss": 3.9881,
+      "epoch": 0.6415094339622641,
+      "grad_norm": 0.602782666683197,
+      "learning_rate": 0.000562121964382083,
+      "loss": 3.9914,
       "step": 5950
     },
     {
-      "epoch": 0.6457862447529867,
-      "grad_norm": 0.617796003818512,
-      "learning_rate": 0.0005618575584527529,
-      "loss": 4.0023,
+      "epoch": 0.6469002695417789,
+      "grad_norm": 0.6626774072647095,
+      "learning_rate": 0.0005617981651376146,
+      "loss": 3.9895,
       "step": 6000
     },
     {
-      "epoch": 0.6457862447529867,
-      "eval_accuracy": 0.329294959016674,
-      "eval_loss": 3.9152743816375732,
-      "eval_runtime": 179.9748,
-      "eval_samples_per_second": 100.075,
-      "eval_steps_per_second": 6.256,
+      "epoch": 0.6469002695417789,
+      "eval_accuracy": 0.3283419645376532,
+      "eval_loss": 3.9216606616973877,
+      "eval_runtime": 185.4587,
+      "eval_samples_per_second": 97.116,
+      "eval_steps_per_second": 6.071,
       "step": 6000
     },
     {
-      "epoch": 0.651167796792595,
-      "grad_norm": 0.667194128036499,
-      "learning_rate": 0.0005615407822432927,
-      "loss": 3.9869,
+      "epoch": 0.6522911051212938,
+      "grad_norm": 0.616033673286438,
+      "learning_rate": 0.0005614743658931463,
+      "loss": 3.9848,
       "step": 6050
     },
     {
-      "epoch": 0.6565493488322032,
-      "grad_norm": 0.6065294146537781,
-      "learning_rate": 0.0005612175412132313,
-      "loss": 3.9867,
+      "epoch": 0.6576819407008087,
+      "grad_norm": 0.6033005714416504,
+      "learning_rate": 0.0005611505666486777,
+      "loss": 4.003,
       "step": 6100
     },
     {
-      "epoch": 0.6619309008718114,
-      "grad_norm": 0.6784459352493286,
-      "learning_rate": 0.0005608943001831699,
-      "loss": 3.9523,
+      "epoch": 0.6630727762803235,
+      "grad_norm": 0.6848000288009644,
+      "learning_rate": 0.0005608267674042094,
+      "loss": 3.9842,
       "step": 6150
     },
     {
-      "epoch": 0.6673124529114196,
-      "grad_norm": 0.6815781593322754,
-      "learning_rate": 0.0005605710591531085,
-      "loss": 3.959,
+      "epoch": 0.6684636118598383,
+      "grad_norm": 0.6065899729728699,
+      "learning_rate": 0.0005605029681597409,
+      "loss": 3.9829,
       "step": 6200
     },
     {
-      "epoch": 0.6726940049510278,
-      "grad_norm": 0.6618684530258179,
-      "learning_rate": 0.000560247818123047,
-      "loss": 3.9849,
+      "epoch": 0.6738544474393531,
+      "grad_norm": 0.6808527708053589,
+      "learning_rate": 0.0005601791689152725,
+      "loss": 3.9738,
       "step": 6250
     },
     {
-      "epoch": 0.6780755569906362,
-      "grad_norm": 0.6926442980766296,
-      "learning_rate": 0.0005599245770929855,
-      "loss": 3.9505,
+      "epoch": 0.6792452830188679,
+      "grad_norm": 0.613246738910675,
+      "learning_rate": 0.000559855369670804,
+      "loss": 3.9753,
       "step": 6300
     },
     {
-      "epoch": 0.6834571090302444,
-      "grad_norm": 0.6575818657875061,
-      "learning_rate": 0.0005596013360629242,
-      "loss": 3.9645,
+      "epoch": 0.6846361185983828,
+      "grad_norm": 0.5766479969024658,
+      "learning_rate": 0.0005595315704263357,
+      "loss": 3.9591,
       "step": 6350
     },
     {
-      "epoch": 0.6888386610698526,
-      "grad_norm": 0.5900609493255615,
-      "learning_rate": 0.0005592780950328628,
-      "loss": 3.9594,
+      "epoch": 0.6900269541778976,
+      "grad_norm": 0.6182272434234619,
+      "learning_rate": 0.0005592077711818672,
+      "loss": 3.9729,
       "step": 6400
     },
     {
-      "epoch": 0.6942202131094608,
-      "grad_norm": 0.6123409867286682,
-      "learning_rate": 0.0005589548540028014,
-      "loss": 3.965,
+      "epoch": 0.6954177897574124,
+      "grad_norm": 0.6838184595108032,
+      "learning_rate": 0.0005588839719373988,
+      "loss": 3.9497,
       "step": 6450
     },
     {
-      "epoch": 0.699601765149069,
-      "grad_norm": 0.554110586643219,
-      "learning_rate": 0.0005586316129727399,
-      "loss": 3.9527,
+      "epoch": 0.7008086253369272,
+      "grad_norm": 0.5696499347686768,
+      "learning_rate": 0.0005585601726929303,
+      "loss": 3.956,
       "step": 6500
     },
     {
-      "epoch": 0.7049833171886772,
-      "grad_norm": 0.6906695365905762,
-      "learning_rate": 0.0005583083719426786,
-      "loss": 3.9451,
+      "epoch": 0.706199460916442,
+      "grad_norm": 0.624225378036499,
+      "learning_rate": 0.0005582363734484619,
+      "loss": 3.939,
       "step": 6550
     },
     {
-      "epoch": 0.7103648692282855,
-      "grad_norm": 0.7299356460571289,
-      "learning_rate": 0.0005579851309126171,
-      "loss": 3.9444,
+      "epoch": 0.7115902964959568,
+      "grad_norm": 0.5950983762741089,
+      "learning_rate": 0.0005579125742039935,
+      "loss": 3.9456,
       "step": 6600
     },
     {
-      "epoch": 0.7157464212678937,
-      "grad_norm": 0.5925883054733276,
-      "learning_rate": 0.0005576618898825558,
-      "loss": 3.932,
+      "epoch": 0.7169811320754716,
+      "grad_norm": 0.6580861806869507,
+      "learning_rate": 0.0005575887749595251,
+      "loss": 3.9493,
       "step": 6650
     },
     {
-      "epoch": 0.7211279733075019,
-      "grad_norm": 0.6869106888771057,
-      "learning_rate": 0.0005573386488524943,
-      "loss": 3.9368,
+      "epoch": 0.7223719676549866,
+      "grad_norm": 0.6185200810432434,
+      "learning_rate": 0.0005572649757150566,
+      "loss": 3.9501,
       "step": 6700
     },
     {
-      "epoch": 0.7265095253471101,
-      "grad_norm": 0.6772210597991943,
-      "learning_rate": 0.0005570154078224328,
-      "loss": 3.9425,
+      "epoch": 0.7277628032345014,
+      "grad_norm": 0.7061867117881775,
+      "learning_rate": 0.0005569411764705882,
+      "loss": 3.9492,
       "step": 6750
     },
     {
-      "epoch": 0.7318910773867183,
-      "grad_norm": 0.6618852615356445,
-      "learning_rate": 0.0005566921667923715,
-      "loss": 3.9328,
+      "epoch": 0.7331536388140162,
+      "grad_norm": 0.5729274153709412,
+      "learning_rate": 0.0005566173772261198,
+      "loss": 3.9163,
       "step": 6800
     },
     {
-      "epoch": 0.7372726294263265,
-      "grad_norm": 0.5648536682128906,
-      "learning_rate": 0.00055636892576231,
-      "loss": 3.9226,
+      "epoch": 0.738544474393531,
+      "grad_norm": 0.5833299160003662,
+      "learning_rate": 0.0005562935779816513,
+      "loss": 3.9305,
       "step": 6850
     },
     {
-      "epoch": 0.7426541814659348,
-      "grad_norm": 0.65449458360672,
-      "learning_rate": 0.0005560456847322487,
-      "loss": 3.9241,
+      "epoch": 0.7439353099730458,
+      "grad_norm": 0.6158512234687805,
+      "learning_rate": 0.0005559697787371828,
+      "loss": 3.947,
       "step": 6900
     },
     {
-      "epoch": 0.748035733505543,
-      "grad_norm": 0.604253888130188,
-      "learning_rate": 0.0005557224437021872,
-      "loss": 3.9116,
+      "epoch": 0.7493261455525606,
+      "grad_norm": 0.6340197324752808,
+      "learning_rate": 0.0005556459794927145,
+      "loss": 3.9268,
       "step": 6950
     },
     {
-      "epoch": 0.7534172855451512,
-      "grad_norm": 1.0373648405075073,
-      "learning_rate": 0.0005553992026721258,
-      "loss": 3.9316,
+      "epoch": 0.7547169811320755,
+      "grad_norm": 0.5887579917907715,
+      "learning_rate": 0.000555322180248246,
+      "loss": 3.9368,
       "step": 7000
     },
     {
-      "epoch": 0.7534172855451512,
-      "eval_accuracy": 0.3344105542807554,
-      "eval_loss": 3.8566644191741943,
-      "eval_runtime": 180.5849,
-      "eval_samples_per_second": 99.737,
-      "eval_steps_per_second": 6.235,
+      "epoch": 0.7547169811320755,
+      "eval_accuracy": 0.33388304451041606,
+      "eval_loss": 3.862060070037842,
+      "eval_runtime": 185.1282,
+      "eval_samples_per_second": 97.289,
+      "eval_steps_per_second": 6.082,
       "step": 7000
     },
     {
-      "epoch": 0.7587988375847594,
-      "grad_norm": 0.5991268754005432,
-      "learning_rate": 0.0005550759616420644,
-      "loss": 3.9329,
+      "epoch": 0.7601078167115903,
+      "grad_norm": 0.7935763001441956,
+      "learning_rate": 0.0005549983810037776,
+      "loss": 3.9071,
       "step": 7050
     },
     {
-      "epoch": 0.7641803896243676,
-      "grad_norm": 0.7183207869529724,
-      "learning_rate": 0.000554752720612003,
-      "loss": 3.9165,
+      "epoch": 0.7654986522911051,
+      "grad_norm": 0.6606423854827881,
+      "learning_rate": 0.0005546745817593091,
+      "loss": 3.9094,
       "step": 7100
     },
     {
-      "epoch": 0.7695619416639758,
-      "grad_norm": 0.5952954888343811,
-      "learning_rate": 0.0005544294795819415,
-      "loss": 3.9215,
+      "epoch": 0.77088948787062,
+      "grad_norm": 0.6245214343070984,
+      "learning_rate": 0.0005543507825148408,
+      "loss": 3.9132,
       "step": 7150
     },
     {
-      "epoch": 0.7749434937035841,
-      "grad_norm": 0.5735570788383484,
-      "learning_rate": 0.0005541062385518801,
-      "loss": 3.9007,
+      "epoch": 0.7762803234501348,
+      "grad_norm": 0.631089448928833,
+      "learning_rate": 0.0005540269832703723,
+      "loss": 3.8915,
       "step": 7200
     },
     {
-      "epoch": 0.7803250457431924,
-      "grad_norm": 0.6777737736701965,
-      "learning_rate": 0.0005537829975218188,
-      "loss": 3.9162,
+      "epoch": 0.7816711590296496,
+      "grad_norm": 0.5624434947967529,
+      "learning_rate": 0.0005537031840259039,
+      "loss": 3.9178,
       "step": 7250
     },
     {
-      "epoch": 0.7857065977828006,
-      "grad_norm": 0.678438663482666,
-      "learning_rate": 0.0005534597564917573,
-      "loss": 3.9188,
+      "epoch": 0.7870619946091644,
+      "grad_norm": 0.5263481736183167,
+      "learning_rate": 0.0005533793847814354,
+      "loss": 3.8948,
       "step": 7300
     },
     {
-      "epoch": 0.7910881498224088,
-      "grad_norm": 0.679999053478241,
-      "learning_rate": 0.0005531365154616959,
-      "loss": 3.9143,
+      "epoch": 0.7924528301886793,
+      "grad_norm": 0.6880287528038025,
+      "learning_rate": 0.000553055585536967,
+      "loss": 3.9139,
       "step": 7350
     },
     {
-      "epoch": 0.796469701862017,
-      "grad_norm": 0.6160652041435242,
-      "learning_rate": 0.0005528132744316344,
-      "loss": 3.9097,
+      "epoch": 0.7978436657681941,
+      "grad_norm": 0.5549008846282959,
+      "learning_rate": 0.0005527317862924987,
+      "loss": 3.9141,
       "step": 7400
     },
     {
-      "epoch": 0.8018512539016253,
-      "grad_norm": 0.6063855886459351,
-      "learning_rate": 0.0005524900334015731,
-      "loss": 3.9028,
+      "epoch": 0.8032345013477089,
+      "grad_norm": 0.5605992674827576,
+      "learning_rate": 0.0005524079870480301,
+      "loss": 3.9102,
       "step": 7450
     },
     {
-      "epoch": 0.8072328059412335,
-      "grad_norm": 0.5554015040397644,
-      "learning_rate": 0.0005521667923715117,
-      "loss": 3.9038,
+      "epoch": 0.8086253369272237,
+      "grad_norm": 0.655853271484375,
+      "learning_rate": 0.0005520841878035618,
+      "loss": 3.9234,
       "step": 7500
     },
     {
-      "epoch": 0.8126143579808417,
-      "grad_norm": 0.6115410327911377,
-      "learning_rate": 0.0005518435513414502,
-      "loss": 3.8952,
+      "epoch": 0.8140161725067385,
+      "grad_norm": 0.5594416856765747,
+      "learning_rate": 0.0005517603885590933,
+      "loss": 3.8961,
       "step": 7550
     },
     {
-      "epoch": 0.8179959100204499,
-      "grad_norm": 0.7529610395431519,
-      "learning_rate": 0.0005515203103113888,
-      "loss": 3.8998,
+      "epoch": 0.8194070080862533,
+      "grad_norm": 0.6285805702209473,
+      "learning_rate": 0.0005514365893146249,
+      "loss": 3.894,
       "step": 7600
     },
     {
-      "epoch": 0.8233774620600581,
-      "grad_norm": 0.553494930267334,
-      "learning_rate": 0.0005511970692813274,
-      "loss": 3.8858,
+      "epoch": 0.8247978436657682,
+      "grad_norm": 0.5235066413879395,
+      "learning_rate": 0.0005511127900701564,
+      "loss": 3.886,
       "step": 7650
     },
     {
-      "epoch": 0.8287590140996663,
-      "grad_norm": 0.6321557760238647,
-      "learning_rate": 0.000550873828251266,
-      "loss": 3.8898,
+      "epoch": 0.8301886792452831,
+      "grad_norm": 0.598432719707489,
+      "learning_rate": 0.000550788990825688,
+      "loss": 3.9134,
       "step": 7700
     },
     {
-      "epoch": 0.8341405661392746,
-      "grad_norm": 0.5359132885932922,
-      "learning_rate": 0.0005505505872212045,
-      "loss": 3.8941,
+      "epoch": 0.8355795148247979,
+      "grad_norm": 0.547592043876648,
+      "learning_rate": 0.0005504651915812196,
+      "loss": 3.9098,
       "step": 7750
     },
     {
-      "epoch": 0.8395221181788828,
-      "grad_norm": 0.638156533241272,
-      "learning_rate": 0.0005502273461911432,
-      "loss": 3.89,
+      "epoch": 0.8409703504043127,
+      "grad_norm": 0.5671661496162415,
+      "learning_rate": 0.0005501413923367512,
+      "loss": 3.872,
       "step": 7800
     },
     {
-      "epoch": 0.844903670218491,
-      "grad_norm": 0.5378808975219727,
-      "learning_rate": 0.0005499041051610817,
-      "loss": 3.907,
+      "epoch": 0.8463611859838275,
+      "grad_norm": 0.5478203892707825,
+      "learning_rate": 0.0005498175930922827,
+      "loss": 3.8859,
       "step": 7850
     },
     {
-      "epoch": 0.8502852222580992,
-      "grad_norm": 0.6681106686592102,
-      "learning_rate": 0.0005495808641310204,
+      "epoch": 0.8517520215633423,
+      "grad_norm": 0.5433807969093323,
+      "learning_rate": 0.0005494937938478143,
       "loss": 3.8922,
       "step": 7900
     },
     {
-      "epoch": 0.8556667742977074,
-      "grad_norm": 0.7690930962562561,
-      "learning_rate": 0.0005492576231009589,
-      "loss": 3.8902,
+      "epoch": 0.8571428571428571,
+      "grad_norm": 0.559604823589325,
+      "learning_rate": 0.0005491699946033459,
+      "loss": 3.8797,
       "step": 7950
     },
     {
-      "epoch": 0.8610483263373157,
-      "grad_norm": 0.6343661546707153,
-      "learning_rate": 0.0005489343820708974,
-      "loss": 3.8661,
+      "epoch": 0.862533692722372,
+      "grad_norm": 0.5964951515197754,
+      "learning_rate": 0.0005488461953588775,
+      "loss": 3.8823,
       "step": 8000
     },
     {
-      "epoch": 0.8610483263373157,
-      "eval_accuracy": 0.3381346320105967,
-      "eval_loss": 3.813588857650757,
-      "eval_runtime": 180.3025,
-      "eval_samples_per_second": 99.893,
-      "eval_steps_per_second": 6.245,
+      "epoch": 0.862533692722372,
+      "eval_accuracy": 0.3377860735464878,
+      "eval_loss": 3.8172028064727783,
+      "eval_runtime": 185.316,
+      "eval_samples_per_second": 97.191,
+      "eval_steps_per_second": 6.076,
       "step": 8000
     },
     {
-      "epoch": 0.8664298783769239,
-      "grad_norm": 0.6980047821998596,
-      "learning_rate": 0.0005486111410408361,
-      "loss": 3.8693,
+      "epoch": 0.8679245283018868,
+      "grad_norm": 0.5655633211135864,
+      "learning_rate": 0.000548522396114409,
+      "loss": 3.8829,
       "step": 8050
     },
     {
-      "epoch": 0.8718114304165321,
-      "grad_norm": 0.6073970794677734,
-      "learning_rate": 0.0005482943648313759,
-      "loss": 3.8802,
+      "epoch": 0.8733153638814016,
+      "grad_norm": 0.5531695485115051,
+      "learning_rate": 0.0005481985968699406,
+      "loss": 3.8906,
       "step": 8100
     },
     {
-      "epoch": 0.8771929824561403,
-      "grad_norm": 0.677251935005188,
-      "learning_rate": 0.0005479711238013145,
-      "loss": 3.872,
+      "epoch": 0.8787061994609164,
+      "grad_norm": 0.5488873720169067,
+      "learning_rate": 0.0005478747976254721,
+      "loss": 3.8692,
       "step": 8150
     },
     {
-      "epoch": 0.8825745344957485,
-      "grad_norm": 0.6714052557945251,
-      "learning_rate": 0.0005476478827712531,
-      "loss": 3.8784,
+      "epoch": 0.8840970350404312,
+      "grad_norm": 0.6054577827453613,
+      "learning_rate": 0.0005475509983810037,
+      "loss": 3.8736,
       "step": 8200
     },
     {
-      "epoch": 0.8879560865353568,
-      "grad_norm": 0.5594466328620911,
-      "learning_rate": 0.0005473246417411916,
-      "loss": 3.8727,
+      "epoch": 0.889487870619946,
+      "grad_norm": 0.5637973546981812,
+      "learning_rate": 0.0005472271991365352,
+      "loss": 3.8679,
       "step": 8250
     },
     {
-      "epoch": 0.8933376385749651,
-      "grad_norm": 0.565110445022583,
-      "learning_rate": 0.0005470014007111302,
-      "loss": 3.8711,
+      "epoch": 0.894878706199461,
+      "grad_norm": 0.676185667514801,
+      "learning_rate": 0.0005469033998920669,
+      "loss": 3.8531,
       "step": 8300
     },
     {
-      "epoch": 0.8987191906145733,
-      "grad_norm": 0.5543609857559204,
-      "learning_rate": 0.0005466781596810688,
-      "loss": 3.8685,
+      "epoch": 0.9002695417789758,
+      "grad_norm": 0.5533009171485901,
+      "learning_rate": 0.0005465796006475984,
+      "loss": 3.8398,
       "step": 8350
     },
     {
-      "epoch": 0.9041007426541815,
-      "grad_norm": 0.6193423867225647,
-      "learning_rate": 0.0005463549186510073,
-      "loss": 3.8681,
+      "epoch": 0.9056603773584906,
+      "grad_norm": 0.5328658819198608,
+      "learning_rate": 0.00054625580140313,
+      "loss": 3.8865,
       "step": 8400
     },
     {
-      "epoch": 0.9094822946937897,
-      "grad_norm": 0.6133995056152344,
-      "learning_rate": 0.000546031677620946,
-      "loss": 3.8575,
+      "epoch": 0.9110512129380054,
+      "grad_norm": 0.660688579082489,
+      "learning_rate": 0.0005459320021586615,
+      "loss": 3.8522,
       "step": 8450
     },
     {
-      "epoch": 0.9148638467333979,
-      "grad_norm": 0.5809963941574097,
-      "learning_rate": 0.0005457084365908845,
-      "loss": 3.8595,
+      "epoch": 0.9164420485175202,
+      "grad_norm": 0.5564950704574585,
+      "learning_rate": 0.0005456082029141932,
+      "loss": 3.8395,
       "step": 8500
     },
     {
-      "epoch": 0.9202453987730062,
-      "grad_norm": 0.5669277906417847,
-      "learning_rate": 0.0005453851955608232,
-      "loss": 3.851,
+      "epoch": 0.921832884097035,
+      "grad_norm": 0.6572705507278442,
+      "learning_rate": 0.0005452844036697248,
+      "loss": 3.8538,
       "step": 8550
     },
     {
-      "epoch": 0.9256269508126144,
-      "grad_norm": 0.6012568473815918,
-      "learning_rate": 0.0005450619545307617,
-      "loss": 3.8583,
+      "epoch": 0.9272237196765498,
+      "grad_norm": 0.7002739310264587,
+      "learning_rate": 0.0005449606044252563,
+      "loss": 3.8541,
       "step": 8600
     },
     {
-      "epoch": 0.9310085028522226,
-      "grad_norm": 0.6442046761512756,
-      "learning_rate": 0.0005447387135007003,
-      "loss": 3.8482,
+      "epoch": 0.9326145552560647,
+      "grad_norm": 0.6021630167961121,
+      "learning_rate": 0.0005446368051807879,
+      "loss": 3.8569,
       "step": 8650
     },
     {
-      "epoch": 0.9363900548918308,
-      "grad_norm": 0.671301007270813,
-      "learning_rate": 0.0005444154724706389,
-      "loss": 3.8524,
+      "epoch": 0.9380053908355795,
+      "grad_norm": 0.5585712194442749,
+      "learning_rate": 0.0005443130059363194,
+      "loss": 3.8355,
       "step": 8700
     },
     {
-      "epoch": 0.941771606931439,
-      "grad_norm": 0.5206149816513062,
-      "learning_rate": 0.0005440922314405775,
-      "loss": 3.8301,
+      "epoch": 0.9433962264150944,
+      "grad_norm": 0.5851734280586243,
+      "learning_rate": 0.0005439892066918511,
+      "loss": 3.8521,
       "step": 8750
     },
     {
-      "epoch": 0.9471531589710472,
-      "grad_norm": 0.6047552227973938,
-      "learning_rate": 0.0005437689904105161,
-      "loss": 3.8374,
+      "epoch": 0.9487870619946092,
+      "grad_norm": 0.5523541569709778,
+      "learning_rate": 0.0005436654074473825,
+      "loss": 3.8391,
       "step": 8800
     },
     {
-      "epoch": 0.9525347110106555,
-      "grad_norm": 0.5780496001243591,
-      "learning_rate": 0.0005434457493804546,
-      "loss": 3.8385,
+      "epoch": 0.954177897574124,
+      "grad_norm": 0.5466392040252686,
+      "learning_rate": 0.0005433416082029142,
+      "loss": 3.8346,
       "step": 8850
     },
     {
-      "epoch": 0.9579162630502637,
-      "grad_norm": 0.5649657249450684,
-      "learning_rate": 0.0005431225083503932,
-      "loss": 3.8445,
+      "epoch": 0.9595687331536388,
+      "grad_norm": 0.6498163938522339,
+      "learning_rate": 0.0005430178089584457,
+      "loss": 3.8249,
       "step": 8900
     },
     {
-      "epoch": 0.9632978150898719,
-      "grad_norm": 0.5616223216056824,
-      "learning_rate": 0.0005427992673203318,
-      "loss": 3.8502,
+      "epoch": 0.9649595687331537,
+      "grad_norm": 0.561964213848114,
+      "learning_rate": 0.0005426940097139773,
+      "loss": 3.8416,
       "step": 8950
     },
     {
-      "epoch": 0.9686793671294801,
-      "grad_norm": 0.5216642022132874,
-      "learning_rate": 0.0005424760262902704,
-      "loss": 3.8455,
+      "epoch": 0.9703504043126685,
+      "grad_norm": 0.7354522347450256,
+      "learning_rate": 0.0005423766864543982,
+      "loss": 3.8417,
       "step": 9000
     },
     {
-      "epoch": 0.9686793671294801,
-      "eval_accuracy": 0.34202397078280383,
-      "eval_loss": 3.775099277496338,
-      "eval_runtime": 181.4282,
-      "eval_samples_per_second": 99.273,
-      "eval_steps_per_second": 6.206,
+      "epoch": 0.9703504043126685,
+      "eval_accuracy": 0.3414106252310911,
+      "eval_loss": 3.777308464050293,
+      "eval_runtime": 184.7863,
+      "eval_samples_per_second": 97.469,
+      "eval_steps_per_second": 6.094,
       "step": 9000
     },
     {
-      "epoch": 0.9740609191690883,
-      "grad_norm": 0.5867716073989868,
-      "learning_rate": 0.000542152785260209,
-      "loss": 3.8377,
+      "epoch": 0.9757412398921833,
+      "grad_norm": 0.5319597721099854,
+      "learning_rate": 0.0005420528872099298,
+      "loss": 3.8258,
       "step": 9050
     },
     {
-      "epoch": 0.9794424712086965,
-      "grad_norm": 0.5908952951431274,
-      "learning_rate": 0.0005418295442301476,
-      "loss": 3.8399,
+      "epoch": 0.9811320754716981,
+      "grad_norm": 0.5643147230148315,
+      "learning_rate": 0.0005417290879654613,
+      "loss": 3.8233,
       "step": 9100
     },
     {
-      "epoch": 0.9848240232483048,
-      "grad_norm": 0.565000057220459,
-      "learning_rate": 0.0005415063032000861,
-      "loss": 3.8336,
+      "epoch": 0.9865229110512129,
+      "grad_norm": 0.6312896609306335,
+      "learning_rate": 0.000541405288720993,
+      "loss": 3.8388,
       "step": 9150
     },
     {
-      "epoch": 0.9902055752879131,
-      "grad_norm": 0.6750233173370361,
-      "learning_rate": 0.0005411830621700248,
-      "loss": 3.8214,
+      "epoch": 0.9919137466307277,
+      "grad_norm": 0.5279715657234192,
+      "learning_rate": 0.0005410814894765245,
+      "loss": 3.8201,
       "step": 9200
     },
     {
-      "epoch": 0.9955871273275213,
-      "grad_norm": 0.5716939568519592,
-      "learning_rate": 0.0005408598211399633,
-      "loss": 3.8249,
+      "epoch": 0.9973045822102425,
+      "grad_norm": 0.5988540649414062,
+      "learning_rate": 0.0005407576902320561,
+      "loss": 3.847,
       "step": 9250
     },
     {
-      "epoch": 1.0009686793671295,
-      "grad_norm": 0.579093337059021,
-      "learning_rate": 0.0005405365801099019,
-      "loss": 3.8114,
+      "epoch": 1.0026954177897573,
+      "grad_norm": 0.6333749890327454,
+      "learning_rate": 0.0005404338909875876,
+      "loss": 3.7852,
       "step": 9300
     },
     {
-      "epoch": 1.0063502314067376,
-      "grad_norm": 0.6211260557174683,
-      "learning_rate": 0.0005402133390798405,
-      "loss": 3.7626,
+      "epoch": 1.0080862533692723,
+      "grad_norm": 0.5815381407737732,
+      "learning_rate": 0.0005401100917431192,
+      "loss": 3.7609,
       "step": 9350
     },
     {
-      "epoch": 1.011731783446346,
-      "grad_norm": 0.5808591842651367,
-      "learning_rate": 0.000539890098049779,
-      "loss": 3.7678,
+      "epoch": 1.013477088948787,
+      "grad_norm": 0.610984206199646,
+      "learning_rate": 0.0005397862924986508,
+      "loss": 3.7739,
       "step": 9400
     },
     {
-      "epoch": 1.017113335485954,
-      "grad_norm": 0.5807240605354309,
-      "learning_rate": 0.0005395668570197177,
-      "loss": 3.7638,
+      "epoch": 1.0188679245283019,
+      "grad_norm": 0.5451633930206299,
+      "learning_rate": 0.0005394624932541824,
+      "loss": 3.7813,
       "step": 9450
     },
     {
-      "epoch": 1.0224948875255624,
-      "grad_norm": 0.6161471605300903,
-      "learning_rate": 0.0005392436159896562,
-      "loss": 3.7686,
+      "epoch": 1.0242587601078168,
+      "grad_norm": 0.5960482954978943,
+      "learning_rate": 0.0005391386940097139,
+      "loss": 3.7504,
       "step": 9500
     },
     {
-      "epoch": 1.0278764395651705,
-      "grad_norm": 0.5693685412406921,
-      "learning_rate": 0.0005389203749595948,
-      "loss": 3.7504,
+      "epoch": 1.0296495956873315,
+      "grad_norm": 0.605932891368866,
+      "learning_rate": 0.0005388148947652455,
+      "loss": 3.7627,
       "step": 9550
     },
     {
-      "epoch": 1.0332579916047788,
-      "grad_norm": 0.5480920076370239,
-      "learning_rate": 0.0005385971339295334,
-      "loss": 3.7575,
+      "epoch": 1.0350404312668464,
+      "grad_norm": 0.5867658257484436,
+      "learning_rate": 0.000538491095520777,
+      "loss": 3.7887,
       "step": 9600
     },
     {
-      "epoch": 1.0386395436443872,
-      "grad_norm": 0.6429648995399475,
-      "learning_rate": 0.000538273892899472,
-      "loss": 3.7475,
+      "epoch": 1.0404312668463611,
+      "grad_norm": 0.6174441576004028,
+      "learning_rate": 0.0005381672962763086,
+      "loss": 3.7492,
       "step": 9650
     },
     {
-      "epoch": 1.0440210956839953,
-      "grad_norm": 0.567170262336731,
-      "learning_rate": 0.0005379506518694106,
-      "loss": 3.7826,
+      "epoch": 1.045822102425876,
+      "grad_norm": 0.594423770904541,
+      "learning_rate": 0.0005378434970318403,
+      "loss": 3.7728,
       "step": 9700
     },
     {
-      "epoch": 1.0494026477236036,
-      "grad_norm": 0.5592578649520874,
-      "learning_rate": 0.0005376274108393491,
-      "loss": 3.751,
+      "epoch": 1.0512129380053907,
+      "grad_norm": 0.5846846103668213,
+      "learning_rate": 0.0005375196977873718,
+      "loss": 3.75,
       "step": 9750
     },
     {
-      "epoch": 1.0547841997632117,
-      "grad_norm": 0.5507737994194031,
-      "learning_rate": 0.0005373041698092877,
-      "loss": 3.777,
+      "epoch": 1.0566037735849056,
+      "grad_norm": 0.6409660577774048,
+      "learning_rate": 0.0005371958985429034,
+      "loss": 3.7615,
       "step": 9800
     },
     {
-      "epoch": 1.06016575180282,
-      "grad_norm": 0.5710098147392273,
-      "learning_rate": 0.0005369809287792263,
-      "loss": 3.7553,
+      "epoch": 1.0619946091644206,
+      "grad_norm": 0.5533095002174377,
+      "learning_rate": 0.0005368720992984349,
+      "loss": 3.764,
       "step": 9850
     },
     {
-      "epoch": 1.0655473038424281,
-      "grad_norm": 0.5798651576042175,
-      "learning_rate": 0.000536657687749165,
-      "loss": 3.7601,
+      "epoch": 1.0673854447439353,
+      "grad_norm": 0.6768531799316406,
+      "learning_rate": 0.0005365483000539665,
+      "loss": 3.7595,
       "step": 9900
     },
     {
-      "epoch": 1.0709288558820365,
-      "grad_norm": 0.5872936248779297,
-      "learning_rate": 0.0005363344467191035,
-      "loss": 3.7599,
+      "epoch": 1.0727762803234502,
+      "grad_norm": 0.597683846950531,
+      "learning_rate": 0.0005362245008094981,
+      "loss": 3.7761,
       "step": 9950
     },
     {
-      "epoch": 1.0763104079216446,
-      "grad_norm": 0.629589855670929,
-      "learning_rate": 0.000536011205689042,
-      "loss": 3.7431,
+      "epoch": 1.0781671159029649,
+      "grad_norm": 0.5655646920204163,
+      "learning_rate": 0.0005359007015650297,
+      "loss": 3.7663,
       "step": 10000
     },
     {
-      "epoch": 1.0763104079216446,
-      "eval_accuracy": 0.3454999939697647,
-      "eval_loss": 3.7408206462860107,
-      "eval_runtime": 180.1755,
-      "eval_samples_per_second": 99.964,
-      "eval_steps_per_second": 6.249,
+      "epoch": 1.0781671159029649,
+      "eval_accuracy": 0.3454923882676177,
+      "eval_loss": 3.7435693740844727,
+      "eval_runtime": 184.0737,
+      "eval_samples_per_second": 97.847,
+      "eval_steps_per_second": 6.117,
       "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": 8.3611359903744e+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:e8878629e9b28f39e30d4d5868e85c181cb6ee54e4ae51c97ee0efcf19286957
+oid sha256:e0ec3c621cbba9d7deb4f98fb044e9dd59a729095fcbb8aec0ad231643d9d8af
 size 5304