Model save

README.md

@@ -56,7 +56,7 @@ The following hyperparameters were used during training:
 
 | Training Loss | Epoch | Step | Validation Loss |
 |:-------------:|:-----:|:----:|:---------------:|
-| 0.9403        | 1.0   | 1084 | 0.9415          |
+| 0.9404        | 1.0   | 1084 | 0.9415          |
 
 
 ### Framework versions

all_results.json

@@ -1,9 +1,9 @@
 {
     "epoch": 1.0,
     "total_flos": 453935093514240.0,
-    "train_loss": 0.9830407258329357,
-    "train_runtime": 11998.8497,
+    "train_loss": 0.9826947098728476,
+    "train_runtime": 5906.6996,
     "train_samples": 207864,
-    "train_samples_per_second": 11.558,
-    "train_steps_per_second": 0.09
+    "train_samples_per_second": 23.48,
+    "train_steps_per_second": 0.184
 }

train_results.json

@@ -1,9 +1,9 @@
 {
     "epoch": 1.0,
     "total_flos": 453935093514240.0,
-    "train_loss": 0.9830407258329357,
-    "train_runtime": 11998.8497,
+    "train_loss": 0.9826947098728476,
+    "train_runtime": 5906.6996,
     "train_samples": 207864,
-    "train_samples_per_second": 11.558,
-    "train_steps_per_second": 0.09
+    "train_samples_per_second": 23.48,
+    "train_steps_per_second": 0.184
 }

trainer_state.json

@@ -10,1539 +10,1539 @@
   "log_history": [
     {
       "epoch": 0.0009225092250922509,
-      "grad_norm": 9.870575577438405,
+      "grad_norm": 9.87069124888532,
       "learning_rate": 1.8348623853211012e-07,
       "loss": 1.1607,
       "step": 1
     },
     {
       "epoch": 0.004612546125461255,
-      "grad_norm": 7.65916956083968,
+      "grad_norm": 7.637481536818217,
       "learning_rate": 9.174311926605506e-07,
       "loss": 1.1303,
       "step": 5
     },
     {
       "epoch": 0.00922509225092251,
-      "grad_norm": 3.7726483117427345,
+      "grad_norm": 3.662821001105185,
       "learning_rate": 1.8348623853211011e-06,
-      "loss": 1.057,
+      "loss": 1.0569,
       "step": 10
     },
     {
       "epoch": 0.013837638376383764,
-      "grad_norm": 2.5379513268735505,
+      "grad_norm": 2.537889977772848,
       "learning_rate": 2.7522935779816517e-06,
-      "loss": 1.0238,
+      "loss": 1.0237,
       "step": 15
     },
     {
       "epoch": 0.01845018450184502,
-      "grad_norm": 2.0372936699866466,
+      "grad_norm": 2.1108929850361218,
       "learning_rate": 3.6697247706422022e-06,
-      "loss": 0.9931,
+      "loss": 0.993,
       "step": 20
     },
     {
       "epoch": 0.023062730627306273,
-      "grad_norm": 1.9015079532614876,
+      "grad_norm": 1.9063826501229393,
       "learning_rate": 4.587155963302753e-06,
-      "loss": 0.9821,
+      "loss": 0.9822,
       "step": 25
     },
     {
       "epoch": 0.027675276752767528,
-      "grad_norm": 2.2588401749573546,
+      "grad_norm": 2.303421644374396,
       "learning_rate": 5.504587155963303e-06,
-      "loss": 0.9798,
+      "loss": 0.9799,
       "step": 30
     },
     {
       "epoch": 0.03228782287822878,
-      "grad_norm": 2.7788236168992237,
+      "grad_norm": 2.4361534479639824,
       "learning_rate": 6.422018348623854e-06,
-      "loss": 0.9812,
+      "loss": 0.9816,
       "step": 35
     },
     {
       "epoch": 0.03690036900369004,
-      "grad_norm": 2.0946702560307022,
+      "grad_norm": 2.0144702285707354,
       "learning_rate": 7.3394495412844045e-06,
-      "loss": 0.9807,
+      "loss": 0.9806,
       "step": 40
     },
     {
       "epoch": 0.04151291512915129,
-      "grad_norm": 2.5421232639973415,
+      "grad_norm": 2.0841326391410573,
       "learning_rate": 8.256880733944956e-06,
-      "loss": 0.9867,
+      "loss": 0.9861,
       "step": 45
     },
     {
       "epoch": 0.046125461254612546,
-      "grad_norm": 2.346175802405767,
+      "grad_norm": 1.9352245609335417,
       "learning_rate": 9.174311926605506e-06,
-      "loss": 0.9966,
+      "loss": 0.9963,
       "step": 50
     },
     {
       "epoch": 0.0507380073800738,
-      "grad_norm": 2.4887196180378783,
+      "grad_norm": 2.3043607004443283,
       "learning_rate": 1.0091743119266055e-05,
-      "loss": 0.9908,
+      "loss": 0.9904,
       "step": 55
     },
     {
       "epoch": 0.055350553505535055,
-      "grad_norm": 2.0063521735753165,
+      "grad_norm": 2.199239666479046,
       "learning_rate": 1.1009174311926607e-05,
-      "loss": 0.9924,
+      "loss": 0.994,
       "step": 60
     },
     {
       "epoch": 0.05996309963099631,
-      "grad_norm": 1.9753528261008033,
+      "grad_norm": 1.9024609720272683,
       "learning_rate": 1.1926605504587156e-05,
-      "loss": 1.0028,
+      "loss": 1.0039,
       "step": 65
     },
     {
       "epoch": 0.06457564575645756,
-      "grad_norm": 2.8188122479797793,
+      "grad_norm": 2.57414786410357,
       "learning_rate": 1.2844036697247708e-05,
-      "loss": 0.9874,
+      "loss": 0.9881,
       "step": 70
     },
     {
       "epoch": 0.06918819188191883,
-      "grad_norm": 2.4283436777324243,
+      "grad_norm": 2.4073690509013734,
       "learning_rate": 1.3761467889908258e-05,
-      "loss": 1.0038,
+      "loss": 1.0037,
       "step": 75
     },
     {
       "epoch": 0.07380073800738007,
-      "grad_norm": 2.2694140697845735,
+      "grad_norm": 2.344151583739395,
       "learning_rate": 1.4678899082568809e-05,
-      "loss": 1.0162,
+      "loss": 1.0155,
       "step": 80
     },
     {
       "epoch": 0.07841328413284133,
-      "grad_norm": 2.2296135287203533,
+      "grad_norm": 2.398529050285677,
       "learning_rate": 1.559633027522936e-05,
-      "loss": 1.0022,
+      "loss": 1.0012,
       "step": 85
     },
     {
       "epoch": 0.08302583025830258,
-      "grad_norm": 2.496334289275322,
+      "grad_norm": 2.418044809142461,
       "learning_rate": 1.6513761467889912e-05,
-      "loss": 0.9963,
+      "loss": 0.9945,
       "step": 90
     },
     {
       "epoch": 0.08763837638376384,
-      "grad_norm": 3.048918418901738,
+      "grad_norm": 2.5129111485618942,
       "learning_rate": 1.743119266055046e-05,
-      "loss": 0.9874,
+      "loss": 0.9857,
       "step": 95
     },
     {
       "epoch": 0.09225092250922509,
-      "grad_norm": 2.219949213385871,
+      "grad_norm": 1.9698114649305183,
       "learning_rate": 1.834862385321101e-05,
-      "loss": 1.0366,
+      "loss": 1.0359,
       "step": 100
     },
     {
       "epoch": 0.09686346863468635,
-      "grad_norm": 2.3152361863940305,
+      "grad_norm": 2.1358990392067705,
       "learning_rate": 1.9266055045871563e-05,
-      "loss": 1.009,
+      "loss": 1.0084,
       "step": 105
     },
     {
       "epoch": 0.1014760147601476,
-      "grad_norm": 2.1557669545307925,
+      "grad_norm": 2.0245472890624825,
       "learning_rate": 1.9999948088910656e-05,
-      "loss": 1.0126,
+      "loss": 1.0141,
       "step": 110
     },
     {
       "epoch": 0.10608856088560886,
-      "grad_norm": 2.5252882972323283,
+      "grad_norm": 2.0374124109562066,
       "learning_rate": 1.9998131257372878e-05,
-      "loss": 1.0228,
+      "loss": 1.0244,
       "step": 115
     },
     {
       "epoch": 0.11070110701107011,
-      "grad_norm": 1.754976914882788,
+      "grad_norm": 1.7325251719074102,
       "learning_rate": 1.999371941029485e-05,
-      "loss": 1.0132,
+      "loss": 1.0162,
       "step": 120
     },
     {
       "epoch": 0.11531365313653137,
-      "grad_norm": 1.8015448977018316,
+      "grad_norm": 1.9546491823840284,
       "learning_rate": 1.9986713692771732e-05,
-      "loss": 1.0354,
+      "loss": 1.0385,
       "step": 125
     },
     {
       "epoch": 0.11992619926199262,
-      "grad_norm": 1.8342985482903624,
+      "grad_norm": 1.7981992444328092,
       "learning_rate": 1.9977115923137912e-05,
-      "loss": 1.0299,
+      "loss": 1.03,
       "step": 130
     },
     {
       "epoch": 0.12453874538745388,
-      "grad_norm": 1.7924526745821787,
+      "grad_norm": 1.787669857072768,
       "learning_rate": 1.9964928592495046e-05,
-      "loss": 1.0315,
+      "loss": 1.0301,
       "step": 135
     },
     {
       "epoch": 0.12915129151291513,
-      "grad_norm": 15.523610268563516,
+      "grad_norm": 1.8171487016580647,
       "learning_rate": 1.9950154864065497e-05,
-      "loss": 1.0376,
+      "loss": 1.0284,
       "step": 140
     },
     {
       "epoch": 0.13376383763837638,
-      "grad_norm": 2.102764923940876,
+      "grad_norm": 1.8557221003169828,
       "learning_rate": 1.993279857237133e-05,
-      "loss": 1.0436,
+      "loss": 1.0318,
       "step": 145
     },
     {
       "epoch": 0.13837638376383765,
-      "grad_norm": 2.1109629634915197,
+      "grad_norm": 2.1551905075917976,
       "learning_rate": 1.9912864222239045e-05,
-      "loss": 1.0436,
+      "loss": 1.0407,
       "step": 150
     },
     {
       "epoch": 0.1429889298892989,
-      "grad_norm": 2.01098830232536,
+      "grad_norm": 2.514388634020912,
       "learning_rate": 1.9890356987630362e-05,
-      "loss": 1.0206,
+      "loss": 1.0212,
       "step": 155
     },
     {
       "epoch": 0.14760147601476015,
-      "grad_norm": 2.0368730493904157,
+      "grad_norm": 1.9323789502583069,
       "learning_rate": 1.986528271029931e-05,
-      "loss": 1.0232,
+      "loss": 1.0247,
       "step": 160
     },
     {
       "epoch": 0.1522140221402214,
-      "grad_norm": 1.6688489601440277,
+      "grad_norm": 2.1684894847599407,
       "learning_rate": 1.9837647898276008e-05,
-      "loss": 1.0236,
+      "loss": 1.0242,
       "step": 165
     },
     {
       "epoch": 0.15682656826568267,
-      "grad_norm": 2.0005420507660334,
+      "grad_norm": 1.9242427718398765,
       "learning_rate": 1.9807459724177497e-05,
-      "loss": 1.0246,
+      "loss": 1.0237,
       "step": 170
     },
     {
       "epoch": 0.16143911439114392,
-      "grad_norm": 1.830621233659763,
+      "grad_norm": 2.1094860370607793,
       "learning_rate": 1.977472602334609e-05,
-      "loss": 1.0322,
+      "loss": 1.0293,
       "step": 175
     },
     {
       "epoch": 0.16605166051660517,
-      "grad_norm": 2.0747249565016426,
+      "grad_norm": 1.6601715392116458,
       "learning_rate": 1.973945529181572e-05,
-      "loss": 1.0356,
+      "loss": 1.0321,
       "step": 180
     },
     {
       "epoch": 0.1706642066420664,
-      "grad_norm": 1.8657819252798358,
+      "grad_norm": 1.7762559074553415,
       "learning_rate": 1.9701656684106764e-05,
-      "loss": 1.0406,
+      "loss": 1.037,
       "step": 185
     },
     {
       "epoch": 0.1752767527675277,
-      "grad_norm": 1.863882659265557,
+      "grad_norm": 1.805704388216368,
       "learning_rate": 1.9661340010850025e-05,
-      "loss": 1.0295,
+      "loss": 1.027,
       "step": 190
     },
     {
       "epoch": 0.17988929889298894,
-      "grad_norm": 1.907383023598496,
+      "grad_norm": 1.7681290984831735,
       "learning_rate": 1.9618515736240353e-05,
-      "loss": 1.0518,
+      "loss": 1.05,
       "step": 195
     },
     {
       "epoch": 0.18450184501845018,
-      "grad_norm": 1.8946641894052252,
+      "grad_norm": 1.9579635085842144,
       "learning_rate": 1.9573194975320672e-05,
-      "loss": 1.0382,
+      "loss": 1.0384,
       "step": 200
     },
     {
       "epoch": 0.18911439114391143,
-      "grad_norm": 1.8166789629573785,
+      "grad_norm": 1.7758217950165514,
       "learning_rate": 1.952538949109708e-05,
-      "loss": 1.0443,
+      "loss": 1.0455,
       "step": 205
     },
     {
       "epoch": 0.1937269372693727,
-      "grad_norm": 1.7352343980788758,
+      "grad_norm": 1.7447577937143957,
       "learning_rate": 1.9475111691485737e-05,
-      "loss": 1.0217,
+      "loss": 1.0223,
       "step": 210
     },
     {
       "epoch": 0.19833948339483395,
-      "grad_norm": 1.6974181401216584,
+      "grad_norm": 1.7070823353641547,
       "learning_rate": 1.9422374626092414e-05,
-      "loss": 1.0305,
+      "loss": 1.0304,
       "step": 215
     },
     {
       "epoch": 0.2029520295202952,
-      "grad_norm": 1.5750696152336092,
+      "grad_norm": 1.608022610830541,
       "learning_rate": 1.936719198282545e-05,
-      "loss": 1.0293,
+      "loss": 1.0295,
       "step": 220
     },
     {
       "epoch": 0.20756457564575645,
-      "grad_norm": 1.623225736738918,
+      "grad_norm": 1.501683364990764,
       "learning_rate": 1.930957808434307e-05,
-      "loss": 1.0269,
+      "loss": 1.0259,
       "step": 225
     },
     {
       "epoch": 0.21217712177121772,
-      "grad_norm": 1.631576712487477,
+      "grad_norm": 1.6270563682677235,
       "learning_rate": 1.9249547884335917e-05,
-      "loss": 1.0305,
+      "loss": 1.0297,
       "step": 230
     },
     {
       "epoch": 0.21678966789667897,
-      "grad_norm": 1.7005295416853186,
+      "grad_norm": 1.7238840286664376,
       "learning_rate": 1.9187116963645845e-05,
-      "loss": 1.0356,
+      "loss": 1.0346,
       "step": 235
     },
     {
       "epoch": 0.22140221402214022,
-      "grad_norm": 1.7890847993052705,
+      "grad_norm": 1.9345013049084874,
       "learning_rate": 1.912230152622189e-05,
-      "loss": 1.033,
+      "loss": 1.0327,
       "step": 240
     },
     {
       "epoch": 0.22601476014760147,
-      "grad_norm": 1.7968201186150952,
+      "grad_norm": 1.8517916832280532,
       "learning_rate": 1.9055118394914545e-05,
       "loss": 1.0367,
       "step": 245
     },
     {
       "epoch": 0.23062730627306274,
-      "grad_norm": 1.5783711589062268,
+      "grad_norm": 1.6298646424255816,
       "learning_rate": 1.898558500710939e-05,
-      "loss": 1.0329,
+      "loss": 1.0328,
       "step": 250
     },
     {
       "epoch": 0.235239852398524,
-      "grad_norm": 1.7821446933403304,
+      "grad_norm": 1.6424022648576855,
       "learning_rate": 1.891371941020121e-05,
-      "loss": 1.0538,
+      "loss": 1.0527,
       "step": 255
     },
     {
       "epoch": 0.23985239852398524,
-      "grad_norm": 1.7571850494815153,
+      "grad_norm": 1.7425295163698495,
       "learning_rate": 1.88395402569098e-05,
-      "loss": 1.0172,
+      "loss": 1.0165,
       "step": 260
     },
     {
       "epoch": 0.2444649446494465,
-      "grad_norm": 1.6777787820198717,
+      "grad_norm": 1.6241841879077579,
       "learning_rate": 1.8763066800438638e-05,
-      "loss": 1.0262,
+      "loss": 1.0261,
       "step": 265
     },
     {
       "epoch": 0.24907749077490776,
-      "grad_norm": 1.7754650640017842,
+      "grad_norm": 1.6945753894174898,
       "learning_rate": 1.868431888947773e-05,
-      "loss": 1.0378,
+      "loss": 1.0365,
       "step": 270
     },
     {
       "epoch": 0.253690036900369,
-      "grad_norm": 1.6204893672219858,
+      "grad_norm": 1.576532058041066,
       "learning_rate": 1.860331696305188e-05,
-      "loss": 1.0272,
+      "loss": 1.0267,
       "step": 275
     },
     {
       "epoch": 0.25830258302583026,
-      "grad_norm": 1.6082347328850755,
+      "grad_norm": 1.584573157237757,
       "learning_rate": 1.852008204521572e-05,
-      "loss": 1.0199,
+      "loss": 1.0196,
       "step": 280
     },
     {
       "epoch": 0.2629151291512915,
-      "grad_norm": 1.5512676811150408,
+      "grad_norm": 1.525407220806812,
       "learning_rate": 1.8434635739596945e-05,
-      "loss": 1.0365,
+      "loss": 1.0359,
       "step": 285
     },
     {
       "epoch": 0.26752767527675275,
-      "grad_norm": 1.7967133178474277,
+      "grad_norm": 1.6099653934208198,
       "learning_rate": 1.834700022378907e-05,
-      "loss": 1.0541,
+      "loss": 1.053,
       "step": 290
     },
     {
       "epoch": 0.272140221402214,
-      "grad_norm": 1.6009722961453385,
+      "grad_norm": 1.6022748273878131,
       "learning_rate": 1.825719824359524e-05,
-      "loss": 1.0399,
+      "loss": 1.0392,
       "step": 295
     },
     {
       "epoch": 0.2767527675276753,
-      "grad_norm": 1.549904165894849,
+      "grad_norm": 1.581538008430849,
       "learning_rate": 1.816525310712456e-05,
-      "loss": 1.0096,
+      "loss": 1.0094,
       "step": 300
     },
     {
       "epoch": 0.28136531365313655,
-      "grad_norm": 1.515466383587945,
+      "grad_norm": 1.547945989389372,
       "learning_rate": 1.8071188678742457e-05,
       "loss": 1.0166,
       "step": 305
     },
     {
       "epoch": 0.2859778597785978,
-      "grad_norm": 1.5677132403955465,
+      "grad_norm": 1.5893709618261977,
       "learning_rate": 1.7975029372876706e-05,
-      "loss": 1.0209,
+      "loss": 1.0215,
       "step": 310
     },
     {
       "epoch": 0.29059040590405905,
-      "grad_norm": 1.542178919337743,
+      "grad_norm": 1.5451010927075632,
       "learning_rate": 1.787680014768065e-05,
       "loss": 1.0434,
       "step": 315
     },
     {
       "epoch": 0.2952029520295203,
-      "grad_norm": 1.6414967467646593,
+      "grad_norm": 1.56800364704446,
       "learning_rate": 1.777652649855531e-05,
-      "loss": 1.0184,
+      "loss": 1.0112,
       "step": 320
     },
     {
       "epoch": 0.29981549815498154,
-      "grad_norm": 1.6372383463858382,
+      "grad_norm": 1.5993537873748278,
       "learning_rate": 1.7674234451532065e-05,
-      "loss": 1.0317,
+      "loss": 1.0309,
       "step": 325
     },
     {
       "epoch": 0.3044280442804428,
-      "grad_norm": 1.650440536079242,
+      "grad_norm": 1.585947275219834,
       "learning_rate": 1.7569950556517566e-05,
       "loss": 1.052,
       "step": 330
     },
     {
       "epoch": 0.30904059040590404,
-      "grad_norm": 1.553654913479007,
+      "grad_norm": 1.5112282757161204,
       "learning_rate": 1.7463701880402738e-05,
-      "loss": 1.0205,
+      "loss": 1.021,
       "step": 335
     },
     {
       "epoch": 0.31365313653136534,
-      "grad_norm": 1.6079774780306095,
+      "grad_norm": 1.589682997809413,
       "learning_rate": 1.7355516000037555e-05,
-      "loss": 1.0289,
+      "loss": 1.0292,
       "step": 340
     },
     {
       "epoch": 0.3182656826568266,
-      "grad_norm": 1.6788267575243834,
+      "grad_norm": 1.7115271953853421,
       "learning_rate": 1.7245420995073453e-05,
-      "loss": 1.0452,
+      "loss": 1.0416,
       "step": 345
     },
     {
       "epoch": 0.32287822878228783,
-      "grad_norm": 1.5790256320964164,
+      "grad_norm": 1.602143873918192,
       "learning_rate": 1.7133445440675268e-05,
       "loss": 1.028,
       "step": 350
     },
     {
       "epoch": 0.3274907749077491,
-      "grad_norm": 1.5621840814508567,
+      "grad_norm": 1.5571424522990809,
       "learning_rate": 1.7019618400104572e-05,
-      "loss": 1.0262,
+      "loss": 1.0271,
       "step": 355
     },
     {
       "epoch": 0.33210332103321033,
-      "grad_norm": 1.519849047216207,
+      "grad_norm": 1.5239968044057388,
       "learning_rate": 1.6903969417176244e-05,
-      "loss": 1.0183,
+      "loss": 1.0184,
       "step": 360
     },
     {
       "epoch": 0.3367158671586716,
-      "grad_norm": 1.5037917545786645,
+      "grad_norm": 1.5569745783828202,
       "learning_rate": 1.6786528508590436e-05,
-      "loss": 1.0334,
+      "loss": 1.0329,
       "step": 365
     },
     {
       "epoch": 0.3413284132841328,
-      "grad_norm": 1.750901636962616,
+      "grad_norm": 1.9748263576909895,
       "learning_rate": 1.666732615614169e-05,
-      "loss": 1.0198,
+      "loss": 1.0202,
       "step": 370
     },
     {
       "epoch": 0.3459409594095941,
-      "grad_norm": 1.4681811166568075,
+      "grad_norm": 1.4426758781209656,
       "learning_rate": 1.6546393298807405e-05,
-      "loss": 1.0181,
+      "loss": 1.0184,
       "step": 375
     },
     {
       "epoch": 0.3505535055350554,
-      "grad_norm": 1.9552415843064699,
+      "grad_norm": 1.956219718617806,
       "learning_rate": 1.6423761324717636e-05,
-      "loss": 1.0237,
+      "loss": 1.0232,
       "step": 380
     },
     {
       "epoch": 0.3551660516605166,
-      "grad_norm": 1.5359322834804126,
+      "grad_norm": 1.552656856196208,
       "learning_rate": 1.6299462063008272e-05,
-      "loss": 1.0164,
+      "loss": 1.0155,
       "step": 385
     },
     {
       "epoch": 0.35977859778597787,
-      "grad_norm": 1.538202612072466,
+      "grad_norm": 1.547453869406189,
       "learning_rate": 1.61735277755598e-05,
-      "loss": 1.0236,
+      "loss": 1.023,
       "step": 390
     },
     {
       "epoch": 0.3643911439114391,
-      "grad_norm": 1.8749904513501805,
+      "grad_norm": 1.9841965884936383,
       "learning_rate": 1.6045991148623752e-05,
-      "loss": 1.0168,
+      "loss": 1.0169,
       "step": 395
     },
     {
       "epoch": 0.36900369003690037,
-      "grad_norm": 1.4674737229917856,
+      "grad_norm": 1.5804913256312618,
       "learning_rate": 1.5916885284338937e-05,
-      "loss": 1.0042,
+      "loss": 1.005,
       "step": 400
     },
     {
       "epoch": 0.3736162361623616,
-      "grad_norm": 1.4565983698821008,
+      "grad_norm": 1.4500292243389075,
       "learning_rate": 1.5786243692139826e-05,
-      "loss": 1.0269,
+      "loss": 1.0268,
       "step": 405
     },
     {
       "epoch": 0.37822878228782286,
-      "grad_norm": 1.5930896810188973,
+      "grad_norm": 1.5270225195154143,
       "learning_rate": 1.5654100280059155e-05,
-      "loss": 1.0039,
+      "loss": 1.0033,
       "step": 410
     },
     {
       "epoch": 0.3828413284132841,
-      "grad_norm": 1.4721927126588246,
+      "grad_norm": 1.4524051988943654,
       "learning_rate": 1.5520489345927095e-05,
-      "loss": 1.0036,
+      "loss": 1.0034,
       "step": 415
     },
     {
       "epoch": 0.3874538745387454,
-      "grad_norm": 1.5395462162350828,
+      "grad_norm": 1.5364744313244658,
       "learning_rate": 1.538544556846925e-05,
-      "loss": 1.0253,
+      "loss": 1.0248,
       "step": 420
     },
     {
       "epoch": 0.39206642066420666,
-      "grad_norm": 1.4973959788085256,
+      "grad_norm": 1.4472396779072696,
       "learning_rate": 1.5249003998305787e-05,
-      "loss": 1.0142,
+      "loss": 1.0135,
       "step": 425
     },
     {
       "epoch": 0.3966789667896679,
-      "grad_norm": 1.6021819916230071,
+      "grad_norm": 1.520494566330917,
       "learning_rate": 1.5111200048854055e-05,
-      "loss": 1.0203,
+      "loss": 1.0201,
       "step": 430
     },
     {
       "epoch": 0.40129151291512916,
-      "grad_norm": 1.5164653354921802,
+      "grad_norm": 1.4719100022352993,
       "learning_rate": 1.4972069487137024e-05,
-      "loss": 1.0097,
+      "loss": 1.0095,
       "step": 435
     },
     {
       "epoch": 0.4059040590405904,
-      "grad_norm": 1.6120537324024575,
+      "grad_norm": 1.598812244722486,
       "learning_rate": 1.4831648424499953e-05,
-      "loss": 1.0156,
+      "loss": 1.0162,
       "step": 440
     },
     {
       "epoch": 0.41051660516605165,
-      "grad_norm": 1.4054842646806989,
+      "grad_norm": 1.4065472391213587,
       "learning_rate": 1.4689973307237687e-05,
-      "loss": 1.0282,
+      "loss": 1.0291,
       "step": 445
     },
     {
       "epoch": 0.4151291512915129,
-      "grad_norm": 1.4671822665223906,
+      "grad_norm": 1.4646473697790325,
       "learning_rate": 1.4547080907135024e-05,
       "loss": 1.0125,
       "step": 450
     },
     {
       "epoch": 0.41974169741697415,
-      "grad_norm": 1.464262893049375,
+      "grad_norm": 1.4001634930874887,
       "learning_rate": 1.4403008311922593e-05,
-      "loss": 1.0106,
+      "loss": 1.0116,
       "step": 455
     },
     {
       "epoch": 0.42435424354243545,
-      "grad_norm": 1.443587186675441,
+      "grad_norm": 1.398806355327192,
       "learning_rate": 1.4257792915650728e-05,
-      "loss": 1.013,
+      "loss": 1.0134,
       "step": 460
     },
     {
       "epoch": 0.4289667896678967,
-      "grad_norm": 1.3761731833498907,
+      "grad_norm": 1.3844222701270295,
       "learning_rate": 1.4111472408983843e-05,
-      "loss": 1.0073,
+      "loss": 1.006,
       "step": 465
     },
     {
       "epoch": 0.43357933579335795,
-      "grad_norm": 1.529603561908737,
+      "grad_norm": 1.5808754908114087,
       "learning_rate": 1.3964084769417823e-05,
-      "loss": 1.0296,
+      "loss": 1.0305,
       "step": 470
     },
     {
       "epoch": 0.4381918819188192,
-      "grad_norm": 1.6733724515060537,
+      "grad_norm": 1.6160615173021982,
       "learning_rate": 1.3815668251422953e-05,
-      "loss": 0.9951,
+      "loss": 0.9955,
       "step": 475
     },
     {
       "epoch": 0.44280442804428044,
-      "grad_norm": 1.4450871420978533,
+      "grad_norm": 1.4667586778867048,
       "learning_rate": 1.3666261376514978e-05,
-      "loss": 0.9927,
+      "loss": 0.9926,
       "step": 480
     },
     {
       "epoch": 0.4474169741697417,
-      "grad_norm": 1.4561810147046073,
+      "grad_norm": 1.4591073794026388,
       "learning_rate": 1.3515902923256832e-05,
-      "loss": 1.0082,
+      "loss": 1.0086,
       "step": 485
     },
     {
       "epoch": 0.45202952029520294,
-      "grad_norm": 1.5238257427571544,
+      "grad_norm": 1.5638562371641933,
       "learning_rate": 1.3364631917193671e-05,
-      "loss": 1.0025,
+      "loss": 1.0009,
       "step": 490
     },
     {
       "epoch": 0.4566420664206642,
-      "grad_norm": 1.5528476103643793,
+      "grad_norm": 1.6344875906653302,
       "learning_rate": 1.321248762072377e-05,
-      "loss": 1.0031,
+      "loss": 1.0035,
       "step": 495
     },
     {
       "epoch": 0.4612546125461255,
-      "grad_norm": 1.4687610879413946,
+      "grad_norm": 1.451893714311935,
       "learning_rate": 1.3059509522907998e-05,
-      "loss": 0.9982,
+      "loss": 0.9977,
       "step": 500
     },
     {
       "epoch": 0.46586715867158673,
-      "grad_norm": 1.4093011733734708,
+      "grad_norm": 1.4139701174616282,
       "learning_rate": 1.2905737329220394e-05,
-      "loss": 0.9966,
+      "loss": 0.9965,
       "step": 505
     },
     {
       "epoch": 0.470479704797048,
-      "grad_norm": 1.44715856596638,
+      "grad_norm": 1.455878372967596,
       "learning_rate": 1.2751210951242636e-05,
       "loss": 1.0052,
       "step": 510
     },
     {
       "epoch": 0.47509225092250923,
-      "grad_norm": 1.3996570148561756,
+      "grad_norm": 1.4569058816755027,
       "learning_rate": 1.2595970496304975e-05,
       "loss": 0.9942,
       "step": 515
     },
     {
       "epoch": 0.4797047970479705,
-      "grad_norm": 1.4325383073010012,
+      "grad_norm": 1.427126351694174,
       "learning_rate": 1.2440056257076376e-05,
-      "loss": 1.0081,
+      "loss": 1.0083,
       "step": 520
     },
     {
       "epoch": 0.4843173431734317,
-      "grad_norm": 1.4383492965547426,
+      "grad_norm": 1.4756361746839437,
       "learning_rate": 1.2283508701106559e-05,
-      "loss": 0.9788,
+      "loss": 0.9791,
       "step": 525
     },
     {
       "epoch": 0.488929889298893,
-      "grad_norm": 1.4099081664313255,
+      "grad_norm": 1.3844767567894687,
       "learning_rate": 1.2126368460322637e-05,
-      "loss": 1.0083,
+      "loss": 1.0081,
       "step": 530
     },
     {
       "epoch": 0.4935424354243542,
-      "grad_norm": 1.3412375300770663,
+      "grad_norm": 1.3341569254535863,
       "learning_rate": 1.1968676320483103e-05,
       "loss": 0.9684,
       "step": 535
     },
     {
       "epoch": 0.4981549815498155,
-      "grad_norm": 1.4515358137974845,
+      "grad_norm": 1.4344785984280697,
       "learning_rate": 1.1810473210591882e-05,
       "loss": 0.9852,
       "step": 540
     },
     {
       "epoch": 0.5027675276752768,
-      "grad_norm": 1.4597275022169718,
+      "grad_norm": 1.754080538395525,
       "learning_rate": 1.1651800192275197e-05,
-      "loss": 0.9793,
+      "loss": 0.98,
       "step": 545
     },
     {
       "epoch": 0.507380073800738,
-      "grad_norm": 1.3868232464146542,
+      "grad_norm": 1.4043415597711335,
       "learning_rate": 1.1492698449124042e-05,
-      "loss": 0.9907,
+      "loss": 0.9885,
       "step": 550
     },
     {
       "epoch": 0.5119926199261993,
-      "grad_norm": 1.4023901770546143,
+      "grad_norm": 1.3966005324424933,
       "learning_rate": 1.1333209276004959e-05,
-      "loss": 0.9967,
+      "loss": 0.9963,
       "step": 555
     },
     {
       "epoch": 0.5166051660516605,
-      "grad_norm": 1.403043028756451,
+      "grad_norm": 1.547684566936734,
       "learning_rate": 1.1173374068341962e-05,
-      "loss": 1.0058,
+      "loss": 1.0059,
       "step": 560
     },
     {
       "epoch": 0.5212177121771218,
-      "grad_norm": 1.5463650175486758,
+      "grad_norm": 1.7439931213982827,
       "learning_rate": 1.1013234311372353e-05,
-      "loss": 0.9843,
+      "loss": 0.9885,
       "step": 565
     },
     {
       "epoch": 0.525830258302583,
-      "grad_norm": 1.566717033090475,
+      "grad_norm": 1.5974123883312268,
       "learning_rate": 1.0852831569379217e-05,
-      "loss": 0.9911,
+      "loss": 0.9919,
       "step": 570
     },
     {
       "epoch": 0.5304428044280443,
-      "grad_norm": 1.4828461511508186,
+      "grad_norm": 1.5564454777648937,
       "learning_rate": 1.0692207474903421e-05,
-      "loss": 0.9909,
+      "loss": 0.9908,
       "step": 575
     },
     {
       "epoch": 0.5350553505535055,
-      "grad_norm": 2.8521203782189253,
+      "grad_norm": 1.380805738943329,
       "learning_rate": 1.0531403717937888e-05,
-      "loss": 0.9927,
+      "loss": 0.9888,
       "step": 580
     },
     {
       "epoch": 0.5396678966789668,
-      "grad_norm": 1.512595456828133,
+      "grad_norm": 1.5491179141261517,
       "learning_rate": 1.037046203510694e-05,
-      "loss": 0.979,
+      "loss": 0.9762,
       "step": 585
     },
     {
       "epoch": 0.544280442804428,
-      "grad_norm": 1.5368674133781381,
+      "grad_norm": 1.4181650465011963,
       "learning_rate": 1.0209424198833571e-05,
-      "loss": 0.9877,
+      "loss": 0.9862,
       "step": 590
     },
     {
       "epoch": 0.5488929889298892,
-      "grad_norm": 1.6413121795161139,
+      "grad_norm": 1.5960637409599259,
       "learning_rate": 1.0048332006497406e-05,
-      "loss": 0.9954,
+      "loss": 0.9933,
       "step": 595
     },
     {
       "epoch": 0.5535055350553506,
-      "grad_norm": 1.4636611051453832,
+      "grad_norm": 1.4796470839876057,
       "learning_rate": 9.887227269586184e-06,
-      "loss": 0.9801,
+      "loss": 0.9802,
       "step": 600
     },
     {
       "epoch": 0.5581180811808119,
-      "grad_norm": 1.4876078711300504,
+      "grad_norm": 1.4873847151380262,
       "learning_rate": 9.7261518028436e-06,
-      "loss": 0.9896,
+      "loss": 0.9897,
       "step": 605
     },
     {
       "epoch": 0.5627306273062731,
-      "grad_norm": 1.4565652096222934,
+      "grad_norm": 1.41471946282249,
       "learning_rate": 9.565147413416266e-06,
-      "loss": 0.9782,
+      "loss": 0.9779,
       "step": 610
     },
     {
       "epoch": 0.5673431734317343,
-      "grad_norm": 1.3376736240483553,
+      "grad_norm": 1.3655515654525507,
       "learning_rate": 9.404255890002677e-06,
-      "loss": 0.996,
+      "loss": 0.9965,
       "step": 615
     },
     {
       "epoch": 0.5719557195571956,
-      "grad_norm": 1.266054229637192,
+      "grad_norm": 1.335274063215355,
       "learning_rate": 9.243518992006944e-06,
-      "loss": 0.974,
+      "loss": 0.9731,
       "step": 620
     },
     {
       "epoch": 0.5765682656826568,
-      "grad_norm": 1.3235752371132756,
+      "grad_norm": 1.3554465338690787,
       "learning_rate": 9.082978438700138e-06,
-      "loss": 0.9679,
+      "loss": 0.9676,
       "step": 625
     },
     {
       "epoch": 0.5811808118081181,
-      "grad_norm": 1.4334968823366236,
+      "grad_norm": 1.4235930731304616,
       "learning_rate": 8.922675898392072e-06,
       "loss": 0.9666,
       "step": 630
     },
     {
       "epoch": 0.5857933579335793,
-      "grad_norm": 1.373620518219274,
+      "grad_norm": 5.2910626769080595,
       "learning_rate": 8.762652977616258e-06,
-      "loss": 0.9862,
+      "loss": 0.986,
       "step": 635
     },
     {
       "epoch": 0.5904059040590406,
-      "grad_norm": 1.4317910304850499,
+      "grad_norm": 1.3831211598811353,
       "learning_rate": 8.602951210330942e-06,
-      "loss": 0.9662,
+      "loss": 0.9666,
       "step": 640
     },
     {
       "epoch": 0.5950184501845018,
-      "grad_norm": 1.3805470019673511,
+      "grad_norm": 1.384620337979853,
       "learning_rate": 8.443612047138965e-06,
-      "loss": 0.9836,
+      "loss": 0.9828,
       "step": 645
     },
     {
       "epoch": 0.5996309963099631,
-      "grad_norm": 1.4344389139852456,
+      "grad_norm": 1.4337206214300657,
       "learning_rate": 8.284676844529258e-06,
-      "loss": 0.9458,
+      "loss": 0.9454,
       "step": 650
     },
     {
       "epoch": 0.6042435424354243,
-      "grad_norm": 1.2923675142084592,
+      "grad_norm": 1.3016737402975005,
       "learning_rate": 8.126186854142752e-06,
-      "loss": 0.953,
+      "loss": 0.9526,
       "step": 655
     },
     {
       "epoch": 0.6088560885608856,
-      "grad_norm": 1.3520769915175057,
+      "grad_norm": 1.339333065167964,
       "learning_rate": 7.968183212065537e-06,
-      "loss": 0.965,
+      "loss": 0.9645,
       "step": 660
     },
     {
       "epoch": 0.6134686346863468,
-      "grad_norm": 1.3540469095893106,
+      "grad_norm": 1.3518100579415706,
       "learning_rate": 7.81070692815195e-06,
-      "loss": 0.962,
+      "loss": 0.9611,
       "step": 665
     },
     {
       "epoch": 0.6180811808118081,
-      "grad_norm": 1.4140291089073116,
+      "grad_norm": 1.3732804507416974,
       "learning_rate": 7.6537988753805e-06,
-      "loss": 0.9566,
+      "loss": 0.9561,
       "step": 670
     },
     {
       "epoch": 0.6226937269372693,
-      "grad_norm": 1.3299616725752847,
+      "grad_norm": 1.3297074852238435,
       "learning_rate": 7.497499779245268e-06,
-      "loss": 0.965,
+      "loss": 0.9647,
       "step": 675
     },
     {
       "epoch": 0.6273062730627307,
-      "grad_norm": 1.3769480821952969,
+      "grad_norm": 1.3980138453757798,
       "learning_rate": 7.3418502071856004e-06,
-      "loss": 0.9736,
+      "loss": 0.9731,
       "step": 680
     },
     {
       "epoch": 0.6319188191881919,
-      "grad_norm": 1.3550016015001538,
+      "grad_norm": 1.3624541986139007,
       "learning_rate": 7.186890558056836e-06,
-      "loss": 0.9528,
+      "loss": 0.953,
       "step": 685
     },
     {
       "epoch": 0.6365313653136532,
-      "grad_norm": 1.4444985355599935,
+      "grad_norm": 1.4662848790274252,
       "learning_rate": 7.0326610516447825e-06,
-      "loss": 0.9518,
+      "loss": 0.9521,
       "step": 690
     },
     {
       "epoch": 0.6411439114391144,
-      "grad_norm": 1.3752886192742222,
+      "grad_norm": 1.4634625847212066,
       "learning_rate": 6.879201718226658e-06,
-      "loss": 0.9437,
+      "loss": 0.9435,
       "step": 695
     },
     {
       "epoch": 0.6457564575645757,
-      "grad_norm": 1.2728918848531818,
+      "grad_norm": 1.2725239803843122,
       "learning_rate": 6.7265523881812335e-06,
-      "loss": 0.9564,
+      "loss": 0.9563,
       "step": 700
     },
     {
       "epoch": 0.6503690036900369,
-      "grad_norm": 1.2776578000630632,
+      "grad_norm": 1.2859613538989152,
       "learning_rate": 6.574752681650864e-06,
-      "loss": 0.9431,
+      "loss": 0.9425,
       "step": 705
     },
     {
       "epoch": 0.6549815498154982,
-      "grad_norm": 1.2934037917676169,
+      "grad_norm": 1.2989719564279842,
       "learning_rate": 6.423841998258069e-06,
-      "loss": 0.9491,
+      "loss": 0.9486,
       "step": 710
     },
     {
       "epoch": 0.6595940959409594,
-      "grad_norm": 1.3237574013597018,
+      "grad_norm": 1.3253272023108238,
       "learning_rate": 6.273859506879365e-06,
-      "loss": 0.9681,
+      "loss": 0.968,
       "step": 715
     },
     {
       "epoch": 0.6642066420664207,
-      "grad_norm": 1.3217921823273022,
+      "grad_norm": 1.3394701310088128,
       "learning_rate": 6.124844135478971e-06,
-      "loss": 0.9649,
+      "loss": 0.965,
       "step": 720
     },
     {
       "epoch": 0.6688191881918819,
-      "grad_norm": 1.2737386646079398,
+      "grad_norm": 1.340333856091027,
       "learning_rate": 5.976834561005069e-06,
-      "loss": 0.9606,
+      "loss": 0.9595,
       "step": 725
     },
     {
       "epoch": 0.6734317343173432,
-      "grad_norm": 1.3636332749577897,
+      "grad_norm": 1.3468413770308945,
       "learning_rate": 5.829869199351188e-06,
-      "loss": 0.9491,
+      "loss": 0.9489,
       "step": 730
     },
     {
       "epoch": 0.6780442804428044,
-      "grad_norm": 1.308891744452198,
+      "grad_norm": 1.4216747763305098,
       "learning_rate": 5.68398619538536e-06,
-      "loss": 0.9431,
+      "loss": 0.943,
       "step": 735
     },
     {
       "epoch": 0.6826568265682657,
-      "grad_norm": 1.3623472799172087,
+      "grad_norm": 1.3645056097489368,
       "learning_rate": 5.53922341304961e-06,
-      "loss": 0.938,
+      "loss": 0.9379,
       "step": 740
     },
     {
       "epoch": 0.6872693726937269,
-      "grad_norm": 1.3779395937417078,
+      "grad_norm": 1.3794510779283775,
       "learning_rate": 5.39561842553239e-06,
-      "loss": 0.9765,
+      "loss": 0.9761,
       "step": 745
     },
     {
       "epoch": 0.6918819188191881,
-      "grad_norm": 1.314594616549353,
+      "grad_norm": 1.350781271725051,
       "learning_rate": 5.2532085055164205e-06,
-      "loss": 0.9495,
+      "loss": 0.9496,
       "step": 750
     },
     {
       "epoch": 0.6964944649446494,
-      "grad_norm": 1.2958945545107472,
+      "grad_norm": 1.3103824946112994,
       "learning_rate": 5.112030615504601e-06,
-      "loss": 0.9526,
+      "loss": 0.9525,
       "step": 755
     },
     {
       "epoch": 0.7011070110701108,
-      "grad_norm": 1.268037176857953,
+      "grad_norm": 1.2630745586647596,
       "learning_rate": 4.972121398226371e-06,
       "loss": 0.9389,
       "step": 760
     },
     {
       "epoch": 0.705719557195572,
-      "grad_norm": 1.3213967171034264,
+      "grad_norm": 1.3222994315961483,
       "learning_rate": 4.833517167127077e-06,
-      "loss": 0.9541,
+      "loss": 0.953,
       "step": 765
     },
     {
       "epoch": 0.7103321033210332,
-      "grad_norm": 1.303739776653261,
+      "grad_norm": 1.2932817703941415,
       "learning_rate": 4.6962538969428416e-06,
-      "loss": 0.9523,
+      "loss": 0.9518,
       "step": 770
     },
     {
       "epoch": 0.7149446494464945,
-      "grad_norm": 1.2626443878609368,
+      "grad_norm": 1.273698406266777,
       "learning_rate": 4.560367214363295e-06,
-      "loss": 0.9463,
+      "loss": 0.9465,
       "step": 775
     },
     {
       "epoch": 0.7195571955719557,
-      "grad_norm": 1.4159625474171884,
+      "grad_norm": 1.3922401933210866,
       "learning_rate": 4.425892388784681e-06,
-      "loss": 0.9486,
+      "loss": 0.9484,
       "step": 780
     },
     {
       "epoch": 0.724169741697417,
-      "grad_norm": 1.4010891439452202,
+      "grad_norm": 1.3825399178312323,
       "learning_rate": 4.292864323155684e-06,
-      "loss": 0.9353,
+      "loss": 0.9356,
       "step": 785
     },
     {
       "epoch": 0.7287822878228782,
-      "grad_norm": 1.3287864388096529,
+      "grad_norm": 1.3059433185103952,
       "learning_rate": 4.161317544918345e-06,
       "loss": 0.9389,
       "step": 790
     },
     {
       "epoch": 0.7333948339483395,
-      "grad_norm": 1.306056357380541,
+      "grad_norm": 1.2856292890537573,
       "learning_rate": 4.031286197046493e-06,
       "loss": 0.9423,
       "step": 795
     },
     {
       "epoch": 0.7380073800738007,
-      "grad_norm": 1.3004419319996845,
+      "grad_norm": 1.3259995341033894,
       "learning_rate": 3.902804029183907e-06,
-      "loss": 0.9448,
+      "loss": 0.9452,
       "step": 800
     },
     {
       "epoch": 0.742619926199262,
-      "grad_norm": 1.3193243689915886,
+      "grad_norm": 1.3497689286484875,
       "learning_rate": 3.775904388884618e-06,
-      "loss": 0.9427,
+      "loss": 0.9426,
       "step": 805
     },
     {
       "epoch": 0.7472324723247232,
-      "grad_norm": 1.296148330228006,
+      "grad_norm": 1.2944563230099602,
       "learning_rate": 3.650620212957524e-06,
-      "loss": 0.9412,
+      "loss": 0.9408,
       "step": 810
     },
     {
       "epoch": 0.7518450184501845,
-      "grad_norm": 1.4326318299445335,
+      "grad_norm": 1.461614769681836,
       "learning_rate": 3.5269840189176616e-06,
-      "loss": 0.9289,
+      "loss": 0.9283,
       "step": 815
     },
     {
       "epoch": 0.7564575645756457,
-      "grad_norm": 1.3055827977236605,
+      "grad_norm": 1.309936562816762,
       "learning_rate": 3.405027896546277e-06,
-      "loss": 0.926,
+      "loss": 0.9264,
       "step": 820
     },
     {
       "epoch": 0.761070110701107,
-      "grad_norm": 1.327865337759123,
+      "grad_norm": 1.3200989144993005,
       "learning_rate": 3.2847834995619067e-06,
-      "loss": 0.9453,
+      "loss": 0.9451,
       "step": 825
     },
     {
       "epoch": 0.7656826568265682,
-      "grad_norm": 1.2543373923102779,
+      "grad_norm": 1.2485940930123258,
       "learning_rate": 3.1662820374046776e-06,
-      "loss": 0.9357,
+      "loss": 0.9354,
       "step": 830
     },
     {
       "epoch": 0.7702952029520295,
-      "grad_norm": 1.311192918139976,
+      "grad_norm": 1.3042562154959694,
       "learning_rate": 3.0495542671358745e-06,
-      "loss": 0.9299,
+      "loss": 0.9293,
       "step": 835
     },
     {
       "epoch": 0.7749077490774908,
-      "grad_norm": 1.3013405217794898,
+      "grad_norm": 1.2841878078681912,
       "learning_rate": 2.934630485454948e-06,
-      "loss": 0.9363,
+      "loss": 0.9354,
       "step": 840
     },
     {
       "epoch": 0.7795202952029521,
-      "grad_norm": 1.3139133572420352,
+      "grad_norm": 1.2965638770448034,
       "learning_rate": 2.8215405208360237e-06,
       "loss": 0.9399,
       "step": 845
     },
     {
       "epoch": 0.7841328413284133,
-      "grad_norm": 1.325489126281033,
+      "grad_norm": 1.309994639049123,
       "learning_rate": 2.7103137257858867e-06,
-      "loss": 0.9359,
+      "loss": 0.9356,
       "step": 850
     },
     {
       "epoch": 0.7887453874538746,
-      "grad_norm": 1.3557387383978536,
+      "grad_norm": 1.363624295286568,
       "learning_rate": 2.600978969225558e-06,
-      "loss": 0.9408,
+      "loss": 0.9404,
       "step": 855
     },
     {
       "epoch": 0.7933579335793358,
-      "grad_norm": 1.2968402542167223,
+      "grad_norm": 1.286633306272539,
       "learning_rate": 2.493564628997369e-06,
-      "loss": 0.9473,
+      "loss": 0.9477,
       "step": 860
     },
     {
       "epoch": 0.7979704797047971,
-      "grad_norm": 1.307351413531213,
+      "grad_norm": 1.3124608084944471,
       "learning_rate": 2.3880985844994674e-06,
-      "loss": 0.9288,
+      "loss": 0.9285,
       "step": 865
     },
     {
       "epoch": 0.8025830258302583,
-      "grad_norm": 1.301711812307168,
+      "grad_norm": 1.3039218311971004,
       "learning_rate": 2.284608209449746e-06,
-      "loss": 0.9367,
+      "loss": 0.9362,
       "step": 870
     },
     {
       "epoch": 0.8071955719557196,
-      "grad_norm": 1.3331160835163591,
+      "grad_norm": 1.360146108241008,
       "learning_rate": 2.183120364780975e-06,
-      "loss": 0.9394,
+      "loss": 0.9391,
       "step": 875
     },
     {
       "epoch": 0.8118081180811808,
-      "grad_norm": 1.2683359981117004,
+      "grad_norm": 1.2717837199196151,
       "learning_rate": 2.083661391669043e-06,
       "loss": 0.9255,
       "step": 880
     },
     {
       "epoch": 0.816420664206642,
-      "grad_norm": 1.2993174833270567,
+      "grad_norm": 1.348042003672959,
       "learning_rate": 1.986257104696121e-06,
-      "loss": 0.9368,
+      "loss": 0.9369,
       "step": 885
     },
     {
       "epoch": 0.8210332103321033,
-      "grad_norm": 1.3745325636574788,
+      "grad_norm": 1.4427511726488103,
       "learning_rate": 1.8909327851504633e-06,
-      "loss": 0.9527,
+      "loss": 0.953,
       "step": 890
     },
     {
       "epoch": 0.8256457564575646,
-      "grad_norm": 1.339181816674666,
+      "grad_norm": 1.3362385077260535,
       "learning_rate": 1.7977131744646724e-06,
-      "loss": 0.9331,
+      "loss": 0.9327,
       "step": 895
     },
     {
       "epoch": 0.8302583025830258,
-      "grad_norm": 1.327094422276523,
+      "grad_norm": 1.3126899684942004,
       "learning_rate": 1.7066224677940313e-06,
-      "loss": 0.9437,
+      "loss": 0.9438,
       "step": 900
     },
     {
       "epoch": 0.834870848708487,
-      "grad_norm": 1.2652133382811892,
+      "grad_norm": 1.263580902374999,
       "learning_rate": 1.6176843077366755e-06,
       "loss": 0.9139,
       "step": 905
     },
     {
       "epoch": 0.8394833948339483,
-      "grad_norm": 1.2892238897012769,
+      "grad_norm": 1.3323883748204894,
       "learning_rate": 1.5309217781971419e-06,
-      "loss": 0.9135,
+      "loss": 0.9137,
       "step": 910
     },
     {
       "epoch": 0.8440959409594095,
-      "grad_norm": 1.3102845800852012,
+      "grad_norm": 1.3049631322592083,
       "learning_rate": 1.446357398394934e-06,
-      "loss": 0.923,
+      "loss": 0.9232,
       "step": 915
     },
     {
       "epoch": 0.8487084870848709,
-      "grad_norm": 1.3117325038831478,
+      "grad_norm": 1.3184435732078046,
       "learning_rate": 1.3640131170196758e-06,
-      "loss": 0.9159,
+      "loss": 0.9163,
       "step": 920
     },
     {
       "epoch": 0.8533210332103321,
-      "grad_norm": 5.1971689200456375,
+      "grad_norm": 1.9046899055595279,
       "learning_rate": 1.2839103065343084e-06,
-      "loss": 0.9295,
+      "loss": 0.9283,
       "step": 925
     },
     {
       "epoch": 0.8579335793357934,
-      "grad_norm": 1.314904030690533,
+      "grad_norm": 1.3121941017876284,
       "learning_rate": 1.2060697576278812e-06,
-      "loss": 0.9336,
+      "loss": 0.9335,
       "step": 930
     },
     {
       "epoch": 0.8625461254612546,
-      "grad_norm": 1.2791863891720787,
+      "grad_norm": 1.2845805796103378,
       "learning_rate": 1.1305116738193211e-06,
-      "loss": 0.9346,
+      "loss": 0.9345,
       "step": 935
     },
     {
       "epoch": 0.8671586715867159,
-      "grad_norm": 1.2890531276807027,
+      "grad_norm": 1.3753219299344046,
       "learning_rate": 1.0572556662136036e-06,
-      "loss": 0.9302,
+      "loss": 0.9303,
       "step": 940
     },
     {
       "epoch": 0.8717712177121771,
-      "grad_norm": 1.2742267464237356,
+      "grad_norm": 1.281057437073945,
       "learning_rate": 9.863207484116987e-07,
-      "loss": 0.9181,
+      "loss": 0.9183,
       "step": 945
     },
     {
       "epoch": 0.8763837638376384,
-      "grad_norm": 1.2644327659379573,
+      "grad_norm": 1.2706888106388057,
       "learning_rate": 9.177253315755796e-07,
-      "loss": 0.9379,
+      "loss": 0.9381,
       "step": 950
     },
     {
       "epoch": 0.8809963099630996,
-      "grad_norm": 1.2421396307830366,
+      "grad_norm": 1.2538061234568678,
       "learning_rate": 8.514872196496182e-07,
       "loss": 0.935,
       "step": 955
     },
     {
       "epoch": 0.8856088560885609,
-      "grad_norm": 1.2203962112131188,
+      "grad_norm": 1.248939276086923,
       "learning_rate": 7.876236047395525e-07,
-      "loss": 0.9264,
+      "loss": 0.9262,
       "step": 960
     },
     {
       "epoch": 0.8902214022140221,
-      "grad_norm": 1.2994699310922848,
+      "grad_norm": 1.282623537519779,
       "learning_rate": 7.26151062650291e-07,
-      "loss": 0.934,
+      "loss": 0.9341,
       "step": 965
     },
     {
       "epoch": 0.8948339483394834,
-      "grad_norm": 1.3062465605195615,
+      "grad_norm": 1.3210618564039815,
       "learning_rate": 6.670855485836525e-07,
-      "loss": 0.9285,
+      "loss": 0.9282,
       "step": 970
     },
     {
       "epoch": 0.8994464944649446,
-      "grad_norm": 1.2690114831578048,
+      "grad_norm": 1.258506432298463,
       "learning_rate": 6.104423929971948e-07,
-      "loss": 0.933,
+      "loss": 0.9324,
       "step": 975
     },
     {
       "epoch": 0.9040590405904059,
-      "grad_norm": 1.2785142678323616,
+      "grad_norm": 1.26397810078899,
       "learning_rate": 5.562362976251901e-07,
-      "loss": 0.9379,
+      "loss": 0.9378,
       "step": 980
     },
     {
       "epoch": 0.9086715867158671,
-      "grad_norm": 1.2131026491304173,
+      "grad_norm": 1.2198200085641175,
       "learning_rate": 5.044813316627994e-07,
-      "loss": 0.9276,
+      "loss": 0.9271,
       "step": 985
     },
     {
       "epoch": 0.9132841328413284,
-      "grad_norm": 1.300615964359426,
+      "grad_norm": 1.2895766328734444,
       "learning_rate": 4.5519092811439627e-07,
-      "loss": 0.9255,
+      "loss": 0.9252,
       "step": 990
     },
     {
       "epoch": 0.9178966789667896,
-      "grad_norm": 1.3050095257612062,
+      "grad_norm": 1.3194370085413256,
       "learning_rate": 4.083778803070504e-07,
-      "loss": 0.9437,
+      "loss": 0.9439,
       "step": 995
     },
     {
       "epoch": 0.922509225092251,
-      "grad_norm": 1.2410745191815336,
+      "grad_norm": 1.2409895536435456,
       "learning_rate": 3.6405433856999684e-07,
-      "loss": 0.957,
+      "loss": 0.9568,
       "step": 1000
     },
     {
       "epoch": 0.9271217712177122,
-      "grad_norm": 1.3063199127205747,
+      "grad_norm": 1.3179871393791862,
       "learning_rate": 3.2223180708102933e-07,
-      "loss": 0.9324,
+      "loss": 0.9325,
       "step": 1005
     },
     {
       "epoch": 0.9317343173431735,
-      "grad_norm": 1.3222002154954164,
+      "grad_norm": 1.285178507362633,
       "learning_rate": 2.829211408805932e-07,
-      "loss": 0.9176,
+      "loss": 0.9178,
       "step": 1010
     },
     {
       "epoch": 0.9363468634686347,
-      "grad_norm": 1.2578620181074625,
+      "grad_norm": 1.2573019061474504,
       "learning_rate": 2.461325430543482e-07,
-      "loss": 0.9322,
+      "loss": 0.9315,
       "step": 1015
     },
     {
       "epoch": 0.940959409594096,
-      "grad_norm": 1.2337296124585635,
+      "grad_norm": 1.22778494354884,
       "learning_rate": 2.1187556208496885e-07,
-      "loss": 0.9218,
+      "loss": 0.921,
       "step": 1020
     },
     {
       "epoch": 0.9455719557195572,
-      "grad_norm": 1.2291744208349304,
+      "grad_norm": 1.24393857404682,
       "learning_rate": 1.8015908937382587e-07,
-      "loss": 0.9331,
+      "loss": 0.933,
       "step": 1025
     },
     {
       "epoch": 0.9501845018450185,
-      "grad_norm": 1.2819268220912796,
+      "grad_norm": 1.285512101839501,
       "learning_rate": 1.5099135693322776e-07,
-      "loss": 0.9337,
+      "loss": 0.9338,
       "step": 1030
     },
     {
       "epoch": 0.9547970479704797,
-      "grad_norm": 1.2311873702844616,
+      "grad_norm": 1.2209551692205383,
       "learning_rate": 1.2437993524979984e-07,
-      "loss": 0.9034,
+      "loss": 0.9032,
       "step": 1035
     },
     {
       "epoch": 0.959409594095941,
-      "grad_norm": 1.306307352211591,
+      "grad_norm": 1.2927034232954018,
       "learning_rate": 1.0033173131956175e-07,
-      "loss": 0.9213,
+      "loss": 0.9212,
       "step": 1040
     },
     {
       "epoch": 0.9640221402214022,
-      "grad_norm": 1.2302162082382204,
+      "grad_norm": 1.237696624670866,
       "learning_rate": 7.885298685522235e-08,
-      "loss": 0.9153,
+      "loss": 0.9155,
       "step": 1045
     },
     {
       "epoch": 0.9686346863468634,
-      "grad_norm": 1.2949224700302513,
+      "grad_norm": 1.2747731851750532,
       "learning_rate": 5.99492766661347e-08,
-      "loss": 0.9207,
+      "loss": 0.9212,
       "step": 1050
     },
     {
       "epoch": 0.9732472324723247,
-      "grad_norm": 1.2979324069801879,
+      "grad_norm": 1.2758933589184267,
       "learning_rate": 4.362550721136338e-08,
-      "loss": 0.9254,
+      "loss": 0.9255,
       "step": 1055
     },
     {
       "epoch": 0.977859778597786,
-      "grad_norm": 1.2464202848956238,
+      "grad_norm": 1.240622077510343,
       "learning_rate": 2.988591532620322e-08,
-      "loss": 0.9155,
+      "loss": 0.9159,
       "step": 1060
     },
     {
       "epoch": 0.9824723247232472,
-      "grad_norm": 1.2826213715670498,
+      "grad_norm": 1.2735821034434907,
       "learning_rate": 1.8734067122514464e-08,
-      "loss": 0.9229,
+      "loss": 0.9223,
       "step": 1065
     },
     {
       "epoch": 0.9870848708487084,
-      "grad_norm": 1.2838870564000777,
+      "grad_norm": 1.4488806574709552,
       "learning_rate": 1.0172857063137643e-08,
-      "loss": 0.9254,
+      "loss": 0.9252,
       "step": 1070
     },
     {
       "epoch": 0.9916974169741697,
-      "grad_norm": 1.2809568528218904,
+      "grad_norm": 1.2832307201478261,
       "learning_rate": 4.204507210633368e-09,
-      "loss": 0.9466,
+      "loss": 0.9462,
       "step": 1075
     },
     {
       "epoch": 0.996309963099631,
-      "grad_norm": 1.2838585438818646,
+      "grad_norm": 1.2810843286313107,
       "learning_rate": 8.30566650548148e-10,
-      "loss": 0.9403,
+      "loss": 0.9404,
       "step": 1080
     },
     {
       "epoch": 1.0,
-      "eval_loss": 0.9415282607078552,
-      "eval_runtime": 141.608,
-      "eval_samples_per_second": 108.398,
-      "eval_steps_per_second": 1.695,
+      "eval_loss": 0.9414571523666382,
+      "eval_runtime": 141.0114,
+      "eval_samples_per_second": 108.856,
+      "eval_steps_per_second": 1.702,
       "step": 1084
     },
     {
       "epoch": 1.0,
       "step": 1084,
       "total_flos": 453935093514240.0,
-      "train_loss": 0.9830407258329357,
-      "train_runtime": 11998.8497,
-      "train_samples_per_second": 11.558,
-      "train_steps_per_second": 0.09
+      "train_loss": 0.9826947098728476,
+      "train_runtime": 5906.6996,
+      "train_samples_per_second": 23.48,
+      "train_steps_per_second": 0.184
     }
   ],
   "logging_steps": 5,