End of training

README.md

@@ -4,7 +4,7 @@ base_model: BridgeTower/bridgetower-large-itm-mlm-itc
 tags:
 - generated_from_trainer
 datasets:
-- newyorker_caption_contest
+- jmhessel/newyorker_caption_contest
 model-index:
 - name: bridgetower
   results: []
@@ -15,7 +15,12 @@ should probably proofread and complete it, then remove this comment. -->
 
 # bridgetower
 
-This model is a fine-tuned version of [BridgeTower/bridgetower-large-itm-mlm-itc](https://huggingface.co/BridgeTower/bridgetower-large-itm-mlm-itc) on the newyorker_caption_contest dataset.
+This model is a fine-tuned version of [BridgeTower/bridgetower-large-itm-mlm-itc](https://huggingface.co/BridgeTower/bridgetower-large-itm-mlm-itc) on the jmhessel/newyorker_caption_contest matching dataset.
+It achieves the following results on the evaluation set:
+- Loss: 0.1309
+- Memory Allocated (gb): 51.27
+- Max Memory Allocated (gb): 57.18
+- Total Memory Available (gb): 94.62
 
 ## Model description
 

all_results.json

@@ -1,15 +1,15 @@
 {
-    "epoch": 5.0,
-    "eval_loss": 0.12431247532367706,
-    "eval_runtime": 8.9978,
-    "eval_samples_per_second": 65.301,
-    "eval_steps_per_second": 4.081,
-    "max_memory_allocated (GB)": 60.52,
+    "epoch": 8.0,
+    "eval_loss": 0.13085901737213135,
+    "eval_runtime": 13.1603,
+    "eval_samples_per_second": 46.572,
+    "eval_steps_per_second": 2.911,
+    "max_memory_allocated (GB)": 57.18,
     "memory_allocated (GB)": 51.27,
-    "total_flos": 3.0598946525952e+16,
+    "total_flos": 4.89583144415232e+16,
     "total_memory_available (GB)": 94.62,
-    "train_loss": 0.06082177159737567,
-    "train_runtime": 1079.1005,
-    "train_samples_per_second": 52.128,
-    "train_steps_per_second": 1.304
+    "train_loss": 0.057532464606421335,
+    "train_runtime": 1666.2328,
+    "train_samples_per_second": 52.538,
+    "train_steps_per_second": 1.315
 }

runs/Oct16_22-15-24_workload-ai-workshop/events.out.tfevents.1729118757.workload-ai-workshop.20257.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e59b885b6ad657e461ee4f7f782e0cd73c5518635475a98a23dc6d66337d35f
+size 998

test_results.json

@@ -1,10 +1,10 @@
 {
-    "epoch": 5.0,
-    "eval_loss": 0.12431247532367706,
-    "eval_runtime": 8.9978,
-    "eval_samples_per_second": 65.301,
-    "eval_steps_per_second": 4.081,
-    "max_memory_allocated (GB)": 60.52,
+    "epoch": 8.0,
+    "eval_loss": 0.13085901737213135,
+    "eval_runtime": 13.1603,
+    "eval_samples_per_second": 46.572,
+    "eval_steps_per_second": 2.911,
+    "max_memory_allocated (GB)": 57.18,
     "memory_allocated (GB)": 51.27,
     "total_memory_available (GB)": 94.62
 }

train_results.json

@@ -1,11 +1,11 @@
 {
-    "epoch": 5.0,
-    "max_memory_allocated (GB)": 60.52,
+    "epoch": 8.0,
+    "max_memory_allocated (GB)": 57.18,
     "memory_allocated (GB)": 50.57,
-    "total_flos": 3.0598946525952e+16,
+    "total_flos": 4.89583144415232e+16,
     "total_memory_available (GB)": 94.62,
-    "train_loss": 0.06082177159737567,
-    "train_runtime": 1079.1005,
-    "train_samples_per_second": 52.128,
-    "train_steps_per_second": 1.304
+    "train_loss": 0.057532464606421335,
+    "train_runtime": 1666.2328,
+    "train_samples_per_second": 52.538,
+    "train_steps_per_second": 1.315
 }

trainer_state.json

@@ -1,18 +1,18 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 5.0,
+  "epoch": 8.0,
   "eval_steps": 500,
-  "global_step": 1225,
+  "global_step": 1960,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
       "epoch": 0.04081632653061224,
-      "grad_norm": 16.731555938720703,
-      "learning_rate": 9.918367346938776e-06,
-      "loss": 0.2616,
+      "grad_norm": 9.978917121887207,
+      "learning_rate": 9.948979591836737e-06,
+      "loss": 0.2617,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 10,
@@ -20,9 +20,9 @@
     },
     {
       "epoch": 0.08163265306122448,
-      "grad_norm": 7.408310890197754,
-      "learning_rate": 9.836734693877552e-06,
-      "loss": 0.1553,
+      "grad_norm": 7.48874044418335,
+      "learning_rate": 9.89795918367347e-06,
+      "loss": 0.1563,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 20,
@@ -30,9 +30,9 @@
     },
     {
       "epoch": 0.12244897959183673,
-      "grad_norm": 5.734114170074463,
-      "learning_rate": 9.755102040816327e-06,
-      "loss": 0.1248,
+      "grad_norm": 5.081777572631836,
+      "learning_rate": 9.846938775510205e-06,
+      "loss": 0.1254,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 30,
@@ -40,9 +40,9 @@
     },
     {
       "epoch": 0.16326530612244897,
-      "grad_norm": 4.232840061187744,
-      "learning_rate": 9.673469387755103e-06,
-      "loss": 0.1106,
+      "grad_norm": 4.443576812744141,
+      "learning_rate": 9.795918367346939e-06,
+      "loss": 0.1113,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 40,
@@ -50,9 +50,9 @@
     },
     {
       "epoch": 0.20408163265306123,
-      "grad_norm": 4.877067565917969,
-      "learning_rate": 9.591836734693878e-06,
-      "loss": 0.1233,
+      "grad_norm": 4.356841087341309,
+      "learning_rate": 9.744897959183674e-06,
+      "loss": 0.1257,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 50,
@@ -60,9 +60,9 @@
     },
     {
       "epoch": 0.24489795918367346,
-      "grad_norm": 2.0087852478027344,
-      "learning_rate": 9.510204081632653e-06,
-      "loss": 0.0799,
+      "grad_norm": 1.98320472240448,
+      "learning_rate": 9.693877551020408e-06,
+      "loss": 0.0819,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 60,
@@ -70,9 +70,9 @@
     },
     {
       "epoch": 0.2857142857142857,
-      "grad_norm": 4.425936698913574,
-      "learning_rate": 9.42857142857143e-06,
-      "loss": 0.1015,
+      "grad_norm": 3.809190273284912,
+      "learning_rate": 9.642857142857144e-06,
+      "loss": 0.1032,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 70,
@@ -80,9 +80,9 @@
     },
     {
       "epoch": 0.32653061224489793,
-      "grad_norm": 1.8348758220672607,
-      "learning_rate": 9.346938775510204e-06,
-      "loss": 0.1094,
+      "grad_norm": 1.645442247390747,
+      "learning_rate": 9.591836734693878e-06,
+      "loss": 0.1124,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 80,
@@ -90,9 +90,9 @@
     },
     {
       "epoch": 0.3673469387755102,
-      "grad_norm": 5.009132385253906,
-      "learning_rate": 9.26530612244898e-06,
-      "loss": 0.0837,
+      "grad_norm": 3.7085306644439697,
+      "learning_rate": 9.540816326530612e-06,
+      "loss": 0.0847,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 90,
@@ -100,9 +100,9 @@
     },
     {
       "epoch": 0.40816326530612246,
-      "grad_norm": 6.888875484466553,
-      "learning_rate": 9.183673469387756e-06,
-      "loss": 0.075,
+      "grad_norm": 3.9240212440490723,
+      "learning_rate": 9.489795918367348e-06,
+      "loss": 0.0753,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 100,
@@ -110,9 +110,9 @@
     },
     {
       "epoch": 0.4489795918367347,
-      "grad_norm": 3.038123369216919,
-      "learning_rate": 9.102040816326532e-06,
-      "loss": 0.0709,
+      "grad_norm": 3.737152338027954,
+      "learning_rate": 9.438775510204082e-06,
+      "loss": 0.0723,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 110,
@@ -120,9 +120,9 @@
     },
     {
       "epoch": 0.4897959183673469,
-      "grad_norm": 2.6441116333007812,
-      "learning_rate": 9.020408163265307e-06,
-      "loss": 0.0706,
+      "grad_norm": 2.308751344680786,
+      "learning_rate": 9.387755102040818e-06,
+      "loss": 0.0699,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 120,
@@ -130,9 +130,9 @@
     },
     {
       "epoch": 0.5306122448979592,
-      "grad_norm": 3.5273971557617188,
-      "learning_rate": 8.938775510204082e-06,
-      "loss": 0.0597,
+      "grad_norm": 2.3706369400024414,
+      "learning_rate": 9.336734693877552e-06,
+      "loss": 0.0589,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 130,
@@ -140,9 +140,9 @@
     },
     {
       "epoch": 0.5714285714285714,
-      "grad_norm": 1.2711573839187622,
-      "learning_rate": 8.857142857142858e-06,
-      "loss": 0.0499,
+      "grad_norm": 0.968673050403595,
+      "learning_rate": 9.285714285714288e-06,
+      "loss": 0.0503,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 140,
@@ -150,9 +150,9 @@
     },
     {
       "epoch": 0.6122448979591837,
-      "grad_norm": 2.2461705207824707,
-      "learning_rate": 8.775510204081633e-06,
-      "loss": 0.0636,
+      "grad_norm": 1.9251790046691895,
+      "learning_rate": 9.234693877551022e-06,
+      "loss": 0.0628,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 150,
@@ -160,8 +160,8 @@
     },
     {
       "epoch": 0.6530612244897959,
-      "grad_norm": 2.0061120986938477,
-      "learning_rate": 8.69387755102041e-06,
+      "grad_norm": 1.7473604679107666,
+      "learning_rate": 9.183673469387756e-06,
       "loss": 0.0708,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
@@ -170,9 +170,9 @@
     },
     {
       "epoch": 0.6938775510204082,
-      "grad_norm": 1.7886881828308105,
-      "learning_rate": 8.612244897959184e-06,
-      "loss": 0.0753,
+      "grad_norm": 1.9279741048812866,
+      "learning_rate": 9.13265306122449e-06,
+      "loss": 0.075,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 170,
@@ -180,9 +180,9 @@
     },
     {
       "epoch": 0.7346938775510204,
-      "grad_norm": 1.8707109689712524,
-      "learning_rate": 8.530612244897961e-06,
-      "loss": 0.0612,
+      "grad_norm": 1.4570097923278809,
+      "learning_rate": 9.081632653061225e-06,
+      "loss": 0.0614,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 180,
@@ -190,9 +190,9 @@
     },
     {
       "epoch": 0.7755102040816326,
-      "grad_norm": 2.3418948650360107,
-      "learning_rate": 8.448979591836736e-06,
-      "loss": 0.0586,
+      "grad_norm": 2.014692544937134,
+      "learning_rate": 9.03061224489796e-06,
+      "loss": 0.058,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 190,
@@ -200,9 +200,9 @@
     },
     {
       "epoch": 0.8163265306122449,
-      "grad_norm": 0.7759063839912415,
-      "learning_rate": 8.36734693877551e-06,
-      "loss": 0.063,
+      "grad_norm": 1.7634117603302002,
+      "learning_rate": 8.979591836734695e-06,
+      "loss": 0.0629,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 200,
@@ -210,9 +210,9 @@
     },
     {
       "epoch": 0.8571428571428571,
-      "grad_norm": 0.9075489044189453,
-      "learning_rate": 8.285714285714287e-06,
-      "loss": 0.063,
+      "grad_norm": 0.7091555595397949,
+      "learning_rate": 8.92857142857143e-06,
+      "loss": 0.0637,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 210,
@@ -220,9 +220,9 @@
     },
     {
       "epoch": 0.8979591836734694,
-      "grad_norm": 1.0522433519363403,
-      "learning_rate": 8.204081632653062e-06,
-      "loss": 0.0605,
+      "grad_norm": 1.073096752166748,
+      "learning_rate": 8.877551020408163e-06,
+      "loss": 0.0603,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 220,
@@ -230,9 +230,9 @@
     },
     {
       "epoch": 0.9387755102040817,
-      "grad_norm": 0.7449356913566589,
-      "learning_rate": 8.122448979591837e-06,
-      "loss": 0.0564,
+      "grad_norm": 0.7938856482505798,
+      "learning_rate": 8.826530612244899e-06,
+      "loss": 0.0538,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 230,
@@ -240,9 +240,9 @@
     },
     {
       "epoch": 0.9795918367346939,
-      "grad_norm": 0.878478467464447,
-      "learning_rate": 8.040816326530613e-06,
-      "loss": 0.0496,
+      "grad_norm": 1.192353367805481,
+      "learning_rate": 8.775510204081633e-06,
+      "loss": 0.0493,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 240,
@@ -250,9 +250,9 @@
     },
     {
       "epoch": 1.0204081632653061,
-      "grad_norm": 1.1043357849121094,
-      "learning_rate": 7.959183673469388e-06,
-      "loss": 0.0593,
+      "grad_norm": 0.9369480013847351,
+      "learning_rate": 8.724489795918369e-06,
+      "loss": 0.0595,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 250,
@@ -260,9 +260,9 @@
     },
     {
       "epoch": 1.0612244897959184,
-      "grad_norm": 1.5091803073883057,
-      "learning_rate": 7.877551020408164e-06,
-      "loss": 0.0542,
+      "grad_norm": 1.2866365909576416,
+      "learning_rate": 8.673469387755103e-06,
+      "loss": 0.0532,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 260,
@@ -270,9 +270,9 @@
     },
     {
       "epoch": 1.1020408163265305,
-      "grad_norm": 0.7768255472183228,
-      "learning_rate": 7.79591836734694e-06,
-      "loss": 0.0636,
+      "grad_norm": 16.09465980529785,
+      "learning_rate": 8.622448979591837e-06,
+      "loss": 0.0663,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 270,
@@ -280,9 +280,9 @@
     },
     {
       "epoch": 1.1428571428571428,
-      "grad_norm": 1.0384677648544312,
-      "learning_rate": 7.714285714285716e-06,
-      "loss": 0.0585,
+      "grad_norm": 2.3071987628936768,
+      "learning_rate": 8.571428571428571e-06,
+      "loss": 0.0633,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 280,
@@ -290,9 +290,9 @@
     },
     {
       "epoch": 1.183673469387755,
-      "grad_norm": 1.9693238735198975,
-      "learning_rate": 7.63265306122449e-06,
-      "loss": 0.0512,
+      "grad_norm": 1.2905592918395996,
+      "learning_rate": 8.520408163265307e-06,
+      "loss": 0.0498,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 290,
@@ -300,9 +300,9 @@
     },
     {
       "epoch": 1.2244897959183674,
-      "grad_norm": 1.0379549264907837,
-      "learning_rate": 7.551020408163265e-06,
-      "loss": 0.0647,
+      "grad_norm": 1.4856091737747192,
+      "learning_rate": 8.469387755102042e-06,
+      "loss": 0.0639,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 300,
@@ -310,9 +310,9 @@
     },
     {
       "epoch": 1.2653061224489797,
-      "grad_norm": 1.01348078250885,
-      "learning_rate": 7.469387755102041e-06,
-      "loss": 0.0668,
+      "grad_norm": 0.5537325739860535,
+      "learning_rate": 8.418367346938776e-06,
+      "loss": 0.0673,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 310,
@@ -320,9 +320,9 @@
     },
     {
       "epoch": 1.306122448979592,
-      "grad_norm": 0.9114493727684021,
-      "learning_rate": 7.387755102040817e-06,
-      "loss": 0.0509,
+      "grad_norm": 1.2954118251800537,
+      "learning_rate": 8.36734693877551e-06,
+      "loss": 0.0505,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 320,
@@ -330,9 +330,9 @@
     },
     {
       "epoch": 1.346938775510204,
-      "grad_norm": 0.5102170705795288,
-      "learning_rate": 7.306122448979592e-06,
-      "loss": 0.0612,
+      "grad_norm": 0.6809917092323303,
+      "learning_rate": 8.316326530612246e-06,
+      "loss": 0.0623,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 330,
@@ -340,9 +340,9 @@
     },
     {
       "epoch": 1.3877551020408163,
-      "grad_norm": 1.097835659980774,
-      "learning_rate": 7.224489795918368e-06,
-      "loss": 0.055,
+      "grad_norm": 2.596815586090088,
+      "learning_rate": 8.26530612244898e-06,
+      "loss": 0.0552,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 340,
@@ -350,9 +350,9 @@
     },
     {
       "epoch": 1.4285714285714286,
-      "grad_norm": 0.5733208656311035,
-      "learning_rate": 7.1428571428571436e-06,
-      "loss": 0.05,
+      "grad_norm": 1.0378236770629883,
+      "learning_rate": 8.214285714285714e-06,
+      "loss": 0.049,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 350,
@@ -360,9 +360,9 @@
     },
     {
       "epoch": 1.469387755102041,
-      "grad_norm": 2.428065538406372,
-      "learning_rate": 7.061224489795919e-06,
-      "loss": 0.0458,
+      "grad_norm": 1.7493040561676025,
+      "learning_rate": 8.16326530612245e-06,
+      "loss": 0.0465,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 360,
@@ -370,9 +370,9 @@
     },
     {
       "epoch": 1.510204081632653,
-      "grad_norm": 0.9827563166618347,
-      "learning_rate": 6.979591836734695e-06,
-      "loss": 0.0588,
+      "grad_norm": 1.149561882019043,
+      "learning_rate": 8.112244897959184e-06,
+      "loss": 0.0584,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 370,
@@ -380,9 +380,9 @@
     },
     {
       "epoch": 1.5510204081632653,
-      "grad_norm": 0.5563956499099731,
-      "learning_rate": 6.8979591836734705e-06,
-      "loss": 0.0473,
+      "grad_norm": 0.8010720014572144,
+      "learning_rate": 8.06122448979592e-06,
+      "loss": 0.047,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 380,
@@ -390,9 +390,9 @@
     },
     {
       "epoch": 1.5918367346938775,
-      "grad_norm": 0.9855480194091797,
-      "learning_rate": 6.816326530612245e-06,
-      "loss": 0.0653,
+      "grad_norm": 0.7010307908058167,
+      "learning_rate": 8.010204081632654e-06,
+      "loss": 0.0649,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 390,
@@ -400,9 +400,9 @@
     },
     {
       "epoch": 1.6326530612244898,
-      "grad_norm": 0.7717217803001404,
-      "learning_rate": 6.734693877551021e-06,
-      "loss": 0.0618,
+      "grad_norm": 2.022503137588501,
+      "learning_rate": 7.959183673469388e-06,
+      "loss": 0.0612,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 400,
@@ -410,9 +410,9 @@
     },
     {
       "epoch": 1.6734693877551021,
-      "grad_norm": 0.5391837358474731,
-      "learning_rate": 6.653061224489797e-06,
-      "loss": 0.061,
+      "grad_norm": 1.3006742000579834,
+      "learning_rate": 7.908163265306124e-06,
+      "loss": 0.0605,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 410,
@@ -420,9 +420,9 @@
     },
     {
       "epoch": 1.7142857142857144,
-      "grad_norm": 1.2298884391784668,
-      "learning_rate": 6.571428571428572e-06,
-      "loss": 0.0534,
+      "grad_norm": 1.513334035873413,
+      "learning_rate": 7.857142857142858e-06,
+      "loss": 0.054,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 420,
@@ -430,9 +430,9 @@
     },
     {
       "epoch": 1.7551020408163265,
-      "grad_norm": 0.22150574624538422,
-      "learning_rate": 6.489795918367348e-06,
-      "loss": 0.0681,
+      "grad_norm": 0.28943702578544617,
+      "learning_rate": 7.806122448979593e-06,
+      "loss": 0.0673,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 430,
@@ -440,9 +440,9 @@
     },
     {
       "epoch": 1.7959183673469388,
-      "grad_norm": 0.8992190957069397,
-      "learning_rate": 6.408163265306124e-06,
-      "loss": 0.0606,
+      "grad_norm": 1.2818681001663208,
+      "learning_rate": 7.755102040816327e-06,
+      "loss": 0.0614,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 440,
@@ -450,9 +450,9 @@
     },
     {
       "epoch": 1.836734693877551,
-      "grad_norm": 0.9748144149780273,
-      "learning_rate": 6.326530612244899e-06,
-      "loss": 0.0436,
+      "grad_norm": 0.5026584267616272,
+      "learning_rate": 7.704081632653061e-06,
+      "loss": 0.0443,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 450,
@@ -460,9 +460,9 @@
     },
     {
       "epoch": 1.8775510204081631,
-      "grad_norm": 0.3272276520729065,
-      "learning_rate": 6.244897959183675e-06,
-      "loss": 0.0545,
+      "grad_norm": 0.400056391954422,
+      "learning_rate": 7.653061224489796e-06,
+      "loss": 0.054,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 460,
@@ -470,9 +470,9 @@
     },
     {
       "epoch": 1.9183673469387754,
-      "grad_norm": 0.7515161633491516,
-      "learning_rate": 6.163265306122449e-06,
-      "loss": 0.0427,
+      "grad_norm": 0.7661588191986084,
+      "learning_rate": 7.602040816326531e-06,
+      "loss": 0.0439,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 470,
@@ -480,9 +480,9 @@
     },
     {
       "epoch": 1.9591836734693877,
-      "grad_norm": 0.35405218601226807,
-      "learning_rate": 6.0816326530612245e-06,
-      "loss": 0.0501,
+      "grad_norm": 0.3066469728946686,
+      "learning_rate": 7.551020408163265e-06,
+      "loss": 0.0511,
       "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 480,
@@ -490,761 +490,1501 @@
     },
     {
       "epoch": 2.0,
-      "grad_norm": 0.9796536564826965,
-      "learning_rate": 6e-06,
-      "loss": 0.0641,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 1.1751477718353271,
+      "learning_rate": 7.500000000000001e-06,
+      "loss": 0.0644,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 490,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.0408163265306123,
-      "grad_norm": 0.639869213104248,
-      "learning_rate": 5.918367346938776e-06,
-      "loss": 0.0592,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.6497346758842468,
+      "learning_rate": 7.448979591836736e-06,
+      "loss": 0.0596,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 500,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.0816326530612246,
-      "grad_norm": 0.8377614617347717,
-      "learning_rate": 5.8367346938775515e-06,
-      "loss": 0.0497,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.585145890712738,
+      "learning_rate": 7.39795918367347e-06,
+      "loss": 0.0502,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 510,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.122448979591837,
-      "grad_norm": 1.07124662399292,
-      "learning_rate": 5.755102040816327e-06,
-      "loss": 0.047,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 1.0224946737289429,
+      "learning_rate": 7.346938775510205e-06,
+      "loss": 0.0462,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 520,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.163265306122449,
-      "grad_norm": 1.0354053974151611,
-      "learning_rate": 5.673469387755103e-06,
-      "loss": 0.0636,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.9922281503677368,
+      "learning_rate": 7.295918367346939e-06,
+      "loss": 0.063,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 530,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.204081632653061,
-      "grad_norm": 0.5396034717559814,
-      "learning_rate": 5.591836734693878e-06,
-      "loss": 0.0602,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.7550894618034363,
+      "learning_rate": 7.244897959183675e-06,
+      "loss": 0.0595,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 540,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.2448979591836733,
-      "grad_norm": 1.1461529731750488,
-      "learning_rate": 5.510204081632653e-06,
-      "loss": 0.0629,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 1.000552773475647,
+      "learning_rate": 7.193877551020409e-06,
+      "loss": 0.0645,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 550,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.2857142857142856,
-      "grad_norm": 1.732596516609192,
-      "learning_rate": 5.428571428571429e-06,
-      "loss": 0.0575,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.7375513315200806,
+      "learning_rate": 7.1428571428571436e-06,
+      "loss": 0.0597,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 560,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.326530612244898,
-      "grad_norm": 0.6468518376350403,
-      "learning_rate": 5.3469387755102045e-06,
-      "loss": 0.0581,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.7129970192909241,
+      "learning_rate": 7.091836734693878e-06,
+      "loss": 0.0603,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 570,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.36734693877551,
-      "grad_norm": 1.7525995969772339,
-      "learning_rate": 5.26530612244898e-06,
-      "loss": 0.0678,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.8948765993118286,
+      "learning_rate": 7.0408163265306125e-06,
+      "loss": 0.0673,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 580,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.4081632653061225,
-      "grad_norm": 0.5057170391082764,
-      "learning_rate": 5.183673469387756e-06,
-      "loss": 0.0533,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.4436047375202179,
+      "learning_rate": 6.989795918367348e-06,
+      "loss": 0.0547,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 590,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.4489795918367347,
-      "grad_norm": 1.2042871713638306,
-      "learning_rate": 5.1020408163265315e-06,
-      "loss": 0.0444,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 2.562260627746582,
+      "learning_rate": 6.938775510204082e-06,
+      "loss": 0.044,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 600,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.489795918367347,
-      "grad_norm": 0.20784775912761688,
-      "learning_rate": 5.020408163265307e-06,
-      "loss": 0.0413,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 2.5978403091430664,
+      "learning_rate": 6.887755102040817e-06,
+      "loss": 0.042,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 610,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.5306122448979593,
-      "grad_norm": 5.924367427825928,
-      "learning_rate": 4.938775510204082e-06,
-      "loss": 0.0424,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.8350633978843689,
+      "learning_rate": 6.836734693877551e-06,
+      "loss": 0.0429,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 620,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.571428571428571,
-      "grad_norm": 0.40434470772743225,
-      "learning_rate": 4.857142857142858e-06,
+      "grad_norm": 1.0908092260360718,
+      "learning_rate": 6.785714285714287e-06,
       "loss": 0.0815,
-      "max_memory_allocated (GB)": 60.52,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 630,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.612244897959184,
-      "grad_norm": 0.7878668904304504,
-      "learning_rate": 4.775510204081633e-06,
-      "loss": 0.0495,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 1.411789059638977,
+      "learning_rate": 6.734693877551021e-06,
+      "loss": 0.0506,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 640,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.6530612244897958,
-      "grad_norm": 0.5937454700469971,
-      "learning_rate": 4.693877551020409e-06,
-      "loss": 0.0471,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.7262634038925171,
+      "learning_rate": 6.683673469387756e-06,
+      "loss": 0.0486,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 650,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.693877551020408,
-      "grad_norm": 0.9869357347488403,
-      "learning_rate": 4.612244897959184e-06,
-      "loss": 0.0472,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.6718008518218994,
+      "learning_rate": 6.63265306122449e-06,
+      "loss": 0.0478,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 660,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.7346938775510203,
-      "grad_norm": 0.7606731653213501,
-      "learning_rate": 4.530612244897959e-06,
-      "loss": 0.0864,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.6992954015731812,
+      "learning_rate": 6.581632653061225e-06,
+      "loss": 0.0876,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 670,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.7755102040816326,
-      "grad_norm": 1.5359723567962646,
-      "learning_rate": 4.448979591836735e-06,
-      "loss": 0.0482,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 3.032949447631836,
+      "learning_rate": 6.530612244897959e-06,
+      "loss": 0.0497,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 680,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.816326530612245,
-      "grad_norm": 0.370749294757843,
-      "learning_rate": 4.367346938775511e-06,
-      "loss": 0.0447,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.544232189655304,
+      "learning_rate": 6.4795918367346946e-06,
+      "loss": 0.0456,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 690,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.857142857142857,
-      "grad_norm": 0.633791983127594,
-      "learning_rate": 4.2857142857142855e-06,
-      "loss": 0.044,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.489704430103302,
+      "learning_rate": 6.4285714285714295e-06,
+      "loss": 0.0438,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 700,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.8979591836734695,
-      "grad_norm": 1.1335562467575073,
-      "learning_rate": 4.204081632653061e-06,
-      "loss": 0.0552,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.7447965741157532,
+      "learning_rate": 6.3775510204081635e-06,
+      "loss": 0.0557,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 710,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.938775510204082,
-      "grad_norm": 1.5698113441467285,
-      "learning_rate": 4.122448979591837e-06,
-      "loss": 0.0576,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 3.607469081878662,
+      "learning_rate": 6.326530612244899e-06,
+      "loss": 0.059,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 720,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 2.979591836734694,
-      "grad_norm": 0.20579344034194946,
-      "learning_rate": 4.040816326530612e-06,
-      "loss": 0.0336,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.2003553807735443,
+      "learning_rate": 6.275510204081633e-06,
+      "loss": 0.0349,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 730,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.020408163265306,
-      "grad_norm": 1.3172132968902588,
-      "learning_rate": 3.959183673469388e-06,
-      "loss": 0.0534,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 1.135377287864685,
+      "learning_rate": 6.224489795918368e-06,
+      "loss": 0.0549,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 740,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.061224489795918,
-      "grad_norm": 1.5193818807601929,
-      "learning_rate": 3.877551020408164e-06,
-      "loss": 0.0605,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.9238697290420532,
+      "learning_rate": 6.173469387755102e-06,
+      "loss": 0.0627,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 750,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.1020408163265305,
-      "grad_norm": 0.7448021769523621,
-      "learning_rate": 3.795918367346939e-06,
-      "loss": 0.0524,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.7442536354064941,
+      "learning_rate": 6.122448979591837e-06,
+      "loss": 0.0528,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 760,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.142857142857143,
-      "grad_norm": 0.6830514073371887,
-      "learning_rate": 3.7142857142857146e-06,
-      "loss": 0.0687,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.6410558819770813,
+      "learning_rate": 6.071428571428571e-06,
+      "loss": 0.0707,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 770,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.183673469387755,
-      "grad_norm": 0.4188525378704071,
-      "learning_rate": 3.6326530612244903e-06,
-      "loss": 0.0646,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.4915910065174103,
+      "learning_rate": 6.020408163265307e-06,
+      "loss": 0.0659,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 780,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.2244897959183674,
-      "grad_norm": 0.562319278717041,
-      "learning_rate": 3.5510204081632655e-06,
-      "loss": 0.0523,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.33948495984077454,
+      "learning_rate": 5.969387755102042e-06,
+      "loss": 0.0535,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 790,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.2653061224489797,
-      "grad_norm": 0.383428156375885,
-      "learning_rate": 3.469387755102041e-06,
-      "loss": 0.0425,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.9314869046211243,
+      "learning_rate": 5.918367346938776e-06,
+      "loss": 0.0443,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 800,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.306122448979592,
-      "grad_norm": 0.6743464469909668,
-      "learning_rate": 3.3877551020408168e-06,
-      "loss": 0.0546,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.9704706072807312,
+      "learning_rate": 5.867346938775511e-06,
+      "loss": 0.0562,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 810,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.3469387755102042,
-      "grad_norm": 0.7104524970054626,
-      "learning_rate": 3.3061224489795924e-06,
-      "loss": 0.0448,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.8564426898956299,
+      "learning_rate": 5.816326530612246e-06,
+      "loss": 0.0466,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 820,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.387755102040816,
-      "grad_norm": 0.5083824396133423,
-      "learning_rate": 3.2244897959183672e-06,
-      "loss": 0.048,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.31214070320129395,
+      "learning_rate": 5.7653061224489805e-06,
+      "loss": 0.0488,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 830,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.4285714285714284,
-      "grad_norm": 0.49159926176071167,
-      "learning_rate": 3.142857142857143e-06,
-      "loss": 0.0521,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.40054649114608765,
+      "learning_rate": 5.7142857142857145e-06,
+      "loss": 0.0536,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 840,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.4693877551020407,
-      "grad_norm": 0.635971188545227,
-      "learning_rate": 3.0612244897959185e-06,
-      "loss": 0.081,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.476951003074646,
+      "learning_rate": 5.663265306122449e-06,
+      "loss": 0.0819,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 850,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.510204081632653,
-      "grad_norm": 5.446665287017822,
-      "learning_rate": 2.979591836734694e-06,
-      "loss": 0.0449,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 1.075916051864624,
+      "learning_rate": 5.6122448979591834e-06,
+      "loss": 0.0451,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 860,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.5510204081632653,
-      "grad_norm": 0.3760354816913605,
-      "learning_rate": 2.8979591836734694e-06,
-      "loss": 0.064,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.4422233998775482,
+      "learning_rate": 5.561224489795919e-06,
+      "loss": 0.065,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 870,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.5918367346938775,
-      "grad_norm": 1.0110430717468262,
-      "learning_rate": 2.816326530612245e-06,
-      "loss": 0.0433,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.7247931361198425,
+      "learning_rate": 5.510204081632653e-06,
+      "loss": 0.0442,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 880,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.63265306122449,
-      "grad_norm": 0.1704891324043274,
-      "learning_rate": 2.7346938775510207e-06,
-      "loss": 0.0282,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.18422362208366394,
+      "learning_rate": 5.459183673469388e-06,
+      "loss": 0.0295,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 890,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.673469387755102,
-      "grad_norm": 0.8289852142333984,
-      "learning_rate": 2.6530612244897964e-06,
-      "loss": 0.0518,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.6566686034202576,
+      "learning_rate": 5.408163265306123e-06,
+      "loss": 0.0527,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 900,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.7142857142857144,
-      "grad_norm": 0.812251091003418,
-      "learning_rate": 2.571428571428571e-06,
-      "loss": 0.0598,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.7151392698287964,
+      "learning_rate": 5.357142857142857e-06,
+      "loss": 0.0614,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 910,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.7551020408163263,
-      "grad_norm": 0.14789150655269623,
-      "learning_rate": 2.489795918367347e-06,
-      "loss": 0.0542,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.1488690972328186,
+      "learning_rate": 5.306122448979593e-06,
+      "loss": 0.0546,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 920,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.795918367346939,
-      "grad_norm": 0.8937461972236633,
-      "learning_rate": 2.4081632653061225e-06,
-      "loss": 0.0499,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.472126841545105,
+      "learning_rate": 5.255102040816327e-06,
+      "loss": 0.0514,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 930,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.836734693877551,
-      "grad_norm": 6.832267761230469,
-      "learning_rate": 2.326530612244898e-06,
-      "loss": 0.0489,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.8530511260032654,
+      "learning_rate": 5.204081632653062e-06,
+      "loss": 0.049,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 940,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.877551020408163,
-      "grad_norm": 0.7444981336593628,
-      "learning_rate": 2.244897959183674e-06,
-      "loss": 0.0591,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 1.6832056045532227,
+      "learning_rate": 5.153061224489796e-06,
+      "loss": 0.0603,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 950,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.9183673469387754,
-      "grad_norm": 0.38545939326286316,
-      "learning_rate": 2.1632653061224495e-06,
-      "loss": 0.0499,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.30192047357559204,
+      "learning_rate": 5.1020408163265315e-06,
+      "loss": 0.0512,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 960,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 3.9591836734693877,
-      "grad_norm": 1.0209099054336548,
-      "learning_rate": 2.0816326530612247e-06,
-      "loss": 0.0736,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 1.3734880685806274,
+      "learning_rate": 5.0510204081632655e-06,
+      "loss": 0.0756,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 970,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.0,
-      "grad_norm": 0.6384561657905579,
-      "learning_rate": 2.0000000000000003e-06,
-      "loss": 0.0705,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.7525829672813416,
+      "learning_rate": 5e-06,
+      "loss": 0.0715,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 980,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.040816326530612,
-      "grad_norm": 0.6284443736076355,
-      "learning_rate": 1.9183673469387756e-06,
-      "loss": 0.0483,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.4748665690422058,
+      "learning_rate": 4.948979591836735e-06,
+      "loss": 0.0487,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 990,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.081632653061225,
-      "grad_norm": 1.0421397686004639,
-      "learning_rate": 1.8367346938775512e-06,
-      "loss": 0.0627,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 1.340325117111206,
+      "learning_rate": 4.897959183673469e-06,
+      "loss": 0.0638,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1000,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.122448979591836,
-      "grad_norm": 0.6968585252761841,
-      "learning_rate": 1.7551020408163267e-06,
-      "loss": 0.0632,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.5442948937416077,
+      "learning_rate": 4.846938775510204e-06,
+      "loss": 0.0642,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1010,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.163265306122449,
-      "grad_norm": 0.35346081852912903,
-      "learning_rate": 1.6734693877551023e-06,
-      "loss": 0.0402,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.3119046688079834,
+      "learning_rate": 4.795918367346939e-06,
+      "loss": 0.0411,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1020,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.204081632653061,
-      "grad_norm": 0.8111013174057007,
-      "learning_rate": 1.5918367346938775e-06,
-      "loss": 0.0528,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.7393902540206909,
+      "learning_rate": 4.744897959183674e-06,
+      "loss": 0.0544,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1030,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.244897959183674,
-      "grad_norm": 0.5146018862724304,
-      "learning_rate": 1.5102040816326532e-06,
-      "loss": 0.0396,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.5412510633468628,
+      "learning_rate": 4.693877551020409e-06,
+      "loss": 0.0406,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1040,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.285714285714286,
-      "grad_norm": 0.6711630821228027,
-      "learning_rate": 1.4285714285714286e-06,
-      "loss": 0.0493,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.6453996300697327,
+      "learning_rate": 4.642857142857144e-06,
+      "loss": 0.0499,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1050,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.326530612244898,
-      "grad_norm": 0.3221682012081146,
-      "learning_rate": 1.3469387755102043e-06,
-      "loss": 0.0454,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.3400985896587372,
+      "learning_rate": 4.591836734693878e-06,
+      "loss": 0.046,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1060,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.36734693877551,
-      "grad_norm": 0.9966460466384888,
-      "learning_rate": 1.2653061224489795e-06,
-      "loss": 0.0485,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.5143836736679077,
+      "learning_rate": 4.540816326530613e-06,
+      "loss": 0.0494,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1070,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.408163265306122,
-      "grad_norm": 0.45073866844177246,
-      "learning_rate": 1.1836734693877552e-06,
-      "loss": 0.052,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.38877835869789124,
+      "learning_rate": 4.489795918367348e-06,
+      "loss": 0.0526,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1080,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.448979591836735,
-      "grad_norm": 0.7129877805709839,
-      "learning_rate": 1.1020408163265308e-06,
-      "loss": 0.0511,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.38251811265945435,
+      "learning_rate": 4.438775510204082e-06,
+      "loss": 0.051,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1090,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.489795918367347,
-      "grad_norm": 0.333344966173172,
-      "learning_rate": 1.0204081632653063e-06,
-      "loss": 0.0352,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.3022618889808655,
+      "learning_rate": 4.3877551020408165e-06,
+      "loss": 0.0368,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1100,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.530612244897959,
-      "grad_norm": 0.11522499471902847,
-      "learning_rate": 9.387755102040817e-07,
-      "loss": 0.0457,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.12300197780132294,
+      "learning_rate": 4.336734693877551e-06,
+      "loss": 0.0474,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1110,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.571428571428571,
-      "grad_norm": 0.3113747239112854,
-      "learning_rate": 8.571428571428572e-07,
-      "loss": 0.0402,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.7275770902633667,
+      "learning_rate": 4.2857142857142855e-06,
+      "loss": 0.0409,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1120,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.612244897959184,
-      "grad_norm": 0.8939142823219299,
-      "learning_rate": 7.755102040816327e-07,
-      "loss": 0.0536,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.46302053332328796,
+      "learning_rate": 4.234693877551021e-06,
+      "loss": 0.0545,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1130,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.653061224489796,
-      "grad_norm": 0.7175144553184509,
-      "learning_rate": 6.938775510204082e-07,
-      "loss": 0.0724,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 1.1193764209747314,
+      "learning_rate": 4.183673469387755e-06,
+      "loss": 0.0736,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1140,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.6938775510204085,
-      "grad_norm": 0.6586973071098328,
-      "learning_rate": 6.122448979591837e-07,
-      "loss": 0.0514,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.936698317527771,
+      "learning_rate": 4.13265306122449e-06,
+      "loss": 0.0532,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1150,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.73469387755102,
-      "grad_norm": 1.190521478652954,
-      "learning_rate": 5.306122448979592e-07,
-      "loss": 0.0608,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 1.091784119606018,
+      "learning_rate": 4.081632653061225e-06,
+      "loss": 0.064,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1160,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.775510204081632,
-      "grad_norm": 0.47001194953918457,
-      "learning_rate": 4.489795918367347e-07,
-      "loss": 0.0548,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.3371049165725708,
+      "learning_rate": 4.03061224489796e-06,
+      "loss": 0.0557,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1170,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.816326530612245,
-      "grad_norm": 0.9076634049415588,
-      "learning_rate": 3.6734693877551025e-07,
-      "loss": 0.0438,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.5533121824264526,
+      "learning_rate": 3.979591836734694e-06,
+      "loss": 0.0449,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1180,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.857142857142857,
-      "grad_norm": 1.4096022844314575,
-      "learning_rate": 2.8571428571428575e-07,
-      "loss": 0.0542,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 1.3483092784881592,
+      "learning_rate": 3.928571428571429e-06,
+      "loss": 0.0551,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1190,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.8979591836734695,
-      "grad_norm": 0.7559374570846558,
-      "learning_rate": 2.0408163265306121e-07,
-      "loss": 0.0715,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 2.4415154457092285,
+      "learning_rate": 3.877551020408164e-06,
+      "loss": 0.0738,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1200,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.938775510204081,
-      "grad_norm": 0.6551031470298767,
-      "learning_rate": 1.2244897959183673e-07,
-      "loss": 0.0647,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 0.4990352690219879,
+      "learning_rate": 3.826530612244898e-06,
+      "loss": 0.0663,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1210,
       "total_memory_available (GB)": 94.62
     },
     {
       "epoch": 4.979591836734694,
-      "grad_norm": 0.9617031812667847,
-      "learning_rate": 4.0816326530612253e-08,
-      "loss": 0.0418,
-      "max_memory_allocated (GB)": 60.52,
+      "grad_norm": 1.045630693435669,
+      "learning_rate": 3.7755102040816327e-06,
+      "loss": 0.0422,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
       "step": 1220,
       "total_memory_available (GB)": 94.62
     },
     {
-      "epoch": 5.0,
-      "max_memory_allocated (GB)": 60.52,
+      "epoch": 5.020408163265306,
+      "grad_norm": 3.719482660293579,
+      "learning_rate": 3.724489795918368e-06,
+      "loss": 0.0531,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1230,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.061224489795919,
+      "grad_norm": 0.6931941509246826,
+      "learning_rate": 3.6734693877551024e-06,
+      "loss": 0.0434,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1240,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.1020408163265305,
+      "grad_norm": 0.945284903049469,
+      "learning_rate": 3.6224489795918373e-06,
+      "loss": 0.0377,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1250,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.142857142857143,
+      "grad_norm": 0.49527707695961,
+      "learning_rate": 3.5714285714285718e-06,
+      "loss": 0.0406,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1260,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.183673469387755,
+      "grad_norm": 1.0614029169082642,
+      "learning_rate": 3.5204081632653062e-06,
+      "loss": 0.0614,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1270,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.224489795918367,
+      "grad_norm": 1.208749771118164,
+      "learning_rate": 3.469387755102041e-06,
+      "loss": 0.0449,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1280,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.26530612244898,
+      "grad_norm": 3.612487554550171,
+      "learning_rate": 3.4183673469387756e-06,
+      "loss": 0.0672,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1290,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.3061224489795915,
+      "grad_norm": 0.6228938102722168,
+      "learning_rate": 3.3673469387755105e-06,
+      "loss": 0.0516,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1300,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.346938775510204,
+      "grad_norm": 0.586557924747467,
+      "learning_rate": 3.316326530612245e-06,
+      "loss": 0.0674,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1310,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.387755102040816,
+      "grad_norm": 0.963624119758606,
+      "learning_rate": 3.2653061224489794e-06,
+      "loss": 0.0621,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1320,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.428571428571429,
+      "grad_norm": 1.1783013343811035,
+      "learning_rate": 3.2142857142857147e-06,
+      "loss": 0.0366,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1330,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.469387755102041,
+      "grad_norm": 4.429933547973633,
+      "learning_rate": 3.1632653061224496e-06,
+      "loss": 0.0511,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1340,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.510204081632653,
+      "grad_norm": 4.795422077178955,
+      "learning_rate": 3.112244897959184e-06,
+      "loss": 0.0601,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1350,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.551020408163265,
+      "grad_norm": 0.19068406522274017,
+      "learning_rate": 3.0612244897959185e-06,
+      "loss": 0.0479,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1360,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.591836734693878,
+      "grad_norm": 3.7448017597198486,
+      "learning_rate": 3.0102040816326534e-06,
+      "loss": 0.0404,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1370,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.63265306122449,
+      "grad_norm": 0.3368137776851654,
+      "learning_rate": 2.959183673469388e-06,
+      "loss": 0.0488,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1380,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.673469387755102,
+      "grad_norm": 0.14420035481452942,
+      "learning_rate": 2.908163265306123e-06,
+      "loss": 0.0582,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1390,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.714285714285714,
+      "grad_norm": 0.372368723154068,
+      "learning_rate": 2.8571428571428573e-06,
+      "loss": 0.0391,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1400,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.755102040816326,
+      "grad_norm": 3.4565131664276123,
+      "learning_rate": 2.8061224489795917e-06,
+      "loss": 0.0616,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1410,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.795918367346939,
+      "grad_norm": 3.389681339263916,
+      "learning_rate": 2.7551020408163266e-06,
+      "loss": 0.0675,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1420,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.836734693877551,
+      "grad_norm": 0.7100503444671631,
+      "learning_rate": 2.7040816326530615e-06,
+      "loss": 0.036,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1430,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.877551020408164,
+      "grad_norm": 0.40696802735328674,
+      "learning_rate": 2.6530612244897964e-06,
+      "loss": 0.0632,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1440,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.918367346938775,
+      "grad_norm": 0.7590793967247009,
+      "learning_rate": 2.602040816326531e-06,
+      "loss": 0.0549,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1450,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 5.959183673469388,
+      "grad_norm": 0.48597386479377747,
+      "learning_rate": 2.5510204081632657e-06,
+      "loss": 0.0393,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1460,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.0,
+      "grad_norm": 2.6455276012420654,
+      "learning_rate": 2.5e-06,
+      "loss": 0.0566,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1470,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.040816326530612,
+      "grad_norm": 2.350471258163452,
+      "learning_rate": 2.4489795918367347e-06,
+      "loss": 0.0509,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1480,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.081632653061225,
+      "grad_norm": 3.315977096557617,
+      "learning_rate": 2.3979591836734696e-06,
+      "loss": 0.0523,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1490,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.122448979591836,
+      "grad_norm": 1.6327887773513794,
+      "learning_rate": 2.3469387755102044e-06,
+      "loss": 0.0432,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1500,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.163265306122449,
+      "grad_norm": 5.029656410217285,
+      "learning_rate": 2.295918367346939e-06,
+      "loss": 0.0468,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1510,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.204081632653061,
+      "grad_norm": 3.1543941497802734,
+      "learning_rate": 2.244897959183674e-06,
+      "loss": 0.0471,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1520,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.244897959183674,
+      "grad_norm": 1.1178909540176392,
+      "learning_rate": 2.1938775510204083e-06,
+      "loss": 0.0538,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1530,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.285714285714286,
+      "grad_norm": 0.7284368276596069,
+      "learning_rate": 2.1428571428571427e-06,
+      "loss": 0.0474,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1540,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.326530612244898,
+      "grad_norm": 0.15386615693569183,
+      "learning_rate": 2.0918367346938776e-06,
+      "loss": 0.0327,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1550,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.36734693877551,
+      "grad_norm": 12.00415325164795,
+      "learning_rate": 2.0408163265306125e-06,
+      "loss": 0.0568,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1560,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.408163265306122,
+      "grad_norm": 0.14763076603412628,
+      "learning_rate": 1.989795918367347e-06,
+      "loss": 0.0389,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1570,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.448979591836735,
+      "grad_norm": 0.10665205121040344,
+      "learning_rate": 1.938775510204082e-06,
+      "loss": 0.0526,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1580,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.489795918367347,
+      "grad_norm": 0.6945566534996033,
+      "learning_rate": 1.8877551020408163e-06,
+      "loss": 0.0276,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1590,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.530612244897959,
+      "grad_norm": 0.6304193735122681,
+      "learning_rate": 1.8367346938775512e-06,
+      "loss": 0.0595,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1600,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.571428571428571,
+      "grad_norm": 0.738591194152832,
+      "learning_rate": 1.7857142857142859e-06,
+      "loss": 0.0591,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1610,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.612244897959184,
+      "grad_norm": 1.1249669790267944,
+      "learning_rate": 1.7346938775510206e-06,
+      "loss": 0.0444,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1620,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.653061224489796,
+      "grad_norm": 0.3204442858695984,
+      "learning_rate": 1.6836734693877552e-06,
+      "loss": 0.041,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1630,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.6938775510204085,
+      "grad_norm": 0.6603041887283325,
+      "learning_rate": 1.6326530612244897e-06,
+      "loss": 0.0485,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1640,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.73469387755102,
+      "grad_norm": 0.9167451858520508,
+      "learning_rate": 1.5816326530612248e-06,
+      "loss": 0.051,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1650,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.775510204081632,
+      "grad_norm": 1.1892409324645996,
+      "learning_rate": 1.5306122448979593e-06,
+      "loss": 0.0577,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1660,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.816326530612245,
+      "grad_norm": 1.1679530143737793,
+      "learning_rate": 1.479591836734694e-06,
+      "loss": 0.0581,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1670,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.857142857142857,
+      "grad_norm": 4.730435848236084,
+      "learning_rate": 1.4285714285714286e-06,
+      "loss": 0.058,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1680,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.8979591836734695,
+      "grad_norm": 2.7492659091949463,
+      "learning_rate": 1.3775510204081633e-06,
+      "loss": 0.0593,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1690,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.938775510204081,
+      "grad_norm": 0.29811447858810425,
+      "learning_rate": 1.3265306122448982e-06,
+      "loss": 0.0425,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1700,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 6.979591836734694,
+      "grad_norm": 0.15881459414958954,
+      "learning_rate": 1.2755102040816329e-06,
+      "loss": 0.0625,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1710,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.020408163265306,
+      "grad_norm": 0.20461368560791016,
+      "learning_rate": 1.2244897959183673e-06,
+      "loss": 0.0581,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1720,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.061224489795919,
+      "grad_norm": 1.1351808309555054,
+      "learning_rate": 1.1734693877551022e-06,
+      "loss": 0.0646,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1730,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.1020408163265305,
+      "grad_norm": 0.1654195487499237,
+      "learning_rate": 1.122448979591837e-06,
+      "loss": 0.0472,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1740,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.142857142857143,
+      "grad_norm": 0.4366483986377716,
+      "learning_rate": 1.0714285714285714e-06,
+      "loss": 0.0461,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1750,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.183673469387755,
+      "grad_norm": 0.5380903482437134,
+      "learning_rate": 1.0204081632653063e-06,
+      "loss": 0.0506,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1760,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.224489795918367,
+      "grad_norm": 1.661912441253662,
+      "learning_rate": 9.69387755102041e-07,
+      "loss": 0.0664,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1770,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.26530612244898,
+      "grad_norm": 0.4192713499069214,
+      "learning_rate": 9.183673469387756e-07,
+      "loss": 0.0394,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1780,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.3061224489795915,
+      "grad_norm": 0.6668973565101624,
+      "learning_rate": 8.673469387755103e-07,
+      "loss": 0.0401,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1790,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.346938775510204,
+      "grad_norm": 0.5573325753211975,
+      "learning_rate": 8.163265306122449e-07,
+      "loss": 0.0526,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1800,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.387755102040816,
+      "grad_norm": 0.39288291335105896,
+      "learning_rate": 7.653061224489796e-07,
+      "loss": 0.0445,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1810,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.428571428571429,
+      "grad_norm": 0.7398673892021179,
+      "learning_rate": 7.142857142857143e-07,
+      "loss": 0.054,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1820,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.469387755102041,
+      "grad_norm": 2.143411636352539,
+      "learning_rate": 6.632653061224491e-07,
+      "loss": 0.0458,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1830,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.510204081632653,
+      "grad_norm": 0.3958425223827362,
+      "learning_rate": 6.122448979591837e-07,
+      "loss": 0.0641,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1840,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.551020408163265,
+      "grad_norm": 2.797384023666382,
+      "learning_rate": 5.612244897959184e-07,
+      "loss": 0.0447,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1850,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.591836734693878,
+      "grad_norm": 1.5026339292526245,
+      "learning_rate": 5.102040816326531e-07,
+      "loss": 0.0274,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1860,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.63265306122449,
+      "grad_norm": 0.993212103843689,
+      "learning_rate": 4.591836734693878e-07,
+      "loss": 0.0393,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1870,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.673469387755102,
+      "grad_norm": 0.16298241913318634,
+      "learning_rate": 4.0816326530612243e-07,
+      "loss": 0.055,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1880,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.714285714285714,
+      "grad_norm": 4.067746639251709,
+      "learning_rate": 3.5714285714285716e-07,
+      "loss": 0.0661,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1890,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.755102040816326,
+      "grad_norm": 1.387778878211975,
+      "learning_rate": 3.0612244897959183e-07,
+      "loss": 0.0586,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1900,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.795918367346939,
+      "grad_norm": 0.6988309621810913,
+      "learning_rate": 2.5510204081632656e-07,
+      "loss": 0.0664,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1910,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.836734693877551,
+      "grad_norm": 0.7139838933944702,
+      "learning_rate": 2.0408163265306121e-07,
+      "loss": 0.053,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1920,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.877551020408164,
+      "grad_norm": 0.5550429224967957,
+      "learning_rate": 1.5306122448979592e-07,
+      "loss": 0.0458,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1930,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.918367346938775,
+      "grad_norm": 1.2351597547531128,
+      "learning_rate": 1.0204081632653061e-07,
+      "loss": 0.0471,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1940,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 7.959183673469388,
+      "grad_norm": 0.6292315125465393,
+      "learning_rate": 5.1020408163265303e-08,
+      "loss": 0.0532,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1950,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 8.0,
+      "grad_norm": 1.6500349044799805,
+      "learning_rate": 0.0,
+      "loss": 0.0453,
+      "max_memory_allocated (GB)": 57.18,
+      "memory_allocated (GB)": 50.57,
+      "step": 1960,
+      "total_memory_available (GB)": 94.62
+    },
+    {
+      "epoch": 8.0,
+      "max_memory_allocated (GB)": 57.18,
       "memory_allocated (GB)": 50.57,
-      "step": 1225,
-      "total_flos": 3.0598946525952e+16,
+      "step": 1960,
+      "total_flos": 4.89583144415232e+16,
       "total_memory_available (GB)": 94.62,
-      "train_loss": 0.06082177159737567,
-      "train_runtime": 1079.1005,
-      "train_samples_per_second": 52.128,
-      "train_steps_per_second": 1.304
+      "train_loss": 0.057532464606421335,
+      "train_runtime": 1666.2328,
+      "train_samples_per_second": 52.538,
+      "train_steps_per_second": 1.315
     }
   ],
   "logging_steps": 10,
-  "max_steps": 1225,
+  "max_steps": 1960,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 5,
+  "num_train_epochs": 8,
   "save_steps": 500,
   "stateful_callbacks": {
     "TrainerControl": {
@@ -1258,7 +1998,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 3.0598946525952e+16,
+  "total_flos": 4.89583144415232e+16,
   "train_batch_size": 40,
   "trial_name": null,
   "trial_params": null

validation_results.json

@@ -1,10 +1,10 @@
 {
-    "epoch": 5.0,
-    "eval_loss": 0.12328128516674042,
-    "eval_runtime": 24.5719,
-    "eval_samples_per_second": 36.301,
-    "eval_steps_per_second": 2.33,
-    "max_memory_allocated (GB)": 60.52,
+    "epoch": 8.0,
+    "eval_loss": 0.12359436601400375,
+    "eval_runtime": 24.2754,
+    "eval_samples_per_second": 44.827,
+    "eval_steps_per_second": 2.877,
+    "max_memory_allocated (GB)": 57.18,
     "memory_allocated (GB)": 51.27,
     "total_memory_available (GB)": 94.62
 }