Training in progress, step 500

all_results.json

@@ -2,16 +2,16 @@
     "epoch": 5.0,
     "eval_accuracy": 0.00733640796431187,
     "eval_bleu": 1.0,
-    "eval_loss": 1.5712791681289673,
-    "eval_perplexity": 4.8128006391188904,
-    "eval_runtime": 30.7765,
+    "eval_loss": 2.2545549869537354,
+    "eval_perplexity": 9.531050924252332,
+    "eval_runtime": 32.1878,
     "eval_samples": 2282,
-    "eval_samples_per_second": 74.148,
-    "eval_steps_per_second": 1.17,
-    "perplexity": 4.8128006391188904,
+    "eval_samples_per_second": 70.896,
+    "eval_steps_per_second": 1.118,
+    "perplexity": 9.531050924252332,
     "total_flos": 2.1154254410092954e+18,
-    "train_loss": 3.7582801998715505,
-    "train_runtime": 52733.3783,
-    "train_samples_per_second": 21.619,
-    "train_steps_per_second": 0.338
+    "train_loss": 5.041201007510048,
+    "train_runtime": 53587.774,
+    "train_samples_per_second": 21.274,
+    "train_steps_per_second": 0.332
 }

eval_metrics.json

@@ -1,10 +1,10 @@
 {
-    "eval_loss": 23.060916900634766,
-    "eval_model_preparation_time": 0.0165,
-    "eval_accuracy": 0.00733640796431187,
-    "eval_perplexity": 10356880316.044792,
+    "eval_loss": 21.727209091186523,
+    "eval_model_preparation_time": 0.0159,
+    "eval_accuracy": 0.007341553311802912,
+    "eval_perplexity": 2729021866.2380643,
     "eval_bleu": 1.0,
-    "eval_runtime": 11.0041,
-    "eval_samples_per_second": 207.378,
-    "eval_steps_per_second": 3.272
+    "eval_runtime": 35.6588,
+    "eval_samples_per_second": 63.995,
+    "eval_steps_per_second": 1.01
 }

eval_results.json

@@ -2,11 +2,11 @@
     "epoch": 5.0,
     "eval_accuracy": 0.00733640796431187,
     "eval_bleu": 1.0,
-    "eval_loss": 1.5712791681289673,
-    "eval_perplexity": 4.8128006391188904,
-    "eval_runtime": 30.7765,
+    "eval_loss": 2.2545549869537354,
+    "eval_perplexity": 9.531050924252332,
+    "eval_runtime": 32.1878,
     "eval_samples": 2282,
-    "eval_samples_per_second": 74.148,
-    "eval_steps_per_second": 1.17,
-    "perplexity": 4.8128006391188904
+    "eval_samples_per_second": 70.896,
+    "eval_steps_per_second": 1.118,
+    "perplexity": 9.531050924252332
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9839d5f4cd171dea6e15ba583be6e1cce05dea4e8444d950b8ccd0ca73da4483
+oid sha256:1aa4939bd0bd098061d24b16a523517a09ce1571e0f03bd10e5237ee28b168a8
 size 1417229824

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 5.0,
     "total_flos": 2.1154254410092954e+18,
-    "train_loss": 3.7582801998715505,
-    "train_runtime": 52733.3783,
-    "train_samples_per_second": 21.619,
-    "train_steps_per_second": 0.338
+    "train_loss": 5.041201007510048,
+    "train_runtime": 53587.774,
+    "train_samples_per_second": 21.274,
+    "train_steps_per_second": 0.332
 }

trainer_state.json

@@ -1,5 +1,5 @@
 {
-  "best_metric": 1.5712791681289673,
+  "best_metric": 2.2545549869537354,
   "best_model_checkpoint": "./output/models/duo-predict-gpt2-medium-wikitext/checkpoint-17500",
   "epoch": 5.0,
   "eval_steps": 500,
@@ -10,1643 +10,1643 @@
   "log_history": [
     {
       "epoch": 0.028066236317709794,
-      "grad_norm": 18.272640228271484,
+      "grad_norm": 40.05345916748047,
       "learning_rate": 5.611672278338946e-06,
-      "loss": 16.6534,
+      "loss": 16.6504,
       "step": 100
     },
     {
       "epoch": 0.05613247263541959,
-      "grad_norm": 11.887374877929688,
+      "grad_norm": 13.785152435302734,
       "learning_rate": 1.1223344556677892e-05,
-      "loss": 11.6666,
+      "loss": 11.9434,
       "step": 200
     },
     {
       "epoch": 0.08419870895312939,
-      "grad_norm": 10.138073921203613,
+      "grad_norm": 11.410269737243652,
       "learning_rate": 1.6835016835016837e-05,
-      "loss": 9.3278,
+      "loss": 9.39,
       "step": 300
     },
     {
       "epoch": 0.11226494527083918,
-      "grad_norm": 6.484872817993164,
+      "grad_norm": 7.723413944244385,
       "learning_rate": 2.2446689113355783e-05,
-      "loss": 8.1162,
+      "loss": 8.1451,
       "step": 400
     },
     {
       "epoch": 0.140331181588549,
-      "grad_norm": 3.6428794860839844,
+      "grad_norm": 3.728137493133545,
       "learning_rate": 2.8058361391694725e-05,
-      "loss": 7.6316,
+      "loss": 7.6654,
       "step": 500
     },
     {
       "epoch": 0.140331181588549,
-      "eval_accuracy": 0.007349271333039477,
+      "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 3.704052448272705,
-      "eval_perplexity": 40.61154753747588,
-      "eval_runtime": 30.7567,
-      "eval_samples_per_second": 74.195,
-      "eval_steps_per_second": 1.17,
+      "eval_loss": 3.7314507961273193,
+      "eval_perplexity": 41.7396199372531,
+      "eval_runtime": 32.1179,
+      "eval_samples_per_second": 71.051,
+      "eval_steps_per_second": 1.121,
       "step": 500
     },
     {
       "epoch": 0.16839741790625878,
-      "grad_norm": 3.5977585315704346,
+      "grad_norm": 2.9916584491729736,
       "learning_rate": 3.3670033670033675e-05,
-      "loss": 7.4262,
+      "loss": 7.4969,
       "step": 600
     },
     {
       "epoch": 0.19646365422396855,
-      "grad_norm": 3.0089902877807617,
+      "grad_norm": 2.379664659500122,
       "learning_rate": 3.9281705948372613e-05,
-      "loss": 7.2448,
+      "loss": 7.3784,
       "step": 700
     },
     {
       "epoch": 0.22452989054167835,
-      "grad_norm": 3.010714530944824,
+      "grad_norm": 2.5484609603881836,
       "learning_rate": 4.4893378226711566e-05,
-      "loss": 7.0773,
+      "loss": 7.2447,
       "step": 800
     },
     {
       "epoch": 0.25259612685938815,
-      "grad_norm": 3.661517858505249,
+      "grad_norm": 2.8146486282348633,
       "learning_rate": 5.050505050505051e-05,
-      "loss": 6.8409,
+      "loss": 7.1428,
       "step": 900
     },
     {
       "epoch": 0.280662363177098,
-      "grad_norm": 2.2237136363983154,
+      "grad_norm": 2.539112091064453,
       "learning_rate": 5.611672278338945e-05,
-      "loss": 6.4879,
+      "loss": 7.0276,
       "step": 1000
     },
     {
       "epoch": 0.280662363177098,
-      "eval_accuracy": 0.007337265522227044,
-      "eval_bleu": 0.9995095586630022,
-      "eval_loss": 3.119645595550537,
-      "eval_perplexity": 22.638355087518736,
-      "eval_runtime": 30.8837,
-      "eval_samples_per_second": 73.89,
-      "eval_steps_per_second": 1.166,
+      "eval_accuracy": 0.00733640796431187,
+      "eval_bleu": 1.0,
+      "eval_loss": 3.4734885692596436,
+      "eval_perplexity": 32.249049478083954,
+      "eval_runtime": 32.1825,
+      "eval_samples_per_second": 70.908,
+      "eval_steps_per_second": 1.119,
       "step": 1000
     },
     {
       "epoch": 0.30872859949480774,
-      "grad_norm": 2.7749340534210205,
+      "grad_norm": 3.6999666690826416,
       "learning_rate": 6.17283950617284e-05,
-      "loss": 6.1614,
+      "loss": 6.9362,
       "step": 1100
     },
     {
       "epoch": 0.33679483581251757,
-      "grad_norm": 1.8694963455200195,
+      "grad_norm": 3.0879666805267334,
       "learning_rate": 6.734006734006735e-05,
-      "loss": 5.8679,
+      "loss": 6.8122,
       "step": 1200
     },
     {
       "epoch": 0.36486107213022734,
-      "grad_norm": 1.7338366508483887,
+      "grad_norm": 2.944125175476074,
       "learning_rate": 7.295173961840629e-05,
-      "loss": 5.6537,
+      "loss": 6.6988,
       "step": 1300
     },
     {
       "epoch": 0.3929273084479371,
-      "grad_norm": 1.5872080326080322,
+      "grad_norm": 1.6308223009109497,
       "learning_rate": 7.856341189674523e-05,
-      "loss": 5.4812,
+      "loss": 6.588,
       "step": 1400
     },
     {
       "epoch": 0.42099354476564693,
-      "grad_norm": 1.8420644998550415,
+      "grad_norm": 1.453412413597107,
       "learning_rate": 8.417508417508418e-05,
-      "loss": 5.3189,
+      "loss": 6.4629,
       "step": 1500
     },
     {
       "epoch": 0.42099354476564693,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 2.5976459980010986,
-      "eval_perplexity": 13.43208164297786,
-      "eval_runtime": 30.857,
-      "eval_samples_per_second": 73.954,
-      "eval_steps_per_second": 1.167,
+      "eval_loss": 3.18629789352417,
+      "eval_perplexity": 24.198675337005124,
+      "eval_runtime": 32.1493,
+      "eval_samples_per_second": 70.981,
+      "eval_steps_per_second": 1.12,
       "step": 1500
     },
     {
       "epoch": 0.4490597810833567,
-      "grad_norm": 1.4730021953582764,
+      "grad_norm": 2.6827428340911865,
       "learning_rate": 8.978675645342313e-05,
-      "loss": 5.1876,
+      "loss": 6.3459,
       "step": 1600
     },
     {
       "epoch": 0.4771260174010665,
-      "grad_norm": 1.5234594345092773,
+      "grad_norm": 1.3705793619155884,
       "learning_rate": 9.539842873176207e-05,
-      "loss": 5.0777,
+      "loss": 6.2517,
       "step": 1700
     },
     {
       "epoch": 0.5051922537187763,
-      "grad_norm": 1.2363477945327759,
+      "grad_norm": 1.0982314348220825,
       "learning_rate": 9.988773155367056e-05,
-      "loss": 4.9567,
+      "loss": 6.1428,
       "step": 1800
     },
     {
       "epoch": 0.5332584900364861,
-      "grad_norm": 1.2880409955978394,
+      "grad_norm": 1.2678226232528687,
       "learning_rate": 9.926401796295142e-05,
-      "loss": 4.8629,
+      "loss": 6.0589,
       "step": 1900
     },
     {
       "epoch": 0.561324726354196,
-      "grad_norm": 1.1858636140823364,
+      "grad_norm": 1.3704299926757812,
       "learning_rate": 9.864030437223227e-05,
-      "loss": 4.7557,
+      "loss": 5.9671,
       "step": 2000
     },
     {
       "epoch": 0.561324726354196,
-      "eval_accuracy": 0.007337265522227044,
+      "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 2.3368639945983887,
-      "eval_perplexity": 10.348731942766573,
-      "eval_runtime": 30.8164,
-      "eval_samples_per_second": 74.051,
-      "eval_steps_per_second": 1.168,
+      "eval_loss": 2.9542465209960938,
+      "eval_perplexity": 19.18726007425907,
+      "eval_runtime": 32.253,
+      "eval_samples_per_second": 70.753,
+      "eval_steps_per_second": 1.116,
       "step": 2000
     },
     {
       "epoch": 0.5893909626719057,
-      "grad_norm": 1.1543203592300415,
+      "grad_norm": 1.1833741664886475,
       "learning_rate": 9.801659078151312e-05,
-      "loss": 4.6862,
+      "loss": 5.9115,
       "step": 2100
     },
     {
       "epoch": 0.6174571989896155,
-      "grad_norm": 1.0767921209335327,
+      "grad_norm": 1.0632543563842773,
       "learning_rate": 9.739287719079399e-05,
-      "loss": 4.6014,
+      "loss": 5.8478,
       "step": 2200
     },
     {
       "epoch": 0.6455234353073253,
-      "grad_norm": 1.0958943367004395,
+      "grad_norm": 1.0249277353286743,
       "learning_rate": 9.676916360007485e-05,
-      "loss": 4.5195,
+      "loss": 5.7905,
       "step": 2300
     },
     {
       "epoch": 0.6735896716250351,
-      "grad_norm": 1.1153371334075928,
+      "grad_norm": 1.1262586116790771,
       "learning_rate": 9.61454500093557e-05,
-      "loss": 4.4267,
+      "loss": 5.7317,
       "step": 2400
     },
     {
       "epoch": 0.7016559079427449,
-      "grad_norm": 1.1394137144088745,
+      "grad_norm": 1.010510802268982,
       "learning_rate": 9.552173641863657e-05,
-      "loss": 4.351,
+      "loss": 5.6969,
       "step": 2500
     },
     {
       "epoch": 0.7016559079427449,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 2.1233675479888916,
-      "eval_perplexity": 8.359240285181098,
-      "eval_runtime": 30.8157,
-      "eval_samples_per_second": 74.053,
-      "eval_steps_per_second": 1.168,
+      "eval_loss": 2.823347568511963,
+      "eval_perplexity": 16.833106436658287,
+      "eval_runtime": 32.2137,
+      "eval_samples_per_second": 70.839,
+      "eval_steps_per_second": 1.118,
       "step": 2500
     },
     {
       "epoch": 0.7297221442604547,
-      "grad_norm": 1.0487689971923828,
+      "grad_norm": 1.0142302513122559,
       "learning_rate": 9.489802282791742e-05,
-      "loss": 4.2704,
+      "loss": 5.6495,
       "step": 2600
     },
     {
       "epoch": 0.7577883805781644,
-      "grad_norm": 1.0785913467407227,
+      "grad_norm": 0.9616004228591919,
       "learning_rate": 9.427430923719829e-05,
-      "loss": 4.2184,
+      "loss": 5.6238,
       "step": 2700
     },
     {
       "epoch": 0.7858546168958742,
-      "grad_norm": 1.016829490661621,
+      "grad_norm": 1.0462982654571533,
       "learning_rate": 9.365059564647915e-05,
-      "loss": 4.1661,
+      "loss": 5.5827,
       "step": 2800
     },
     {
       "epoch": 0.8139208532135841,
-      "grad_norm": 1.0146821737289429,
+      "grad_norm": 0.9683243036270142,
       "learning_rate": 9.302688205576e-05,
-      "loss": 4.1221,
+      "loss": 5.5456,
       "step": 2900
     },
     {
       "epoch": 0.8419870895312939,
-      "grad_norm": 1.0396431684494019,
+      "grad_norm": 1.0990265607833862,
       "learning_rate": 9.240316846504086e-05,
-      "loss": 4.0701,
+      "loss": 5.5077,
       "step": 3000
     },
     {
       "epoch": 0.8419870895312939,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 2.002086877822876,
-      "eval_perplexity": 7.404492257321689,
-      "eval_runtime": 30.948,
-      "eval_samples_per_second": 73.736,
-      "eval_steps_per_second": 1.163,
+      "eval_loss": 2.7350916862487793,
+      "eval_perplexity": 15.41115636227158,
+      "eval_runtime": 32.161,
+      "eval_samples_per_second": 70.956,
+      "eval_steps_per_second": 1.119,
       "step": 3000
     },
     {
       "epoch": 0.8700533258490036,
-      "grad_norm": 1.023743987083435,
+      "grad_norm": 0.8678885102272034,
       "learning_rate": 9.177945487432171e-05,
-      "loss": 4.024,
+      "loss": 5.4624,
       "step": 3100
     },
     {
       "epoch": 0.8981195621667134,
-      "grad_norm": 1.0103009939193726,
+      "grad_norm": 0.8358529806137085,
       "learning_rate": 9.115574128360257e-05,
-      "loss": 4.0087,
+      "loss": 5.4515,
       "step": 3200
     },
     {
       "epoch": 0.9261857984844233,
-      "grad_norm": 0.9384715557098389,
+      "grad_norm": 1.0331616401672363,
       "learning_rate": 9.053202769288343e-05,
-      "loss": 3.9665,
+      "loss": 5.414,
       "step": 3300
     },
     {
       "epoch": 0.954252034802133,
-      "grad_norm": 0.9419653415679932,
+      "grad_norm": 1.0942983627319336,
       "learning_rate": 8.990831410216429e-05,
-      "loss": 3.9352,
+      "loss": 5.3818,
       "step": 3400
     },
     {
       "epoch": 0.9823182711198428,
-      "grad_norm": 0.9483797550201416,
+      "grad_norm": 1.0206348896026611,
       "learning_rate": 8.928460051144514e-05,
-      "loss": 3.9112,
+      "loss": 5.3536,
       "step": 3500
     },
     {
       "epoch": 0.9823182711198428,
-      "eval_accuracy": 0.00733640796431187,
+      "eval_accuracy": 0.007337265522227044,
       "eval_bleu": 1.0,
-      "eval_loss": 1.9279801845550537,
-      "eval_perplexity": 6.8756087463407365,
-      "eval_runtime": 30.8769,
-      "eval_samples_per_second": 73.906,
-      "eval_steps_per_second": 1.166,
+      "eval_loss": 2.660673141479492,
+      "eval_perplexity": 14.305915763562952,
+      "eval_runtime": 32.161,
+      "eval_samples_per_second": 70.956,
+      "eval_steps_per_second": 1.119,
       "step": 3500
     },
     {
       "epoch": 1.0103845074375526,
-      "grad_norm": 0.9196979403495789,
+      "grad_norm": 0.994918167591095,
       "learning_rate": 8.866088692072601e-05,
-      "loss": 3.8761,
+      "loss": 5.3179,
       "step": 3600
     },
     {
       "epoch": 1.0384507437552624,
-      "grad_norm": 0.8850469589233398,
+      "grad_norm": 0.9707785844802856,
       "learning_rate": 8.803717333000687e-05,
-      "loss": 3.8219,
+      "loss": 5.2636,
       "step": 3700
     },
     {
       "epoch": 1.0665169800729721,
-      "grad_norm": 0.8743171095848083,
+      "grad_norm": 0.9952827095985413,
       "learning_rate": 8.741345973928773e-05,
-      "loss": 3.8114,
+      "loss": 5.2529,
       "step": 3800
     },
     {
       "epoch": 1.094583216390682,
-      "grad_norm": 0.8881787657737732,
+      "grad_norm": 1.023485541343689,
       "learning_rate": 8.678974614856859e-05,
-      "loss": 3.7866,
+      "loss": 5.2212,
       "step": 3900
     },
     {
       "epoch": 1.122649452708392,
-      "grad_norm": 0.9794315695762634,
+      "grad_norm": 1.0113359689712524,
       "learning_rate": 8.616603255784944e-05,
-      "loss": 3.7775,
+      "loss": 5.2099,
       "step": 4000
     },
     {
       "epoch": 1.122649452708392,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.87688148021698,
-      "eval_perplexity": 6.533099461490326,
-      "eval_runtime": 30.8839,
-      "eval_samples_per_second": 73.89,
-      "eval_steps_per_second": 1.166,
+      "eval_loss": 2.6000266075134277,
+      "eval_perplexity": 13.464096276358179,
+      "eval_runtime": 32.2235,
+      "eval_samples_per_second": 70.818,
+      "eval_steps_per_second": 1.117,
       "step": 4000
     },
     {
       "epoch": 1.1507156890261017,
-      "grad_norm": 0.8849213719367981,
+      "grad_norm": 1.0399143695831299,
       "learning_rate": 8.55423189671303e-05,
-      "loss": 3.761,
+      "loss": 5.1888,
       "step": 4100
     },
     {
       "epoch": 1.1787819253438114,
-      "grad_norm": 0.9264663457870483,
+      "grad_norm": 1.063310146331787,
       "learning_rate": 8.491860537641115e-05,
-      "loss": 3.7411,
+      "loss": 5.1639,
       "step": 4200
     },
     {
       "epoch": 1.2068481616615212,
-      "grad_norm": 0.8461829423904419,
+      "grad_norm": 1.0906972885131836,
       "learning_rate": 8.429489178569201e-05,
-      "loss": 3.7261,
+      "loss": 5.1483,
       "step": 4300
     },
     {
       "epoch": 1.234914397979231,
-      "grad_norm": 0.9400424957275391,
+      "grad_norm": 1.0262458324432373,
       "learning_rate": 8.367117819497288e-05,
-      "loss": 3.7196,
+      "loss": 5.1362,
       "step": 4400
     },
     {
       "epoch": 1.2629806342969407,
-      "grad_norm": 0.8834152817726135,
+      "grad_norm": 1.0694169998168945,
       "learning_rate": 8.304746460425373e-05,
-      "loss": 3.703,
+      "loss": 5.1158,
       "step": 4500
     },
     {
       "epoch": 1.2629806342969407,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.8368678092956543,
-      "eval_perplexity": 6.276847156661428,
-      "eval_runtime": 30.8747,
-      "eval_samples_per_second": 73.912,
-      "eval_steps_per_second": 1.166,
+      "eval_loss": 2.549288511276245,
+      "eval_perplexity": 12.797994913550793,
+      "eval_runtime": 32.2254,
+      "eval_samples_per_second": 70.814,
+      "eval_steps_per_second": 1.117,
       "step": 4500
     },
     {
       "epoch": 1.2910468706146505,
-      "grad_norm": 0.8625416159629822,
+      "grad_norm": 0.9079988598823547,
       "learning_rate": 8.242375101353458e-05,
-      "loss": 3.6845,
+      "loss": 5.0967,
       "step": 4600
     },
     {
       "epoch": 1.3191131069323605,
-      "grad_norm": 0.8428071141242981,
+      "grad_norm": 1.0364301204681396,
       "learning_rate": 8.180003742281544e-05,
-      "loss": 3.6765,
+      "loss": 5.087,
       "step": 4700
     },
     {
       "epoch": 1.3471793432500703,
-      "grad_norm": 0.8716861605644226,
+      "grad_norm": 0.9813855290412903,
       "learning_rate": 8.117632383209631e-05,
-      "loss": 3.6565,
+      "loss": 5.0592,
       "step": 4800
     },
     {
       "epoch": 1.37524557956778,
-      "grad_norm": 0.878508985042572,
+      "grad_norm": 1.0680088996887207,
       "learning_rate": 8.055261024137716e-05,
-      "loss": 3.6412,
+      "loss": 5.0446,
       "step": 4900
     },
     {
       "epoch": 1.4033118158854898,
-      "grad_norm": 0.8190825581550598,
+      "grad_norm": 0.983411431312561,
       "learning_rate": 7.992889665065803e-05,
-      "loss": 3.6443,
+      "loss": 5.0453,
       "step": 5000
     },
     {
       "epoch": 1.4033118158854898,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.8060888051986694,
-      "eval_perplexity": 6.086594958242813,
-      "eval_runtime": 30.8156,
-      "eval_samples_per_second": 74.053,
-      "eval_steps_per_second": 1.168,
+      "eval_loss": 2.5125412940979004,
+      "eval_perplexity": 12.336240274028102,
+      "eval_runtime": 32.159,
+      "eval_samples_per_second": 70.96,
+      "eval_steps_per_second": 1.119,
       "step": 5000
     },
     {
       "epoch": 1.4313780522031996,
-      "grad_norm": 0.8404738903045654,
+      "grad_norm": 1.0499087572097778,
       "learning_rate": 7.930518305993888e-05,
-      "loss": 3.6201,
+      "loss": 5.0223,
       "step": 5100
     },
     {
       "epoch": 1.4594442885209093,
-      "grad_norm": 0.8295940160751343,
+      "grad_norm": 0.9792733788490295,
       "learning_rate": 7.868146946921974e-05,
-      "loss": 3.6238,
+      "loss": 5.0227,
       "step": 5200
     },
     {
       "epoch": 1.4875105248386191,
-      "grad_norm": 0.8289036750793457,
+      "grad_norm": 1.020256757736206,
       "learning_rate": 7.80577558785006e-05,
-      "loss": 3.605,
+      "loss": 5.0,
       "step": 5300
     },
     {
       "epoch": 1.5155767611563289,
-      "grad_norm": 0.8368122577667236,
+      "grad_norm": 1.1104633808135986,
       "learning_rate": 7.743404228778145e-05,
-      "loss": 3.5954,
+      "loss": 4.992,
       "step": 5400
     },
     {
       "epoch": 1.5436429974740387,
-      "grad_norm": 0.8075991272926331,
+      "grad_norm": 0.9350359439849854,
       "learning_rate": 7.681032869706232e-05,
-      "loss": 3.5659,
+      "loss": 4.955,
       "step": 5500
     },
     {
       "epoch": 1.5436429974740387,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.7788798809051514,
-      "eval_perplexity": 5.9232179917953625,
-      "eval_runtime": 30.8647,
-      "eval_samples_per_second": 73.936,
-      "eval_steps_per_second": 1.166,
+      "eval_loss": 2.4806432723999023,
+      "eval_perplexity": 11.948948374842802,
+      "eval_runtime": 32.229,
+      "eval_samples_per_second": 70.806,
+      "eval_steps_per_second": 1.117,
       "step": 5500
     },
     {
       "epoch": 1.5717092337917484,
-      "grad_norm": 0.8155397176742554,
+      "grad_norm": 0.9568005800247192,
       "learning_rate": 7.618661510634317e-05,
-      "loss": 3.5588,
+      "loss": 4.9496,
       "step": 5600
     },
     {
       "epoch": 1.5997754701094582,
-      "grad_norm": 0.8227369785308838,
+      "grad_norm": 0.9825177192687988,
       "learning_rate": 7.556290151562403e-05,
-      "loss": 3.5663,
+      "loss": 4.9622,
       "step": 5700
     },
     {
       "epoch": 1.627841706427168,
-      "grad_norm": 0.8452321290969849,
+      "grad_norm": 1.073524832725525,
       "learning_rate": 7.493918792490488e-05,
-      "loss": 3.5497,
+      "loss": 4.9399,
       "step": 5800
     },
     {
       "epoch": 1.655907942744878,
-      "grad_norm": 0.8576854467391968,
+      "grad_norm": 0.9796638488769531,
       "learning_rate": 7.431547433418574e-05,
-      "loss": 3.541,
+      "loss": 4.9266,
       "step": 5900
     },
     {
       "epoch": 1.6839741790625877,
-      "grad_norm": 0.8500741720199585,
+      "grad_norm": 0.9600309729576111,
       "learning_rate": 7.36917607434666e-05,
-      "loss": 3.5303,
+      "loss": 4.9157,
       "step": 6000
     },
     {
       "epoch": 1.6839741790625877,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.7552289962768555,
-      "eval_perplexity": 5.784772281608607,
-      "eval_runtime": 30.8476,
-      "eval_samples_per_second": 73.977,
-      "eval_steps_per_second": 1.167,
+      "eval_loss": 2.4536778926849365,
+      "eval_perplexity": 11.631045888200894,
+      "eval_runtime": 32.1946,
+      "eval_samples_per_second": 70.881,
+      "eval_steps_per_second": 1.118,
       "step": 6000
     },
     {
       "epoch": 1.7120404153802975,
-      "grad_norm": 0.7958677411079407,
+      "grad_norm": 1.0209360122680664,
       "learning_rate": 7.306804715274747e-05,
-      "loss": 3.5277,
+      "loss": 4.917,
       "step": 6100
     },
     {
       "epoch": 1.7401066516980073,
-      "grad_norm": 0.8098702430725098,
+      "grad_norm": 1.0189194679260254,
       "learning_rate": 7.244433356202833e-05,
-      "loss": 3.5095,
+      "loss": 4.8932,
       "step": 6200
     },
     {
       "epoch": 1.768172888015717,
-      "grad_norm": 0.8205238580703735,
+      "grad_norm": 0.9920120239257812,
       "learning_rate": 7.182061997130918e-05,
-      "loss": 3.4993,
+      "loss": 4.8782,
       "step": 6300
     },
     {
       "epoch": 1.796239124333427,
-      "grad_norm": 0.8508559465408325,
+      "grad_norm": 1.0284303426742554,
       "learning_rate": 7.119690638059004e-05,
-      "loss": 3.4999,
+      "loss": 4.8821,
       "step": 6400
     },
     {
       "epoch": 1.8243053606511368,
-      "grad_norm": 0.7979349493980408,
+      "grad_norm": 0.9442264437675476,
       "learning_rate": 7.057319278987089e-05,
-      "loss": 3.4926,
+      "loss": 4.8756,
       "step": 6500
     },
     {
       "epoch": 1.8243053606511368,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.7347623109817505,
-      "eval_perplexity": 5.667580522044148,
-      "eval_runtime": 30.8642,
-      "eval_samples_per_second": 73.937,
-      "eval_steps_per_second": 1.166,
+      "eval_loss": 2.4299607276916504,
+      "eval_perplexity": 11.358435999241271,
+      "eval_runtime": 32.2208,
+      "eval_samples_per_second": 70.824,
+      "eval_steps_per_second": 1.117,
       "step": 6500
     },
     {
       "epoch": 1.8523715969688466,
-      "grad_norm": 0.7983048558235168,
+      "grad_norm": 0.980950117111206,
       "learning_rate": 6.994947919915176e-05,
-      "loss": 3.4827,
+      "loss": 4.8649,
       "step": 6600
     },
     {
       "epoch": 1.8804378332865563,
-      "grad_norm": 0.8478196263313293,
+      "grad_norm": 0.9818932414054871,
       "learning_rate": 6.932576560843261e-05,
-      "loss": 3.4791,
+      "loss": 4.8603,
       "step": 6700
     },
     {
       "epoch": 1.908504069604266,
-      "grad_norm": 0.812839925289154,
+      "grad_norm": 0.8722342848777771,
       "learning_rate": 6.870205201771347e-05,
-      "loss": 3.4678,
+      "loss": 4.8513,
       "step": 6800
     },
     {
       "epoch": 1.9365703059219759,
-      "grad_norm": 0.760836660861969,
+      "grad_norm": 0.8945412635803223,
       "learning_rate": 6.807833842699432e-05,
-      "loss": 3.4628,
+      "loss": 4.8385,
       "step": 6900
     },
     {
       "epoch": 1.9646365422396856,
-      "grad_norm": 0.8636198043823242,
+      "grad_norm": 0.946018636226654,
       "learning_rate": 6.745462483627518e-05,
-      "loss": 3.464,
+      "loss": 4.844,
       "step": 7000
     },
     {
       "epoch": 1.9646365422396856,
-      "eval_accuracy": 0.007337265522227044,
+      "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.7166798114776611,
-      "eval_perplexity": 5.566017523877059,
-      "eval_runtime": 30.9572,
-      "eval_samples_per_second": 73.715,
-      "eval_steps_per_second": 1.163,
+      "eval_loss": 2.4100253582000732,
+      "eval_perplexity": 11.134243485859441,
+      "eval_runtime": 32.2405,
+      "eval_samples_per_second": 70.781,
+      "eval_steps_per_second": 1.117,
       "step": 7000
     },
     {
       "epoch": 1.9927027785573954,
-      "grad_norm": 0.8247271776199341,
+      "grad_norm": 0.9749892354011536,
       "learning_rate": 6.683091124555604e-05,
-      "loss": 3.4484,
+      "loss": 4.8242,
       "step": 7100
     },
     {
       "epoch": 2.020769014875105,
-      "grad_norm": 0.7949260473251343,
+      "grad_norm": 0.9573102593421936,
       "learning_rate": 6.62071976548369e-05,
-      "loss": 3.3852,
+      "loss": 4.7617,
       "step": 7200
     },
     {
       "epoch": 2.048835251192815,
-      "grad_norm": 0.7939648032188416,
+      "grad_norm": 0.973569393157959,
       "learning_rate": 6.558348406411777e-05,
-      "loss": 3.3462,
+      "loss": 4.7183,
       "step": 7300
     },
     {
       "epoch": 2.0769014875105247,
-      "grad_norm": 0.811790943145752,
+      "grad_norm": 0.9861319661140442,
       "learning_rate": 6.495977047339862e-05,
-      "loss": 3.3487,
+      "loss": 4.7262,
       "step": 7400
     },
     {
       "epoch": 2.1049677238282345,
-      "grad_norm": 0.8714494705200195,
+      "grad_norm": 0.9737097024917603,
       "learning_rate": 6.433605688267948e-05,
-      "loss": 3.3432,
+      "loss": 4.7136,
       "step": 7500
     },
     {
       "epoch": 2.1049677238282345,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.7015873193740845,
-      "eval_perplexity": 5.482643194156826,
-      "eval_runtime": 30.8028,
-      "eval_samples_per_second": 74.084,
-      "eval_steps_per_second": 1.169,
+      "eval_loss": 2.394768476486206,
+      "eval_perplexity": 10.96565895726802,
+      "eval_runtime": 32.1525,
+      "eval_samples_per_second": 70.974,
+      "eval_steps_per_second": 1.12,
       "step": 7500
     },
     {
       "epoch": 2.1330339601459443,
-      "grad_norm": 0.8343294262886047,
+      "grad_norm": 1.123484492301941,
       "learning_rate": 6.371234329196033e-05,
-      "loss": 3.3477,
+      "loss": 4.7211,
       "step": 7600
     },
     {
       "epoch": 2.161100196463654,
-      "grad_norm": 0.8184479475021362,
+      "grad_norm": 0.9870331883430481,
       "learning_rate": 6.30886297012412e-05,
-      "loss": 3.3313,
+      "loss": 4.7004,
       "step": 7700
     },
     {
       "epoch": 2.189166432781364,
-      "grad_norm": 0.8070995807647705,
+      "grad_norm": 0.977360725402832,
       "learning_rate": 6.246491611052205e-05,
-      "loss": 3.3352,
+      "loss": 4.7039,
       "step": 7800
     },
     {
       "epoch": 2.217232669099074,
-      "grad_norm": 0.8009788393974304,
+      "grad_norm": 0.9247573614120483,
       "learning_rate": 6.184120251980291e-05,
-      "loss": 3.3306,
+      "loss": 4.702,
       "step": 7900
     },
     {
       "epoch": 2.245298905416784,
-      "grad_norm": 0.8022483587265015,
+      "grad_norm": 1.0123528242111206,
       "learning_rate": 6.121748892908376e-05,
-      "loss": 3.3215,
+      "loss": 4.6911,
       "step": 8000
     },
     {
       "epoch": 2.245298905416784,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.688325047492981,
-      "eval_perplexity": 5.410410930178965,
-      "eval_runtime": 30.8326,
-      "eval_samples_per_second": 74.013,
-      "eval_steps_per_second": 1.168,
+      "eval_loss": 2.380514144897461,
+      "eval_perplexity": 10.810459577961986,
+      "eval_runtime": 32.2153,
+      "eval_samples_per_second": 70.836,
+      "eval_steps_per_second": 1.117,
       "step": 8000
     },
     {
       "epoch": 2.2733651417344936,
-      "grad_norm": 0.8439472317695618,
+      "grad_norm": 0.9982300400733948,
       "learning_rate": 6.0593775338364625e-05,
-      "loss": 3.3238,
+      "loss": 4.6946,
       "step": 8100
     },
     {
       "epoch": 2.3014313780522033,
-      "grad_norm": 0.8445388674736023,
+      "grad_norm": 1.065459132194519,
       "learning_rate": 5.9970061747645486e-05,
-      "loss": 3.3158,
+      "loss": 4.6873,
       "step": 8200
     },
     {
       "epoch": 2.329497614369913,
-      "grad_norm": 0.8093782663345337,
+      "grad_norm": 0.9361574053764343,
       "learning_rate": 5.934634815692635e-05,
-      "loss": 3.3138,
+      "loss": 4.685,
       "step": 8300
     },
     {
       "epoch": 2.357563850687623,
-      "grad_norm": 0.8225380778312683,
+      "grad_norm": 0.9473018646240234,
       "learning_rate": 5.87226345662072e-05,
-      "loss": 3.3111,
+      "loss": 4.6763,
       "step": 8400
     },
     {
       "epoch": 2.3856300870053326,
-      "grad_norm": 0.8095957040786743,
+      "grad_norm": 0.9295810461044312,
       "learning_rate": 5.809892097548806e-05,
-      "loss": 3.3122,
+      "loss": 4.6741,
       "step": 8500
     },
     {
       "epoch": 2.3856300870053326,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.676782488822937,
-      "eval_perplexity": 5.348319978183154,
-      "eval_runtime": 30.8301,
-      "eval_samples_per_second": 74.018,
-      "eval_steps_per_second": 1.168,
+      "eval_loss": 2.36684513092041,
+      "eval_perplexity": 10.66369659265753,
+      "eval_runtime": 32.3295,
+      "eval_samples_per_second": 70.586,
+      "eval_steps_per_second": 1.114,
       "step": 8500
     },
     {
       "epoch": 2.4136963233230424,
-      "grad_norm": 0.8112284541130066,
+      "grad_norm": 0.9082818031311035,
       "learning_rate": 5.747520738476891e-05,
-      "loss": 3.2971,
+      "loss": 4.6627,
       "step": 8600
     },
     {
       "epoch": 2.441762559640752,
-      "grad_norm": 0.7994459867477417,
+      "grad_norm": 0.9860712885856628,
       "learning_rate": 5.685149379404977e-05,
-      "loss": 3.2922,
+      "loss": 4.6598,
       "step": 8700
     },
     {
       "epoch": 2.469828795958462,
-      "grad_norm": 0.8026987910270691,
+      "grad_norm": 0.9747830629348755,
       "learning_rate": 5.622778020333064e-05,
-      "loss": 3.2983,
+      "loss": 4.6669,
       "step": 8800
     },
     {
       "epoch": 2.4978950322761717,
-      "grad_norm": 0.8162180781364441,
+      "grad_norm": 0.9759917259216309,
       "learning_rate": 5.5604066612611496e-05,
-      "loss": 3.2879,
+      "loss": 4.6527,
       "step": 8900
     },
     {
       "epoch": 2.5259612685938815,
-      "grad_norm": 0.7989441156387329,
+      "grad_norm": 0.9565300941467285,
       "learning_rate": 5.498035302189235e-05,
-      "loss": 3.2836,
+      "loss": 4.6485,
       "step": 9000
     },
     {
       "epoch": 2.5259612685938815,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.6650645732879639,
-      "eval_perplexity": 5.286014574081701,
-      "eval_runtime": 30.7972,
-      "eval_samples_per_second": 74.098,
-      "eval_steps_per_second": 1.169,
+      "eval_loss": 2.3538177013397217,
+      "eval_perplexity": 10.525677008415643,
+      "eval_runtime": 32.2172,
+      "eval_samples_per_second": 70.832,
+      "eval_steps_per_second": 1.117,
       "step": 9000
     },
     {
       "epoch": 2.5540275049115913,
-      "grad_norm": 0.7936413288116455,
+      "grad_norm": 0.9805787801742554,
       "learning_rate": 5.4356639431173205e-05,
-      "loss": 3.2837,
+      "loss": 4.652,
       "step": 9100
     },
     {
       "epoch": 2.582093741229301,
-      "grad_norm": 0.8317012190818787,
+      "grad_norm": 0.9719738960266113,
       "learning_rate": 5.373292584045406e-05,
-      "loss": 3.2767,
+      "loss": 4.6414,
       "step": 9200
     },
     {
       "epoch": 2.610159977547011,
-      "grad_norm": 0.8149669170379639,
+      "grad_norm": 0.9390770792961121,
       "learning_rate": 5.310921224973493e-05,
-      "loss": 3.2627,
+      "loss": 4.6219,
       "step": 9300
     },
     {
       "epoch": 2.638226213864721,
-      "grad_norm": 0.8095958828926086,
+      "grad_norm": 0.972922146320343,
       "learning_rate": 5.248549865901579e-05,
-      "loss": 3.2753,
+      "loss": 4.641,
       "step": 9400
     },
     {
       "epoch": 2.6662924501824303,
-      "grad_norm": 0.8062021136283875,
+      "grad_norm": 1.0112109184265137,
       "learning_rate": 5.1861785068296644e-05,
-      "loss": 3.2582,
+      "loss": 4.623,
       "step": 9500
     },
     {
       "epoch": 2.6662924501824303,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.6540542840957642,
-      "eval_perplexity": 5.22813325470726,
-      "eval_runtime": 30.8763,
-      "eval_samples_per_second": 73.908,
-      "eval_steps_per_second": 1.166,
+      "eval_loss": 2.3415746688842773,
+      "eval_perplexity": 10.39759645024994,
+      "eval_runtime": 32.3216,
+      "eval_samples_per_second": 70.603,
+      "eval_steps_per_second": 1.114,
       "step": 9500
     },
     {
       "epoch": 2.6943586865001405,
-      "grad_norm": 0.8398740291595459,
+      "grad_norm": 1.0064200162887573,
       "learning_rate": 5.12380714775775e-05,
-      "loss": 3.258,
+      "loss": 4.6216,
       "step": 9600
     },
     {
       "epoch": 2.72242492281785,
-      "grad_norm": 0.7982565760612488,
+      "grad_norm": 0.9441614747047424,
       "learning_rate": 5.061435788685835e-05,
-      "loss": 3.2541,
+      "loss": 4.6191,
       "step": 9700
     },
     {
       "epoch": 2.75049115913556,
-      "grad_norm": 0.8176411390304565,
+      "grad_norm": 0.9196330904960632,
       "learning_rate": 4.9990644296139214e-05,
-      "loss": 3.2556,
+      "loss": 4.619,
       "step": 9800
     },
     {
       "epoch": 2.77855739545327,
-      "grad_norm": 0.7924163937568665,
+      "grad_norm": 0.9170246720314026,
       "learning_rate": 4.9366930705420076e-05,
-      "loss": 3.2548,
+      "loss": 4.6202,
       "step": 9900
     },
     {
       "epoch": 2.8066236317709796,
-      "grad_norm": 0.8106917142868042,
+      "grad_norm": 0.915802538394928,
       "learning_rate": 4.874321711470094e-05,
-      "loss": 3.2387,
+      "loss": 4.6016,
       "step": 10000
     },
     {
       "epoch": 2.8066236317709796,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.6433695554733276,
-      "eval_perplexity": 5.172569440429115,
-      "eval_runtime": 30.8104,
-      "eval_samples_per_second": 74.066,
-      "eval_steps_per_second": 1.168,
+      "eval_loss": 2.330256223678589,
+      "eval_perplexity": 10.28057532243656,
+      "eval_runtime": 32.2844,
+      "eval_samples_per_second": 70.684,
+      "eval_steps_per_second": 1.115,
       "step": 10000
     },
     {
       "epoch": 2.8346898680886894,
-      "grad_norm": 0.8256287574768066,
+      "grad_norm": 0.9382310509681702,
       "learning_rate": 4.811950352398179e-05,
-      "loss": 3.2405,
+      "loss": 4.6025,
       "step": 10100
     },
     {
       "epoch": 2.862756104406399,
-      "grad_norm": 0.802058756351471,
+      "grad_norm": 0.9588121771812439,
       "learning_rate": 4.7495789933262646e-05,
-      "loss": 3.2469,
+      "loss": 4.6119,
       "step": 10200
     },
     {
       "epoch": 2.890822340724109,
-      "grad_norm": 0.816027820110321,
+      "grad_norm": 0.97861647605896,
       "learning_rate": 4.687207634254351e-05,
-      "loss": 3.2417,
+      "loss": 4.6048,
       "step": 10300
     },
     {
       "epoch": 2.9188885770418187,
-      "grad_norm": 0.8111074566841125,
+      "grad_norm": 0.9339620471000671,
       "learning_rate": 4.624836275182436e-05,
-      "loss": 3.2294,
+      "loss": 4.5877,
       "step": 10400
     },
     {
       "epoch": 2.9469548133595285,
-      "grad_norm": 0.8160965442657471,
+      "grad_norm": 0.9401476979255676,
       "learning_rate": 4.5624649161105224e-05,
-      "loss": 3.223,
+      "loss": 4.5823,
       "step": 10500
     },
     {
       "epoch": 2.9469548133595285,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.6337802410125732,
-      "eval_perplexity": 5.1232051088182144,
-      "eval_runtime": 30.8817,
-      "eval_samples_per_second": 73.895,
-      "eval_steps_per_second": 1.166,
+      "eval_loss": 2.3201913833618164,
+      "eval_perplexity": 10.177621947197888,
+      "eval_runtime": 32.2222,
+      "eval_samples_per_second": 70.821,
+      "eval_steps_per_second": 1.117,
       "step": 10500
     },
     {
       "epoch": 2.9750210496772382,
-      "grad_norm": 0.7960518598556519,
+      "grad_norm": 0.9543293118476868,
       "learning_rate": 4.5000935570386085e-05,
-      "loss": 3.2314,
+      "loss": 4.5929,
       "step": 10600
     },
     {
       "epoch": 3.003087285994948,
-      "grad_norm": 0.8226578235626221,
+      "grad_norm": 0.9759846925735474,
       "learning_rate": 4.437722197966694e-05,
-      "loss": 3.2255,
+      "loss": 4.5889,
       "step": 10700
     },
     {
       "epoch": 3.0311535223126578,
-      "grad_norm": 0.8036494255065918,
+      "grad_norm": 0.9409395456314087,
       "learning_rate": 4.3753508388947794e-05,
-      "loss": 3.1286,
+      "loss": 4.4832,
       "step": 10800
     },
     {
       "epoch": 3.0592197586303675,
-      "grad_norm": 0.8422260880470276,
+      "grad_norm": 0.9838731288909912,
       "learning_rate": 4.3129794798228656e-05,
-      "loss": 3.1179,
+      "loss": 4.4707,
       "step": 10900
     },
     {
       "epoch": 3.0872859949480773,
-      "grad_norm": 0.8350833058357239,
+      "grad_norm": 0.9607841372489929,
       "learning_rate": 4.250608120750951e-05,
-      "loss": 3.126,
+      "loss": 4.4802,
       "step": 11000
     },
     {
       "epoch": 3.0872859949480773,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.6289175748825073,
-      "eval_perplexity": 5.098353145228774,
-      "eval_runtime": 30.868,
-      "eval_samples_per_second": 73.928,
-      "eval_steps_per_second": 1.166,
+      "eval_loss": 2.3143389225006104,
+      "eval_perplexity": 10.118231771945648,
+      "eval_runtime": 32.1721,
+      "eval_samples_per_second": 70.931,
+      "eval_steps_per_second": 1.119,
       "step": 11000
     },
     {
       "epoch": 3.115352231265787,
-      "grad_norm": 0.8042276501655579,
+      "grad_norm": 0.9209914803504944,
       "learning_rate": 4.188236761679037e-05,
-      "loss": 3.1171,
+      "loss": 4.4671,
       "step": 11100
     },
     {
       "epoch": 3.143418467583497,
-      "grad_norm": 0.8350549340248108,
+      "grad_norm": 0.9684066772460938,
       "learning_rate": 4.125865402607123e-05,
-      "loss": 3.122,
+      "loss": 4.4784,
       "step": 11200
     },
     {
       "epoch": 3.171484703901207,
-      "grad_norm": 0.9750901460647583,
+      "grad_norm": 0.9463008642196655,
       "learning_rate": 4.063494043535209e-05,
-      "loss": 3.1236,
+      "loss": 4.4771,
       "step": 11300
     },
     {
       "epoch": 3.199550940218917,
-      "grad_norm": 0.8534584045410156,
+      "grad_norm": 0.9767867922782898,
       "learning_rate": 4.001122684463295e-05,
-      "loss": 3.1242,
+      "loss": 4.4748,
       "step": 11400
     },
     {
       "epoch": 3.2276171765366266,
-      "grad_norm": 0.8372871279716492,
+      "grad_norm": 0.9567114114761353,
       "learning_rate": 3.9387513253913804e-05,
-      "loss": 3.1149,
+      "loss": 4.4671,
       "step": 11500
     },
     {
       "epoch": 3.2276171765366266,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.6211658716201782,
-      "eval_perplexity": 5.058985006716426,
-      "eval_runtime": 30.8572,
-      "eval_samples_per_second": 73.953,
-      "eval_steps_per_second": 1.167,
+      "eval_loss": 2.307267904281616,
+      "eval_perplexity": 10.046937927830804,
+      "eval_runtime": 32.2168,
+      "eval_samples_per_second": 70.833,
+      "eval_steps_per_second": 1.117,
       "step": 11500
     },
     {
       "epoch": 3.2556834128543364,
-      "grad_norm": 0.8385348916053772,
+      "grad_norm": 0.9912461638450623,
       "learning_rate": 3.876379966319466e-05,
-      "loss": 3.1165,
+      "loss": 4.4684,
       "step": 11600
     },
     {
       "epoch": 3.283749649172046,
-      "grad_norm": 0.8371408581733704,
+      "grad_norm": 0.9675642251968384,
       "learning_rate": 3.814008607247552e-05,
-      "loss": 3.1205,
+      "loss": 4.4732,
       "step": 11700
     },
     {
       "epoch": 3.311815885489756,
-      "grad_norm": 1.0450332164764404,
+      "grad_norm": 0.9854398965835571,
       "learning_rate": 3.751637248175638e-05,
-      "loss": 3.1143,
+      "loss": 4.4635,
       "step": 11800
     },
     {
       "epoch": 3.3398821218074657,
-      "grad_norm": 0.8579763174057007,
+      "grad_norm": 1.0165531635284424,
       "learning_rate": 3.6892658891037236e-05,
-      "loss": 3.1081,
+      "loss": 4.459,
       "step": 11900
     },
     {
       "epoch": 3.3679483581251755,
-      "grad_norm": 0.8332571983337402,
+      "grad_norm": 0.9304424524307251,
       "learning_rate": 3.62689453003181e-05,
-      "loss": 3.1048,
+      "loss": 4.4557,
       "step": 12000
     },
     {
       "epoch": 3.3679483581251755,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.6149429082870483,
-      "eval_perplexity": 5.027600880928623,
-      "eval_runtime": 30.7824,
-      "eval_samples_per_second": 74.133,
-      "eval_steps_per_second": 1.169,
+      "eval_loss": 2.3005871772766113,
+      "eval_perplexity": 9.980040787876671,
+      "eval_runtime": 32.266,
+      "eval_samples_per_second": 70.725,
+      "eval_steps_per_second": 1.116,
       "step": 12000
     },
     {
       "epoch": 3.3960145944428852,
-      "grad_norm": 0.8373920321464539,
+      "grad_norm": 0.9791555404663086,
       "learning_rate": 3.564523170959895e-05,
-      "loss": 3.1223,
+      "loss": 4.476,
       "step": 12100
     },
     {
       "epoch": 3.424080830760595,
-      "grad_norm": 0.8379535675048828,
+      "grad_norm": 1.014588713645935,
       "learning_rate": 3.502151811887981e-05,
-      "loss": 3.1178,
+      "loss": 4.4673,
       "step": 12200
     },
     {
       "epoch": 3.4521470670783048,
-      "grad_norm": 0.8385716080665588,
+      "grad_norm": 0.9664494395256042,
       "learning_rate": 3.439780452816067e-05,
-      "loss": 3.1028,
+      "loss": 4.4511,
       "step": 12300
     },
     {
       "epoch": 3.4802133033960145,
-      "grad_norm": 0.8330587148666382,
+      "grad_norm": 0.9452320337295532,
       "learning_rate": 3.377409093744153e-05,
-      "loss": 3.1069,
+      "loss": 4.4579,
       "step": 12400
     },
     {
       "epoch": 3.5082795397137243,
-      "grad_norm": 0.8396536111831665,
+      "grad_norm": 0.9321849346160889,
       "learning_rate": 3.315037734672239e-05,
-      "loss": 3.0966,
+      "loss": 4.4437,
       "step": 12500
     },
     {
       "epoch": 3.5082795397137243,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.6073817014694214,
-      "eval_perplexity": 4.989729507944433,
-      "eval_runtime": 30.8556,
-      "eval_samples_per_second": 73.957,
-      "eval_steps_per_second": 1.167,
+      "eval_loss": 2.2927675247192383,
+      "eval_perplexity": 9.902304667244572,
+      "eval_runtime": 32.1615,
+      "eval_samples_per_second": 70.954,
+      "eval_steps_per_second": 1.119,
       "step": 12500
     },
     {
       "epoch": 3.536345776031434,
-      "grad_norm": 0.8449435234069824,
+      "grad_norm": 0.9828037619590759,
       "learning_rate": 3.2526663756003245e-05,
-      "loss": 3.1004,
+      "loss": 4.4508,
       "step": 12600
     },
     {
       "epoch": 3.564412012349144,
-      "grad_norm": 0.8464480042457581,
+      "grad_norm": 0.9577271342277527,
       "learning_rate": 3.19029501652841e-05,
-      "loss": 3.1037,
+      "loss": 4.4563,
       "step": 12700
     },
     {
       "epoch": 3.5924782486668536,
-      "grad_norm": 0.8337912559509277,
+      "grad_norm": 0.9698833227157593,
       "learning_rate": 3.127923657456496e-05,
-      "loss": 3.0935,
+      "loss": 4.4441,
       "step": 12800
     },
     {
       "epoch": 3.6205444849845634,
-      "grad_norm": 0.8544827103614807,
+      "grad_norm": 1.0039185285568237,
       "learning_rate": 3.065552298384582e-05,
-      "loss": 3.0967,
+      "loss": 4.4479,
       "step": 12900
     },
     {
       "epoch": 3.6486107213022736,
-      "grad_norm": 0.8739194869995117,
+      "grad_norm": 0.9685314893722534,
       "learning_rate": 3.0031809393126674e-05,
-      "loss": 3.0904,
+      "loss": 4.4402,
       "step": 13000
     },
     {
       "epoch": 3.6486107213022736,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.6013461351394653,
-      "eval_perplexity": 4.959704365047878,
-      "eval_runtime": 30.8424,
-      "eval_samples_per_second": 73.989,
-      "eval_steps_per_second": 1.167,
+      "eval_loss": 2.286196708679199,
+      "eval_perplexity": 9.837451746548071,
+      "eval_runtime": 32.1948,
+      "eval_samples_per_second": 70.881,
+      "eval_steps_per_second": 1.118,
       "step": 13000
     },
     {
       "epoch": 3.676676957619983,
-      "grad_norm": 1.3021411895751953,
+      "grad_norm": 0.9649451375007629,
       "learning_rate": 2.940809580240754e-05,
-      "loss": 3.0893,
+      "loss": 4.4369,
       "step": 13100
     },
     {
       "epoch": 3.704743193937693,
-      "grad_norm": 0.8261712193489075,
+      "grad_norm": 0.9450793266296387,
       "learning_rate": 2.8784382211688393e-05,
-      "loss": 3.0913,
+      "loss": 4.4417,
       "step": 13200
     },
     {
       "epoch": 3.732809430255403,
-      "grad_norm": 0.8515812754631042,
+      "grad_norm": 0.9554929733276367,
       "learning_rate": 2.8160668620969254e-05,
-      "loss": 3.0891,
+      "loss": 4.4376,
       "step": 13300
     },
     {
       "epoch": 3.7608756665731127,
-      "grad_norm": 0.8673535585403442,
+      "grad_norm": 0.9607460498809814,
       "learning_rate": 2.7536955030250112e-05,
-      "loss": 3.08,
+      "loss": 4.4279,
       "step": 13400
     },
     {
       "epoch": 3.7889419028908224,
-      "grad_norm": 0.8489654660224915,
+      "grad_norm": 0.9623512029647827,
       "learning_rate": 2.6913241439530967e-05,
-      "loss": 3.0979,
+      "loss": 4.4482,
       "step": 13500
     },
     {
       "epoch": 3.7889419028908224,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.595861554145813,
-      "eval_perplexity": 4.932576924077616,
-      "eval_runtime": 30.8403,
-      "eval_samples_per_second": 73.994,
-      "eval_steps_per_second": 1.167,
+      "eval_loss": 2.2799580097198486,
+      "eval_perplexity": 9.77626989259848,
+      "eval_runtime": 32.3051,
+      "eval_samples_per_second": 70.639,
+      "eval_steps_per_second": 1.114,
       "step": 13500
     },
     {
       "epoch": 3.817008139208532,
-      "grad_norm": 0.8460671901702881,
+      "grad_norm": 0.9740085601806641,
       "learning_rate": 2.628952784881183e-05,
-      "loss": 3.0687,
+      "loss": 4.4128,
       "step": 13600
     },
     {
       "epoch": 3.845074375526242,
-      "grad_norm": 0.8813645839691162,
+      "grad_norm": 0.9849666953086853,
       "learning_rate": 2.5665814258092686e-05,
-      "loss": 3.0696,
+      "loss": 4.4165,
       "step": 13700
     },
     {
       "epoch": 3.8731406118439518,
-      "grad_norm": 0.8696038126945496,
+      "grad_norm": 0.946613073348999,
       "learning_rate": 2.504210066737354e-05,
-      "loss": 3.07,
+      "loss": 4.4181,
       "step": 13800
     },
     {
       "epoch": 3.9012068481616615,
-      "grad_norm": 0.8538317680358887,
+      "grad_norm": 0.9492228031158447,
       "learning_rate": 2.44183870766544e-05,
-      "loss": 3.0845,
+      "loss": 4.4349,
       "step": 13900
     },
     {
       "epoch": 3.9292730844793713,
-      "grad_norm": 0.8570790886878967,
+      "grad_norm": 0.97336345911026,
       "learning_rate": 2.379467348593526e-05,
-      "loss": 3.0772,
+      "loss": 4.4279,
       "step": 14000
     },
     {
       "epoch": 3.9292730844793713,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.5907403230667114,
-      "eval_perplexity": 4.907380630922354,
-      "eval_runtime": 30.8034,
-      "eval_samples_per_second": 74.083,
-      "eval_steps_per_second": 1.169,
+      "eval_loss": 2.2752432823181152,
+      "eval_perplexity": 9.730285931150744,
+      "eval_runtime": 32.3352,
+      "eval_samples_per_second": 70.573,
+      "eval_steps_per_second": 1.113,
       "step": 14000
     },
     {
       "epoch": 3.957339320797081,
-      "grad_norm": 0.8513247966766357,
+      "grad_norm": 0.950677752494812,
       "learning_rate": 2.317095989521612e-05,
-      "loss": 3.0714,
+      "loss": 4.4203,
       "step": 14100
     },
     {
       "epoch": 3.985405557114791,
-      "grad_norm": 0.8460662961006165,
+      "grad_norm": 1.0031086206436157,
       "learning_rate": 2.2547246304496976e-05,
-      "loss": 3.0662,
+      "loss": 4.4137,
       "step": 14200
     },
     {
       "epoch": 4.013471793432501,
-      "grad_norm": 0.8479135632514954,
+      "grad_norm": 0.9779632687568665,
       "learning_rate": 2.1923532713777834e-05,
-      "loss": 3.0293,
+      "loss": 4.3757,
       "step": 14300
     },
     {
       "epoch": 4.04153802975021,
-      "grad_norm": 0.8859968185424805,
+      "grad_norm": 0.9976330399513245,
       "learning_rate": 2.1299819123058692e-05,
-      "loss": 2.9782,
+      "loss": 4.3132,
       "step": 14400
     },
     {
       "epoch": 4.069604266067921,
-      "grad_norm": 0.8683611750602722,
+      "grad_norm": 1.0129767656326294,
       "learning_rate": 2.067610553233955e-05,
-      "loss": 2.9795,
+      "loss": 4.3188,
       "step": 14500
     },
     {
       "epoch": 4.069604266067921,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.5887154340744019,
-      "eval_perplexity": 4.897453783675403,
-      "eval_runtime": 30.7941,
-      "eval_samples_per_second": 74.105,
-      "eval_steps_per_second": 1.169,
+      "eval_loss": 2.273023843765259,
+      "eval_perplexity": 9.708714106949042,
+      "eval_runtime": 32.217,
+      "eval_samples_per_second": 70.832,
+      "eval_steps_per_second": 1.117,
       "step": 14500
     },
     {
       "epoch": 4.09767050238563,
-      "grad_norm": 0.8947647213935852,
+      "grad_norm": 0.9901084899902344,
       "learning_rate": 2.005239194162041e-05,
-      "loss": 2.9837,
+      "loss": 4.3242,
       "step": 14600
     },
     {
       "epoch": 4.12573673870334,
-      "grad_norm": 0.8813712000846863,
+      "grad_norm": 1.0149174928665161,
       "learning_rate": 1.9428678350901266e-05,
-      "loss": 2.9733,
+      "loss": 4.3121,
       "step": 14700
     },
     {
       "epoch": 4.1538029750210494,
-      "grad_norm": 0.8821486234664917,
+      "grad_norm": 0.9936476945877075,
       "learning_rate": 1.8804964760182124e-05,
-      "loss": 2.9878,
+      "loss": 4.3303,
       "step": 14800
     },
     {
       "epoch": 4.18186921133876,
-      "grad_norm": 0.9034985899925232,
+      "grad_norm": 0.9973461627960205,
       "learning_rate": 1.8181251169462986e-05,
-      "loss": 2.9819,
+      "loss": 4.3193,
       "step": 14900
     },
     {
       "epoch": 4.209935447656469,
-      "grad_norm": 0.9136458039283752,
+      "grad_norm": 0.9823731780052185,
       "learning_rate": 1.755753757874384e-05,
-      "loss": 2.9807,
+      "loss": 4.3193,
       "step": 15000
     },
     {
       "epoch": 4.209935447656469,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.5847208499908447,
-      "eval_perplexity": 4.877929514370786,
-      "eval_runtime": 30.8617,
-      "eval_samples_per_second": 73.943,
-      "eval_steps_per_second": 1.166,
+      "eval_loss": 2.2690694332122803,
+      "eval_perplexity": 9.67039767481112,
+      "eval_runtime": 32.165,
+      "eval_samples_per_second": 70.947,
+      "eval_steps_per_second": 1.119,
       "step": 15000
     },
     {
       "epoch": 4.238001683974179,
-      "grad_norm": 0.900557816028595,
+      "grad_norm": 1.0172957181930542,
       "learning_rate": 1.69338239880247e-05,
-      "loss": 2.9784,
+      "loss": 4.3189,
       "step": 15100
     },
     {
       "epoch": 4.2660679202918885,
-      "grad_norm": 0.8973119258880615,
+      "grad_norm": 0.983648955821991,
       "learning_rate": 1.631011039730556e-05,
-      "loss": 2.9765,
+      "loss": 4.3141,
       "step": 15200
     },
     {
       "epoch": 4.294134156609599,
-      "grad_norm": 0.913901686668396,
+      "grad_norm": 0.9884918928146362,
       "learning_rate": 1.5686396806586418e-05,
-      "loss": 2.9763,
+      "loss": 4.3171,
       "step": 15300
     },
     {
       "epoch": 4.322200392927308,
-      "grad_norm": 0.8976321816444397,
+      "grad_norm": 0.9933040738105774,
       "learning_rate": 1.5062683215867274e-05,
-      "loss": 2.9718,
+      "loss": 4.3079,
       "step": 15400
     },
     {
       "epoch": 4.350266629245018,
-      "grad_norm": 0.904545247554779,
+      "grad_norm": 1.0270668268203735,
       "learning_rate": 1.4438969625148132e-05,
-      "loss": 2.976,
+      "loss": 4.3158,
       "step": 15500
     },
     {
       "epoch": 4.350266629245018,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.5817325115203857,
-      "eval_perplexity": 4.86337436862944,
-      "eval_runtime": 30.8917,
-      "eval_samples_per_second": 73.871,
-      "eval_steps_per_second": 1.165,
+      "eval_loss": 2.265188694000244,
+      "eval_perplexity": 9.632942108009049,
+      "eval_runtime": 32.2172,
+      "eval_samples_per_second": 70.832,
+      "eval_steps_per_second": 1.117,
       "step": 15500
     },
     {
       "epoch": 4.378332865562728,
-      "grad_norm": 0.9036585092544556,
+      "grad_norm": 1.007458209991455,
       "learning_rate": 1.3815256034428992e-05,
-      "loss": 2.9729,
+      "loss": 4.3108,
       "step": 15600
     },
     {
       "epoch": 4.406399101880438,
-      "grad_norm": 0.8927172422409058,
+      "grad_norm": 0.9828687310218811,
       "learning_rate": 1.3191542443709848e-05,
-      "loss": 2.9622,
+      "loss": 4.2975,
       "step": 15700
     },
     {
       "epoch": 4.434465338198148,
-      "grad_norm": 0.8881782293319702,
+      "grad_norm": 0.9771233201026917,
       "learning_rate": 1.2567828852990706e-05,
-      "loss": 2.974,
+      "loss": 4.3164,
       "step": 15800
     },
     {
       "epoch": 4.462531574515857,
-      "grad_norm": 0.8996316194534302,
+      "grad_norm": 1.0265594720840454,
       "learning_rate": 1.1944115262271566e-05,
-      "loss": 2.9673,
+      "loss": 4.3054,
       "step": 15900
     },
     {
       "epoch": 4.490597810833568,
-      "grad_norm": 0.9018428921699524,
+      "grad_norm": 1.0030221939086914,
       "learning_rate": 1.1320401671552424e-05,
-      "loss": 2.9795,
+      "loss": 4.3196,
       "step": 16000
     },
     {
       "epoch": 4.490597810833568,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.5778179168701172,
-      "eval_perplexity": 4.844373444066827,
-      "eval_runtime": 30.8539,
-      "eval_samples_per_second": 73.961,
-      "eval_steps_per_second": 1.167,
+      "eval_loss": 2.2618846893310547,
+      "eval_perplexity": 9.601167343197428,
+      "eval_runtime": 32.2414,
+      "eval_samples_per_second": 70.779,
+      "eval_steps_per_second": 1.117,
       "step": 16000
     },
     {
       "epoch": 4.518664047151277,
-      "grad_norm": 0.8891787528991699,
+      "grad_norm": 1.0003290176391602,
       "learning_rate": 1.0696688080833282e-05,
-      "loss": 2.9647,
+      "loss": 4.3014,
       "step": 16100
     },
     {
       "epoch": 4.546730283468987,
-      "grad_norm": 0.9130581021308899,
+      "grad_norm": 1.0033382177352905,
       "learning_rate": 1.007297449011414e-05,
-      "loss": 2.9789,
+      "loss": 4.3222,
       "step": 16200
     },
     {
       "epoch": 4.574796519786696,
-      "grad_norm": 0.8879275918006897,
+      "grad_norm": 0.9857285022735596,
       "learning_rate": 9.449260899395e-06,
-      "loss": 2.9773,
+      "loss": 4.317,
       "step": 16300
     },
     {
       "epoch": 4.602862756104407,
-      "grad_norm": 0.9074363112449646,
+      "grad_norm": 1.0158944129943848,
       "learning_rate": 8.825547308675856e-06,
-      "loss": 2.9711,
+      "loss": 4.3128,
       "step": 16400
     },
     {
       "epoch": 4.630928992422116,
-      "grad_norm": 0.8947946429252625,
+      "grad_norm": 1.0046790838241577,
       "learning_rate": 8.201833717956715e-06,
-      "loss": 2.9594,
+      "loss": 4.2946,
       "step": 16500
     },
     {
       "epoch": 4.630928992422116,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.5754330158233643,
-      "eval_perplexity": 4.832833858623437,
-      "eval_runtime": 30.8924,
-      "eval_samples_per_second": 73.869,
-      "eval_steps_per_second": 1.165,
+      "eval_loss": 2.2588629722595215,
+      "eval_perplexity": 9.572199120857682,
+      "eval_runtime": 32.1819,
+      "eval_samples_per_second": 70.909,
+      "eval_steps_per_second": 1.119,
       "step": 16500
     },
     {
       "epoch": 4.658995228739826,
-      "grad_norm": 0.9059449434280396,
+      "grad_norm": 1.0288445949554443,
       "learning_rate": 7.578120127237573e-06,
-      "loss": 2.9653,
+      "loss": 4.3035,
       "step": 16600
     },
     {
       "epoch": 4.6870614650575355,
-      "grad_norm": 0.9055649638175964,
+      "grad_norm": 1.042069673538208,
       "learning_rate": 6.954406536518431e-06,
-      "loss": 2.9653,
+      "loss": 4.306,
       "step": 16700
     },
     {
       "epoch": 4.715127701375246,
-      "grad_norm": 0.9104346036911011,
+      "grad_norm": 1.0207899808883667,
       "learning_rate": 6.3306929457992894e-06,
-      "loss": 2.9689,
+      "loss": 4.3115,
       "step": 16800
     },
     {
       "epoch": 4.743193937692955,
-      "grad_norm": 0.8993477821350098,
+      "grad_norm": 0.9933531284332275,
       "learning_rate": 5.7069793550801474e-06,
-      "loss": 2.9637,
+      "loss": 4.3009,
       "step": 16900
     },
     {
       "epoch": 4.771260174010665,
-      "grad_norm": 0.9155908226966858,
+      "grad_norm": 1.0096008777618408,
       "learning_rate": 5.083265764361006e-06,
-      "loss": 2.9698,
+      "loss": 4.3078,
       "step": 17000
     },
     {
       "epoch": 4.771260174010665,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.5731316804885864,
-      "eval_perplexity": 4.821724675178664,
-      "eval_runtime": 30.8071,
-      "eval_samples_per_second": 74.074,
-      "eval_steps_per_second": 1.169,
+      "eval_loss": 2.2564101219177246,
+      "eval_perplexity": 9.54874872089052,
+      "eval_runtime": 32.2438,
+      "eval_samples_per_second": 70.773,
+      "eval_steps_per_second": 1.116,
       "step": 17000
     },
     {
       "epoch": 4.799326410328375,
-      "grad_norm": 0.9122122526168823,
+      "grad_norm": 1.0053852796554565,
       "learning_rate": 4.4595521736418634e-06,
-      "loss": 2.9619,
+      "loss": 4.3014,
       "step": 17100
     },
     {
       "epoch": 4.827392646646085,
-      "grad_norm": 0.9053720831871033,
+      "grad_norm": 1.0015292167663574,
       "learning_rate": 3.8358385829227214e-06,
-      "loss": 2.9557,
+      "loss": 4.2933,
       "step": 17200
     },
     {
       "epoch": 4.855458882963795,
-      "grad_norm": 0.9079867601394653,
+      "grad_norm": 1.004568099975586,
       "learning_rate": 3.2121249922035803e-06,
-      "loss": 2.9681,
+      "loss": 4.308,
       "step": 17300
     },
     {
       "epoch": 4.883525119281504,
-      "grad_norm": 0.911987841129303,
+      "grad_norm": 1.0016422271728516,
       "learning_rate": 2.5884114014844383e-06,
-      "loss": 2.9475,
+      "loss": 4.2804,
       "step": 17400
     },
     {
       "epoch": 4.911591355599214,
-      "grad_norm": 0.9014911651611328,
+      "grad_norm": 1.0021764039993286,
       "learning_rate": 1.9646978107652967e-06,
-      "loss": 2.9604,
+      "loss": 4.2974,
       "step": 17500
     },
     {
       "epoch": 4.911591355599214,
       "eval_accuracy": 0.00733640796431187,
       "eval_bleu": 1.0,
-      "eval_loss": 1.5712791681289673,
-      "eval_perplexity": 4.8128006391188904,
-      "eval_runtime": 30.8544,
-      "eval_samples_per_second": 73.96,
-      "eval_steps_per_second": 1.167,
+      "eval_loss": 2.2545549869537354,
+      "eval_perplexity": 9.531050924252332,
+      "eval_runtime": 32.1894,
+      "eval_samples_per_second": 70.893,
+      "eval_steps_per_second": 1.118,
       "step": 17500
     },
     {
       "epoch": 4.939657591916924,
-      "grad_norm": 0.9085986614227295,
+      "grad_norm": 0.9952707886695862,
       "learning_rate": 1.340984220046155e-06,
-      "loss": 2.9556,
+      "loss": 4.2967,
       "step": 17600
     },
     {
       "epoch": 4.967723828234634,
-      "grad_norm": 0.8967222571372986,
+      "grad_norm": 1.0102629661560059,
       "learning_rate": 7.172706293270131e-07,
-      "loss": 2.972,
+      "loss": 4.3179,
       "step": 17700
     },
     {
       "epoch": 4.995790064552343,
-      "grad_norm": 0.8958012461662292,
+      "grad_norm": 0.9876748919487,
       "learning_rate": 9.355703860787128e-08,
-      "loss": 2.9616,
+      "loss": 4.3018,
       "step": 17800
     },
     {
       "epoch": 5.0,
       "step": 17815,
       "total_flos": 2.1154254410092954e+18,
-      "train_loss": 3.7582801998715505,
-      "train_runtime": 52733.3783,
-      "train_samples_per_second": 21.619,
-      "train_steps_per_second": 0.338
+      "train_loss": 5.041201007510048,
+      "train_runtime": 53587.774,
+      "train_samples_per_second": 21.274,
+      "train_steps_per_second": 0.332
     }
   ],
   "logging_steps": 100,

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:06b88e540f475a3df78ab599583a624266c4e626b709d17699358ae255edf5c7
+oid sha256:7ff3a68d59702567c595400f51ab343dabb05938e1473a95a4984ee25c0af6a5
 size 5560