dhritic99/vit-base-brain-alzheimer-detection

README.md

@@ -17,8 +17,8 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model is a fine-tuned version of [google/vit-base-patch16-224-in21k](https://huggingface.co/google/vit-base-patch16-224-in21k) on an unknown dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.2156
-- Accuracy: 0.9688
+- Loss: 0.3405
+- Accuracy: 0.9484
 
 ## Model description
 

all_results.json

@@ -1,13 +1,13 @@
 {
-    "epoch": 20.0,
-    "eval_accuracy": 0.95234375,
-    "eval_loss": 0.23598822951316833,
-    "eval_runtime": 6.8172,
-    "eval_samples_per_second": 187.761,
-    "eval_steps_per_second": 23.47,
-    "total_flos": 6.348257582037074e+18,
-    "train_loss": 0.15641047188546509,
-    "train_runtime": 1165.0601,
-    "train_samples_per_second": 70.314,
-    "train_steps_per_second": 4.395
+    "epoch": 13.671875,
+    "eval_accuracy": 0.9484375,
+    "eval_loss": 0.3405180871486664,
+    "eval_runtime": 7.1229,
+    "eval_samples_per_second": 179.702,
+    "eval_steps_per_second": 22.463,
+    "total_flos": 4.339629206470656e+18,
+    "train_loss": 0.02788944040451731,
+    "train_runtime": 959.4355,
+    "train_samples_per_second": 85.384,
+    "train_steps_per_second": 5.336
 }

eval_results.json

@@ -1,8 +1,8 @@
 {
-    "epoch": 20.0,
-    "eval_accuracy": 0.95234375,
-    "eval_loss": 0.23598822951316833,
-    "eval_runtime": 6.8172,
-    "eval_samples_per_second": 187.761,
-    "eval_steps_per_second": 23.47
+    "epoch": 13.671875,
+    "eval_accuracy": 0.9484375,
+    "eval_loss": 0.3405180871486664,
+    "eval_runtime": 7.1229,
+    "eval_samples_per_second": 179.702,
+    "eval_steps_per_second": 22.463
 }

train_results.json

@@ -1,8 +1,8 @@
 {
-    "epoch": 20.0,
-    "total_flos": 6.348257582037074e+18,
-    "train_loss": 0.15641047188546509,
-    "train_runtime": 1165.0601,
-    "train_samples_per_second": 70.314,
-    "train_steps_per_second": 4.395
+    "epoch": 13.671875,
+    "total_flos": 4.339629206470656e+18,
+    "train_loss": 0.02788944040451731,
+    "train_runtime": 959.4355,
+    "train_samples_per_second": 85.384,
+    "train_steps_per_second": 5.336
 }

trainer_state.json

@@ -1,825 +1,574 @@
 {
-  "best_metric": 0.15439653396606445,
-  "best_model_checkpoint": "./vit-base-brain-alzheimer-detection/checkpoint-4000",
-  "epoch": 20.0,
+  "best_metric": 0.20142494142055511,
+  "best_model_checkpoint": "./vit-base-brain-alzheimer-detection/checkpoint-2000",
+  "epoch": 13.671875,
   "eval_steps": 500,
-  "global_step": 5120,
+  "global_step": 3500,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
       "epoch": 0.1953125,
-      "grad_norm": 3.019329786300659,
+      "grad_norm": 0.01695910282433033,
       "learning_rate": 2.0000000000000003e-06,
-      "loss": 1.3778,
+      "loss": 0.0035,
       "step": 50
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 2.534724712371826,
+      "grad_norm": 0.013198412954807281,
       "learning_rate": 4.000000000000001e-06,
-      "loss": 1.1895,
+      "loss": 0.0031,
       "step": 100
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 2.774301052093506,
+      "grad_norm": 0.01613868959248066,
       "learning_rate": 6e-06,
-      "loss": 1.0423,
+      "loss": 0.0031,
       "step": 150
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 3.057741165161133,
+      "grad_norm": 0.011624956503510475,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.9859,
+      "loss": 0.0031,
       "step": 200
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 2.8204987049102783,
+      "grad_norm": 0.026181865483522415,
       "learning_rate": 1e-05,
-      "loss": 0.9008,
+      "loss": 0.0026,
       "step": 250
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 1.7199639081954956,
+      "grad_norm": 0.010051410645246506,
       "learning_rate": 1.2e-05,
-      "loss": 0.8892,
+      "loss": 0.0026,
       "step": 300
     },
     {
       "epoch": 1.3671875,
-      "grad_norm": 3.138557195663452,
+      "grad_norm": 0.009718737564980984,
       "learning_rate": 1.4e-05,
-      "loss": 0.8273,
+      "loss": 0.0022,
       "step": 350
     },
     {
       "epoch": 1.5625,
-      "grad_norm": 3.1744067668914795,
+      "grad_norm": 0.00813582818955183,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.7807,
+      "loss": 0.0019,
       "step": 400
     },
     {
       "epoch": 1.7578125,
-      "grad_norm": 5.47639799118042,
+      "grad_norm": 14.344162940979004,
       "learning_rate": 1.8e-05,
-      "loss": 0.7546,
+      "loss": 0.3794,
       "step": 450
     },
     {
       "epoch": 1.953125,
-      "grad_norm": 2.743036985397339,
-      "learning_rate": 2e-05,
-      "loss": 0.7278,
+      "grad_norm": 27.69124984741211,
+      "learning_rate": 1.9960000000000002e-05,
+      "loss": 0.4213,
       "step": 500
     },
     {
       "epoch": 1.953125,
-      "eval_accuracy": 0.705078125,
-      "eval_loss": 0.7124308347702026,
-      "eval_runtime": 5.7619,
-      "eval_samples_per_second": 177.72,
-      "eval_steps_per_second": 22.215,
+      "eval_accuracy": 0.8955078125,
+      "eval_loss": 0.3758355975151062,
+      "eval_runtime": 7.1938,
+      "eval_samples_per_second": 142.346,
+      "eval_steps_per_second": 17.793,
       "step": 500
     },
     {
       "epoch": 2.1484375,
-      "grad_norm": 12.242274284362793,
-      "learning_rate": 1.9783549783549785e-05,
-      "loss": 0.6108,
+      "grad_norm": 0.10575103759765625,
+      "learning_rate": 1.978787878787879e-05,
+      "loss": 0.219,
       "step": 550
     },
     {
       "epoch": 2.34375,
-      "grad_norm": 9.961933135986328,
-      "learning_rate": 1.9567099567099568e-05,
-      "loss": 0.5815,
+      "grad_norm": 10.299068450927734,
+      "learning_rate": 1.9571428571428572e-05,
+      "loss": 0.2295,
       "step": 600
     },
     {
       "epoch": 2.5390625,
-      "grad_norm": 8.35198974609375,
-      "learning_rate": 1.9350649350649354e-05,
-      "loss": 0.5702,
+      "grad_norm": 30.670997619628906,
+      "learning_rate": 1.935497835497836e-05,
+      "loss": 0.0989,
       "step": 650
     },
     {
       "epoch": 2.734375,
-      "grad_norm": 4.211214542388916,
-      "learning_rate": 1.9134199134199138e-05,
-      "loss": 0.4907,
+      "grad_norm": 10.104991912841797,
+      "learning_rate": 1.9138528138528142e-05,
+      "loss": 0.0813,
       "step": 700
     },
     {
       "epoch": 2.9296875,
-      "grad_norm": 5.052712917327881,
-      "learning_rate": 1.891774891774892e-05,
-      "loss": 0.4993,
+      "grad_norm": 0.01667148247361183,
+      "learning_rate": 1.8922077922077925e-05,
+      "loss": 0.0195,
       "step": 750
     },
     {
       "epoch": 3.125,
-      "grad_norm": 2.4637234210968018,
-      "learning_rate": 1.8701298701298704e-05,
-      "loss": 0.3788,
+      "grad_norm": 4.969385147094727,
+      "learning_rate": 1.8705627705627708e-05,
+      "loss": 0.02,
       "step": 800
     },
     {
       "epoch": 3.3203125,
-      "grad_norm": 5.4455790519714355,
-      "learning_rate": 1.8484848484848487e-05,
-      "loss": 0.358,
+      "grad_norm": 0.010836214758455753,
+      "learning_rate": 1.848917748917749e-05,
+      "loss": 0.0306,
       "step": 850
     },
     {
       "epoch": 3.515625,
-      "grad_norm": 13.753146171569824,
-      "learning_rate": 1.826839826839827e-05,
-      "loss": 0.273,
+      "grad_norm": 0.009891550056636333,
+      "learning_rate": 1.8272727272727275e-05,
+      "loss": 0.0433,
       "step": 900
     },
     {
       "epoch": 3.7109375,
-      "grad_norm": 7.315534591674805,
-      "learning_rate": 1.8051948051948053e-05,
-      "loss": 0.2588,
+      "grad_norm": 0.08000622689723969,
+      "learning_rate": 1.8056277056277058e-05,
+      "loss": 0.0294,
       "step": 950
     },
     {
       "epoch": 3.90625,
-      "grad_norm": 15.05672836303711,
-      "learning_rate": 1.7835497835497836e-05,
-      "loss": 0.3023,
+      "grad_norm": 0.015158270485699177,
+      "learning_rate": 1.783982683982684e-05,
+      "loss": 0.0224,
       "step": 1000
     },
     {
       "epoch": 3.90625,
-      "eval_accuracy": 0.8828125,
-      "eval_loss": 0.3776271939277649,
-      "eval_runtime": 6.9418,
-      "eval_samples_per_second": 147.512,
-      "eval_steps_per_second": 18.439,
+      "eval_accuracy": 0.9501953125,
+      "eval_loss": 0.2260357141494751,
+      "eval_runtime": 5.6203,
+      "eval_samples_per_second": 182.196,
+      "eval_steps_per_second": 22.774,
       "step": 1000
     },
     {
       "epoch": 4.1015625,
-      "grad_norm": 0.9234067797660828,
-      "learning_rate": 1.761904761904762e-05,
-      "loss": 0.2126,
+      "grad_norm": 0.011108304373919964,
+      "learning_rate": 1.7623376623376624e-05,
+      "loss": 0.0121,
       "step": 1050
     },
     {
       "epoch": 4.296875,
-      "grad_norm": 4.910510540008545,
-      "learning_rate": 1.7402597402597403e-05,
-      "loss": 0.1539,
+      "grad_norm": 0.014846593141555786,
+      "learning_rate": 1.7406926406926407e-05,
+      "loss": 0.0205,
       "step": 1100
     },
     {
       "epoch": 4.4921875,
-      "grad_norm": 6.401501655578613,
-      "learning_rate": 1.718614718614719e-05,
-      "loss": 0.2101,
+      "grad_norm": 0.032500285655260086,
+      "learning_rate": 1.7190476190476194e-05,
+      "loss": 0.0261,
       "step": 1150
     },
     {
       "epoch": 4.6875,
-      "grad_norm": 1.9736899137496948,
-      "learning_rate": 1.6969696969696972e-05,
-      "loss": 0.1984,
+      "grad_norm": 0.01257646270096302,
+      "learning_rate": 1.6974025974025977e-05,
+      "loss": 0.0252,
       "step": 1200
     },
     {
       "epoch": 4.8828125,
-      "grad_norm": 8.743949890136719,
-      "learning_rate": 1.6753246753246756e-05,
-      "loss": 0.1249,
+      "grad_norm": 0.06579085439443588,
+      "learning_rate": 1.675757575757576e-05,
+      "loss": 0.024,
       "step": 1250
     },
     {
       "epoch": 5.078125,
-      "grad_norm": 14.517168998718262,
-      "learning_rate": 1.653679653679654e-05,
-      "loss": 0.1279,
+      "grad_norm": 2.4693167209625244,
+      "learning_rate": 1.6541125541125543e-05,
+      "loss": 0.0324,
       "step": 1300
     },
     {
       "epoch": 5.2734375,
-      "grad_norm": 0.22373303771018982,
-      "learning_rate": 1.6320346320346322e-05,
-      "loss": 0.0986,
+      "grad_norm": 0.005182833410799503,
+      "learning_rate": 1.6324675324675326e-05,
+      "loss": 0.018,
       "step": 1350
     },
     {
       "epoch": 5.46875,
-      "grad_norm": 0.18486973643302917,
-      "learning_rate": 1.6103896103896105e-05,
-      "loss": 0.0922,
+      "grad_norm": 0.006213045679032803,
+      "learning_rate": 1.610822510822511e-05,
+      "loss": 0.0479,
       "step": 1400
     },
     {
       "epoch": 5.6640625,
-      "grad_norm": 0.4893491268157959,
-      "learning_rate": 1.5887445887445888e-05,
-      "loss": 0.0666,
+      "grad_norm": 50.79215621948242,
+      "learning_rate": 1.5891774891774893e-05,
+      "loss": 0.0251,
       "step": 1450
     },
     {
       "epoch": 5.859375,
-      "grad_norm": 17.192604064941406,
-      "learning_rate": 1.567099567099567e-05,
-      "loss": 0.0997,
+      "grad_norm": 0.021290121600031853,
+      "learning_rate": 1.5675324675324676e-05,
+      "loss": 0.0012,
       "step": 1500
     },
     {
       "epoch": 5.859375,
-      "eval_accuracy": 0.9130859375,
-      "eval_loss": 0.2808457016944885,
-      "eval_runtime": 5.8114,
-      "eval_samples_per_second": 176.206,
-      "eval_steps_per_second": 22.026,
+      "eval_accuracy": 0.95703125,
+      "eval_loss": 0.21266639232635498,
+      "eval_runtime": 5.9949,
+      "eval_samples_per_second": 170.813,
+      "eval_steps_per_second": 21.352,
       "step": 1500
     },
     {
       "epoch": 6.0546875,
-      "grad_norm": 0.13603505492210388,
-      "learning_rate": 1.5454545454545454e-05,
-      "loss": 0.0481,
+      "grad_norm": 12.610783576965332,
+      "learning_rate": 1.545887445887446e-05,
+      "loss": 0.0256,
       "step": 1550
     },
     {
       "epoch": 6.25,
-      "grad_norm": 44.868900299072266,
-      "learning_rate": 1.523809523809524e-05,
-      "loss": 0.0465,
+      "grad_norm": 0.013099271804094315,
+      "learning_rate": 1.5242424242424244e-05,
+      "loss": 0.0143,
       "step": 1600
     },
     {
       "epoch": 6.4453125,
-      "grad_norm": 3.908478021621704,
-      "learning_rate": 1.5021645021645024e-05,
-      "loss": 0.0485,
+      "grad_norm": 0.04105505719780922,
+      "learning_rate": 1.5025974025974027e-05,
+      "loss": 0.0109,
       "step": 1650
     },
     {
       "epoch": 6.640625,
-      "grad_norm": 0.24283438920974731,
-      "learning_rate": 1.4805194805194807e-05,
-      "loss": 0.0387,
+      "grad_norm": 0.005291896406561136,
+      "learning_rate": 1.4809523809523812e-05,
+      "loss": 0.0222,
       "step": 1700
     },
     {
       "epoch": 6.8359375,
-      "grad_norm": 21.719118118286133,
-      "learning_rate": 1.458874458874459e-05,
-      "loss": 0.0803,
+      "grad_norm": 0.005415532272309065,
+      "learning_rate": 1.4593073593073595e-05,
+      "loss": 0.0052,
       "step": 1750
     },
     {
       "epoch": 7.03125,
-      "grad_norm": 0.08735662698745728,
-      "learning_rate": 1.4372294372294374e-05,
-      "loss": 0.0266,
+      "grad_norm": 0.004291017074137926,
+      "learning_rate": 1.4376623376623378e-05,
+      "loss": 0.0077,
       "step": 1800
     },
     {
       "epoch": 7.2265625,
-      "grad_norm": 0.16133151948451996,
-      "learning_rate": 1.4155844155844157e-05,
-      "loss": 0.0252,
+      "grad_norm": 0.004927146248519421,
+      "learning_rate": 1.4160173160173161e-05,
+      "loss": 0.0009,
       "step": 1850
     },
     {
       "epoch": 7.421875,
-      "grad_norm": 0.12098684161901474,
-      "learning_rate": 1.3939393939393942e-05,
-      "loss": 0.0174,
+      "grad_norm": 0.004201130475848913,
+      "learning_rate": 1.3943722943722944e-05,
+      "loss": 0.0008,
       "step": 1900
     },
     {
       "epoch": 7.6171875,
-      "grad_norm": 0.14388597011566162,
-      "learning_rate": 1.3722943722943725e-05,
-      "loss": 0.016,
+      "grad_norm": 0.003896114882081747,
+      "learning_rate": 1.3727272727272729e-05,
+      "loss": 0.0007,
       "step": 1950
     },
     {
       "epoch": 7.8125,
-      "grad_norm": 0.07232346385717392,
-      "learning_rate": 1.3506493506493508e-05,
-      "loss": 0.0424,
+      "grad_norm": 0.0028053205460309982,
+      "learning_rate": 1.3510822510822512e-05,
+      "loss": 0.0007,
       "step": 2000
     },
     {
       "epoch": 7.8125,
-      "eval_accuracy": 0.95703125,
-      "eval_loss": 0.1914406716823578,
-      "eval_runtime": 5.2231,
-      "eval_samples_per_second": 196.054,
-      "eval_steps_per_second": 24.507,
+      "eval_accuracy": 0.9677734375,
+      "eval_loss": 0.20142494142055511,
+      "eval_runtime": 5.4298,
+      "eval_samples_per_second": 188.588,
+      "eval_steps_per_second": 23.573,
       "step": 2000
     },
     {
       "epoch": 8.0078125,
-      "grad_norm": 0.06997977197170258,
+      "grad_norm": 0.003189160255715251,
       "learning_rate": 1.3294372294372295e-05,
-      "loss": 0.031,
+      "loss": 0.0006,
       "step": 2050
     },
     {
       "epoch": 8.203125,
-      "grad_norm": 0.6767904162406921,
+      "grad_norm": 0.005525187123566866,
       "learning_rate": 1.3077922077922078e-05,
-      "loss": 0.016,
+      "loss": 0.0006,
       "step": 2100
     },
     {
       "epoch": 8.3984375,
-      "grad_norm": 0.055935341864824295,
+      "grad_norm": 0.002676301635801792,
       "learning_rate": 1.2861471861471862e-05,
-      "loss": 0.0128,
+      "loss": 0.0006,
       "step": 2150
     },
     {
       "epoch": 8.59375,
-      "grad_norm": 0.05341877415776253,
+      "grad_norm": 0.0033032067585736513,
       "learning_rate": 1.2645021645021646e-05,
-      "loss": 0.0123,
+      "loss": 0.0006,
       "step": 2200
     },
     {
       "epoch": 8.7890625,
-      "grad_norm": 0.051278505474328995,
+      "grad_norm": 0.003413607832044363,
       "learning_rate": 1.242857142857143e-05,
-      "loss": 0.0243,
+      "loss": 0.0006,
       "step": 2250
     },
     {
       "epoch": 8.984375,
-      "grad_norm": 0.04905688017606735,
+      "grad_norm": 0.00263944361358881,
       "learning_rate": 1.2212121212121213e-05,
-      "loss": 0.0116,
+      "loss": 0.0006,
       "step": 2300
     },
     {
       "epoch": 9.1796875,
-      "grad_norm": 0.05154794827103615,
+      "grad_norm": 0.002298458479344845,
       "learning_rate": 1.1995670995670996e-05,
-      "loss": 0.0104,
+      "loss": 0.0006,
       "step": 2350
     },
     {
       "epoch": 9.375,
-      "grad_norm": 0.04436648637056351,
+      "grad_norm": 0.002362418919801712,
       "learning_rate": 1.1779220779220779e-05,
-      "loss": 0.0096,
+      "loss": 0.0005,
       "step": 2400
     },
     {
       "epoch": 9.5703125,
-      "grad_norm": 0.04592837765812874,
+      "grad_norm": 0.002904605120420456,
       "learning_rate": 1.1562770562770564e-05,
-      "loss": 0.0096,
+      "loss": 0.0005,
       "step": 2450
     },
     {
       "epoch": 9.765625,
-      "grad_norm": 0.5957643985748291,
+      "grad_norm": 0.004212340340018272,
       "learning_rate": 1.1346320346320347e-05,
-      "loss": 0.0108,
+      "loss": 0.0005,
       "step": 2500
     },
     {
       "epoch": 9.765625,
-      "eval_accuracy": 0.89453125,
-      "eval_loss": 0.4534318447113037,
-      "eval_runtime": 5.8562,
-      "eval_samples_per_second": 174.857,
-      "eval_steps_per_second": 21.857,
+      "eval_accuracy": 0.9697265625,
+      "eval_loss": 0.20150402188301086,
+      "eval_runtime": 10.8662,
+      "eval_samples_per_second": 94.237,
+      "eval_steps_per_second": 11.78,
       "step": 2500
     },
     {
       "epoch": 9.9609375,
-      "grad_norm": 0.3423336446285248,
+      "grad_norm": 0.0020910820458084345,
       "learning_rate": 1.112987012987013e-05,
-      "loss": 0.0183,
+      "loss": 0.0005,
       "step": 2550
     },
     {
       "epoch": 10.15625,
-      "grad_norm": 0.04016263037919998,
+      "grad_norm": 0.0021248324774205685,
       "learning_rate": 1.0913419913419913e-05,
-      "loss": 0.016,
+      "loss": 0.0005,
       "step": 2600
     },
     {
       "epoch": 10.3515625,
-      "grad_norm": 0.04166042432188988,
+      "grad_norm": 0.0025855612475425005,
       "learning_rate": 1.0696969696969696e-05,
-      "loss": 0.0077,
+      "loss": 0.0005,
       "step": 2650
     },
     {
       "epoch": 10.546875,
-      "grad_norm": 0.036698199808597565,
+      "grad_norm": 0.0029177917167544365,
       "learning_rate": 1.0480519480519481e-05,
-      "loss": 0.0077,
+      "loss": 0.0005,
       "step": 2700
     },
     {
       "epoch": 10.7421875,
-      "grad_norm": 0.03544982895255089,
+      "grad_norm": 0.002265964401885867,
       "learning_rate": 1.0264069264069264e-05,
-      "loss": 0.0076,
+      "loss": 0.0005,
       "step": 2750
     },
     {
       "epoch": 10.9375,
-      "grad_norm": 0.033364105969667435,
+      "grad_norm": 0.0022889028768986464,
       "learning_rate": 1.0047619047619048e-05,
-      "loss": 0.0072,
+      "loss": 0.0004,
       "step": 2800
     },
     {
       "epoch": 11.1328125,
-      "grad_norm": 0.034085117280483246,
+      "grad_norm": 0.0023099584504961967,
       "learning_rate": 9.831168831168832e-06,
-      "loss": 0.007,
+      "loss": 0.0004,
       "step": 2850
     },
     {
       "epoch": 11.328125,
-      "grad_norm": 0.03140813857316971,
+      "grad_norm": 0.0022109313867986202,
       "learning_rate": 9.614718614718616e-06,
-      "loss": 0.0065,
+      "loss": 0.0004,
       "step": 2900
     },
     {
       "epoch": 11.5234375,
-      "grad_norm": 0.035150136798620224,
+      "grad_norm": 0.002484516939148307,
       "learning_rate": 9.398268398268399e-06,
-      "loss": 0.014,
+      "loss": 0.0004,
       "step": 2950
     },
     {
       "epoch": 11.71875,
-      "grad_norm": 0.029343638569116592,
+      "grad_norm": 0.001686307368800044,
       "learning_rate": 9.181818181818184e-06,
-      "loss": 0.0088,
+      "loss": 0.0004,
       "step": 3000
     },
     {
       "epoch": 11.71875,
-      "eval_accuracy": 0.9580078125,
-      "eval_loss": 0.1553695797920227,
-      "eval_runtime": 5.2716,
-      "eval_samples_per_second": 194.25,
-      "eval_steps_per_second": 24.281,
+      "eval_accuracy": 0.96875,
+      "eval_loss": 0.20902524888515472,
+      "eval_runtime": 6.1763,
+      "eval_samples_per_second": 165.796,
+      "eval_steps_per_second": 20.725,
       "step": 3000
     },
     {
       "epoch": 11.9140625,
-      "grad_norm": 0.02912583015859127,
+      "grad_norm": 0.002134295180439949,
       "learning_rate": 8.965367965367967e-06,
-      "loss": 0.0063,
+      "loss": 0.0004,
       "step": 3050
     },
     {
       "epoch": 12.109375,
-      "grad_norm": 0.02827904000878334,
+      "grad_norm": 0.002223787596449256,
       "learning_rate": 8.74891774891775e-06,
-      "loss": 0.0067,
+      "loss": 0.0004,
       "step": 3100
     },
     {
       "epoch": 12.3046875,
-      "grad_norm": 0.02774711512029171,
-      "learning_rate": 8.536796536796537e-06,
-      "loss": 0.0088,
+      "grad_norm": 0.0019202354596927762,
+      "learning_rate": 8.532467532467533e-06,
+      "loss": 0.0004,
       "step": 3150
     },
     {
       "epoch": 12.5,
-      "grad_norm": 0.027258064597845078,
-      "learning_rate": 8.32034632034632e-06,
-      "loss": 0.0055,
+      "grad_norm": 0.0023374806623905897,
+      "learning_rate": 8.316017316017316e-06,
+      "loss": 0.0004,
       "step": 3200
     },
     {
       "epoch": 12.6953125,
-      "grad_norm": 0.02587069943547249,
-      "learning_rate": 8.103896103896105e-06,
-      "loss": 0.0053,
+      "grad_norm": 0.0015720732044428587,
+      "learning_rate": 8.099567099567101e-06,
+      "loss": 0.0004,
       "step": 3250
     },
     {
       "epoch": 12.890625,
-      "grad_norm": 0.02641950733959675,
-      "learning_rate": 7.887445887445888e-06,
-      "loss": 0.0054,
+      "grad_norm": 0.0014028995065018535,
+      "learning_rate": 7.883116883116884e-06,
+      "loss": 0.0004,
       "step": 3300
     },
     {
       "epoch": 13.0859375,
-      "grad_norm": 0.028370605781674385,
-      "learning_rate": 7.670995670995672e-06,
-      "loss": 0.0052,
+      "grad_norm": 0.002278986619785428,
+      "learning_rate": 7.666666666666667e-06,
+      "loss": 0.0004,
       "step": 3350
     },
     {
       "epoch": 13.28125,
-      "grad_norm": 0.02395276166498661,
-      "learning_rate": 7.454545454545456e-06,
-      "loss": 0.0049,
+      "grad_norm": 0.0019616258796304464,
+      "learning_rate": 7.450216450216451e-06,
+      "loss": 0.0003,
       "step": 3400
     },
     {
       "epoch": 13.4765625,
-      "grad_norm": 0.022906098514795303,
-      "learning_rate": 7.238095238095239e-06,
-      "loss": 0.0049,
+      "grad_norm": 0.0017822927329689264,
+      "learning_rate": 7.233766233766234e-06,
+      "loss": 0.0003,
       "step": 3450
     },
     {
       "epoch": 13.671875,
-      "grad_norm": 0.023122085258364677,
-      "learning_rate": 7.021645021645023e-06,
-      "loss": 0.0051,
+      "grad_norm": 0.0021076672710478306,
+      "learning_rate": 7.017316017316018e-06,
+      "loss": 0.0004,
       "step": 3500
     },
     {
       "epoch": 13.671875,
-      "eval_accuracy": 0.958984375,
-      "eval_loss": 0.16660287976264954,
-      "eval_runtime": 5.1728,
-      "eval_samples_per_second": 197.959,
-      "eval_steps_per_second": 24.745,
+      "eval_accuracy": 0.96875,
+      "eval_loss": 0.21559299528598785,
+      "eval_runtime": 6.0759,
+      "eval_samples_per_second": 168.535,
+      "eval_steps_per_second": 21.067,
       "step": 3500
     },
     {
-      "epoch": 13.8671875,
-      "grad_norm": 0.022511709481477737,
-      "learning_rate": 6.805194805194806e-06,
-      "loss": 0.0047,
-      "step": 3550
-    },
-    {
-      "epoch": 14.0625,
-      "grad_norm": 0.06705322116613388,
-      "learning_rate": 6.588744588744589e-06,
-      "loss": 0.0061,
-      "step": 3600
-    },
-    {
-      "epoch": 14.2578125,
-      "grad_norm": 0.032343845814466476,
-      "learning_rate": 6.372294372294373e-06,
-      "loss": 0.0167,
-      "step": 3650
-    },
-    {
-      "epoch": 14.453125,
-      "grad_norm": 0.040268197655677795,
-      "learning_rate": 6.155844155844156e-06,
-      "loss": 0.0059,
-      "step": 3700
-    },
-    {
-      "epoch": 14.6484375,
-      "grad_norm": 0.020453229546546936,
-      "learning_rate": 5.93939393939394e-06,
-      "loss": 0.0044,
-      "step": 3750
-    },
-    {
-      "epoch": 14.84375,
-      "grad_norm": 0.019762134179472923,
-      "learning_rate": 5.722943722943723e-06,
-      "loss": 0.0044,
-      "step": 3800
-    },
-    {
-      "epoch": 15.0390625,
-      "grad_norm": 0.020851831883192062,
-      "learning_rate": 5.5064935064935065e-06,
-      "loss": 0.0041,
-      "step": 3850
-    },
-    {
-      "epoch": 15.234375,
-      "grad_norm": 0.01954171247780323,
-      "learning_rate": 5.2900432900432905e-06,
-      "loss": 0.0041,
-      "step": 3900
-    },
-    {
-      "epoch": 15.4296875,
-      "grad_norm": 0.020014379173517227,
-      "learning_rate": 5.073593073593074e-06,
-      "loss": 0.0042,
-      "step": 3950
-    },
-    {
-      "epoch": 15.625,
-      "grad_norm": 0.01890181563794613,
-      "learning_rate": 4.857142857142858e-06,
-      "loss": 0.0039,
-      "step": 4000
-    },
-    {
-      "epoch": 15.625,
-      "eval_accuracy": 0.96484375,
-      "eval_loss": 0.15439653396606445,
-      "eval_runtime": 5.142,
-      "eval_samples_per_second": 199.146,
-      "eval_steps_per_second": 24.893,
-      "step": 4000
-    },
-    {
-      "epoch": 15.8203125,
-      "grad_norm": 0.017986547201871872,
-      "learning_rate": 4.640692640692641e-06,
-      "loss": 0.0038,
-      "step": 4050
-    },
-    {
-      "epoch": 16.015625,
-      "grad_norm": 0.02049080841243267,
-      "learning_rate": 4.424242424242425e-06,
-      "loss": 0.0037,
-      "step": 4100
-    },
-    {
-      "epoch": 16.2109375,
-      "grad_norm": 0.018345637246966362,
-      "learning_rate": 4.207792207792208e-06,
-      "loss": 0.0038,
-      "step": 4150
-    },
-    {
-      "epoch": 16.40625,
-      "grad_norm": 0.018377136439085007,
-      "learning_rate": 3.991341991341992e-06,
-      "loss": 0.0037,
-      "step": 4200
-    },
-    {
-      "epoch": 16.6015625,
-      "grad_norm": 0.01728520356118679,
-      "learning_rate": 3.774891774891775e-06,
-      "loss": 0.0037,
-      "step": 4250
-    },
-    {
-      "epoch": 16.796875,
-      "grad_norm": 0.017723916098475456,
-      "learning_rate": 3.558441558441559e-06,
-      "loss": 0.0035,
-      "step": 4300
-    },
-    {
-      "epoch": 16.9921875,
-      "grad_norm": 0.017535051330924034,
-      "learning_rate": 3.3419913419913426e-06,
-      "loss": 0.0036,
-      "step": 4350
-    },
-    {
-      "epoch": 17.1875,
-      "grad_norm": 0.01705988124012947,
-      "learning_rate": 3.1255411255411257e-06,
-      "loss": 0.0036,
-      "step": 4400
-    },
-    {
-      "epoch": 17.3828125,
-      "grad_norm": 0.01828162744641304,
-      "learning_rate": 2.9090909090909093e-06,
-      "loss": 0.0035,
-      "step": 4450
-    },
-    {
-      "epoch": 17.578125,
-      "grad_norm": 0.026976697146892548,
-      "learning_rate": 2.692640692640693e-06,
-      "loss": 0.0034,
-      "step": 4500
-    },
-    {
-      "epoch": 17.578125,
-      "eval_accuracy": 0.96484375,
-      "eval_loss": 0.15748074650764465,
-      "eval_runtime": 5.486,
-      "eval_samples_per_second": 186.655,
-      "eval_steps_per_second": 23.332,
-      "step": 4500
-    },
-    {
-      "epoch": 17.7734375,
-      "grad_norm": 0.01693248189985752,
-      "learning_rate": 2.4761904761904764e-06,
-      "loss": 0.0033,
-      "step": 4550
-    },
-    {
-      "epoch": 17.96875,
-      "grad_norm": 0.016402654349803925,
-      "learning_rate": 2.25974025974026e-06,
-      "loss": 0.0034,
-      "step": 4600
-    },
-    {
-      "epoch": 18.1640625,
-      "grad_norm": 0.01663408987224102,
-      "learning_rate": 2.0432900432900435e-06,
-      "loss": 0.0033,
-      "step": 4650
-    },
-    {
-      "epoch": 18.359375,
-      "grad_norm": 0.02396741323173046,
-      "learning_rate": 1.8268398268398269e-06,
-      "loss": 0.0034,
-      "step": 4700
-    },
-    {
-      "epoch": 18.5546875,
-      "grad_norm": 0.016884848475456238,
-      "learning_rate": 1.6103896103896105e-06,
-      "loss": 0.0032,
-      "step": 4750
-    },
-    {
-      "epoch": 18.75,
-      "grad_norm": 0.016911398619413376,
-      "learning_rate": 1.3939393939393942e-06,
-      "loss": 0.0033,
-      "step": 4800
-    },
-    {
-      "epoch": 18.9453125,
-      "grad_norm": 0.01589149609208107,
-      "learning_rate": 1.1774891774891776e-06,
-      "loss": 0.0033,
-      "step": 4850
-    },
-    {
-      "epoch": 19.140625,
-      "grad_norm": 0.017035914584994316,
-      "learning_rate": 9.610389610389612e-07,
-      "loss": 0.0032,
-      "step": 4900
-    },
-    {
-      "epoch": 19.3359375,
-      "grad_norm": 0.01621602661907673,
-      "learning_rate": 7.445887445887446e-07,
-      "loss": 0.0033,
-      "step": 4950
-    },
-    {
-      "epoch": 19.53125,
-      "grad_norm": 0.015999143943190575,
-      "learning_rate": 5.281385281385282e-07,
-      "loss": 0.003,
-      "step": 5000
-    },
-    {
-      "epoch": 19.53125,
-      "eval_accuracy": 0.9658203125,
-      "eval_loss": 0.15924891829490662,
-      "eval_runtime": 5.5508,
-      "eval_samples_per_second": 184.477,
-      "eval_steps_per_second": 23.06,
-      "step": 5000
-    },
-    {
-      "epoch": 19.7265625,
-      "grad_norm": 0.014971093274652958,
-      "learning_rate": 3.116883116883117e-07,
-      "loss": 0.0033,
-      "step": 5050
-    },
-    {
-      "epoch": 19.921875,
-      "grad_norm": 0.015599260106682777,
-      "learning_rate": 9.523809523809525e-08,
-      "loss": 0.0031,
-      "step": 5100
-    },
-    {
-      "epoch": 20.0,
-      "step": 5120,
-      "total_flos": 6.348257582037074e+18,
-      "train_loss": 0.15641047188546509,
-      "train_runtime": 1165.0601,
-      "train_samples_per_second": 70.314,
-      "train_steps_per_second": 4.395
+      "epoch": 13.671875,
+      "step": 3500,
+      "total_flos": 4.339629206470656e+18,
+      "train_loss": 0.02788944040451731,
+      "train_runtime": 959.4355,
+      "train_samples_per_second": 85.384,
+      "train_steps_per_second": 5.336
     }
   ],
   "logging_steps": 50,
@@ -848,7 +597,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 6.348257582037074e+18,
+  "total_flos": 4.339629206470656e+18,
   "train_batch_size": 16,
   "trial_name": null,
   "trial_params": null