diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,7710 +1,606 @@
 {
-  "best_metric": 0.3460787236690521,
-  "best_model_checkpoint": "outputs/checkpoint-672",
-  "epoch": 24.041085840058695,
-  "eval_steps": 8,
-  "global_step": 4096,
+  "best_metric": 0.43469348549842834,
+  "best_model_checkpoint": "outputs/checkpoint-512",
+  "epoch": 3.002932551319648,
+  "eval_steps": 32,
+  "global_step": 512,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
       "epoch": 0.05,
-      "grad_norm": 0.55859375,
-      "learning_rate": 1.25e-05,
-      "loss": 0.981,
-      "step": 8
-    },
-    {
-      "epoch": 0.05,
-      "eval_loss": 0.9143704771995544,
-      "eval_runtime": 4.8662,
-      "eval_samples_per_second": 9.864,
-      "eval_steps_per_second": 1.233,
-      "step": 8
-    },
-    {
-      "epoch": 0.09,
-      "grad_norm": 0.494140625,
+      "grad_norm": 0.103515625,
       "learning_rate": 2.5e-05,
-      "loss": 0.9604,
-      "step": 16
+      "loss": 1.2622,
+      "step": 8
     },
     {
       "epoch": 0.09,
-      "eval_loss": 0.8369872570037842,
-      "eval_runtime": 4.8785,
-      "eval_samples_per_second": 9.839,
-      "eval_steps_per_second": 1.23,
+      "grad_norm": 0.08984375,
+      "learning_rate": 5e-05,
+      "loss": 1.2695,
       "step": 16
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.447265625,
-      "learning_rate": 3.7500000000000003e-05,
-      "loss": 0.8553,
-      "step": 24
-    },
-    {
-      "epoch": 0.14,
-      "eval_loss": 0.7113822102546692,
-      "eval_runtime": 4.8736,
-      "eval_samples_per_second": 9.849,
-      "eval_steps_per_second": 1.231,
+      "grad_norm": 0.1181640625,
+      "learning_rate": 7.500000000000001e-05,
+      "loss": 1.1917,
       "step": 24
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.4453125,
-      "learning_rate": 5e-05,
-      "loss": 0.7358,
+      "grad_norm": 0.1689453125,
+      "learning_rate": 0.0001,
+      "loss": 1.0581,
       "step": 32
     },
     {
       "epoch": 0.19,
-      "eval_loss": 0.5980567336082458,
-      "eval_runtime": 4.8799,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.23,
+      "eval_loss": 0.8670921325683594,
+      "eval_runtime": 2.7914,
+      "eval_samples_per_second": 17.195,
+      "eval_steps_per_second": 2.149,
       "step": 32
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.33984375,
-      "learning_rate": 4.9901574803149606e-05,
-      "loss": 0.6122,
+      "grad_norm": 0.12158203125,
+      "learning_rate": 9.833333333333333e-05,
+      "loss": 0.8582,
       "step": 40
     },
-    {
-      "epoch": 0.23,
-      "eval_loss": 0.5387040972709656,
-      "eval_runtime": 4.8813,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
-      "step": 40
-    },
-    {
-      "epoch": 0.28,
-      "grad_norm": 0.353515625,
-      "learning_rate": 4.9803149606299216e-05,
-      "loss": 0.5819,
-      "step": 48
-    },
     {
       "epoch": 0.28,
-      "eval_loss": 0.5024240016937256,
-      "eval_runtime": 4.8808,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
+      "grad_norm": 0.09619140625,
+      "learning_rate": 9.666666666666667e-05,
+      "loss": 0.7709,
       "step": 48
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.328125,
-      "learning_rate": 4.970472440944882e-05,
-      "loss": 0.547,
-      "step": 56
-    },
-    {
-      "epoch": 0.33,
-      "eval_loss": 0.47695469856262207,
-      "eval_runtime": 4.8838,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.229,
+      "grad_norm": 0.08642578125,
+      "learning_rate": 9.5e-05,
+      "loss": 0.7284,
       "step": 56
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.345703125,
-      "learning_rate": 4.960629921259843e-05,
-      "loss": 0.5432,
+      "grad_norm": 0.1318359375,
+      "learning_rate": 9.333333333333334e-05,
+      "loss": 0.7043,
       "step": 64
     },
     {
       "epoch": 0.38,
-      "eval_loss": 0.46160802245140076,
-      "eval_runtime": 4.8825,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
+      "eval_loss": 0.6009237170219421,
+      "eval_runtime": 2.1819,
+      "eval_samples_per_second": 22.0,
+      "eval_steps_per_second": 2.75,
       "step": 64
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.328125,
-      "learning_rate": 4.950787401574803e-05,
-      "loss": 0.5187,
-      "step": 72
-    },
-    {
-      "epoch": 0.42,
-      "eval_loss": 0.4508754014968872,
-      "eval_runtime": 4.8793,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
+      "grad_norm": 0.0986328125,
+      "learning_rate": 9.166666666666667e-05,
+      "loss": 0.6693,
       "step": 72
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.341796875,
-      "learning_rate": 4.940944881889764e-05,
-      "loss": 0.5056,
+      "grad_norm": 0.09814453125,
+      "learning_rate": 9e-05,
+      "loss": 0.6409,
       "step": 80
     },
-    {
-      "epoch": 0.47,
-      "eval_loss": 0.4417424201965332,
-      "eval_runtime": 4.8827,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 80
-    },
-    {
-      "epoch": 0.52,
-      "grad_norm": 0.32421875,
-      "learning_rate": 4.9311023622047246e-05,
-      "loss": 0.4723,
-      "step": 88
-    },
     {
       "epoch": 0.52,
-      "eval_loss": 0.4341174364089966,
-      "eval_runtime": 4.8789,
-      "eval_samples_per_second": 9.838,
-      "eval_steps_per_second": 1.23,
+      "grad_norm": 0.1279296875,
+      "learning_rate": 8.833333333333333e-05,
+      "loss": 0.6038,
       "step": 88
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.349609375,
-      "learning_rate": 4.9212598425196856e-05,
-      "loss": 0.4813,
+      "grad_norm": 0.0947265625,
+      "learning_rate": 8.666666666666667e-05,
+      "loss": 0.6093,
       "step": 96
     },
     {
       "epoch": 0.56,
-      "eval_loss": 0.429939866065979,
-      "eval_runtime": 4.878,
-      "eval_samples_per_second": 9.84,
-      "eval_steps_per_second": 1.23,
+      "eval_loss": 0.5342517495155334,
+      "eval_runtime": 2.1808,
+      "eval_samples_per_second": 22.01,
+      "eval_steps_per_second": 2.751,
       "step": 96
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.333984375,
-      "learning_rate": 4.911417322834646e-05,
-      "loss": 0.4873,
+      "grad_norm": 0.1005859375,
+      "learning_rate": 8.5e-05,
+      "loss": 0.5974,
       "step": 104
     },
-    {
-      "epoch": 0.61,
-      "eval_loss": 0.42384615540504456,
-      "eval_runtime": 4.8764,
-      "eval_samples_per_second": 9.843,
-      "eval_steps_per_second": 1.23,
-      "step": 104
-    },
-    {
-      "epoch": 0.66,
-      "grad_norm": 0.376953125,
-      "learning_rate": 4.901574803149607e-05,
-      "loss": 0.464,
-      "step": 112
-    },
     {
       "epoch": 0.66,
-      "eval_loss": 0.41946908831596375,
-      "eval_runtime": 4.8802,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.229,
+      "grad_norm": 0.11474609375,
+      "learning_rate": 8.333333333333334e-05,
+      "loss": 0.5732,
       "step": 112
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.37109375,
-      "learning_rate": 4.891732283464567e-05,
-      "loss": 0.4859,
-      "step": 120
-    },
-    {
-      "epoch": 0.7,
-      "eval_loss": 0.41445252299308777,
-      "eval_runtime": 4.8881,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.227,
+      "grad_norm": 0.11669921875,
+      "learning_rate": 8.166666666666667e-05,
+      "loss": 0.5907,
       "step": 120
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.35546875,
-      "learning_rate": 4.881889763779528e-05,
-      "loss": 0.4579,
+      "grad_norm": 0.0966796875,
+      "learning_rate": 8e-05,
+      "loss": 0.5548,
       "step": 128
     },
     {
       "epoch": 0.75,
-      "eval_loss": 0.4118093252182007,
-      "eval_runtime": 4.8842,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.228,
+      "eval_loss": 0.5006037950515747,
+      "eval_runtime": 2.1797,
+      "eval_samples_per_second": 22.021,
+      "eval_steps_per_second": 2.753,
       "step": 128
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.33984375,
-      "learning_rate": 4.872047244094489e-05,
-      "loss": 0.4549,
-      "step": 136
-    },
-    {
-      "epoch": 0.8,
-      "eval_loss": 0.40753862261772156,
-      "eval_runtime": 4.8904,
-      "eval_samples_per_second": 9.815,
-      "eval_steps_per_second": 1.227,
+      "grad_norm": 0.095703125,
+      "learning_rate": 7.833333333333333e-05,
+      "loss": 0.5563,
       "step": 136
     },
     {
-      "epoch": 0.85,
-      "grad_norm": 0.365234375,
-      "learning_rate": 4.862204724409449e-05,
-      "loss": 0.4374,
+      "epoch": 0.84,
+      "grad_norm": 0.1142578125,
+      "learning_rate": 7.666666666666667e-05,
+      "loss": 0.5311,
       "step": 144
     },
-    {
-      "epoch": 0.85,
-      "eval_loss": 0.4049430787563324,
-      "eval_runtime": 4.8885,
-      "eval_samples_per_second": 9.819,
-      "eval_steps_per_second": 1.227,
-      "step": 144
-    },
-    {
-      "epoch": 0.89,
-      "grad_norm": 0.369140625,
-      "learning_rate": 4.85236220472441e-05,
-      "loss": 0.438,
-      "step": 152
-    },
     {
       "epoch": 0.89,
-      "eval_loss": 0.4011856019496918,
-      "eval_runtime": 4.8787,
-      "eval_samples_per_second": 9.839,
-      "eval_steps_per_second": 1.23,
+      "grad_norm": 0.126953125,
+      "learning_rate": 7.500000000000001e-05,
+      "loss": 0.5284,
       "step": 152
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.333984375,
-      "learning_rate": 4.84251968503937e-05,
-      "loss": 0.4347,
+      "grad_norm": 0.09814453125,
+      "learning_rate": 7.333333333333333e-05,
+      "loss": 0.5316,
       "step": 160
     },
     {
       "epoch": 0.94,
-      "eval_loss": 0.39831605553627014,
-      "eval_runtime": 4.8796,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
+      "eval_loss": 0.48313403129577637,
+      "eval_runtime": 2.1847,
+      "eval_samples_per_second": 21.971,
+      "eval_steps_per_second": 2.746,
       "step": 160
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.37109375,
-      "learning_rate": 4.832677165354331e-05,
-      "loss": 0.4388,
-      "step": 168
-    },
-    {
-      "epoch": 0.99,
-      "eval_loss": 0.39434146881103516,
-      "eval_runtime": 4.8874,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
+      "grad_norm": 0.09912109375,
+      "learning_rate": 7.166666666666667e-05,
+      "loss": 0.5324,
       "step": 168
     },
     {
       "epoch": 1.03,
-      "grad_norm": 0.38671875,
-      "learning_rate": 4.822834645669291e-05,
-      "loss": 0.4519,
-      "step": 176
-    },
-    {
-      "epoch": 1.03,
-      "eval_loss": 0.39422157406806946,
-      "eval_runtime": 4.8818,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
+      "grad_norm": 0.1455078125,
+      "learning_rate": 7e-05,
+      "loss": 0.5456,
       "step": 176
     },
     {
       "epoch": 1.08,
-      "grad_norm": 0.34375,
-      "learning_rate": 4.812992125984252e-05,
-      "loss": 0.4273,
-      "step": 184
-    },
-    {
-      "epoch": 1.08,
-      "eval_loss": 0.3909244239330292,
-      "eval_runtime": 4.8834,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
+      "grad_norm": 0.1025390625,
+      "learning_rate": 6.833333333333333e-05,
+      "loss": 0.5274,
       "step": 184
     },
     {
       "epoch": 1.13,
-      "grad_norm": 0.3828125,
-      "learning_rate": 4.8031496062992124e-05,
-      "loss": 0.4169,
+      "grad_norm": 0.11181640625,
+      "learning_rate": 6.666666666666667e-05,
+      "loss": 0.5161,
       "step": 192
     },
     {
       "epoch": 1.13,
-      "eval_loss": 0.38786780834198,
-      "eval_runtime": 4.8897,
-      "eval_samples_per_second": 9.816,
-      "eval_steps_per_second": 1.227,
+      "eval_loss": 0.4703657329082489,
+      "eval_runtime": 2.1798,
+      "eval_samples_per_second": 22.021,
+      "eval_steps_per_second": 2.753,
       "step": 192
     },
     {
       "epoch": 1.17,
-      "grad_norm": 0.3828125,
-      "learning_rate": 4.7933070866141734e-05,
-      "loss": 0.4217,
-      "step": 200
-    },
-    {
-      "epoch": 1.17,
-      "eval_loss": 0.38564667105674744,
-      "eval_runtime": 4.884,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.229,
+      "grad_norm": 0.12060546875,
+      "learning_rate": 6.500000000000001e-05,
+      "loss": 0.5129,
       "step": 200
     },
     {
       "epoch": 1.22,
-      "grad_norm": 0.404296875,
-      "learning_rate": 4.783464566929134e-05,
-      "loss": 0.4473,
-      "step": 208
-    },
-    {
-      "epoch": 1.22,
-      "eval_loss": 0.3840738534927368,
-      "eval_runtime": 4.8763,
-      "eval_samples_per_second": 9.844,
-      "eval_steps_per_second": 1.23,
+      "grad_norm": 0.11083984375,
+      "learning_rate": 6.333333333333333e-05,
+      "loss": 0.5455,
       "step": 208
     },
     {
       "epoch": 1.27,
-      "grad_norm": 0.357421875,
-      "learning_rate": 4.773622047244095e-05,
-      "loss": 0.4158,
-      "step": 216
-    },
-    {
-      "epoch": 1.27,
-      "eval_loss": 0.3802422285079956,
-      "eval_runtime": 4.8799,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.23,
+      "grad_norm": 0.1142578125,
+      "learning_rate": 6.166666666666667e-05,
+      "loss": 0.5113,
       "step": 216
     },
     {
       "epoch": 1.31,
-      "grad_norm": 0.3671875,
-      "learning_rate": 4.763779527559055e-05,
-      "loss": 0.4112,
+      "grad_norm": 0.134765625,
+      "learning_rate": 6e-05,
+      "loss": 0.4937,
       "step": 224
     },
     {
       "epoch": 1.31,
-      "eval_loss": 0.38026484847068787,
-      "eval_runtime": 4.8808,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
+      "eval_loss": 0.4608299434185028,
+      "eval_runtime": 2.1784,
+      "eval_samples_per_second": 22.035,
+      "eval_steps_per_second": 2.754,
       "step": 224
     },
     {
       "epoch": 1.36,
-      "grad_norm": 0.37109375,
-      "learning_rate": 4.753937007874016e-05,
-      "loss": 0.4317,
-      "step": 232
-    },
-    {
-      "epoch": 1.36,
-      "eval_loss": 0.3779226839542389,
-      "eval_runtime": 4.8807,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
+      "grad_norm": 0.10791015625,
+      "learning_rate": 5.833333333333334e-05,
+      "loss": 0.5279,
       "step": 232
     },
     {
       "epoch": 1.41,
-      "grad_norm": 0.408203125,
-      "learning_rate": 4.7440944881889765e-05,
-      "loss": 0.4105,
-      "step": 240
-    },
-    {
-      "epoch": 1.41,
-      "eval_loss": 0.37829089164733887,
-      "eval_runtime": 4.8824,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
+      "grad_norm": 0.1181640625,
+      "learning_rate": 5.666666666666667e-05,
+      "loss": 0.5057,
       "step": 240
     },
     {
-      "epoch": 1.46,
-      "grad_norm": 0.41015625,
-      "learning_rate": 4.7342519685039375e-05,
-      "loss": 0.4015,
-      "step": 248
-    },
-    {
-      "epoch": 1.46,
-      "eval_loss": 0.3772072494029999,
-      "eval_runtime": 4.8791,
-      "eval_samples_per_second": 9.838,
-      "eval_steps_per_second": 1.23,
+      "epoch": 1.45,
+      "grad_norm": 0.11328125,
+      "learning_rate": 5.500000000000001e-05,
+      "loss": 0.4863,
       "step": 248
     },
     {
       "epoch": 1.5,
-      "grad_norm": 0.451171875,
-      "learning_rate": 4.724409448818898e-05,
-      "loss": 0.4262,
+      "grad_norm": 0.1376953125,
+      "learning_rate": 5.333333333333333e-05,
+      "loss": 0.5236,
       "step": 256
     },
     {
       "epoch": 1.5,
-      "eval_loss": 0.37624046206474304,
-      "eval_runtime": 4.8828,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
+      "eval_loss": 0.454235315322876,
+      "eval_runtime": 2.1818,
+      "eval_samples_per_second": 22.0,
+      "eval_steps_per_second": 2.75,
       "step": 256
     },
     {
       "epoch": 1.55,
-      "grad_norm": 0.361328125,
-      "learning_rate": 4.714566929133858e-05,
-      "loss": 0.4135,
-      "step": 264
-    },
-    {
-      "epoch": 1.55,
-      "eval_loss": 0.37519559264183044,
-      "eval_runtime": 4.8797,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
+      "grad_norm": 0.11181640625,
+      "learning_rate": 5.166666666666667e-05,
+      "loss": 0.5,
       "step": 264
     },
     {
       "epoch": 1.6,
-      "grad_norm": 0.4453125,
-      "learning_rate": 4.704724409448819e-05,
-      "loss": 0.4099,
-      "step": 272
-    },
-    {
-      "epoch": 1.6,
-      "eval_loss": 0.37314102053642273,
-      "eval_runtime": 4.8876,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
+      "grad_norm": 0.1259765625,
+      "learning_rate": 5e-05,
+      "loss": 0.4989,
       "step": 272
     },
     {
       "epoch": 1.64,
-      "grad_norm": 0.408203125,
-      "learning_rate": 4.6948818897637795e-05,
-      "loss": 0.4074,
-      "step": 280
-    },
-    {
-      "epoch": 1.64,
-      "eval_loss": 0.3711419999599457,
-      "eval_runtime": 4.8865,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
+      "grad_norm": 0.12451171875,
+      "learning_rate": 4.8333333333333334e-05,
+      "loss": 0.4944,
       "step": 280
     },
     {
       "epoch": 1.69,
-      "grad_norm": 0.41015625,
-      "learning_rate": 4.6850393700787405e-05,
-      "loss": 0.4203,
+      "grad_norm": 0.11962890625,
+      "learning_rate": 4.666666666666667e-05,
+      "loss": 0.5064,
       "step": 288
     },
     {
       "epoch": 1.69,
-      "eval_loss": 0.3701172173023224,
-      "eval_runtime": 4.8785,
-      "eval_samples_per_second": 9.839,
-      "eval_steps_per_second": 1.23,
+      "eval_loss": 0.4474702775478363,
+      "eval_runtime": 2.1767,
+      "eval_samples_per_second": 22.052,
+      "eval_steps_per_second": 2.756,
       "step": 288
     },
     {
       "epoch": 1.74,
-      "grad_norm": 0.35546875,
-      "learning_rate": 4.675196850393701e-05,
-      "loss": 0.4384,
-      "step": 296
-    },
-    {
-      "epoch": 1.74,
-      "eval_loss": 0.36824676394462585,
-      "eval_runtime": 4.8786,
-      "eval_samples_per_second": 9.839,
-      "eval_steps_per_second": 1.23,
+      "grad_norm": 0.103515625,
+      "learning_rate": 4.5e-05,
+      "loss": 0.5323,
       "step": 296
     },
     {
       "epoch": 1.78,
-      "grad_norm": 0.43359375,
-      "learning_rate": 4.665354330708662e-05,
-      "loss": 0.4047,
-      "step": 304
-    },
-    {
-      "epoch": 1.78,
-      "eval_loss": 0.36743244528770447,
-      "eval_runtime": 4.8787,
-      "eval_samples_per_second": 9.839,
-      "eval_steps_per_second": 1.23,
+      "grad_norm": 0.1328125,
+      "learning_rate": 4.3333333333333334e-05,
+      "loss": 0.4955,
       "step": 304
     },
     {
       "epoch": 1.83,
-      "grad_norm": 0.33984375,
-      "learning_rate": 4.655511811023622e-05,
-      "loss": 0.4196,
-      "step": 312
-    },
-    {
-      "epoch": 1.83,
-      "eval_loss": 0.36576294898986816,
-      "eval_runtime": 4.8821,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
+      "grad_norm": 0.10546875,
+      "learning_rate": 4.166666666666667e-05,
+      "loss": 0.5137,
       "step": 312
     },
     {
       "epoch": 1.88,
-      "grad_norm": 0.423828125,
-      "learning_rate": 4.645669291338583e-05,
-      "loss": 0.4044,
+      "grad_norm": 0.1337890625,
+      "learning_rate": 4e-05,
+      "loss": 0.497,
       "step": 320
     },
     {
       "epoch": 1.88,
-      "eval_loss": 0.3646230697631836,
-      "eval_runtime": 4.8839,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.229,
+      "eval_loss": 0.442803293466568,
+      "eval_runtime": 2.1775,
+      "eval_samples_per_second": 22.044,
+      "eval_steps_per_second": 2.755,
       "step": 320
     },
     {
-      "epoch": 1.93,
-      "grad_norm": 0.39453125,
-      "learning_rate": 4.6358267716535436e-05,
-      "loss": 0.4169,
-      "step": 328
-    },
-    {
-      "epoch": 1.93,
-      "eval_loss": 0.363146036863327,
-      "eval_runtime": 4.8824,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
+      "epoch": 1.92,
+      "grad_norm": 0.10791015625,
+      "learning_rate": 3.8333333333333334e-05,
+      "loss": 0.5032,
       "step": 328
     },
     {
       "epoch": 1.97,
-      "grad_norm": 0.41015625,
-      "learning_rate": 4.6259842519685046e-05,
-      "loss": 0.4047,
-      "step": 336
-    },
-    {
-      "epoch": 1.97,
-      "eval_loss": 0.36223217844963074,
-      "eval_runtime": 4.8806,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
+      "grad_norm": 0.130859375,
+      "learning_rate": 3.6666666666666666e-05,
+      "loss": 0.4914,
       "step": 336
     },
     {
       "epoch": 2.02,
-      "grad_norm": 0.3515625,
-      "learning_rate": 4.616141732283465e-05,
-      "loss": 0.4016,
-      "step": 344
-    },
-    {
-      "epoch": 2.02,
-      "eval_loss": 0.3615216314792633,
-      "eval_runtime": 4.8812,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
+      "grad_norm": 0.1171875,
+      "learning_rate": 3.5e-05,
+      "loss": 0.5013,
       "step": 344
     },
     {
-      "epoch": 2.07,
-      "grad_norm": 0.447265625,
-      "learning_rate": 4.606299212598425e-05,
-      "loss": 0.3953,
+      "epoch": 2.06,
+      "grad_norm": 0.1328125,
+      "learning_rate": 3.3333333333333335e-05,
+      "loss": 0.4933,
       "step": 352
     },
     {
-      "epoch": 2.07,
-      "eval_loss": 0.3632339537143707,
-      "eval_runtime": 4.8829,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
+      "epoch": 2.06,
+      "eval_loss": 0.4411480128765106,
+      "eval_runtime": 2.1813,
+      "eval_samples_per_second": 22.005,
+      "eval_steps_per_second": 2.751,
       "step": 352
     },
     {
       "epoch": 2.11,
-      "grad_norm": 0.384765625,
-      "learning_rate": 4.596456692913386e-05,
-      "loss": 0.3771,
-      "step": 360
-    },
-    {
-      "epoch": 2.11,
-      "eval_loss": 0.36078453063964844,
-      "eval_runtime": 4.8875,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
+      "grad_norm": 0.1201171875,
+      "learning_rate": 3.1666666666666666e-05,
+      "loss": 0.4741,
       "step": 360
     },
     {
       "epoch": 2.16,
-      "grad_norm": 0.40625,
-      "learning_rate": 4.5866141732283466e-05,
-      "loss": 0.3837,
-      "step": 368
-    },
-    {
-      "epoch": 2.16,
-      "eval_loss": 0.36075982451438904,
-      "eval_runtime": 4.8905,
-      "eval_samples_per_second": 9.815,
-      "eval_steps_per_second": 1.227,
+      "grad_norm": 0.1181640625,
+      "learning_rate": 3e-05,
+      "loss": 0.4802,
       "step": 368
     },
     {
       "epoch": 2.21,
-      "grad_norm": 0.42578125,
-      "learning_rate": 4.5767716535433076e-05,
-      "loss": 0.396,
-      "step": 376
-    },
-    {
-      "epoch": 2.21,
-      "eval_loss": 0.36021366715431213,
-      "eval_runtime": 4.8856,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
+      "grad_norm": 0.1181640625,
+      "learning_rate": 2.8333333333333335e-05,
+      "loss": 0.4911,
       "step": 376
     },
     {
       "epoch": 2.25,
-      "grad_norm": 0.423828125,
-      "learning_rate": 4.566929133858268e-05,
-      "loss": 0.4079,
+      "grad_norm": 0.1455078125,
+      "learning_rate": 2.6666666666666667e-05,
+      "loss": 0.517,
       "step": 384
     },
     {
       "epoch": 2.25,
-      "eval_loss": 0.36072564125061035,
-      "eval_runtime": 4.8735,
-      "eval_samples_per_second": 9.849,
-      "eval_steps_per_second": 1.231,
+      "eval_loss": 0.4384828805923462,
+      "eval_runtime": 2.1808,
+      "eval_samples_per_second": 22.01,
+      "eval_steps_per_second": 2.751,
       "step": 384
     },
     {
       "epoch": 2.3,
-      "grad_norm": 0.4609375,
-      "learning_rate": 4.557086614173229e-05,
-      "loss": 0.4058,
-      "step": 392
-    },
-    {
-      "epoch": 2.3,
-      "eval_loss": 0.35943296551704407,
-      "eval_runtime": 4.8875,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
+      "grad_norm": 0.1318359375,
+      "learning_rate": 2.5e-05,
+      "loss": 0.5024,
       "step": 392
     },
     {
       "epoch": 2.35,
-      "grad_norm": 0.4140625,
-      "learning_rate": 4.547244094488189e-05,
-      "loss": 0.3831,
-      "step": 400
-    },
-    {
-      "epoch": 2.35,
-      "eval_loss": 0.35869550704956055,
-      "eval_runtime": 4.8806,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
+      "grad_norm": 0.1259765625,
+      "learning_rate": 2.3333333333333336e-05,
+      "loss": 0.4802,
       "step": 400
     },
     {
       "epoch": 2.39,
-      "grad_norm": 0.404296875,
-      "learning_rate": 4.53740157480315e-05,
-      "loss": 0.3945,
-      "step": 408
-    },
-    {
-      "epoch": 2.39,
-      "eval_loss": 0.3580925762653351,
-      "eval_runtime": 4.8786,
-      "eval_samples_per_second": 9.839,
-      "eval_steps_per_second": 1.23,
+      "grad_norm": 0.12109375,
+      "learning_rate": 2.1666666666666667e-05,
+      "loss": 0.4857,
       "step": 408
     },
     {
       "epoch": 2.44,
-      "grad_norm": 0.40625,
-      "learning_rate": 4.52755905511811e-05,
-      "loss": 0.3953,
+      "grad_norm": 0.11962890625,
+      "learning_rate": 2e-05,
+      "loss": 0.4949,
       "step": 416
     },
     {
       "epoch": 2.44,
-      "eval_loss": 0.3581734001636505,
-      "eval_runtime": 4.8757,
-      "eval_samples_per_second": 9.845,
-      "eval_steps_per_second": 1.231,
+      "eval_loss": 0.4360348880290985,
+      "eval_runtime": 2.179,
+      "eval_samples_per_second": 22.029,
+      "eval_steps_per_second": 2.754,
       "step": 416
     },
     {
       "epoch": 2.49,
-      "grad_norm": 0.435546875,
-      "learning_rate": 4.517716535433071e-05,
-      "loss": 0.4043,
-      "step": 424
-    },
-    {
-      "epoch": 2.49,
-      "eval_loss": 0.3573249280452728,
-      "eval_runtime": 4.8833,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
+      "grad_norm": 0.12890625,
+      "learning_rate": 1.8333333333333333e-05,
+      "loss": 0.501,
       "step": 424
     },
     {
-      "epoch": 2.54,
-      "grad_norm": 0.404296875,
-      "learning_rate": 4.507874015748031e-05,
-      "loss": 0.3674,
-      "step": 432
-    },
-    {
-      "epoch": 2.54,
-      "eval_loss": 0.358001708984375,
-      "eval_runtime": 4.8899,
-      "eval_samples_per_second": 9.816,
-      "eval_steps_per_second": 1.227,
+      "epoch": 2.53,
+      "grad_norm": 0.1318359375,
+      "learning_rate": 1.6666666666666667e-05,
+      "loss": 0.4595,
       "step": 432
     },
     {
       "epoch": 2.58,
-      "grad_norm": 0.427734375,
-      "learning_rate": 4.4980314960629924e-05,
-      "loss": 0.3783,
-      "step": 440
-    },
-    {
-      "epoch": 2.58,
-      "eval_loss": 0.35629966855049133,
-      "eval_runtime": 4.8821,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
+      "grad_norm": 0.11962890625,
+      "learning_rate": 1.5e-05,
+      "loss": 0.4715,
       "step": 440
     },
     {
       "epoch": 2.63,
-      "grad_norm": 0.44921875,
-      "learning_rate": 4.488188976377953e-05,
-      "loss": 0.3981,
+      "grad_norm": 0.12255859375,
+      "learning_rate": 1.3333333333333333e-05,
+      "loss": 0.4946,
       "step": 448
     },
     {
       "epoch": 2.63,
-      "eval_loss": 0.35627031326293945,
-      "eval_runtime": 4.8857,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
+      "eval_loss": 0.4355371296405792,
+      "eval_runtime": 2.1775,
+      "eval_samples_per_second": 22.044,
+      "eval_steps_per_second": 2.755,
       "step": 448
     },
     {
-      "epoch": 2.68,
-      "grad_norm": 0.4609375,
-      "learning_rate": 4.478346456692914e-05,
-      "loss": 0.3795,
-      "step": 456
-    },
-    {
-      "epoch": 2.68,
-      "eval_loss": 0.3570912182331085,
-      "eval_runtime": 4.883,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
+      "epoch": 2.67,
+      "grad_norm": 0.123046875,
+      "learning_rate": 1.1666666666666668e-05,
+      "loss": 0.4692,
       "step": 456
     },
     {
       "epoch": 2.72,
-      "grad_norm": 0.453125,
-      "learning_rate": 4.468503937007874e-05,
-      "loss": 0.3744,
-      "step": 464
-    },
-    {
-      "epoch": 2.72,
-      "eval_loss": 0.35693350434303284,
-      "eval_runtime": 4.8794,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
+      "grad_norm": 0.125,
+      "learning_rate": 1e-05,
+      "loss": 0.4751,
       "step": 464
     },
     {
       "epoch": 2.77,
-      "grad_norm": 0.416015625,
-      "learning_rate": 4.4586614173228344e-05,
-      "loss": 0.4174,
-      "step": 472
-    },
-    {
-      "epoch": 2.77,
-      "eval_loss": 0.35543546080589294,
-      "eval_runtime": 4.88,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.23,
+      "grad_norm": 0.12060546875,
+      "learning_rate": 8.333333333333334e-06,
+      "loss": 0.5144,
       "step": 472
     },
     {
       "epoch": 2.82,
-      "grad_norm": 0.423828125,
-      "learning_rate": 4.4488188976377954e-05,
-      "loss": 0.3639,
+      "grad_norm": 0.119140625,
+      "learning_rate": 6.666666666666667e-06,
+      "loss": 0.4589,
       "step": 480
     },
     {
       "epoch": 2.82,
-      "eval_loss": 0.35410675406455994,
-      "eval_runtime": 4.8843,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
+      "eval_loss": 0.4347504675388336,
+      "eval_runtime": 2.1766,
+      "eval_samples_per_second": 22.053,
+      "eval_steps_per_second": 2.757,
       "step": 480
     },
     {
       "epoch": 2.86,
-      "grad_norm": 0.40625,
-      "learning_rate": 4.438976377952756e-05,
-      "loss": 0.3877,
-      "step": 488
-    },
-    {
-      "epoch": 2.86,
-      "eval_loss": 0.35232722759246826,
-      "eval_runtime": 4.8764,
-      "eval_samples_per_second": 9.843,
-      "eval_steps_per_second": 1.23,
+      "grad_norm": 0.1162109375,
+      "learning_rate": 5e-06,
+      "loss": 0.4768,
       "step": 488
     },
     {
       "epoch": 2.91,
-      "grad_norm": 0.484375,
-      "learning_rate": 4.429133858267717e-05,
-      "loss": 0.3805,
-      "step": 496
-    },
-    {
-      "epoch": 2.91,
-      "eval_loss": 0.3522258996963501,
-      "eval_runtime": 4.8775,
-      "eval_samples_per_second": 9.841,
-      "eval_steps_per_second": 1.23,
+      "grad_norm": 0.130859375,
+      "learning_rate": 3.3333333333333333e-06,
+      "loss": 0.4795,
       "step": 496
     },
     {
       "epoch": 2.96,
-      "grad_norm": 0.423828125,
-      "learning_rate": 4.419291338582677e-05,
-      "loss": 0.3964,
-      "step": 504
-    },
-    {
-      "epoch": 2.96,
-      "eval_loss": 0.3513607680797577,
-      "eval_runtime": 4.8776,
-      "eval_samples_per_second": 9.841,
-      "eval_steps_per_second": 1.23,
+      "grad_norm": 0.1201171875,
+      "learning_rate": 1.6666666666666667e-06,
+      "loss": 0.4992,
       "step": 504
     },
     {
-      "epoch": 3.01,
-      "grad_norm": 0.42578125,
-      "learning_rate": 4.409448818897638e-05,
-      "loss": 0.3824,
+      "epoch": 3.0,
+      "grad_norm": 0.11328125,
+      "learning_rate": 0.0,
+      "loss": 0.466,
       "step": 512
     },
     {
-      "epoch": 3.01,
-      "eval_loss": 0.350473016500473,
-      "eval_runtime": 4.8912,
-      "eval_samples_per_second": 9.814,
-      "eval_steps_per_second": 1.227,
+      "epoch": 3.0,
+      "eval_loss": 0.43469348549842834,
+      "eval_runtime": 2.179,
+      "eval_samples_per_second": 22.029,
+      "eval_steps_per_second": 2.754,
       "step": 512
     },
     {
-      "epoch": 3.05,
-      "grad_norm": 0.5078125,
-      "learning_rate": 4.3996062992125984e-05,
-      "loss": 0.3683,
-      "step": 520
-    },
-    {
-      "epoch": 3.05,
-      "eval_loss": 0.3539038896560669,
-      "eval_runtime": 4.8794,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
-      "step": 520
-    },
-    {
-      "epoch": 3.1,
-      "grad_norm": 0.44140625,
-      "learning_rate": 4.3897637795275594e-05,
-      "loss": 0.3829,
-      "step": 528
-    },
-    {
-      "epoch": 3.1,
-      "eval_loss": 0.3512038290500641,
-      "eval_runtime": 4.8797,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
-      "step": 528
-    },
-    {
-      "epoch": 3.15,
-      "grad_norm": 0.44921875,
-      "learning_rate": 4.37992125984252e-05,
-      "loss": 0.3675,
-      "step": 536
-    },
-    {
-      "epoch": 3.15,
-      "eval_loss": 0.3524076044559479,
-      "eval_runtime": 4.8842,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.228,
-      "step": 536
-    },
-    {
-      "epoch": 3.19,
-      "grad_norm": 0.46484375,
-      "learning_rate": 4.370078740157481e-05,
-      "loss": 0.3751,
-      "step": 544
-    },
-    {
-      "epoch": 3.19,
-      "eval_loss": 0.351447194814682,
-      "eval_runtime": 4.8802,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.229,
-      "step": 544
-    },
-    {
-      "epoch": 3.24,
-      "grad_norm": 0.44140625,
-      "learning_rate": 4.360236220472441e-05,
-      "loss": 0.3767,
-      "step": 552
-    },
-    {
-      "epoch": 3.24,
-      "eval_loss": 0.35135617852211,
-      "eval_runtime": 4.8875,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 552
-    },
-    {
-      "epoch": 3.29,
-      "grad_norm": 0.474609375,
-      "learning_rate": 4.350393700787402e-05,
-      "loss": 0.347,
-      "step": 560
-    },
-    {
-      "epoch": 3.29,
-      "eval_loss": 0.35124826431274414,
-      "eval_runtime": 4.879,
-      "eval_samples_per_second": 9.838,
-      "eval_steps_per_second": 1.23,
-      "step": 560
-    },
-    {
-      "epoch": 3.33,
-      "grad_norm": 0.458984375,
-      "learning_rate": 4.3405511811023625e-05,
-      "loss": 0.3565,
-      "step": 568
-    },
-    {
-      "epoch": 3.33,
-      "eval_loss": 0.3507702648639679,
-      "eval_runtime": 4.88,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.23,
-      "step": 568
-    },
-    {
-      "epoch": 3.38,
-      "grad_norm": 0.44921875,
-      "learning_rate": 4.330708661417323e-05,
-      "loss": 0.3986,
-      "step": 576
-    },
-    {
-      "epoch": 3.38,
-      "eval_loss": 0.35253670811653137,
-      "eval_runtime": 4.887,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 576
-    },
-    {
-      "epoch": 3.43,
-      "grad_norm": 0.458984375,
-      "learning_rate": 4.320866141732284e-05,
-      "loss": 0.3802,
-      "step": 584
-    },
-    {
-      "epoch": 3.43,
-      "eval_loss": 0.35116323828697205,
-      "eval_runtime": 4.8807,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 584
-    },
-    {
-      "epoch": 3.47,
-      "grad_norm": 0.451171875,
-      "learning_rate": 4.311023622047244e-05,
-      "loss": 0.3556,
-      "step": 592
-    },
-    {
-      "epoch": 3.47,
-      "eval_loss": 0.35001620650291443,
-      "eval_runtime": 4.8879,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.228,
-      "step": 592
-    },
-    {
-      "epoch": 3.52,
-      "grad_norm": 0.451171875,
-      "learning_rate": 4.301181102362205e-05,
-      "loss": 0.3681,
-      "step": 600
-    },
-    {
-      "epoch": 3.52,
-      "eval_loss": 0.3492867946624756,
-      "eval_runtime": 4.8869,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 600
-    },
-    {
-      "epoch": 3.57,
-      "grad_norm": 0.478515625,
-      "learning_rate": 4.2913385826771655e-05,
-      "loss": 0.3753,
-      "step": 608
-    },
-    {
-      "epoch": 3.57,
-      "eval_loss": 0.3498481810092926,
-      "eval_runtime": 4.8796,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
-      "step": 608
-    },
-    {
-      "epoch": 3.62,
-      "grad_norm": 0.56640625,
-      "learning_rate": 4.2814960629921265e-05,
-      "loss": 0.3694,
-      "step": 616
-    },
-    {
-      "epoch": 3.62,
-      "eval_loss": 0.3518795073032379,
-      "eval_runtime": 4.8802,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.229,
-      "step": 616
-    },
-    {
-      "epoch": 3.66,
-      "grad_norm": 0.470703125,
-      "learning_rate": 4.271653543307087e-05,
-      "loss": 0.3619,
-      "step": 624
-    },
-    {
-      "epoch": 3.66,
-      "eval_loss": 0.35094156861305237,
-      "eval_runtime": 4.8819,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 624
-    },
-    {
-      "epoch": 3.71,
-      "grad_norm": 0.494140625,
-      "learning_rate": 4.261811023622048e-05,
-      "loss": 0.3703,
-      "step": 632
-    },
-    {
-      "epoch": 3.71,
-      "eval_loss": 0.3496703803539276,
-      "eval_runtime": 4.8781,
-      "eval_samples_per_second": 9.84,
-      "eval_steps_per_second": 1.23,
-      "step": 632
-    },
-    {
-      "epoch": 3.76,
-      "grad_norm": 0.5234375,
-      "learning_rate": 4.251968503937008e-05,
-      "loss": 0.3391,
-      "step": 640
-    },
-    {
-      "epoch": 3.76,
-      "eval_loss": 0.3481089770793915,
-      "eval_runtime": 4.8794,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
-      "step": 640
-    },
-    {
-      "epoch": 3.8,
-      "grad_norm": 0.46875,
-      "learning_rate": 4.242125984251969e-05,
-      "loss": 0.3731,
-      "step": 648
-    },
-    {
-      "epoch": 3.8,
-      "eval_loss": 0.3471480906009674,
-      "eval_runtime": 4.8836,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 648
-    },
-    {
-      "epoch": 3.85,
-      "grad_norm": 0.47265625,
-      "learning_rate": 4.2322834645669296e-05,
-      "loss": 0.3825,
-      "step": 656
-    },
-    {
-      "epoch": 3.85,
-      "eval_loss": 0.34693393111228943,
-      "eval_runtime": 4.884,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.228,
-      "step": 656
-    },
-    {
-      "epoch": 3.9,
-      "grad_norm": 0.48046875,
-      "learning_rate": 4.22244094488189e-05,
-      "loss": 0.3597,
-      "step": 664
-    },
-    {
-      "epoch": 3.9,
-      "eval_loss": 0.34674397110939026,
-      "eval_runtime": 4.8793,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
-      "step": 664
-    },
-    {
-      "epoch": 3.94,
-      "grad_norm": 0.470703125,
-      "learning_rate": 4.21259842519685e-05,
-      "loss": 0.3801,
-      "step": 672
-    },
-    {
-      "epoch": 3.94,
-      "eval_loss": 0.3460787236690521,
-      "eval_runtime": 4.8876,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 672
-    },
-    {
-      "epoch": 3.99,
-      "grad_norm": 0.5,
-      "learning_rate": 4.202755905511811e-05,
-      "loss": 0.3605,
-      "step": 680
-    },
-    {
-      "epoch": 3.99,
-      "eval_loss": 0.3465350568294525,
-      "eval_runtime": 4.8783,
-      "eval_samples_per_second": 9.84,
-      "eval_steps_per_second": 1.23,
-      "step": 680
-    },
-    {
-      "epoch": 4.04,
-      "grad_norm": 0.49609375,
-      "learning_rate": 4.1929133858267716e-05,
-      "loss": 0.3543,
-      "step": 688
-    },
-    {
-      "epoch": 4.04,
-      "eval_loss": 0.3508842885494232,
-      "eval_runtime": 4.8828,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 688
-    },
-    {
-      "epoch": 4.09,
-      "grad_norm": 0.4609375,
-      "learning_rate": 4.183070866141732e-05,
-      "loss": 0.3491,
-      "step": 696
-    },
-    {
-      "epoch": 4.09,
-      "eval_loss": 0.3495161533355713,
-      "eval_runtime": 4.8837,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 696
-    },
-    {
-      "epoch": 4.13,
-      "grad_norm": 0.5234375,
-      "learning_rate": 4.173228346456693e-05,
-      "loss": 0.3486,
-      "step": 704
-    },
-    {
-      "epoch": 4.13,
-      "eval_loss": 0.352257639169693,
-      "eval_runtime": 4.8831,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 704
-    },
-    {
-      "epoch": 4.18,
-      "grad_norm": 0.51953125,
-      "learning_rate": 4.163385826771653e-05,
-      "loss": 0.3514,
-      "step": 712
-    },
-    {
-      "epoch": 4.18,
-      "eval_loss": 0.3515308201313019,
-      "eval_runtime": 4.8782,
-      "eval_samples_per_second": 9.84,
-      "eval_steps_per_second": 1.23,
-      "step": 712
-    },
-    {
-      "epoch": 4.23,
-      "grad_norm": 0.546875,
-      "learning_rate": 4.153543307086614e-05,
-      "loss": 0.3599,
-      "step": 720
-    },
-    {
-      "epoch": 4.23,
-      "eval_loss": 0.3499787747859955,
-      "eval_runtime": 4.8802,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.229,
-      "step": 720
-    },
-    {
-      "epoch": 4.27,
-      "grad_norm": 0.54296875,
-      "learning_rate": 4.1437007874015747e-05,
-      "loss": 0.337,
-      "step": 728
-    },
-    {
-      "epoch": 4.27,
-      "eval_loss": 0.35002824664115906,
-      "eval_runtime": 4.8833,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 728
-    },
-    {
-      "epoch": 4.32,
-      "grad_norm": 0.5703125,
-      "learning_rate": 4.133858267716536e-05,
-      "loss": 0.3466,
-      "step": 736
-    },
-    {
-      "epoch": 4.32,
-      "eval_loss": 0.3474964201450348,
-      "eval_runtime": 4.8824,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 736
-    },
-    {
-      "epoch": 4.37,
-      "grad_norm": 0.53125,
-      "learning_rate": 4.124015748031496e-05,
-      "loss": 0.3566,
-      "step": 744
-    },
-    {
-      "epoch": 4.37,
-      "eval_loss": 0.34953927993774414,
-      "eval_runtime": 4.8828,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 744
-    },
-    {
-      "epoch": 4.41,
-      "grad_norm": 0.53515625,
-      "learning_rate": 4.114173228346457e-05,
-      "loss": 0.3448,
-      "step": 752
-    },
-    {
-      "epoch": 4.41,
-      "eval_loss": 0.34765908122062683,
-      "eval_runtime": 4.8762,
-      "eval_samples_per_second": 9.844,
-      "eval_steps_per_second": 1.23,
-      "step": 752
-    },
-    {
-      "epoch": 4.46,
-      "grad_norm": 0.5,
-      "learning_rate": 4.1043307086614174e-05,
-      "loss": 0.3518,
-      "step": 760
-    },
-    {
-      "epoch": 4.46,
-      "eval_loss": 0.3473018407821655,
-      "eval_runtime": 4.8872,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 760
-    },
-    {
-      "epoch": 4.51,
-      "grad_norm": 0.55078125,
-      "learning_rate": 4.0944881889763784e-05,
-      "loss": 0.3608,
-      "step": 768
-    },
-    {
-      "epoch": 4.51,
-      "eval_loss": 0.3483521044254303,
-      "eval_runtime": 4.8829,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 768
-    },
-    {
-      "epoch": 4.55,
-      "grad_norm": 0.578125,
-      "learning_rate": 4.084645669291339e-05,
-      "loss": 0.3535,
-      "step": 776
-    },
-    {
-      "epoch": 4.55,
-      "eval_loss": 0.3491772413253784,
-      "eval_runtime": 4.8826,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 776
-    },
-    {
-      "epoch": 4.6,
-      "grad_norm": 0.5390625,
-      "learning_rate": 4.074803149606299e-05,
-      "loss": 0.3513,
-      "step": 784
-    },
-    {
-      "epoch": 4.6,
-      "eval_loss": 0.34816303849220276,
-      "eval_runtime": 4.8828,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 784
-    },
-    {
-      "epoch": 4.65,
-      "grad_norm": 0.53515625,
-      "learning_rate": 4.06496062992126e-05,
-      "loss": 0.3424,
-      "step": 792
-    },
-    {
-      "epoch": 4.65,
-      "eval_loss": 0.3502413332462311,
-      "eval_runtime": 4.881,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
-      "step": 792
-    },
-    {
-      "epoch": 4.7,
-      "grad_norm": 0.515625,
-      "learning_rate": 4.0551181102362204e-05,
-      "loss": 0.3481,
-      "step": 800
-    },
-    {
-      "epoch": 4.7,
-      "eval_loss": 0.3473189175128937,
-      "eval_runtime": 4.8816,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 800
-    },
-    {
-      "epoch": 4.74,
-      "grad_norm": 0.5078125,
-      "learning_rate": 4.0452755905511814e-05,
-      "loss": 0.3772,
-      "step": 808
-    },
-    {
-      "epoch": 4.74,
-      "eval_loss": 0.3471733033657074,
-      "eval_runtime": 4.8804,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 808
-    },
-    {
-      "epoch": 4.79,
-      "grad_norm": 0.515625,
-      "learning_rate": 4.035433070866142e-05,
-      "loss": 0.3542,
-      "step": 816
-    },
-    {
-      "epoch": 4.79,
-      "eval_loss": 0.3472949266433716,
-      "eval_runtime": 4.8862,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 816
-    },
-    {
-      "epoch": 4.84,
-      "grad_norm": 0.51171875,
-      "learning_rate": 4.025590551181103e-05,
-      "loss": 0.3712,
-      "step": 824
-    },
-    {
-      "epoch": 4.84,
-      "eval_loss": 0.3476013243198395,
-      "eval_runtime": 4.8816,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 824
-    },
-    {
-      "epoch": 4.88,
-      "grad_norm": 0.5234375,
-      "learning_rate": 4.015748031496063e-05,
-      "loss": 0.3496,
-      "step": 832
-    },
-    {
-      "epoch": 4.88,
-      "eval_loss": 0.3474799394607544,
-      "eval_runtime": 4.8815,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 832
-    },
-    {
-      "epoch": 4.93,
-      "grad_norm": 0.58984375,
-      "learning_rate": 4.005905511811024e-05,
-      "loss": 0.3481,
-      "step": 840
-    },
-    {
-      "epoch": 4.93,
-      "eval_loss": 0.3481266498565674,
-      "eval_runtime": 4.8813,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 840
-    },
-    {
-      "epoch": 4.98,
-      "grad_norm": 0.53125,
-      "learning_rate": 3.9960629921259845e-05,
-      "loss": 0.3416,
-      "step": 848
-    },
-    {
-      "epoch": 4.98,
-      "eval_loss": 0.3473891317844391,
-      "eval_runtime": 4.8872,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 848
-    },
-    {
-      "epoch": 5.02,
-      "grad_norm": 0.54296875,
-      "learning_rate": 3.9862204724409455e-05,
-      "loss": 0.3518,
-      "step": 856
-    },
-    {
-      "epoch": 5.02,
-      "eval_loss": 0.34982213377952576,
-      "eval_runtime": 4.8796,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
-      "step": 856
-    },
-    {
-      "epoch": 5.07,
-      "grad_norm": 0.62890625,
-      "learning_rate": 3.976377952755906e-05,
-      "loss": 0.3239,
-      "step": 864
-    },
-    {
-      "epoch": 5.07,
-      "eval_loss": 0.35387638211250305,
-      "eval_runtime": 4.8826,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 864
-    },
-    {
-      "epoch": 5.12,
-      "grad_norm": 0.578125,
-      "learning_rate": 3.966535433070867e-05,
-      "loss": 0.3389,
-      "step": 872
-    },
-    {
-      "epoch": 5.12,
-      "eval_loss": 0.3555631935596466,
-      "eval_runtime": 4.8804,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 872
-    },
-    {
-      "epoch": 5.17,
-      "grad_norm": 0.5703125,
-      "learning_rate": 3.956692913385827e-05,
-      "loss": 0.3357,
-      "step": 880
-    },
-    {
-      "epoch": 5.17,
-      "eval_loss": 0.35232648253440857,
-      "eval_runtime": 4.8819,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 880
-    },
-    {
-      "epoch": 5.21,
-      "grad_norm": 0.5859375,
-      "learning_rate": 3.9468503937007875e-05,
-      "loss": 0.3305,
-      "step": 888
-    },
-    {
-      "epoch": 5.21,
-      "eval_loss": 0.3552100956439972,
-      "eval_runtime": 4.8799,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.23,
-      "step": 888
-    },
-    {
-      "epoch": 5.26,
-      "grad_norm": 0.56640625,
-      "learning_rate": 3.9370078740157485e-05,
-      "loss": 0.3308,
-      "step": 896
-    },
-    {
-      "epoch": 5.26,
-      "eval_loss": 0.35367465019226074,
-      "eval_runtime": 4.8807,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 896
-    },
-    {
-      "epoch": 5.31,
-      "grad_norm": 0.625,
-      "learning_rate": 3.927165354330709e-05,
-      "loss": 0.3646,
-      "step": 904
-    },
-    {
-      "epoch": 5.31,
-      "eval_loss": 0.3549775183200836,
-      "eval_runtime": 4.8817,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 904
-    },
-    {
-      "epoch": 5.35,
-      "grad_norm": 0.60546875,
-      "learning_rate": 3.91732283464567e-05,
-      "loss": 0.3369,
-      "step": 912
-    },
-    {
-      "epoch": 5.35,
-      "eval_loss": 0.352167010307312,
-      "eval_runtime": 4.8815,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 912
-    },
-    {
-      "epoch": 5.4,
-      "grad_norm": 0.66015625,
-      "learning_rate": 3.90748031496063e-05,
-      "loss": 0.3391,
-      "step": 920
-    },
-    {
-      "epoch": 5.4,
-      "eval_loss": 0.35382068157196045,
-      "eval_runtime": 4.8737,
-      "eval_samples_per_second": 9.849,
-      "eval_steps_per_second": 1.231,
-      "step": 920
-    },
-    {
-      "epoch": 5.45,
-      "grad_norm": 0.5703125,
-      "learning_rate": 3.8976377952755905e-05,
-      "loss": 0.3343,
-      "step": 928
-    },
-    {
-      "epoch": 5.45,
-      "eval_loss": 0.35162171721458435,
-      "eval_runtime": 4.8953,
-      "eval_samples_per_second": 9.805,
-      "eval_steps_per_second": 1.226,
-      "step": 928
-    },
-    {
-      "epoch": 5.49,
-      "grad_norm": 0.61328125,
-      "learning_rate": 3.887795275590551e-05,
-      "loss": 0.3333,
-      "step": 936
-    },
-    {
-      "epoch": 5.49,
-      "eval_loss": 0.3531843423843384,
-      "eval_runtime": 4.8803,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 936
-    },
-    {
-      "epoch": 5.54,
-      "grad_norm": 0.58984375,
-      "learning_rate": 3.877952755905512e-05,
-      "loss": 0.3337,
-      "step": 944
-    },
-    {
-      "epoch": 5.54,
-      "eval_loss": 0.3522440493106842,
-      "eval_runtime": 4.8853,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 944
-    },
-    {
-      "epoch": 5.59,
-      "grad_norm": 0.64453125,
-      "learning_rate": 3.868110236220472e-05,
-      "loss": 0.3281,
-      "step": 952
-    },
-    {
-      "epoch": 5.59,
-      "eval_loss": 0.3523660898208618,
-      "eval_runtime": 4.8767,
-      "eval_samples_per_second": 9.843,
-      "eval_steps_per_second": 1.23,
-      "step": 952
-    },
-    {
-      "epoch": 5.63,
-      "grad_norm": 0.5625,
-      "learning_rate": 3.858267716535433e-05,
-      "loss": 0.3268,
-      "step": 960
-    },
-    {
-      "epoch": 5.63,
-      "eval_loss": 0.35167932510375977,
-      "eval_runtime": 4.8799,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.23,
-      "step": 960
-    },
-    {
-      "epoch": 5.68,
-      "grad_norm": 0.55078125,
-      "learning_rate": 3.8484251968503936e-05,
-      "loss": 0.3407,
-      "step": 968
-    },
-    {
-      "epoch": 5.68,
-      "eval_loss": 0.3501236140727997,
-      "eval_runtime": 4.8881,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.227,
-      "step": 968
-    },
-    {
-      "epoch": 5.73,
-      "grad_norm": 0.625,
-      "learning_rate": 3.8385826771653546e-05,
-      "loss": 0.3221,
-      "step": 976
-    },
-    {
-      "epoch": 5.73,
-      "eval_loss": 0.35152295231819153,
-      "eval_runtime": 4.8854,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 976
-    },
-    {
-      "epoch": 5.78,
-      "grad_norm": 0.57421875,
-      "learning_rate": 3.828740157480315e-05,
-      "loss": 0.3506,
-      "step": 984
-    },
-    {
-      "epoch": 5.78,
-      "eval_loss": 0.3493311107158661,
-      "eval_runtime": 4.8909,
-      "eval_samples_per_second": 9.814,
-      "eval_steps_per_second": 1.227,
-      "step": 984
-    },
-    {
-      "epoch": 5.82,
-      "grad_norm": 0.546875,
-      "learning_rate": 3.818897637795276e-05,
-      "loss": 0.3393,
-      "step": 992
-    },
-    {
-      "epoch": 5.82,
-      "eval_loss": 0.34996774792671204,
-      "eval_runtime": 4.8838,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 992
-    },
-    {
-      "epoch": 5.87,
-      "grad_norm": 0.61328125,
-      "learning_rate": 3.809055118110236e-05,
-      "loss": 0.3431,
-      "step": 1000
-    },
-    {
-      "epoch": 5.87,
-      "eval_loss": 0.3525948226451874,
-      "eval_runtime": 4.889,
-      "eval_samples_per_second": 9.818,
-      "eval_steps_per_second": 1.227,
-      "step": 1000
-    },
-    {
-      "epoch": 5.92,
-      "grad_norm": 0.5625,
-      "learning_rate": 3.7992125984251966e-05,
-      "loss": 0.3454,
-      "step": 1008
-    },
-    {
-      "epoch": 5.92,
-      "eval_loss": 0.349923700094223,
-      "eval_runtime": 4.8841,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.228,
-      "step": 1008
-    },
-    {
-      "epoch": 5.96,
-      "grad_norm": 0.6015625,
-      "learning_rate": 3.7893700787401576e-05,
-      "loss": 0.3276,
-      "step": 1016
-    },
-    {
-      "epoch": 5.96,
-      "eval_loss": 0.3507179021835327,
-      "eval_runtime": 4.8837,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 1016
-    },
-    {
-      "epoch": 6.01,
-      "grad_norm": 0.61328125,
-      "learning_rate": 3.779527559055118e-05,
-      "loss": 0.3146,
-      "step": 1024
-    },
-    {
-      "epoch": 6.01,
-      "eval_loss": 0.3517909049987793,
-      "eval_runtime": 4.8786,
-      "eval_samples_per_second": 9.839,
-      "eval_steps_per_second": 1.23,
-      "step": 1024
-    },
-    {
-      "epoch": 6.06,
-      "grad_norm": 0.6875,
-      "learning_rate": 3.769685039370079e-05,
-      "loss": 0.3186,
-      "step": 1032
-    },
-    {
-      "epoch": 6.06,
-      "eval_loss": 0.3593278229236603,
-      "eval_runtime": 4.8816,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 1032
-    },
-    {
-      "epoch": 6.1,
-      "grad_norm": 0.63671875,
-      "learning_rate": 3.759842519685039e-05,
-      "loss": 0.3197,
-      "step": 1040
-    },
-    {
-      "epoch": 6.1,
-      "eval_loss": 0.3572365343570709,
-      "eval_runtime": 4.8841,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.228,
-      "step": 1040
-    },
-    {
-      "epoch": 6.15,
-      "grad_norm": 0.625,
-      "learning_rate": 3.7500000000000003e-05,
-      "loss": 0.3081,
-      "step": 1048
-    },
-    {
-      "epoch": 6.15,
-      "eval_loss": 0.35952886939048767,
-      "eval_runtime": 4.8823,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 1048
-    },
-    {
-      "epoch": 6.2,
-      "grad_norm": 0.69921875,
-      "learning_rate": 3.740157480314961e-05,
-      "loss": 0.3225,
-      "step": 1056
-    },
-    {
-      "epoch": 6.2,
-      "eval_loss": 0.3600703179836273,
-      "eval_runtime": 4.8852,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 1056
-    },
-    {
-      "epoch": 6.25,
-      "grad_norm": 0.734375,
-      "learning_rate": 3.730314960629922e-05,
-      "loss": 0.3154,
-      "step": 1064
-    },
-    {
-      "epoch": 6.25,
-      "eval_loss": 0.3606225550174713,
-      "eval_runtime": 4.8812,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
-      "step": 1064
-    },
-    {
-      "epoch": 6.29,
-      "grad_norm": 0.62890625,
-      "learning_rate": 3.720472440944882e-05,
-      "loss": 0.3043,
-      "step": 1072
-    },
-    {
-      "epoch": 6.29,
-      "eval_loss": 0.35906192660331726,
-      "eval_runtime": 4.8866,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 1072
-    },
-    {
-      "epoch": 6.34,
-      "grad_norm": 0.7109375,
-      "learning_rate": 3.710629921259843e-05,
-      "loss": 0.3131,
-      "step": 1080
-    },
-    {
-      "epoch": 6.34,
-      "eval_loss": 0.3606855869293213,
-      "eval_runtime": 4.8845,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 1080
-    },
-    {
-      "epoch": 6.39,
-      "grad_norm": 0.6640625,
-      "learning_rate": 3.7007874015748034e-05,
-      "loss": 0.3131,
-      "step": 1088
-    },
-    {
-      "epoch": 6.39,
-      "eval_loss": 0.35807374119758606,
-      "eval_runtime": 4.8819,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 1088
-    },
-    {
-      "epoch": 6.43,
-      "grad_norm": 0.70703125,
-      "learning_rate": 3.690944881889764e-05,
-      "loss": 0.3277,
-      "step": 1096
-    },
-    {
-      "epoch": 6.43,
-      "eval_loss": 0.35696378350257874,
-      "eval_runtime": 4.89,
-      "eval_samples_per_second": 9.816,
-      "eval_steps_per_second": 1.227,
-      "step": 1096
-    },
-    {
-      "epoch": 6.48,
-      "grad_norm": 0.66796875,
-      "learning_rate": 3.681102362204725e-05,
-      "loss": 0.3051,
-      "step": 1104
-    },
-    {
-      "epoch": 6.48,
-      "eval_loss": 0.359957218170166,
-      "eval_runtime": 4.8831,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 1104
-    },
-    {
-      "epoch": 6.53,
-      "grad_norm": 0.73828125,
-      "learning_rate": 3.671259842519685e-05,
-      "loss": 0.3308,
-      "step": 1112
-    },
-    {
-      "epoch": 6.53,
-      "eval_loss": 0.35956406593322754,
-      "eval_runtime": 4.8844,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 1112
-    },
-    {
-      "epoch": 6.57,
-      "grad_norm": 0.68359375,
-      "learning_rate": 3.661417322834646e-05,
-      "loss": 0.3162,
-      "step": 1120
-    },
-    {
-      "epoch": 6.57,
-      "eval_loss": 0.3570359945297241,
-      "eval_runtime": 4.8827,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 1120
-    },
-    {
-      "epoch": 6.62,
-      "grad_norm": 0.76953125,
-      "learning_rate": 3.6515748031496064e-05,
-      "loss": 0.3212,
-      "step": 1128
-    },
-    {
-      "epoch": 6.62,
-      "eval_loss": 0.3584875762462616,
-      "eval_runtime": 4.8859,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 1128
-    },
-    {
-      "epoch": 6.67,
-      "grad_norm": 0.6875,
-      "learning_rate": 3.6417322834645674e-05,
-      "loss": 0.3191,
-      "step": 1136
-    },
-    {
-      "epoch": 6.67,
-      "eval_loss": 0.3554946482181549,
-      "eval_runtime": 4.8817,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 1136
-    },
-    {
-      "epoch": 6.71,
-      "grad_norm": 0.66015625,
-      "learning_rate": 3.631889763779528e-05,
-      "loss": 0.3206,
-      "step": 1144
-    },
-    {
-      "epoch": 6.71,
-      "eval_loss": 0.3560527265071869,
-      "eval_runtime": 4.8883,
-      "eval_samples_per_second": 9.819,
-      "eval_steps_per_second": 1.227,
-      "step": 1144
-    },
-    {
-      "epoch": 6.76,
-      "grad_norm": 0.66796875,
-      "learning_rate": 3.622047244094489e-05,
-      "loss": 0.3411,
-      "step": 1152
-    },
-    {
-      "epoch": 6.76,
-      "eval_loss": 0.3585592210292816,
-      "eval_runtime": 4.8873,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 1152
-    },
-    {
-      "epoch": 6.81,
-      "grad_norm": 0.62890625,
-      "learning_rate": 3.612204724409449e-05,
-      "loss": 0.3278,
-      "step": 1160
-    },
-    {
-      "epoch": 6.81,
-      "eval_loss": 0.3558274507522583,
-      "eval_runtime": 4.8914,
-      "eval_samples_per_second": 9.813,
-      "eval_steps_per_second": 1.227,
-      "step": 1160
-    },
-    {
-      "epoch": 6.86,
-      "grad_norm": 0.62890625,
-      "learning_rate": 3.60236220472441e-05,
-      "loss": 0.3273,
-      "step": 1168
-    },
-    {
-      "epoch": 6.86,
-      "eval_loss": 0.3541262447834015,
-      "eval_runtime": 4.8855,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 1168
-    },
-    {
-      "epoch": 6.9,
-      "grad_norm": 0.72265625,
-      "learning_rate": 3.59251968503937e-05,
-      "loss": 0.3275,
-      "step": 1176
-    },
-    {
-      "epoch": 6.9,
-      "eval_loss": 0.35615813732147217,
-      "eval_runtime": 4.8897,
-      "eval_samples_per_second": 9.817,
-      "eval_steps_per_second": 1.227,
-      "step": 1176
-    },
-    {
-      "epoch": 6.95,
-      "grad_norm": 0.69140625,
-      "learning_rate": 3.582677165354331e-05,
-      "loss": 0.3283,
-      "step": 1184
-    },
-    {
-      "epoch": 6.95,
-      "eval_loss": 0.35849082469940186,
-      "eval_runtime": 4.8881,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.227,
-      "step": 1184
-    },
-    {
-      "epoch": 7.0,
-      "grad_norm": 0.6328125,
-      "learning_rate": 3.572834645669291e-05,
-      "loss": 0.3107,
-      "step": 1192
-    },
-    {
-      "epoch": 7.0,
-      "eval_loss": 0.3558361530303955,
-      "eval_runtime": 4.8883,
-      "eval_samples_per_second": 9.819,
-      "eval_steps_per_second": 1.227,
-      "step": 1192
-    },
-    {
-      "epoch": 7.04,
-      "grad_norm": 0.859375,
-      "learning_rate": 3.562992125984252e-05,
-      "loss": 0.3027,
-      "step": 1200
-    },
-    {
-      "epoch": 7.04,
-      "eval_loss": 0.37191081047058105,
-      "eval_runtime": 4.8846,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 1200
-    },
-    {
-      "epoch": 7.09,
-      "grad_norm": 0.7890625,
-      "learning_rate": 3.5531496062992125e-05,
-      "loss": 0.2883,
-      "step": 1208
-    },
-    {
-      "epoch": 7.09,
-      "eval_loss": 0.3696741759777069,
-      "eval_runtime": 4.8811,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
-      "step": 1208
-    },
-    {
-      "epoch": 7.14,
-      "grad_norm": 0.73046875,
-      "learning_rate": 3.5433070866141735e-05,
-      "loss": 0.2871,
-      "step": 1216
-    },
-    {
-      "epoch": 7.14,
-      "eval_loss": 0.36751437187194824,
-      "eval_runtime": 4.8847,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 1216
-    },
-    {
-      "epoch": 7.18,
-      "grad_norm": 0.74609375,
-      "learning_rate": 3.533464566929134e-05,
-      "loss": 0.3094,
-      "step": 1224
-    },
-    {
-      "epoch": 7.18,
-      "eval_loss": 0.3699875771999359,
-      "eval_runtime": 4.8886,
-      "eval_samples_per_second": 9.819,
-      "eval_steps_per_second": 1.227,
-      "step": 1224
-    },
-    {
-      "epoch": 7.23,
-      "grad_norm": 0.765625,
-      "learning_rate": 3.523622047244094e-05,
-      "loss": 0.3026,
-      "step": 1232
-    },
-    {
-      "epoch": 7.23,
-      "eval_loss": 0.368384450674057,
-      "eval_runtime": 4.8856,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 1232
-    },
-    {
-      "epoch": 7.28,
-      "grad_norm": 0.7890625,
-      "learning_rate": 3.513779527559055e-05,
-      "loss": 0.2984,
-      "step": 1240
-    },
-    {
-      "epoch": 7.28,
-      "eval_loss": 0.3673759400844574,
-      "eval_runtime": 4.8881,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.227,
-      "step": 1240
-    },
-    {
-      "epoch": 7.33,
-      "grad_norm": 0.76953125,
-      "learning_rate": 3.5039370078740156e-05,
-      "loss": 0.3057,
-      "step": 1248
-    },
-    {
-      "epoch": 7.33,
-      "eval_loss": 0.37147435545921326,
-      "eval_runtime": 4.8851,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 1248
-    },
-    {
-      "epoch": 7.37,
-      "grad_norm": 0.77734375,
-      "learning_rate": 3.4940944881889766e-05,
-      "loss": 0.277,
-      "step": 1256
-    },
-    {
-      "epoch": 7.37,
-      "eval_loss": 0.3669518232345581,
-      "eval_runtime": 4.8848,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 1256
-    },
-    {
-      "epoch": 7.42,
-      "grad_norm": 0.71484375,
-      "learning_rate": 3.484251968503937e-05,
-      "loss": 0.2999,
-      "step": 1264
-    },
-    {
-      "epoch": 7.42,
-      "eval_loss": 0.3678833246231079,
-      "eval_runtime": 4.8848,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 1264
-    },
-    {
-      "epoch": 7.47,
-      "grad_norm": 0.66015625,
-      "learning_rate": 3.474409448818898e-05,
-      "loss": 0.3124,
-      "step": 1272
-    },
-    {
-      "epoch": 7.47,
-      "eval_loss": 0.3651003837585449,
-      "eval_runtime": 4.8819,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 1272
-    },
-    {
-      "epoch": 7.51,
-      "grad_norm": 0.8359375,
-      "learning_rate": 3.464566929133858e-05,
-      "loss": 0.2919,
-      "step": 1280
-    },
-    {
-      "epoch": 7.51,
-      "eval_loss": 0.36933448910713196,
-      "eval_runtime": 4.8856,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 1280
-    },
-    {
-      "epoch": 7.56,
-      "grad_norm": 0.84375,
-      "learning_rate": 3.454724409448819e-05,
-      "loss": 0.3111,
-      "step": 1288
-    },
-    {
-      "epoch": 7.56,
-      "eval_loss": 0.3658764362335205,
-      "eval_runtime": 4.8883,
-      "eval_samples_per_second": 9.819,
-      "eval_steps_per_second": 1.227,
-      "step": 1288
-    },
-    {
-      "epoch": 7.61,
-      "grad_norm": 0.7890625,
-      "learning_rate": 3.4448818897637796e-05,
-      "loss": 0.3025,
-      "step": 1296
-    },
-    {
-      "epoch": 7.61,
-      "eval_loss": 0.36660221219062805,
-      "eval_runtime": 4.8886,
-      "eval_samples_per_second": 9.819,
-      "eval_steps_per_second": 1.227,
-      "step": 1296
-    },
-    {
-      "epoch": 7.65,
-      "grad_norm": 0.78125,
-      "learning_rate": 3.4350393700787406e-05,
-      "loss": 0.3028,
-      "step": 1304
-    },
-    {
-      "epoch": 7.65,
-      "eval_loss": 0.36641809344291687,
-      "eval_runtime": 4.8816,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 1304
-    },
-    {
-      "epoch": 7.7,
-      "grad_norm": 0.765625,
-      "learning_rate": 3.425196850393701e-05,
-      "loss": 0.3075,
-      "step": 1312
-    },
-    {
-      "epoch": 7.7,
-      "eval_loss": 0.362444132566452,
-      "eval_runtime": 4.8873,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 1312
-    },
-    {
-      "epoch": 7.75,
-      "grad_norm": 0.765625,
-      "learning_rate": 3.415354330708661e-05,
-      "loss": 0.3134,
-      "step": 1320
-    },
-    {
-      "epoch": 7.75,
-      "eval_loss": 0.36705541610717773,
-      "eval_runtime": 4.8843,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 1320
-    },
-    {
-      "epoch": 7.79,
-      "grad_norm": 0.7265625,
-      "learning_rate": 3.405511811023622e-05,
-      "loss": 0.299,
-      "step": 1328
-    },
-    {
-      "epoch": 7.79,
-      "eval_loss": 0.366720050573349,
-      "eval_runtime": 4.8831,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 1328
-    },
-    {
-      "epoch": 7.84,
-      "grad_norm": 0.671875,
-      "learning_rate": 3.3956692913385827e-05,
-      "loss": 0.3067,
-      "step": 1336
-    },
-    {
-      "epoch": 7.84,
-      "eval_loss": 0.3674195110797882,
-      "eval_runtime": 4.8885,
-      "eval_samples_per_second": 9.819,
-      "eval_steps_per_second": 1.227,
-      "step": 1336
-    },
-    {
-      "epoch": 7.89,
-      "grad_norm": 0.75,
-      "learning_rate": 3.385826771653544e-05,
-      "loss": 0.3073,
-      "step": 1344
-    },
-    {
-      "epoch": 7.89,
-      "eval_loss": 0.3649815618991852,
-      "eval_runtime": 4.8905,
-      "eval_samples_per_second": 9.815,
-      "eval_steps_per_second": 1.227,
-      "step": 1344
-    },
-    {
-      "epoch": 7.94,
-      "grad_norm": 0.796875,
-      "learning_rate": 3.375984251968504e-05,
-      "loss": 0.3187,
-      "step": 1352
-    },
-    {
-      "epoch": 7.94,
-      "eval_loss": 0.3641921281814575,
-      "eval_runtime": 4.8875,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 1352
-    },
-    {
-      "epoch": 7.98,
-      "grad_norm": 0.83203125,
-      "learning_rate": 3.366141732283465e-05,
-      "loss": 0.2982,
-      "step": 1360
-    },
-    {
-      "epoch": 7.98,
-      "eval_loss": 0.36615875363349915,
-      "eval_runtime": 4.8874,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 1360
-    },
-    {
-      "epoch": 8.03,
-      "grad_norm": 0.77734375,
-      "learning_rate": 3.3562992125984254e-05,
-      "loss": 0.2816,
-      "step": 1368
-    },
-    {
-      "epoch": 8.03,
-      "eval_loss": 0.3784830570220947,
-      "eval_runtime": 4.8814,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 1368
-    },
-    {
-      "epoch": 8.08,
-      "grad_norm": 0.8984375,
-      "learning_rate": 3.3464566929133864e-05,
-      "loss": 0.2811,
-      "step": 1376
-    },
-    {
-      "epoch": 8.08,
-      "eval_loss": 0.37378013134002686,
-      "eval_runtime": 4.8805,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 1376
-    },
-    {
-      "epoch": 8.12,
-      "grad_norm": 0.86328125,
-      "learning_rate": 3.336614173228347e-05,
-      "loss": 0.2771,
-      "step": 1384
-    },
-    {
-      "epoch": 8.12,
-      "eval_loss": 0.37527838349342346,
-      "eval_runtime": 4.8863,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 1384
-    },
-    {
-      "epoch": 8.17,
-      "grad_norm": 0.95703125,
-      "learning_rate": 3.326771653543308e-05,
-      "loss": 0.2884,
-      "step": 1392
-    },
-    {
-      "epoch": 8.17,
-      "eval_loss": 0.37870660424232483,
-      "eval_runtime": 4.8831,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 1392
-    },
-    {
-      "epoch": 8.22,
-      "grad_norm": 0.99609375,
-      "learning_rate": 3.316929133858268e-05,
-      "loss": 0.2671,
-      "step": 1400
-    },
-    {
-      "epoch": 8.22,
-      "eval_loss": 0.3793374001979828,
-      "eval_runtime": 4.8799,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.23,
-      "step": 1400
-    },
-    {
-      "epoch": 8.26,
-      "grad_norm": 0.95703125,
-      "learning_rate": 3.3070866141732284e-05,
-      "loss": 0.2861,
-      "step": 1408
-    },
-    {
-      "epoch": 8.26,
-      "eval_loss": 0.38111141324043274,
-      "eval_runtime": 4.8809,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
-      "step": 1408
-    },
-    {
-      "epoch": 8.31,
-      "grad_norm": 0.83203125,
-      "learning_rate": 3.2972440944881894e-05,
-      "loss": 0.29,
-      "step": 1416
-    },
-    {
-      "epoch": 8.31,
-      "eval_loss": 0.3800075948238373,
-      "eval_runtime": 4.8832,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 1416
-    },
-    {
-      "epoch": 8.36,
-      "grad_norm": 0.8515625,
-      "learning_rate": 3.28740157480315e-05,
-      "loss": 0.2739,
-      "step": 1424
-    },
-    {
-      "epoch": 8.36,
-      "eval_loss": 0.37995681166648865,
-      "eval_runtime": 4.8873,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 1424
-    },
-    {
-      "epoch": 8.4,
-      "grad_norm": 0.84375,
-      "learning_rate": 3.27755905511811e-05,
-      "loss": 0.2803,
-      "step": 1432
-    },
-    {
-      "epoch": 8.4,
-      "eval_loss": 0.37614905834198,
-      "eval_runtime": 4.8889,
-      "eval_samples_per_second": 9.818,
-      "eval_steps_per_second": 1.227,
-      "step": 1432
-    },
-    {
-      "epoch": 8.45,
-      "grad_norm": 1.0703125,
-      "learning_rate": 3.2677165354330704e-05,
-      "loss": 0.282,
-      "step": 1440
-    },
-    {
-      "epoch": 8.45,
-      "eval_loss": 0.37973472476005554,
-      "eval_runtime": 4.8822,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 1440
-    },
-    {
-      "epoch": 8.5,
-      "grad_norm": 0.87890625,
-      "learning_rate": 3.2578740157480314e-05,
-      "loss": 0.2858,
-      "step": 1448
-    },
-    {
-      "epoch": 8.5,
-      "eval_loss": 0.3786061704158783,
-      "eval_runtime": 4.8817,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 1448
-    },
-    {
-      "epoch": 8.55,
-      "grad_norm": 0.87109375,
-      "learning_rate": 3.248031496062992e-05,
-      "loss": 0.2853,
-      "step": 1456
-    },
-    {
-      "epoch": 8.55,
-      "eval_loss": 0.38181185722351074,
-      "eval_runtime": 4.8849,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 1456
-    },
-    {
-      "epoch": 8.59,
-      "grad_norm": 0.79296875,
-      "learning_rate": 3.238188976377953e-05,
-      "loss": 0.2978,
-      "step": 1464
-    },
-    {
-      "epoch": 8.59,
-      "eval_loss": 0.3762815296649933,
-      "eval_runtime": 4.8805,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 1464
-    },
-    {
-      "epoch": 8.64,
-      "grad_norm": 0.90625,
-      "learning_rate": 3.228346456692913e-05,
-      "loss": 0.2948,
-      "step": 1472
-    },
-    {
-      "epoch": 8.64,
-      "eval_loss": 0.3758275508880615,
-      "eval_runtime": 4.8797,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
-      "step": 1472
-    },
-    {
-      "epoch": 8.69,
-      "grad_norm": 0.953125,
-      "learning_rate": 3.218503937007874e-05,
-      "loss": 0.2876,
-      "step": 1480
-    },
-    {
-      "epoch": 8.69,
-      "eval_loss": 0.3739815950393677,
-      "eval_runtime": 4.8829,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 1480
-    },
-    {
-      "epoch": 8.73,
-      "grad_norm": 0.94921875,
-      "learning_rate": 3.2086614173228345e-05,
-      "loss": 0.27,
-      "step": 1488
-    },
-    {
-      "epoch": 8.73,
-      "eval_loss": 0.38057199120521545,
-      "eval_runtime": 4.8809,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
-      "step": 1488
-    },
-    {
-      "epoch": 8.78,
-      "grad_norm": 0.9609375,
-      "learning_rate": 3.1988188976377955e-05,
-      "loss": 0.2898,
-      "step": 1496
-    },
-    {
-      "epoch": 8.78,
-      "eval_loss": 0.3779284954071045,
-      "eval_runtime": 4.8853,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 1496
-    },
-    {
-      "epoch": 8.83,
-      "grad_norm": 0.80078125,
-      "learning_rate": 3.188976377952756e-05,
-      "loss": 0.2696,
-      "step": 1504
-    },
-    {
-      "epoch": 8.83,
-      "eval_loss": 0.3726881742477417,
-      "eval_runtime": 4.8862,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 1504
-    },
-    {
-      "epoch": 8.87,
-      "grad_norm": 0.92578125,
-      "learning_rate": 3.179133858267717e-05,
-      "loss": 0.2998,
-      "step": 1512
-    },
-    {
-      "epoch": 8.87,
-      "eval_loss": 0.37989136576652527,
-      "eval_runtime": 4.8764,
-      "eval_samples_per_second": 9.843,
-      "eval_steps_per_second": 1.23,
-      "step": 1512
-    },
-    {
-      "epoch": 8.92,
-      "grad_norm": 0.96875,
-      "learning_rate": 3.169291338582677e-05,
-      "loss": 0.2733,
-      "step": 1520
-    },
-    {
-      "epoch": 8.92,
-      "eval_loss": 0.37501704692840576,
-      "eval_runtime": 4.8844,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 1520
-    },
-    {
-      "epoch": 8.97,
-      "grad_norm": 0.87109375,
-      "learning_rate": 3.159448818897638e-05,
-      "loss": 0.2873,
-      "step": 1528
-    },
-    {
-      "epoch": 8.97,
-      "eval_loss": 0.37481260299682617,
-      "eval_runtime": 4.88,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.23,
-      "step": 1528
-    },
-    {
-      "epoch": 9.02,
-      "grad_norm": 0.76171875,
-      "learning_rate": 3.1496062992125985e-05,
-      "loss": 0.2747,
-      "step": 1536
-    },
-    {
-      "epoch": 9.02,
-      "eval_loss": 0.3823326528072357,
-      "eval_runtime": 4.8824,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 1536
-    },
-    {
-      "epoch": 9.06,
-      "grad_norm": 1.0625,
-      "learning_rate": 3.139763779527559e-05,
-      "loss": 0.266,
-      "step": 1544
-    },
-    {
-      "epoch": 9.06,
-      "eval_loss": 0.38916265964508057,
-      "eval_runtime": 4.88,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.229,
-      "step": 1544
-    },
-    {
-      "epoch": 9.11,
-      "grad_norm": 1.0234375,
-      "learning_rate": 3.12992125984252e-05,
-      "loss": 0.264,
-      "step": 1552
-    },
-    {
-      "epoch": 9.11,
-      "eval_loss": 0.3951665163040161,
-      "eval_runtime": 4.8839,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.229,
-      "step": 1552
-    },
-    {
-      "epoch": 9.16,
-      "grad_norm": 1.046875,
-      "learning_rate": 3.12007874015748e-05,
-      "loss": 0.2638,
-      "step": 1560
-    },
-    {
-      "epoch": 9.16,
-      "eval_loss": 0.3967585861682892,
-      "eval_runtime": 4.8805,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 1560
-    },
-    {
-      "epoch": 9.2,
-      "grad_norm": 0.88671875,
-      "learning_rate": 3.110236220472441e-05,
-      "loss": 0.2597,
-      "step": 1568
-    },
-    {
-      "epoch": 9.2,
-      "eval_loss": 0.39212462306022644,
-      "eval_runtime": 4.8807,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 1568
-    },
-    {
-      "epoch": 9.25,
-      "grad_norm": 1.1015625,
-      "learning_rate": 3.1003937007874016e-05,
-      "loss": 0.2663,
-      "step": 1576
-    },
-    {
-      "epoch": 9.25,
-      "eval_loss": 0.3920614421367645,
-      "eval_runtime": 4.8778,
-      "eval_samples_per_second": 9.841,
-      "eval_steps_per_second": 1.23,
-      "step": 1576
-    },
-    {
-      "epoch": 9.3,
-      "grad_norm": 1.0546875,
-      "learning_rate": 3.0905511811023626e-05,
-      "loss": 0.2518,
-      "step": 1584
-    },
-    {
-      "epoch": 9.3,
-      "eval_loss": 0.393242746591568,
-      "eval_runtime": 4.8823,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 1584
-    },
-    {
-      "epoch": 9.34,
-      "grad_norm": 0.98046875,
-      "learning_rate": 3.080708661417323e-05,
-      "loss": 0.261,
-      "step": 1592
-    },
-    {
-      "epoch": 9.34,
-      "eval_loss": 0.39824220538139343,
-      "eval_runtime": 4.8878,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.228,
-      "step": 1592
-    },
-    {
-      "epoch": 9.39,
-      "grad_norm": 0.9609375,
-      "learning_rate": 3.070866141732284e-05,
-      "loss": 0.2499,
-      "step": 1600
-    },
-    {
-      "epoch": 9.39,
-      "eval_loss": 0.3865773677825928,
-      "eval_runtime": 4.8853,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 1600
-    },
-    {
-      "epoch": 9.44,
-      "grad_norm": 1.046875,
-      "learning_rate": 3.061023622047244e-05,
-      "loss": 0.2721,
-      "step": 1608
-    },
-    {
-      "epoch": 9.44,
-      "eval_loss": 0.387246698141098,
-      "eval_runtime": 4.8866,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 1608
-    },
-    {
-      "epoch": 9.48,
-      "grad_norm": 1.0703125,
-      "learning_rate": 3.051181102362205e-05,
-      "loss": 0.2621,
-      "step": 1616
-    },
-    {
-      "epoch": 9.48,
-      "eval_loss": 0.3960321843624115,
-      "eval_runtime": 4.8825,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 1616
-    },
-    {
-      "epoch": 9.53,
-      "grad_norm": 0.8671875,
-      "learning_rate": 3.0413385826771656e-05,
-      "loss": 0.2597,
-      "step": 1624
-    },
-    {
-      "epoch": 9.53,
-      "eval_loss": 0.3888339102268219,
-      "eval_runtime": 4.8833,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 1624
-    },
-    {
-      "epoch": 9.58,
-      "grad_norm": 1.0703125,
-      "learning_rate": 3.0314960629921263e-05,
-      "loss": 0.2549,
-      "step": 1632
-    },
-    {
-      "epoch": 9.58,
-      "eval_loss": 0.3892790377140045,
-      "eval_runtime": 4.8822,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 1632
-    },
-    {
-      "epoch": 9.63,
-      "grad_norm": 1.1015625,
-      "learning_rate": 3.021653543307087e-05,
-      "loss": 0.2712,
-      "step": 1640
-    },
-    {
-      "epoch": 9.63,
-      "eval_loss": 0.3891643285751343,
-      "eval_runtime": 4.8808,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
-      "step": 1640
-    },
-    {
-      "epoch": 9.67,
-      "grad_norm": 0.9921875,
-      "learning_rate": 3.0118110236220477e-05,
-      "loss": 0.2586,
-      "step": 1648
-    },
-    {
-      "epoch": 9.67,
-      "eval_loss": 0.38737013936042786,
-      "eval_runtime": 4.88,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.23,
-      "step": 1648
-    },
-    {
-      "epoch": 9.72,
-      "grad_norm": 1.0625,
-      "learning_rate": 3.0019685039370083e-05,
-      "loss": 0.2572,
-      "step": 1656
-    },
-    {
-      "epoch": 9.72,
-      "eval_loss": 0.3896384537220001,
-      "eval_runtime": 4.879,
-      "eval_samples_per_second": 9.838,
-      "eval_steps_per_second": 1.23,
-      "step": 1656
-    },
-    {
-      "epoch": 9.77,
-      "grad_norm": 0.98046875,
-      "learning_rate": 2.992125984251969e-05,
-      "loss": 0.2642,
-      "step": 1664
-    },
-    {
-      "epoch": 9.77,
-      "eval_loss": 0.39220455288887024,
-      "eval_runtime": 4.8805,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 1664
-    },
-    {
-      "epoch": 9.81,
-      "grad_norm": 0.9609375,
-      "learning_rate": 2.9822834645669297e-05,
-      "loss": 0.2555,
-      "step": 1672
-    },
-    {
-      "epoch": 9.81,
-      "eval_loss": 0.3866095244884491,
-      "eval_runtime": 4.8806,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 1672
-    },
-    {
-      "epoch": 9.86,
-      "grad_norm": 0.97265625,
-      "learning_rate": 2.97244094488189e-05,
-      "loss": 0.2639,
-      "step": 1680
-    },
-    {
-      "epoch": 9.86,
-      "eval_loss": 0.3891311585903168,
-      "eval_runtime": 4.8849,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 1680
-    },
-    {
-      "epoch": 9.91,
-      "grad_norm": 1.171875,
-      "learning_rate": 2.9625984251968504e-05,
-      "loss": 0.2634,
-      "step": 1688
-    },
-    {
-      "epoch": 9.91,
-      "eval_loss": 0.38789328932762146,
-      "eval_runtime": 4.882,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 1688
-    },
-    {
-      "epoch": 9.95,
-      "grad_norm": 0.953125,
-      "learning_rate": 2.952755905511811e-05,
-      "loss": 0.2638,
-      "step": 1696
-    },
-    {
-      "epoch": 9.95,
-      "eval_loss": 0.38731780648231506,
-      "eval_runtime": 4.8797,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
-      "step": 1696
-    },
-    {
-      "epoch": 10.0,
-      "grad_norm": 1.03125,
-      "learning_rate": 2.9429133858267717e-05,
-      "loss": 0.2873,
-      "step": 1704
-    },
-    {
-      "epoch": 10.0,
-      "eval_loss": 0.3895009458065033,
-      "eval_runtime": 4.8797,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
-      "step": 1704
-    },
-    {
-      "epoch": 10.05,
-      "grad_norm": 1.1796875,
-      "learning_rate": 2.933070866141732e-05,
-      "loss": 0.23,
-      "step": 1712
-    },
-    {
-      "epoch": 10.05,
-      "eval_loss": 0.41275954246520996,
-      "eval_runtime": 4.8847,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 1712
-    },
-    {
-      "epoch": 10.1,
-      "grad_norm": 1.09375,
-      "learning_rate": 2.9232283464566927e-05,
-      "loss": 0.2327,
-      "step": 1720
-    },
-    {
-      "epoch": 10.1,
-      "eval_loss": 0.41290533542633057,
-      "eval_runtime": 4.8825,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 1720
-    },
-    {
-      "epoch": 10.14,
-      "grad_norm": 1.1015625,
-      "learning_rate": 2.9133858267716534e-05,
-      "loss": 0.2434,
-      "step": 1728
-    },
-    {
-      "epoch": 10.14,
-      "eval_loss": 0.4091669023036957,
-      "eval_runtime": 4.8855,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 1728
-    },
-    {
-      "epoch": 10.19,
-      "grad_norm": 1.21875,
-      "learning_rate": 2.903543307086614e-05,
-      "loss": 0.2254,
-      "step": 1736
-    },
-    {
-      "epoch": 10.19,
-      "eval_loss": 0.4104800224304199,
-      "eval_runtime": 4.8818,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 1736
-    },
-    {
-      "epoch": 10.24,
-      "grad_norm": 1.015625,
-      "learning_rate": 2.8937007874015748e-05,
-      "loss": 0.2461,
-      "step": 1744
-    },
-    {
-      "epoch": 10.24,
-      "eval_loss": 0.4064279794692993,
-      "eval_runtime": 4.883,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 1744
-    },
-    {
-      "epoch": 10.28,
-      "grad_norm": 1.046875,
-      "learning_rate": 2.8838582677165354e-05,
-      "loss": 0.2318,
-      "step": 1752
-    },
-    {
-      "epoch": 10.28,
-      "eval_loss": 0.4084310829639435,
-      "eval_runtime": 4.8848,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 1752
-    },
-    {
-      "epoch": 10.33,
-      "grad_norm": 1.1171875,
-      "learning_rate": 2.874015748031496e-05,
-      "loss": 0.2486,
-      "step": 1760
-    },
-    {
-      "epoch": 10.33,
-      "eval_loss": 0.40453585982322693,
-      "eval_runtime": 4.8825,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 1760
-    },
-    {
-      "epoch": 10.38,
-      "grad_norm": 1.0546875,
-      "learning_rate": 2.8641732283464568e-05,
-      "loss": 0.2425,
-      "step": 1768
-    },
-    {
-      "epoch": 10.38,
-      "eval_loss": 0.4052884578704834,
-      "eval_runtime": 4.8845,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 1768
-    },
-    {
-      "epoch": 10.42,
-      "grad_norm": 1.2734375,
-      "learning_rate": 2.8543307086614175e-05,
-      "loss": 0.2369,
-      "step": 1776
-    },
-    {
-      "epoch": 10.42,
-      "eval_loss": 0.4092380702495575,
-      "eval_runtime": 4.8838,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.229,
-      "step": 1776
-    },
-    {
-      "epoch": 10.47,
-      "grad_norm": 1.234375,
-      "learning_rate": 2.844488188976378e-05,
-      "loss": 0.226,
-      "step": 1784
-    },
-    {
-      "epoch": 10.47,
-      "eval_loss": 0.40878424048423767,
-      "eval_runtime": 4.8818,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 1784
-    },
-    {
-      "epoch": 10.52,
-      "grad_norm": 1.0234375,
-      "learning_rate": 2.8346456692913388e-05,
-      "loss": 0.242,
-      "step": 1792
-    },
-    {
-      "epoch": 10.52,
-      "eval_loss": 0.40601250529289246,
-      "eval_runtime": 4.889,
-      "eval_samples_per_second": 9.818,
-      "eval_steps_per_second": 1.227,
-      "step": 1792
-    },
-    {
-      "epoch": 10.56,
-      "grad_norm": 1.09375,
-      "learning_rate": 2.824803149606299e-05,
-      "loss": 0.261,
-      "step": 1800
-    },
-    {
-      "epoch": 10.56,
-      "eval_loss": 0.4090254604816437,
-      "eval_runtime": 4.877,
-      "eval_samples_per_second": 9.842,
-      "eval_steps_per_second": 1.23,
-      "step": 1800
-    },
-    {
-      "epoch": 10.61,
-      "grad_norm": 1.1015625,
-      "learning_rate": 2.81496062992126e-05,
-      "loss": 0.2375,
-      "step": 1808
-    },
-    {
-      "epoch": 10.61,
-      "eval_loss": 0.4072287082672119,
-      "eval_runtime": 4.8878,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.228,
-      "step": 1808
-    },
-    {
-      "epoch": 10.66,
-      "grad_norm": 1.046875,
-      "learning_rate": 2.8051181102362205e-05,
-      "loss": 0.2557,
-      "step": 1816
-    },
-    {
-      "epoch": 10.66,
-      "eval_loss": 0.4026179611682892,
-      "eval_runtime": 4.8857,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 1816
-    },
-    {
-      "epoch": 10.71,
-      "grad_norm": 1.15625,
-      "learning_rate": 2.7952755905511812e-05,
-      "loss": 0.2389,
-      "step": 1824
-    },
-    {
-      "epoch": 10.71,
-      "eval_loss": 0.4051065742969513,
-      "eval_runtime": 4.879,
-      "eval_samples_per_second": 9.838,
-      "eval_steps_per_second": 1.23,
-      "step": 1824
-    },
-    {
-      "epoch": 10.75,
-      "grad_norm": 1.375,
-      "learning_rate": 2.785433070866142e-05,
-      "loss": 0.2481,
-      "step": 1832
-    },
-    {
-      "epoch": 10.75,
-      "eval_loss": 0.4018687307834625,
-      "eval_runtime": 4.8831,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 1832
-    },
-    {
-      "epoch": 10.8,
-      "grad_norm": 1.046875,
-      "learning_rate": 2.7755905511811025e-05,
-      "loss": 0.2612,
-      "step": 1840
-    },
-    {
-      "epoch": 10.8,
-      "eval_loss": 0.40441223978996277,
-      "eval_runtime": 4.8793,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
-      "step": 1840
-    },
-    {
-      "epoch": 10.85,
-      "grad_norm": 1.2890625,
-      "learning_rate": 2.7657480314960632e-05,
-      "loss": 0.2487,
-      "step": 1848
-    },
-    {
-      "epoch": 10.85,
-      "eval_loss": 0.40596404671669006,
-      "eval_runtime": 4.8789,
-      "eval_samples_per_second": 9.838,
-      "eval_steps_per_second": 1.23,
-      "step": 1848
-    },
-    {
-      "epoch": 10.89,
-      "grad_norm": 1.109375,
-      "learning_rate": 2.755905511811024e-05,
-      "loss": 0.2339,
-      "step": 1856
-    },
-    {
-      "epoch": 10.89,
-      "eval_loss": 0.4074425995349884,
-      "eval_runtime": 4.8799,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.23,
-      "step": 1856
-    },
-    {
-      "epoch": 10.94,
-      "grad_norm": 1.046875,
-      "learning_rate": 2.7460629921259846e-05,
-      "loss": 0.2356,
-      "step": 1864
-    },
-    {
-      "epoch": 10.94,
-      "eval_loss": 0.4032224714756012,
-      "eval_runtime": 4.8798,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.23,
-      "step": 1864
-    },
-    {
-      "epoch": 10.99,
-      "grad_norm": 1.125,
-      "learning_rate": 2.7362204724409452e-05,
-      "loss": 0.2452,
-      "step": 1872
-    },
-    {
-      "epoch": 10.99,
-      "eval_loss": 0.40676093101501465,
-      "eval_runtime": 4.8865,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 1872
-    },
-    {
-      "epoch": 11.03,
-      "grad_norm": 1.2890625,
-      "learning_rate": 2.726377952755906e-05,
-      "loss": 0.2151,
-      "step": 1880
-    },
-    {
-      "epoch": 11.03,
-      "eval_loss": 0.4335891306400299,
-      "eval_runtime": 4.8848,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 1880
-    },
-    {
-      "epoch": 11.08,
-      "grad_norm": 1.1796875,
-      "learning_rate": 2.7165354330708666e-05,
-      "loss": 0.2191,
-      "step": 1888
-    },
-    {
-      "epoch": 11.08,
-      "eval_loss": 0.4158717095851898,
-      "eval_runtime": 4.8824,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 1888
-    },
-    {
-      "epoch": 11.13,
-      "grad_norm": 1.15625,
-      "learning_rate": 2.706692913385827e-05,
-      "loss": 0.2186,
-      "step": 1896
-    },
-    {
-      "epoch": 11.13,
-      "eval_loss": 0.41939032077789307,
-      "eval_runtime": 4.8848,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 1896
-    },
-    {
-      "epoch": 11.18,
-      "grad_norm": 1.1875,
-      "learning_rate": 2.6968503937007876e-05,
-      "loss": 0.2177,
-      "step": 1904
-    },
-    {
-      "epoch": 11.18,
-      "eval_loss": 0.4248329699039459,
-      "eval_runtime": 4.8806,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 1904
-    },
-    {
-      "epoch": 11.22,
-      "grad_norm": 1.28125,
-      "learning_rate": 2.6870078740157483e-05,
-      "loss": 0.2203,
-      "step": 1912
-    },
-    {
-      "epoch": 11.22,
-      "eval_loss": 0.4253775179386139,
-      "eval_runtime": 4.878,
-      "eval_samples_per_second": 9.84,
-      "eval_steps_per_second": 1.23,
-      "step": 1912
-    },
-    {
-      "epoch": 11.27,
-      "grad_norm": 1.265625,
-      "learning_rate": 2.677165354330709e-05,
-      "loss": 0.2151,
-      "step": 1920
-    },
-    {
-      "epoch": 11.27,
-      "eval_loss": 0.4222745895385742,
-      "eval_runtime": 4.8805,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 1920
-    },
-    {
-      "epoch": 11.32,
-      "grad_norm": 1.2109375,
-      "learning_rate": 2.6673228346456696e-05,
-      "loss": 0.2197,
-      "step": 1928
-    },
-    {
-      "epoch": 11.32,
-      "eval_loss": 0.42230936884880066,
-      "eval_runtime": 4.8826,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 1928
-    },
-    {
-      "epoch": 11.36,
-      "grad_norm": 1.3515625,
-      "learning_rate": 2.6574803149606303e-05,
-      "loss": 0.2138,
-      "step": 1936
-    },
-    {
-      "epoch": 11.36,
-      "eval_loss": 0.4267289936542511,
-      "eval_runtime": 4.8854,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 1936
-    },
-    {
-      "epoch": 11.41,
-      "grad_norm": 1.2421875,
-      "learning_rate": 2.6476377952755903e-05,
-      "loss": 0.2217,
-      "step": 1944
-    },
-    {
-      "epoch": 11.41,
-      "eval_loss": 0.4230763018131256,
-      "eval_runtime": 4.8832,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 1944
-    },
-    {
-      "epoch": 11.46,
-      "grad_norm": 1.1640625,
-      "learning_rate": 2.637795275590551e-05,
-      "loss": 0.2189,
-      "step": 1952
-    },
-    {
-      "epoch": 11.46,
-      "eval_loss": 0.42412805557250977,
-      "eval_runtime": 4.8848,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 1952
-    },
-    {
-      "epoch": 11.5,
-      "grad_norm": 1.2265625,
-      "learning_rate": 2.6279527559055117e-05,
-      "loss": 0.2336,
-      "step": 1960
-    },
-    {
-      "epoch": 11.5,
-      "eval_loss": 0.42926904559135437,
-      "eval_runtime": 4.8945,
-      "eval_samples_per_second": 9.807,
-      "eval_steps_per_second": 1.226,
-      "step": 1960
-    },
-    {
-      "epoch": 11.55,
-      "grad_norm": 1.390625,
-      "learning_rate": 2.6181102362204723e-05,
-      "loss": 0.2146,
-      "step": 1968
-    },
-    {
-      "epoch": 11.55,
-      "eval_loss": 0.4218541085720062,
-      "eval_runtime": 4.8946,
-      "eval_samples_per_second": 9.807,
-      "eval_steps_per_second": 1.226,
-      "step": 1968
-    },
-    {
-      "epoch": 11.6,
-      "grad_norm": 1.28125,
-      "learning_rate": 2.608267716535433e-05,
-      "loss": 0.2332,
-      "step": 1976
-    },
-    {
-      "epoch": 11.6,
-      "eval_loss": 0.42434871196746826,
-      "eval_runtime": 4.8844,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 1976
-    },
-    {
-      "epoch": 11.64,
-      "grad_norm": 0.98828125,
-      "learning_rate": 2.5984251968503937e-05,
-      "loss": 0.214,
-      "step": 1984
-    },
-    {
-      "epoch": 11.64,
-      "eval_loss": 0.42285728454589844,
-      "eval_runtime": 4.8863,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 1984
-    },
-    {
-      "epoch": 11.69,
-      "grad_norm": 1.2265625,
-      "learning_rate": 2.5885826771653544e-05,
-      "loss": 0.2278,
-      "step": 1992
-    },
-    {
-      "epoch": 11.69,
-      "eval_loss": 0.4248940050601959,
-      "eval_runtime": 4.8801,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.229,
-      "step": 1992
-    },
-    {
-      "epoch": 11.74,
-      "grad_norm": 1.4140625,
-      "learning_rate": 2.578740157480315e-05,
-      "loss": 0.2165,
-      "step": 2000
-    },
-    {
-      "epoch": 11.74,
-      "eval_loss": 0.4275287687778473,
-      "eval_runtime": 4.8821,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 2000
-    },
-    {
-      "epoch": 11.79,
-      "grad_norm": 1.171875,
-      "learning_rate": 2.5688976377952757e-05,
-      "loss": 0.2221,
-      "step": 2008
-    },
-    {
-      "epoch": 11.79,
-      "eval_loss": 0.4152206480503082,
-      "eval_runtime": 4.8863,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 2008
-    },
-    {
-      "epoch": 11.83,
-      "grad_norm": 1.2109375,
-      "learning_rate": 2.5590551181102364e-05,
-      "loss": 0.2238,
-      "step": 2016
-    },
-    {
-      "epoch": 11.83,
-      "eval_loss": 0.4216756522655487,
-      "eval_runtime": 4.8871,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 2016
-    },
-    {
-      "epoch": 11.88,
-      "grad_norm": 1.171875,
-      "learning_rate": 2.5492125984251967e-05,
-      "loss": 0.2166,
-      "step": 2024
-    },
-    {
-      "epoch": 11.88,
-      "eval_loss": 0.4253371059894562,
-      "eval_runtime": 4.8804,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 2024
-    },
-    {
-      "epoch": 11.93,
-      "grad_norm": 1.234375,
-      "learning_rate": 2.5393700787401574e-05,
-      "loss": 0.2366,
-      "step": 2032
-    },
-    {
-      "epoch": 11.93,
-      "eval_loss": 0.42123886942863464,
-      "eval_runtime": 4.8787,
-      "eval_samples_per_second": 9.839,
-      "eval_steps_per_second": 1.23,
-      "step": 2032
-    },
-    {
-      "epoch": 11.97,
-      "grad_norm": 1.4140625,
-      "learning_rate": 2.529527559055118e-05,
-      "loss": 0.2384,
-      "step": 2040
-    },
-    {
-      "epoch": 11.97,
-      "eval_loss": 0.42442429065704346,
-      "eval_runtime": 4.8882,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.227,
-      "step": 2040
-    },
-    {
-      "epoch": 12.02,
-      "grad_norm": 1.328125,
-      "learning_rate": 2.5196850393700788e-05,
-      "loss": 0.2215,
-      "step": 2048
-    },
-    {
-      "epoch": 12.02,
-      "eval_loss": 0.4381261169910431,
-      "eval_runtime": 4.8867,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 2048
-    },
-    {
-      "epoch": 12.07,
-      "grad_norm": 1.3671875,
-      "learning_rate": 2.5098425196850394e-05,
-      "loss": 0.1966,
-      "step": 2056
-    },
-    {
-      "epoch": 12.07,
-      "eval_loss": 0.4352915585041046,
-      "eval_runtime": 4.8879,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.228,
-      "step": 2056
-    },
-    {
-      "epoch": 12.11,
-      "grad_norm": 1.2265625,
-      "learning_rate": 2.5e-05,
-      "loss": 0.1956,
-      "step": 2064
-    },
-    {
-      "epoch": 12.11,
-      "eval_loss": 0.44628992676734924,
-      "eval_runtime": 4.8866,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 2064
-    },
-    {
-      "epoch": 12.16,
-      "grad_norm": 1.28125,
-      "learning_rate": 2.4901574803149608e-05,
-      "loss": 0.193,
-      "step": 2072
-    },
-    {
-      "epoch": 12.16,
-      "eval_loss": 0.44343388080596924,
-      "eval_runtime": 4.8863,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 2072
-    },
-    {
-      "epoch": 12.21,
-      "grad_norm": 1.390625,
-      "learning_rate": 2.4803149606299215e-05,
-      "loss": 0.1983,
-      "step": 2080
-    },
-    {
-      "epoch": 12.21,
-      "eval_loss": 0.4414764642715454,
-      "eval_runtime": 4.8847,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 2080
-    },
-    {
-      "epoch": 12.26,
-      "grad_norm": 1.2578125,
-      "learning_rate": 2.470472440944882e-05,
-      "loss": 0.205,
-      "step": 2088
-    },
-    {
-      "epoch": 12.26,
-      "eval_loss": 0.44560471177101135,
-      "eval_runtime": 4.8846,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 2088
-    },
-    {
-      "epoch": 12.3,
-      "grad_norm": 1.2578125,
-      "learning_rate": 2.4606299212598428e-05,
-      "loss": 0.2081,
-      "step": 2096
-    },
-    {
-      "epoch": 12.3,
-      "eval_loss": 0.44146040081977844,
-      "eval_runtime": 4.8849,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 2096
-    },
-    {
-      "epoch": 12.35,
-      "grad_norm": 1.6953125,
-      "learning_rate": 2.4507874015748035e-05,
-      "loss": 0.1933,
-      "step": 2104
-    },
-    {
-      "epoch": 12.35,
-      "eval_loss": 0.44217804074287415,
-      "eval_runtime": 4.8814,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 2104
-    },
-    {
-      "epoch": 12.4,
-      "grad_norm": 1.3984375,
-      "learning_rate": 2.440944881889764e-05,
-      "loss": 0.2039,
-      "step": 2112
-    },
-    {
-      "epoch": 12.4,
-      "eval_loss": 0.4430825710296631,
-      "eval_runtime": 4.8902,
-      "eval_samples_per_second": 9.815,
-      "eval_steps_per_second": 1.227,
-      "step": 2112
-    },
-    {
-      "epoch": 12.44,
-      "grad_norm": 1.3046875,
-      "learning_rate": 2.4311023622047245e-05,
-      "loss": 0.1976,
-      "step": 2120
-    },
-    {
-      "epoch": 12.44,
-      "eval_loss": 0.43755796551704407,
-      "eval_runtime": 4.8815,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 2120
-    },
-    {
-      "epoch": 12.49,
-      "grad_norm": 1.296875,
-      "learning_rate": 2.421259842519685e-05,
-      "loss": 0.2136,
-      "step": 2128
-    },
-    {
-      "epoch": 12.49,
-      "eval_loss": 0.43662428855895996,
-      "eval_runtime": 4.8797,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
-      "step": 2128
-    },
-    {
-      "epoch": 12.54,
-      "grad_norm": 1.328125,
-      "learning_rate": 2.4114173228346455e-05,
-      "loss": 0.2026,
-      "step": 2136
-    },
-    {
-      "epoch": 12.54,
-      "eval_loss": 0.44358769059181213,
-      "eval_runtime": 4.8878,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.228,
-      "step": 2136
-    },
-    {
-      "epoch": 12.58,
-      "grad_norm": 1.2421875,
-      "learning_rate": 2.4015748031496062e-05,
-      "loss": 0.2116,
-      "step": 2144
-    },
-    {
-      "epoch": 12.58,
-      "eval_loss": 0.44140565395355225,
-      "eval_runtime": 4.8848,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 2144
-    },
-    {
-      "epoch": 12.63,
-      "grad_norm": 1.3046875,
-      "learning_rate": 2.391732283464567e-05,
-      "loss": 0.2021,
-      "step": 2152
-    },
-    {
-      "epoch": 12.63,
-      "eval_loss": 0.4431957006454468,
-      "eval_runtime": 4.8844,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 2152
-    },
-    {
-      "epoch": 12.68,
-      "grad_norm": 1.4453125,
-      "learning_rate": 2.3818897637795276e-05,
-      "loss": 0.1976,
-      "step": 2160
-    },
-    {
-      "epoch": 12.68,
-      "eval_loss": 0.4439380466938019,
-      "eval_runtime": 4.8845,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 2160
-    },
-    {
-      "epoch": 12.72,
-      "grad_norm": 1.21875,
-      "learning_rate": 2.3720472440944882e-05,
-      "loss": 0.2031,
-      "step": 2168
-    },
-    {
-      "epoch": 12.72,
-      "eval_loss": 0.4403323233127594,
-      "eval_runtime": 4.8836,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 2168
-    },
-    {
-      "epoch": 12.77,
-      "grad_norm": 1.375,
-      "learning_rate": 2.362204724409449e-05,
-      "loss": 0.222,
-      "step": 2176
-    },
-    {
-      "epoch": 12.77,
-      "eval_loss": 0.4379936158657074,
-      "eval_runtime": 4.8834,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 2176
-    },
-    {
-      "epoch": 12.82,
-      "grad_norm": 1.2890625,
-      "learning_rate": 2.3523622047244096e-05,
-      "loss": 0.2128,
-      "step": 2184
-    },
-    {
-      "epoch": 12.82,
-      "eval_loss": 0.43629226088523865,
-      "eval_runtime": 4.8828,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 2184
-    },
-    {
-      "epoch": 12.87,
-      "grad_norm": 1.453125,
-      "learning_rate": 2.3425196850393703e-05,
-      "loss": 0.2014,
-      "step": 2192
-    },
-    {
-      "epoch": 12.87,
-      "eval_loss": 0.4432964324951172,
-      "eval_runtime": 4.887,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 2192
-    },
-    {
-      "epoch": 12.91,
-      "grad_norm": 1.296875,
-      "learning_rate": 2.332677165354331e-05,
-      "loss": 0.2151,
-      "step": 2200
-    },
-    {
-      "epoch": 12.91,
-      "eval_loss": 0.4419555366039276,
-      "eval_runtime": 4.8841,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.228,
-      "step": 2200
-    },
-    {
-      "epoch": 12.96,
-      "grad_norm": 1.4765625,
-      "learning_rate": 2.3228346456692916e-05,
-      "loss": 0.194,
-      "step": 2208
-    },
-    {
-      "epoch": 12.96,
-      "eval_loss": 0.44093310832977295,
-      "eval_runtime": 4.8894,
-      "eval_samples_per_second": 9.817,
-      "eval_steps_per_second": 1.227,
-      "step": 2208
-    },
-    {
-      "epoch": 13.01,
-      "grad_norm": 1.1875,
-      "learning_rate": 2.3129921259842523e-05,
-      "loss": 0.2048,
-      "step": 2216
-    },
-    {
-      "epoch": 13.01,
-      "eval_loss": 0.44801631569862366,
-      "eval_runtime": 4.8873,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 2216
-    },
-    {
-      "epoch": 13.05,
-      "grad_norm": 1.25,
-      "learning_rate": 2.3031496062992126e-05,
-      "loss": 0.1913,
-      "step": 2224
-    },
-    {
-      "epoch": 13.05,
-      "eval_loss": 0.45641574263572693,
-      "eval_runtime": 4.8946,
-      "eval_samples_per_second": 9.807,
-      "eval_steps_per_second": 1.226,
-      "step": 2224
-    },
-    {
-      "epoch": 13.1,
-      "grad_norm": 1.5625,
-      "learning_rate": 2.2933070866141733e-05,
-      "loss": 0.1817,
-      "step": 2232
-    },
-    {
-      "epoch": 13.1,
-      "eval_loss": 0.4665268361568451,
-      "eval_runtime": 4.8876,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 2232
-    },
-    {
-      "epoch": 13.15,
-      "grad_norm": 1.5,
-      "learning_rate": 2.283464566929134e-05,
-      "loss": 0.1855,
-      "step": 2240
-    },
-    {
-      "epoch": 13.15,
-      "eval_loss": 0.462589293718338,
-      "eval_runtime": 4.8824,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 2240
-    },
-    {
-      "epoch": 13.19,
-      "grad_norm": 1.3671875,
-      "learning_rate": 2.2736220472440947e-05,
-      "loss": 0.1776,
-      "step": 2248
-    },
-    {
-      "epoch": 13.19,
-      "eval_loss": 0.45993486046791077,
-      "eval_runtime": 4.8821,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 2248
-    },
-    {
-      "epoch": 13.24,
-      "grad_norm": 1.3828125,
-      "learning_rate": 2.263779527559055e-05,
-      "loss": 0.1895,
-      "step": 2256
-    },
-    {
-      "epoch": 13.24,
-      "eval_loss": 0.46107491850852966,
-      "eval_runtime": 4.8877,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 2256
-    },
-    {
-      "epoch": 13.29,
-      "grad_norm": 1.421875,
-      "learning_rate": 2.2539370078740157e-05,
-      "loss": 0.1871,
-      "step": 2264
-    },
-    {
-      "epoch": 13.29,
-      "eval_loss": 0.4591566324234009,
-      "eval_runtime": 4.8853,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 2264
-    },
-    {
-      "epoch": 13.34,
-      "grad_norm": 1.3125,
-      "learning_rate": 2.2440944881889763e-05,
-      "loss": 0.1834,
-      "step": 2272
-    },
-    {
-      "epoch": 13.34,
-      "eval_loss": 0.45331981778144836,
-      "eval_runtime": 4.8872,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 2272
-    },
-    {
-      "epoch": 13.38,
-      "grad_norm": 1.390625,
-      "learning_rate": 2.234251968503937e-05,
-      "loss": 0.1726,
-      "step": 2280
-    },
-    {
-      "epoch": 13.38,
-      "eval_loss": 0.46644964814186096,
-      "eval_runtime": 4.8868,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 2280
-    },
-    {
-      "epoch": 13.43,
-      "grad_norm": 1.390625,
-      "learning_rate": 2.2244094488188977e-05,
-      "loss": 0.1846,
-      "step": 2288
-    },
-    {
-      "epoch": 13.43,
-      "eval_loss": 0.45397841930389404,
-      "eval_runtime": 4.8828,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 2288
-    },
-    {
-      "epoch": 13.48,
-      "grad_norm": 1.5703125,
-      "learning_rate": 2.2145669291338584e-05,
-      "loss": 0.187,
-      "step": 2296
-    },
-    {
-      "epoch": 13.48,
-      "eval_loss": 0.459914892911911,
-      "eval_runtime": 4.8861,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 2296
-    },
-    {
-      "epoch": 13.52,
-      "grad_norm": 1.4140625,
-      "learning_rate": 2.204724409448819e-05,
-      "loss": 0.1731,
-      "step": 2304
-    },
-    {
-      "epoch": 13.52,
-      "eval_loss": 0.45958980917930603,
-      "eval_runtime": 4.8898,
-      "eval_samples_per_second": 9.816,
-      "eval_steps_per_second": 1.227,
-      "step": 2304
-    },
-    {
-      "epoch": 13.57,
-      "grad_norm": 1.578125,
-      "learning_rate": 2.1948818897637797e-05,
-      "loss": 0.1923,
-      "step": 2312
-    },
-    {
-      "epoch": 13.57,
-      "eval_loss": 0.4603557884693146,
-      "eval_runtime": 4.8813,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 2312
-    },
-    {
-      "epoch": 13.62,
-      "grad_norm": 1.328125,
-      "learning_rate": 2.1850393700787404e-05,
-      "loss": 0.1867,
-      "step": 2320
-    },
-    {
-      "epoch": 13.62,
-      "eval_loss": 0.4594487249851227,
-      "eval_runtime": 4.8808,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
-      "step": 2320
-    },
-    {
-      "epoch": 13.66,
-      "grad_norm": 1.390625,
-      "learning_rate": 2.175196850393701e-05,
-      "loss": 0.2047,
-      "step": 2328
-    },
-    {
-      "epoch": 13.66,
-      "eval_loss": 0.46039581298828125,
-      "eval_runtime": 4.8857,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 2328
-    },
-    {
-      "epoch": 13.71,
-      "grad_norm": 1.59375,
-      "learning_rate": 2.1653543307086614e-05,
-      "loss": 0.1851,
-      "step": 2336
-    },
-    {
-      "epoch": 13.71,
-      "eval_loss": 0.4601996839046478,
-      "eval_runtime": 4.8881,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.227,
-      "step": 2336
-    },
-    {
-      "epoch": 13.76,
-      "grad_norm": 1.4765625,
-      "learning_rate": 2.155511811023622e-05,
-      "loss": 0.1905,
-      "step": 2344
-    },
-    {
-      "epoch": 13.76,
-      "eval_loss": 0.4591521918773651,
-      "eval_runtime": 4.8801,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.229,
-      "step": 2344
-    },
-    {
-      "epoch": 13.8,
-      "grad_norm": 1.3515625,
-      "learning_rate": 2.1456692913385828e-05,
-      "loss": 0.1998,
-      "step": 2352
-    },
-    {
-      "epoch": 13.8,
-      "eval_loss": 0.4632813036441803,
-      "eval_runtime": 4.8892,
-      "eval_samples_per_second": 9.818,
-      "eval_steps_per_second": 1.227,
-      "step": 2352
-    },
-    {
-      "epoch": 13.85,
-      "grad_norm": 1.34375,
-      "learning_rate": 2.1358267716535434e-05,
-      "loss": 0.1889,
-      "step": 2360
-    },
-    {
-      "epoch": 13.85,
-      "eval_loss": 0.45855093002319336,
-      "eval_runtime": 4.8879,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.228,
-      "step": 2360
-    },
-    {
-      "epoch": 13.9,
-      "grad_norm": 1.3984375,
-      "learning_rate": 2.125984251968504e-05,
-      "loss": 0.2007,
-      "step": 2368
-    },
-    {
-      "epoch": 13.9,
-      "eval_loss": 0.46005454659461975,
-      "eval_runtime": 4.8909,
-      "eval_samples_per_second": 9.814,
-      "eval_steps_per_second": 1.227,
-      "step": 2368
-    },
-    {
-      "epoch": 13.95,
-      "grad_norm": 1.484375,
-      "learning_rate": 2.1161417322834648e-05,
-      "loss": 0.1929,
-      "step": 2376
-    },
-    {
-      "epoch": 13.95,
-      "eval_loss": 0.46431541442871094,
-      "eval_runtime": 4.8851,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 2376
-    },
-    {
-      "epoch": 13.99,
-      "grad_norm": 1.375,
-      "learning_rate": 2.106299212598425e-05,
-      "loss": 0.1928,
-      "step": 2384
-    },
-    {
-      "epoch": 13.99,
-      "eval_loss": 0.4559873342514038,
-      "eval_runtime": 4.8813,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 2384
-    },
-    {
-      "epoch": 14.04,
-      "grad_norm": 1.234375,
-      "learning_rate": 2.0964566929133858e-05,
-      "loss": 0.1781,
-      "step": 2392
-    },
-    {
-      "epoch": 14.04,
-      "eval_loss": 0.4926118850708008,
-      "eval_runtime": 4.8853,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 2392
-    },
-    {
-      "epoch": 14.09,
-      "grad_norm": 1.4453125,
-      "learning_rate": 2.0866141732283465e-05,
-      "loss": 0.1716,
-      "step": 2400
-    },
-    {
-      "epoch": 14.09,
-      "eval_loss": 0.4850791394710541,
-      "eval_runtime": 4.8833,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 2400
-    },
-    {
-      "epoch": 14.13,
-      "grad_norm": 1.421875,
-      "learning_rate": 2.076771653543307e-05,
-      "loss": 0.1757,
-      "step": 2408
-    },
-    {
-      "epoch": 14.13,
-      "eval_loss": 0.4749085605144501,
-      "eval_runtime": 4.8853,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 2408
-    },
-    {
-      "epoch": 14.18,
-      "grad_norm": 1.515625,
-      "learning_rate": 2.066929133858268e-05,
-      "loss": 0.1738,
-      "step": 2416
-    },
-    {
-      "epoch": 14.18,
-      "eval_loss": 0.4851483404636383,
-      "eval_runtime": 4.891,
-      "eval_samples_per_second": 9.814,
-      "eval_steps_per_second": 1.227,
-      "step": 2416
-    },
-    {
-      "epoch": 14.23,
-      "grad_norm": 1.4296875,
-      "learning_rate": 2.0570866141732285e-05,
-      "loss": 0.1738,
-      "step": 2424
-    },
-    {
-      "epoch": 14.23,
-      "eval_loss": 0.4827621281147003,
-      "eval_runtime": 4.8822,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 2424
-    },
-    {
-      "epoch": 14.27,
-      "grad_norm": 1.4609375,
-      "learning_rate": 2.0472440944881892e-05,
-      "loss": 0.1717,
-      "step": 2432
-    },
-    {
-      "epoch": 14.27,
-      "eval_loss": 0.4806636571884155,
-      "eval_runtime": 4.8847,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 2432
-    },
-    {
-      "epoch": 14.32,
-      "grad_norm": 2.1875,
-      "learning_rate": 2.0374015748031495e-05,
-      "loss": 0.1725,
-      "step": 2440
-    },
-    {
-      "epoch": 14.32,
-      "eval_loss": 0.48100754618644714,
-      "eval_runtime": 4.8883,
-      "eval_samples_per_second": 9.819,
-      "eval_steps_per_second": 1.227,
-      "step": 2440
-    },
-    {
-      "epoch": 14.37,
-      "grad_norm": 1.59375,
-      "learning_rate": 2.0275590551181102e-05,
-      "loss": 0.1798,
-      "step": 2448
-    },
-    {
-      "epoch": 14.37,
-      "eval_loss": 0.48213496804237366,
-      "eval_runtime": 4.8914,
-      "eval_samples_per_second": 9.813,
-      "eval_steps_per_second": 1.227,
-      "step": 2448
-    },
-    {
-      "epoch": 14.42,
-      "grad_norm": 1.3046875,
-      "learning_rate": 2.017716535433071e-05,
-      "loss": 0.1827,
-      "step": 2456
-    },
-    {
-      "epoch": 14.42,
-      "eval_loss": 0.48051026463508606,
-      "eval_runtime": 4.8852,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 2456
-    },
-    {
-      "epoch": 14.46,
-      "grad_norm": 1.4375,
-      "learning_rate": 2.0078740157480316e-05,
-      "loss": 0.1718,
-      "step": 2464
-    },
-    {
-      "epoch": 14.46,
-      "eval_loss": 0.4809204638004303,
-      "eval_runtime": 4.8859,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 2464
-    },
-    {
-      "epoch": 14.51,
-      "grad_norm": 1.453125,
-      "learning_rate": 1.9980314960629922e-05,
-      "loss": 0.1827,
-      "step": 2472
-    },
-    {
-      "epoch": 14.51,
-      "eval_loss": 0.48105254769325256,
-      "eval_runtime": 4.8876,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 2472
-    },
-    {
-      "epoch": 14.56,
-      "grad_norm": 1.4921875,
-      "learning_rate": 1.988188976377953e-05,
-      "loss": 0.169,
-      "step": 2480
-    },
-    {
-      "epoch": 14.56,
-      "eval_loss": 0.4779665768146515,
-      "eval_runtime": 4.8866,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 2480
-    },
-    {
-      "epoch": 14.6,
-      "grad_norm": 1.53125,
-      "learning_rate": 1.9783464566929136e-05,
-      "loss": 0.1802,
-      "step": 2488
-    },
-    {
-      "epoch": 14.6,
-      "eval_loss": 0.4763511121273041,
-      "eval_runtime": 4.882,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 2488
-    },
-    {
-      "epoch": 14.65,
-      "grad_norm": 1.484375,
-      "learning_rate": 1.9685039370078743e-05,
-      "loss": 0.1742,
-      "step": 2496
-    },
-    {
-      "epoch": 14.65,
-      "eval_loss": 0.4802992641925812,
-      "eval_runtime": 4.8869,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 2496
-    },
-    {
-      "epoch": 14.7,
-      "grad_norm": 1.5390625,
-      "learning_rate": 1.958661417322835e-05,
-      "loss": 0.1724,
-      "step": 2504
-    },
-    {
-      "epoch": 14.7,
-      "eval_loss": 0.47574201226234436,
-      "eval_runtime": 4.8863,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 2504
-    },
-    {
-      "epoch": 14.74,
-      "grad_norm": 1.5234375,
-      "learning_rate": 1.9488188976377953e-05,
-      "loss": 0.1697,
-      "step": 2512
-    },
-    {
-      "epoch": 14.74,
-      "eval_loss": 0.47717785835266113,
-      "eval_runtime": 4.8813,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 2512
-    },
-    {
-      "epoch": 14.79,
-      "grad_norm": 1.40625,
-      "learning_rate": 1.938976377952756e-05,
-      "loss": 0.1757,
-      "step": 2520
-    },
-    {
-      "epoch": 14.79,
-      "eval_loss": 0.48033222556114197,
-      "eval_runtime": 4.8883,
-      "eval_samples_per_second": 9.819,
-      "eval_steps_per_second": 1.227,
-      "step": 2520
-    },
-    {
-      "epoch": 14.84,
-      "grad_norm": 1.515625,
-      "learning_rate": 1.9291338582677166e-05,
-      "loss": 0.1758,
-      "step": 2528
-    },
-    {
-      "epoch": 14.84,
-      "eval_loss": 0.4787493050098419,
-      "eval_runtime": 4.8859,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 2528
-    },
-    {
-      "epoch": 14.88,
-      "grad_norm": 1.3828125,
-      "learning_rate": 1.9192913385826773e-05,
-      "loss": 0.1643,
-      "step": 2536
-    },
-    {
-      "epoch": 14.88,
-      "eval_loss": 0.4781215488910675,
-      "eval_runtime": 4.8837,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 2536
-    },
-    {
-      "epoch": 14.93,
-      "grad_norm": 1.4453125,
-      "learning_rate": 1.909448818897638e-05,
-      "loss": 0.1714,
-      "step": 2544
-    },
-    {
-      "epoch": 14.93,
-      "eval_loss": 0.47745776176452637,
-      "eval_runtime": 4.8844,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 2544
-    },
-    {
-      "epoch": 14.98,
-      "grad_norm": 1.5,
-      "learning_rate": 1.8996062992125983e-05,
-      "loss": 0.176,
-      "step": 2552
-    },
-    {
-      "epoch": 14.98,
-      "eval_loss": 0.47671830654144287,
-      "eval_runtime": 4.8827,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 2552
-    },
-    {
-      "epoch": 15.03,
-      "grad_norm": 1.703125,
-      "learning_rate": 1.889763779527559e-05,
-      "loss": 0.1609,
-      "step": 2560
-    },
-    {
-      "epoch": 15.03,
-      "eval_loss": 0.5071304440498352,
-      "eval_runtime": 4.8789,
-      "eval_samples_per_second": 9.838,
-      "eval_steps_per_second": 1.23,
-      "step": 2560
-    },
-    {
-      "epoch": 15.07,
-      "grad_norm": 1.28125,
-      "learning_rate": 1.8799212598425197e-05,
-      "loss": 0.1545,
-      "step": 2568
-    },
-    {
-      "epoch": 15.07,
-      "eval_loss": 0.49048110842704773,
-      "eval_runtime": 4.8841,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.228,
-      "step": 2568
-    },
-    {
-      "epoch": 15.12,
-      "grad_norm": 1.4140625,
-      "learning_rate": 1.8700787401574803e-05,
-      "loss": 0.1591,
-      "step": 2576
-    },
-    {
-      "epoch": 15.12,
-      "eval_loss": 0.5035098195075989,
-      "eval_runtime": 4.8866,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 2576
-    },
-    {
-      "epoch": 15.17,
-      "grad_norm": 1.3828125,
-      "learning_rate": 1.860236220472441e-05,
-      "loss": 0.1488,
-      "step": 2584
-    },
-    {
-      "epoch": 15.17,
-      "eval_loss": 0.49650588631629944,
-      "eval_runtime": 4.8876,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 2584
-    },
-    {
-      "epoch": 15.21,
-      "grad_norm": 1.546875,
-      "learning_rate": 1.8503937007874017e-05,
-      "loss": 0.1605,
-      "step": 2592
-    },
-    {
-      "epoch": 15.21,
-      "eval_loss": 0.5021148920059204,
-      "eval_runtime": 4.8938,
-      "eval_samples_per_second": 9.808,
-      "eval_steps_per_second": 1.226,
-      "step": 2592
-    },
-    {
-      "epoch": 15.26,
-      "grad_norm": 1.53125,
-      "learning_rate": 1.8405511811023624e-05,
-      "loss": 0.1453,
-      "step": 2600
-    },
-    {
-      "epoch": 15.26,
-      "eval_loss": 0.49250075221061707,
-      "eval_runtime": 4.8904,
-      "eval_samples_per_second": 9.815,
-      "eval_steps_per_second": 1.227,
-      "step": 2600
-    },
-    {
-      "epoch": 15.31,
-      "grad_norm": 1.6328125,
-      "learning_rate": 1.830708661417323e-05,
-      "loss": 0.1577,
-      "step": 2608
-    },
-    {
-      "epoch": 15.31,
-      "eval_loss": 0.503433883190155,
-      "eval_runtime": 4.8812,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
-      "step": 2608
-    },
-    {
-      "epoch": 15.35,
-      "grad_norm": 1.421875,
-      "learning_rate": 1.8208661417322837e-05,
-      "loss": 0.1643,
-      "step": 2616
-    },
-    {
-      "epoch": 15.35,
-      "eval_loss": 0.4985417127609253,
-      "eval_runtime": 4.8869,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 2616
-    },
-    {
-      "epoch": 15.4,
-      "grad_norm": 1.59375,
-      "learning_rate": 1.8110236220472444e-05,
-      "loss": 0.1581,
-      "step": 2624
-    },
-    {
-      "epoch": 15.4,
-      "eval_loss": 0.500441312789917,
-      "eval_runtime": 4.889,
-      "eval_samples_per_second": 9.818,
-      "eval_steps_per_second": 1.227,
-      "step": 2624
-    },
-    {
-      "epoch": 15.45,
-      "grad_norm": 1.53125,
-      "learning_rate": 1.801181102362205e-05,
-      "loss": 0.1597,
-      "step": 2632
-    },
-    {
-      "epoch": 15.45,
-      "eval_loss": 0.5028078556060791,
-      "eval_runtime": 4.8809,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
-      "step": 2632
-    },
-    {
-      "epoch": 15.5,
-      "grad_norm": 1.671875,
-      "learning_rate": 1.7913385826771654e-05,
-      "loss": 0.1689,
-      "step": 2640
-    },
-    {
-      "epoch": 15.5,
-      "eval_loss": 0.49254146218299866,
-      "eval_runtime": 4.8836,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 2640
-    },
-    {
-      "epoch": 15.54,
-      "grad_norm": 1.5625,
-      "learning_rate": 1.781496062992126e-05,
-      "loss": 0.1709,
-      "step": 2648
-    },
-    {
-      "epoch": 15.54,
-      "eval_loss": 0.4996674358844757,
-      "eval_runtime": 4.8881,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.227,
-      "step": 2648
-    },
-    {
-      "epoch": 15.59,
-      "grad_norm": 1.5,
-      "learning_rate": 1.7716535433070868e-05,
-      "loss": 0.1579,
-      "step": 2656
-    },
-    {
-      "epoch": 15.59,
-      "eval_loss": 0.49186739325523376,
-      "eval_runtime": 4.8863,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 2656
-    },
-    {
-      "epoch": 15.64,
-      "grad_norm": 1.40625,
-      "learning_rate": 1.761811023622047e-05,
-      "loss": 0.1564,
-      "step": 2664
-    },
-    {
-      "epoch": 15.64,
-      "eval_loss": 0.5025432705879211,
-      "eval_runtime": 4.8881,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.227,
-      "step": 2664
-    },
-    {
-      "epoch": 15.68,
-      "grad_norm": 1.640625,
-      "learning_rate": 1.7519685039370078e-05,
-      "loss": 0.1708,
-      "step": 2672
-    },
-    {
-      "epoch": 15.68,
-      "eval_loss": 0.49223020672798157,
-      "eval_runtime": 4.8883,
-      "eval_samples_per_second": 9.819,
-      "eval_steps_per_second": 1.227,
-      "step": 2672
-    },
-    {
-      "epoch": 15.73,
-      "grad_norm": 1.5390625,
-      "learning_rate": 1.7421259842519685e-05,
-      "loss": 0.1723,
-      "step": 2680
-    },
-    {
-      "epoch": 15.73,
-      "eval_loss": 0.49755749106407166,
-      "eval_runtime": 4.8924,
-      "eval_samples_per_second": 9.811,
-      "eval_steps_per_second": 1.226,
-      "step": 2680
-    },
-    {
-      "epoch": 15.78,
-      "grad_norm": 1.546875,
-      "learning_rate": 1.732283464566929e-05,
-      "loss": 0.1603,
-      "step": 2688
-    },
-    {
-      "epoch": 15.78,
-      "eval_loss": 0.4947017431259155,
-      "eval_runtime": 4.8859,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 2688
-    },
-    {
-      "epoch": 15.82,
-      "grad_norm": 1.6328125,
-      "learning_rate": 1.7224409448818898e-05,
-      "loss": 0.1679,
-      "step": 2696
-    },
-    {
-      "epoch": 15.82,
-      "eval_loss": 0.4993188679218292,
-      "eval_runtime": 4.8796,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
-      "step": 2696
-    },
-    {
-      "epoch": 15.87,
-      "grad_norm": 1.4140625,
-      "learning_rate": 1.7125984251968505e-05,
-      "loss": 0.1681,
-      "step": 2704
-    },
-    {
-      "epoch": 15.87,
-      "eval_loss": 0.4942636489868164,
-      "eval_runtime": 4.8866,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 2704
-    },
-    {
-      "epoch": 15.92,
-      "grad_norm": 1.4765625,
-      "learning_rate": 1.702755905511811e-05,
-      "loss": 0.1762,
-      "step": 2712
-    },
-    {
-      "epoch": 15.92,
-      "eval_loss": 0.49583515524864197,
-      "eval_runtime": 4.8908,
-      "eval_samples_per_second": 9.814,
-      "eval_steps_per_second": 1.227,
-      "step": 2712
-    },
-    {
-      "epoch": 15.96,
-      "grad_norm": 1.6015625,
-      "learning_rate": 1.692913385826772e-05,
-      "loss": 0.1639,
-      "step": 2720
-    },
-    {
-      "epoch": 15.96,
-      "eval_loss": 0.4938255846500397,
-      "eval_runtime": 4.8932,
-      "eval_samples_per_second": 9.809,
-      "eval_steps_per_second": 1.226,
-      "step": 2720
-    },
-    {
-      "epoch": 16.01,
-      "grad_norm": 1.2890625,
-      "learning_rate": 1.6830708661417325e-05,
-      "loss": 0.1606,
-      "step": 2728
-    },
-    {
-      "epoch": 16.01,
-      "eval_loss": 0.50501948595047,
-      "eval_runtime": 4.8843,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 2728
-    },
-    {
-      "epoch": 16.06,
-      "grad_norm": 1.5546875,
-      "learning_rate": 1.6732283464566932e-05,
-      "loss": 0.1362,
-      "step": 2736
-    },
-    {
-      "epoch": 16.06,
-      "eval_loss": 0.5190303921699524,
-      "eval_runtime": 4.8864,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 2736
-    },
-    {
-      "epoch": 16.11,
-      "grad_norm": 1.3359375,
-      "learning_rate": 1.663385826771654e-05,
-      "loss": 0.151,
-      "step": 2744
-    },
-    {
-      "epoch": 16.11,
-      "eval_loss": 0.5132401585578918,
-      "eval_runtime": 4.8954,
-      "eval_samples_per_second": 9.805,
-      "eval_steps_per_second": 1.226,
-      "step": 2744
-    },
-    {
-      "epoch": 16.15,
-      "grad_norm": 1.5078125,
-      "learning_rate": 1.6535433070866142e-05,
-      "loss": 0.152,
-      "step": 2752
-    },
-    {
-      "epoch": 16.15,
-      "eval_loss": 0.5155924558639526,
-      "eval_runtime": 4.8785,
-      "eval_samples_per_second": 9.839,
-      "eval_steps_per_second": 1.23,
-      "step": 2752
-    },
-    {
-      "epoch": 16.2,
-      "grad_norm": 1.546875,
-      "learning_rate": 1.643700787401575e-05,
-      "loss": 0.1403,
-      "step": 2760
-    },
-    {
-      "epoch": 16.2,
-      "eval_loss": 0.5209617018699646,
-      "eval_runtime": 4.8897,
-      "eval_samples_per_second": 9.817,
-      "eval_steps_per_second": 1.227,
-      "step": 2760
-    },
-    {
-      "epoch": 16.25,
-      "grad_norm": 1.453125,
-      "learning_rate": 1.6338582677165352e-05,
-      "loss": 0.1486,
-      "step": 2768
-    },
-    {
-      "epoch": 16.25,
-      "eval_loss": 0.5135500431060791,
-      "eval_runtime": 4.8876,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 2768
-    },
-    {
-      "epoch": 16.29,
-      "grad_norm": 1.4921875,
-      "learning_rate": 1.624015748031496e-05,
-      "loss": 0.1498,
-      "step": 2776
-    },
-    {
-      "epoch": 16.29,
-      "eval_loss": 0.5194461345672607,
-      "eval_runtime": 4.8786,
-      "eval_samples_per_second": 9.839,
-      "eval_steps_per_second": 1.23,
-      "step": 2776
-    },
-    {
-      "epoch": 16.34,
-      "grad_norm": 1.4921875,
-      "learning_rate": 1.6141732283464566e-05,
-      "loss": 0.1471,
-      "step": 2784
-    },
-    {
-      "epoch": 16.34,
-      "eval_loss": 0.5094698071479797,
-      "eval_runtime": 4.8841,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.228,
-      "step": 2784
-    },
-    {
-      "epoch": 16.39,
-      "grad_norm": 1.5078125,
-      "learning_rate": 1.6043307086614172e-05,
-      "loss": 0.1514,
-      "step": 2792
-    },
-    {
-      "epoch": 16.39,
-      "eval_loss": 0.5184895992279053,
-      "eval_runtime": 4.8815,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 2792
-    },
-    {
-      "epoch": 16.43,
-      "grad_norm": 1.53125,
-      "learning_rate": 1.594488188976378e-05,
-      "loss": 0.1508,
-      "step": 2800
-    },
-    {
-      "epoch": 16.43,
-      "eval_loss": 0.5169212222099304,
-      "eval_runtime": 4.8823,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 2800
-    },
-    {
-      "epoch": 16.48,
-      "grad_norm": 1.5390625,
-      "learning_rate": 1.5846456692913386e-05,
-      "loss": 0.1459,
-      "step": 2808
-    },
-    {
-      "epoch": 16.48,
-      "eval_loss": 0.5152599215507507,
-      "eval_runtime": 4.8859,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 2808
-    },
-    {
-      "epoch": 16.53,
-      "grad_norm": 1.6015625,
-      "learning_rate": 1.5748031496062993e-05,
-      "loss": 0.1647,
-      "step": 2816
-    },
-    {
-      "epoch": 16.53,
-      "eval_loss": 0.5134853720664978,
-      "eval_runtime": 4.8828,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 2816
-    },
-    {
-      "epoch": 16.58,
-      "grad_norm": 1.625,
-      "learning_rate": 1.56496062992126e-05,
-      "loss": 0.1541,
-      "step": 2824
-    },
-    {
-      "epoch": 16.58,
-      "eval_loss": 0.520317018032074,
-      "eval_runtime": 4.8877,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 2824
-    },
-    {
-      "epoch": 16.62,
-      "grad_norm": 1.53125,
-      "learning_rate": 1.5551181102362206e-05,
-      "loss": 0.1538,
-      "step": 2832
-    },
-    {
-      "epoch": 16.62,
-      "eval_loss": 0.5117902159690857,
-      "eval_runtime": 4.8856,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 2832
-    },
-    {
-      "epoch": 16.67,
-      "grad_norm": 1.640625,
-      "learning_rate": 1.5452755905511813e-05,
-      "loss": 0.1645,
-      "step": 2840
-    },
-    {
-      "epoch": 16.67,
-      "eval_loss": 0.5128079056739807,
-      "eval_runtime": 4.886,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 2840
-    },
-    {
-      "epoch": 16.72,
-      "grad_norm": 1.4765625,
-      "learning_rate": 1.535433070866142e-05,
-      "loss": 0.1544,
-      "step": 2848
-    },
-    {
-      "epoch": 16.72,
-      "eval_loss": 0.5158179402351379,
-      "eval_runtime": 4.8867,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 2848
-    },
-    {
-      "epoch": 16.76,
-      "grad_norm": 1.6875,
-      "learning_rate": 1.5255905511811025e-05,
-      "loss": 0.1552,
-      "step": 2856
-    },
-    {
-      "epoch": 16.76,
-      "eval_loss": 0.5202960968017578,
-      "eval_runtime": 4.8838,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.229,
-      "step": 2856
-    },
-    {
-      "epoch": 16.81,
-      "grad_norm": 1.4921875,
-      "learning_rate": 1.5157480314960632e-05,
-      "loss": 0.1481,
-      "step": 2864
-    },
-    {
-      "epoch": 16.81,
-      "eval_loss": 0.5155553221702576,
-      "eval_runtime": 4.8836,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 2864
-    },
-    {
-      "epoch": 16.86,
-      "grad_norm": 1.3984375,
-      "learning_rate": 1.5059055118110238e-05,
-      "loss": 0.1477,
-      "step": 2872
-    },
-    {
-      "epoch": 16.86,
-      "eval_loss": 0.5139450430870056,
-      "eval_runtime": 4.8861,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 2872
-    },
-    {
-      "epoch": 16.9,
-      "grad_norm": 1.546875,
-      "learning_rate": 1.4960629921259845e-05,
-      "loss": 0.1561,
-      "step": 2880
-    },
-    {
-      "epoch": 16.9,
-      "eval_loss": 0.518673837184906,
-      "eval_runtime": 4.8833,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 2880
-    },
-    {
-      "epoch": 16.95,
-      "grad_norm": 1.4375,
-      "learning_rate": 1.486220472440945e-05,
-      "loss": 0.1398,
-      "step": 2888
-    },
-    {
-      "epoch": 16.95,
-      "eval_loss": 0.5137103796005249,
-      "eval_runtime": 4.8804,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 2888
-    },
-    {
-      "epoch": 17.0,
-      "grad_norm": 1.5859375,
-      "learning_rate": 1.4763779527559055e-05,
-      "loss": 0.16,
-      "step": 2896
-    },
-    {
-      "epoch": 17.0,
-      "eval_loss": 0.5180995464324951,
-      "eval_runtime": 4.8791,
-      "eval_samples_per_second": 9.838,
-      "eval_steps_per_second": 1.23,
-      "step": 2896
-    },
-    {
-      "epoch": 17.04,
-      "grad_norm": 1.8125,
-      "learning_rate": 1.466535433070866e-05,
-      "loss": 0.1346,
-      "step": 2904
-    },
-    {
-      "epoch": 17.04,
-      "eval_loss": 0.5409136414527893,
-      "eval_runtime": 4.8801,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.229,
-      "step": 2904
-    },
-    {
-      "epoch": 17.09,
-      "grad_norm": 1.296875,
-      "learning_rate": 1.4566929133858267e-05,
-      "loss": 0.1391,
-      "step": 2912
-    },
-    {
-      "epoch": 17.09,
-      "eval_loss": 0.527400553226471,
-      "eval_runtime": 4.8828,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 2912
-    },
-    {
-      "epoch": 17.14,
-      "grad_norm": 1.5234375,
-      "learning_rate": 1.4468503937007874e-05,
-      "loss": 0.1392,
-      "step": 2920
-    },
-    {
-      "epoch": 17.14,
-      "eval_loss": 0.5358954071998596,
-      "eval_runtime": 4.885,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 2920
-    },
-    {
-      "epoch": 17.19,
-      "grad_norm": 1.65625,
-      "learning_rate": 1.437007874015748e-05,
-      "loss": 0.1457,
-      "step": 2928
-    },
-    {
-      "epoch": 17.19,
-      "eval_loss": 0.5281966328620911,
-      "eval_runtime": 4.885,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 2928
-    },
-    {
-      "epoch": 17.23,
-      "grad_norm": 1.796875,
-      "learning_rate": 1.4271653543307087e-05,
-      "loss": 0.1423,
-      "step": 2936
-    },
-    {
-      "epoch": 17.23,
-      "eval_loss": 0.539531409740448,
-      "eval_runtime": 4.8847,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 2936
-    },
-    {
-      "epoch": 17.28,
-      "grad_norm": 1.25,
-      "learning_rate": 1.4173228346456694e-05,
-      "loss": 0.1406,
-      "step": 2944
-    },
-    {
-      "epoch": 17.28,
-      "eval_loss": 0.5257437825202942,
-      "eval_runtime": 4.8836,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 2944
-    },
-    {
-      "epoch": 17.33,
-      "grad_norm": 1.3984375,
-      "learning_rate": 1.40748031496063e-05,
-      "loss": 0.1457,
-      "step": 2952
-    },
-    {
-      "epoch": 17.33,
-      "eval_loss": 0.5385013222694397,
-      "eval_runtime": 4.8868,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 2952
-    },
-    {
-      "epoch": 17.37,
-      "grad_norm": 1.4140625,
-      "learning_rate": 1.3976377952755906e-05,
-      "loss": 0.1472,
-      "step": 2960
-    },
-    {
-      "epoch": 17.37,
-      "eval_loss": 0.530069887638092,
-      "eval_runtime": 4.8843,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 2960
-    },
-    {
-      "epoch": 17.42,
-      "grad_norm": 1.6015625,
-      "learning_rate": 1.3877952755905513e-05,
-      "loss": 0.1429,
-      "step": 2968
-    },
-    {
-      "epoch": 17.42,
-      "eval_loss": 0.5348724722862244,
-      "eval_runtime": 4.8847,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 2968
-    },
-    {
-      "epoch": 17.47,
-      "grad_norm": 1.7734375,
-      "learning_rate": 1.377952755905512e-05,
-      "loss": 0.141,
-      "step": 2976
-    },
-    {
-      "epoch": 17.47,
-      "eval_loss": 0.5355264544487,
-      "eval_runtime": 4.8845,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 2976
-    },
-    {
-      "epoch": 17.51,
-      "grad_norm": 1.515625,
-      "learning_rate": 1.3681102362204726e-05,
-      "loss": 0.1391,
-      "step": 2984
-    },
-    {
-      "epoch": 17.51,
-      "eval_loss": 0.530957043170929,
-      "eval_runtime": 4.8859,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 2984
-    },
-    {
-      "epoch": 17.56,
-      "grad_norm": 1.5859375,
-      "learning_rate": 1.3582677165354333e-05,
-      "loss": 0.145,
-      "step": 2992
-    },
-    {
-      "epoch": 17.56,
-      "eval_loss": 0.5390105843544006,
-      "eval_runtime": 4.8852,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 2992
-    },
-    {
-      "epoch": 17.61,
-      "grad_norm": 1.3125,
-      "learning_rate": 1.3484251968503938e-05,
-      "loss": 0.1378,
-      "step": 3000
-    },
-    {
-      "epoch": 17.61,
-      "eval_loss": 0.5294584631919861,
-      "eval_runtime": 4.8791,
-      "eval_samples_per_second": 9.838,
-      "eval_steps_per_second": 1.23,
-      "step": 3000
-    },
-    {
-      "epoch": 17.66,
-      "grad_norm": 1.7265625,
-      "learning_rate": 1.3385826771653545e-05,
-      "loss": 0.1464,
-      "step": 3008
-    },
-    {
-      "epoch": 17.66,
-      "eval_loss": 0.534781813621521,
-      "eval_runtime": 4.8826,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 3008
-    },
-    {
-      "epoch": 17.7,
-      "grad_norm": 1.6171875,
-      "learning_rate": 1.3287401574803152e-05,
-      "loss": 0.142,
-      "step": 3016
-    },
-    {
-      "epoch": 17.7,
-      "eval_loss": 0.5333996415138245,
-      "eval_runtime": 4.8814,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 3016
-    },
-    {
-      "epoch": 17.75,
-      "grad_norm": 1.484375,
-      "learning_rate": 1.3188976377952755e-05,
-      "loss": 0.1484,
-      "step": 3024
-    },
-    {
-      "epoch": 17.75,
-      "eval_loss": 0.5344775319099426,
-      "eval_runtime": 4.8867,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 3024
-    },
-    {
-      "epoch": 17.8,
-      "grad_norm": 1.4296875,
-      "learning_rate": 1.3090551181102362e-05,
-      "loss": 0.1424,
-      "step": 3032
-    },
-    {
-      "epoch": 17.8,
-      "eval_loss": 0.5302088260650635,
-      "eval_runtime": 4.8838,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.229,
-      "step": 3032
-    },
-    {
-      "epoch": 17.84,
-      "grad_norm": 1.421875,
-      "learning_rate": 1.2992125984251968e-05,
-      "loss": 0.1373,
-      "step": 3040
-    },
-    {
-      "epoch": 17.84,
-      "eval_loss": 0.5336248278617859,
-      "eval_runtime": 4.8857,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 3040
-    },
-    {
-      "epoch": 17.89,
-      "grad_norm": 1.4765625,
-      "learning_rate": 1.2893700787401575e-05,
-      "loss": 0.1415,
-      "step": 3048
-    },
-    {
-      "epoch": 17.89,
-      "eval_loss": 0.5314746499061584,
-      "eval_runtime": 4.8897,
-      "eval_samples_per_second": 9.817,
-      "eval_steps_per_second": 1.227,
-      "step": 3048
-    },
-    {
-      "epoch": 17.94,
-      "grad_norm": 1.546875,
-      "learning_rate": 1.2795275590551182e-05,
-      "loss": 0.1497,
-      "step": 3056
-    },
-    {
-      "epoch": 17.94,
-      "eval_loss": 0.534315824508667,
-      "eval_runtime": 4.8827,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 3056
-    },
-    {
-      "epoch": 17.98,
-      "grad_norm": 1.3828125,
-      "learning_rate": 1.2696850393700787e-05,
-      "loss": 0.1413,
-      "step": 3064
-    },
-    {
-      "epoch": 17.98,
-      "eval_loss": 0.5319940447807312,
-      "eval_runtime": 4.8902,
-      "eval_samples_per_second": 9.815,
-      "eval_steps_per_second": 1.227,
-      "step": 3064
-    },
-    {
-      "epoch": 18.03,
-      "grad_norm": 1.375,
-      "learning_rate": 1.2598425196850394e-05,
-      "loss": 0.1353,
-      "step": 3072
-    },
-    {
-      "epoch": 18.03,
-      "eval_loss": 0.5491126179695129,
-      "eval_runtime": 4.884,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.228,
-      "step": 3072
-    },
-    {
-      "epoch": 18.08,
-      "grad_norm": 1.421875,
-      "learning_rate": 1.25e-05,
-      "loss": 0.1313,
-      "step": 3080
-    },
-    {
-      "epoch": 18.08,
-      "eval_loss": 0.5500498414039612,
-      "eval_runtime": 4.8805,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 3080
-    },
-    {
-      "epoch": 18.12,
-      "grad_norm": 1.578125,
-      "learning_rate": 1.2401574803149607e-05,
-      "loss": 0.1281,
-      "step": 3088
-    },
-    {
-      "epoch": 18.12,
-      "eval_loss": 0.5494930744171143,
-      "eval_runtime": 4.8801,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.229,
-      "step": 3088
-    },
-    {
-      "epoch": 18.17,
-      "grad_norm": 1.4375,
-      "learning_rate": 1.2303149606299214e-05,
-      "loss": 0.1309,
-      "step": 3096
-    },
-    {
-      "epoch": 18.17,
-      "eval_loss": 0.5453672409057617,
-      "eval_runtime": 4.8903,
-      "eval_samples_per_second": 9.815,
-      "eval_steps_per_second": 1.227,
-      "step": 3096
-    },
-    {
-      "epoch": 18.22,
-      "grad_norm": 1.3984375,
-      "learning_rate": 1.220472440944882e-05,
-      "loss": 0.1342,
-      "step": 3104
-    },
-    {
-      "epoch": 18.22,
-      "eval_loss": 0.5458211302757263,
-      "eval_runtime": 4.8823,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 3104
-    },
-    {
-      "epoch": 18.27,
-      "grad_norm": 1.6015625,
-      "learning_rate": 1.2106299212598424e-05,
-      "loss": 0.1272,
-      "step": 3112
-    },
-    {
-      "epoch": 18.27,
-      "eval_loss": 0.5490582585334778,
-      "eval_runtime": 4.8824,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 3112
-    },
-    {
-      "epoch": 18.31,
-      "grad_norm": 1.3984375,
-      "learning_rate": 1.2007874015748031e-05,
-      "loss": 0.1406,
-      "step": 3120
-    },
-    {
-      "epoch": 18.31,
-      "eval_loss": 0.5500710010528564,
-      "eval_runtime": 4.8802,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.229,
-      "step": 3120
-    },
-    {
-      "epoch": 18.36,
-      "grad_norm": 1.453125,
-      "learning_rate": 1.1909448818897638e-05,
-      "loss": 0.1358,
-      "step": 3128
-    },
-    {
-      "epoch": 18.36,
-      "eval_loss": 0.5460704565048218,
-      "eval_runtime": 4.8824,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 3128
-    },
-    {
-      "epoch": 18.41,
-      "grad_norm": 1.5078125,
-      "learning_rate": 1.1811023622047245e-05,
-      "loss": 0.1401,
-      "step": 3136
-    },
-    {
-      "epoch": 18.41,
-      "eval_loss": 0.5513778328895569,
-      "eval_runtime": 4.8857,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 3136
-    },
-    {
-      "epoch": 18.45,
-      "grad_norm": 1.5234375,
-      "learning_rate": 1.1712598425196851e-05,
-      "loss": 0.1353,
-      "step": 3144
-    },
-    {
-      "epoch": 18.45,
-      "eval_loss": 0.5427709817886353,
-      "eval_runtime": 4.8814,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 3144
-    },
-    {
-      "epoch": 18.5,
-      "grad_norm": 1.5234375,
-      "learning_rate": 1.1614173228346458e-05,
-      "loss": 0.1386,
-      "step": 3152
-    },
-    {
-      "epoch": 18.5,
-      "eval_loss": 0.5440021753311157,
-      "eval_runtime": 4.8807,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 3152
-    },
-    {
-      "epoch": 18.55,
-      "grad_norm": 1.40625,
-      "learning_rate": 1.1515748031496063e-05,
-      "loss": 0.1394,
-      "step": 3160
-    },
-    {
-      "epoch": 18.55,
-      "eval_loss": 0.548543393611908,
-      "eval_runtime": 4.8841,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.228,
-      "step": 3160
-    },
-    {
-      "epoch": 18.59,
-      "grad_norm": 1.46875,
-      "learning_rate": 1.141732283464567e-05,
-      "loss": 0.1316,
-      "step": 3168
-    },
-    {
-      "epoch": 18.59,
-      "eval_loss": 0.5458658933639526,
-      "eval_runtime": 4.8863,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 3168
-    },
-    {
-      "epoch": 18.64,
-      "grad_norm": 1.484375,
-      "learning_rate": 1.1318897637795275e-05,
-      "loss": 0.1397,
-      "step": 3176
-    },
-    {
-      "epoch": 18.64,
-      "eval_loss": 0.5435895323753357,
-      "eval_runtime": 4.8847,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 3176
-    },
-    {
-      "epoch": 18.69,
-      "grad_norm": 1.515625,
-      "learning_rate": 1.1220472440944882e-05,
-      "loss": 0.1248,
-      "step": 3184
-    },
-    {
-      "epoch": 18.69,
-      "eval_loss": 0.5472026467323303,
-      "eval_runtime": 4.8816,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 3184
-    },
-    {
-      "epoch": 18.74,
-      "grad_norm": 1.3984375,
-      "learning_rate": 1.1122047244094488e-05,
-      "loss": 0.1339,
-      "step": 3192
-    },
-    {
-      "epoch": 18.74,
-      "eval_loss": 0.5473195910453796,
-      "eval_runtime": 4.8846,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 3192
-    },
-    {
-      "epoch": 18.78,
-      "grad_norm": 1.4609375,
-      "learning_rate": 1.1023622047244095e-05,
-      "loss": 0.1374,
-      "step": 3200
-    },
-    {
-      "epoch": 18.78,
-      "eval_loss": 0.5464627146720886,
-      "eval_runtime": 4.8843,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 3200
-    },
-    {
-      "epoch": 18.83,
-      "grad_norm": 1.640625,
-      "learning_rate": 1.0925196850393702e-05,
-      "loss": 0.1439,
-      "step": 3208
-    },
-    {
-      "epoch": 18.83,
-      "eval_loss": 0.5440012812614441,
-      "eval_runtime": 4.8794,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
-      "step": 3208
-    },
-    {
-      "epoch": 18.88,
-      "grad_norm": 1.453125,
-      "learning_rate": 1.0826771653543307e-05,
-      "loss": 0.1377,
-      "step": 3216
-    },
-    {
-      "epoch": 18.88,
-      "eval_loss": 0.5485637187957764,
-      "eval_runtime": 4.8976,
-      "eval_samples_per_second": 9.801,
-      "eval_steps_per_second": 1.225,
-      "step": 3216
-    },
-    {
-      "epoch": 18.92,
-      "grad_norm": 1.515625,
-      "learning_rate": 1.0728346456692914e-05,
-      "loss": 0.1379,
-      "step": 3224
-    },
-    {
-      "epoch": 18.92,
-      "eval_loss": 0.5467141270637512,
-      "eval_runtime": 4.8886,
-      "eval_samples_per_second": 9.819,
-      "eval_steps_per_second": 1.227,
-      "step": 3224
-    },
-    {
-      "epoch": 18.97,
-      "grad_norm": 1.640625,
-      "learning_rate": 1.062992125984252e-05,
-      "loss": 0.1436,
-      "step": 3232
-    },
-    {
-      "epoch": 18.97,
-      "eval_loss": 0.547528088092804,
-      "eval_runtime": 4.8826,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 3232
-    },
-    {
-      "epoch": 19.02,
-      "grad_norm": 1.3359375,
-      "learning_rate": 1.0531496062992126e-05,
-      "loss": 0.1378,
-      "step": 3240
-    },
-    {
-      "epoch": 19.02,
-      "eval_loss": 0.551363468170166,
-      "eval_runtime": 4.881,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
-      "step": 3240
-    },
-    {
-      "epoch": 19.06,
-      "grad_norm": 1.53125,
-      "learning_rate": 1.0433070866141732e-05,
-      "loss": 0.1222,
-      "step": 3248
-    },
-    {
-      "epoch": 19.06,
-      "eval_loss": 0.5650167465209961,
-      "eval_runtime": 4.8822,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 3248
-    },
-    {
-      "epoch": 19.11,
-      "grad_norm": 1.484375,
-      "learning_rate": 1.033464566929134e-05,
-      "loss": 0.1295,
-      "step": 3256
-    },
-    {
-      "epoch": 19.11,
-      "eval_loss": 0.5542344450950623,
-      "eval_runtime": 4.8865,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 3256
-    },
-    {
-      "epoch": 19.16,
-      "grad_norm": 1.5,
-      "learning_rate": 1.0236220472440946e-05,
-      "loss": 0.1308,
-      "step": 3264
-    },
-    {
-      "epoch": 19.16,
-      "eval_loss": 0.5623123645782471,
-      "eval_runtime": 4.8834,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 3264
-    },
-    {
-      "epoch": 19.2,
-      "grad_norm": 1.3515625,
-      "learning_rate": 1.0137795275590551e-05,
-      "loss": 0.1353,
-      "step": 3272
-    },
-    {
-      "epoch": 19.2,
-      "eval_loss": 0.5574969053268433,
-      "eval_runtime": 4.8843,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 3272
-    },
-    {
-      "epoch": 19.25,
-      "grad_norm": 1.3203125,
-      "learning_rate": 1.0039370078740158e-05,
-      "loss": 0.1175,
-      "step": 3280
-    },
-    {
-      "epoch": 19.25,
-      "eval_loss": 0.5596539378166199,
-      "eval_runtime": 4.884,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.228,
-      "step": 3280
-    },
-    {
-      "epoch": 19.3,
-      "grad_norm": 1.34375,
-      "learning_rate": 9.940944881889765e-06,
-      "loss": 0.1293,
-      "step": 3288
-    },
-    {
-      "epoch": 19.3,
-      "eval_loss": 0.5600309371948242,
-      "eval_runtime": 4.8839,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.229,
-      "step": 3288
-    },
-    {
-      "epoch": 19.35,
-      "grad_norm": 1.484375,
-      "learning_rate": 9.842519685039371e-06,
-      "loss": 0.1274,
-      "step": 3296
-    },
-    {
-      "epoch": 19.35,
-      "eval_loss": 0.554668128490448,
-      "eval_runtime": 4.8842,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.228,
-      "step": 3296
-    },
-    {
-      "epoch": 19.39,
-      "grad_norm": 1.4375,
-      "learning_rate": 9.744094488188976e-06,
-      "loss": 0.1303,
-      "step": 3304
-    },
-    {
-      "epoch": 19.39,
-      "eval_loss": 0.5577480792999268,
-      "eval_runtime": 4.8832,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 3304
-    },
-    {
-      "epoch": 19.44,
-      "grad_norm": 1.4609375,
-      "learning_rate": 9.645669291338583e-06,
-      "loss": 0.1341,
-      "step": 3312
-    },
-    {
-      "epoch": 19.44,
-      "eval_loss": 0.5581305027008057,
-      "eval_runtime": 4.8831,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 3312
-    },
-    {
-      "epoch": 19.49,
-      "grad_norm": 1.3671875,
-      "learning_rate": 9.54724409448819e-06,
-      "loss": 0.1256,
-      "step": 3320
-    },
-    {
-      "epoch": 19.49,
-      "eval_loss": 0.5582204461097717,
-      "eval_runtime": 4.8865,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 3320
-    },
-    {
-      "epoch": 19.53,
-      "grad_norm": 1.4453125,
-      "learning_rate": 9.448818897637795e-06,
-      "loss": 0.1267,
-      "step": 3328
-    },
-    {
-      "epoch": 19.53,
-      "eval_loss": 0.5566105842590332,
-      "eval_runtime": 4.8844,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 3328
-    },
-    {
-      "epoch": 19.58,
-      "grad_norm": 1.4140625,
-      "learning_rate": 9.350393700787402e-06,
-      "loss": 0.126,
-      "step": 3336
-    },
-    {
-      "epoch": 19.58,
-      "eval_loss": 0.5577941536903381,
-      "eval_runtime": 4.8857,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 3336
-    },
-    {
-      "epoch": 19.63,
-      "grad_norm": 1.390625,
-      "learning_rate": 9.251968503937008e-06,
-      "loss": 0.1335,
-      "step": 3344
-    },
-    {
-      "epoch": 19.63,
-      "eval_loss": 0.559036374092102,
-      "eval_runtime": 4.8877,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 3344
-    },
-    {
-      "epoch": 19.67,
-      "grad_norm": 1.484375,
-      "learning_rate": 9.153543307086615e-06,
-      "loss": 0.1254,
-      "step": 3352
-    },
-    {
-      "epoch": 19.67,
-      "eval_loss": 0.5568073987960815,
-      "eval_runtime": 4.8832,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 3352
-    },
-    {
-      "epoch": 19.72,
-      "grad_norm": 1.3515625,
-      "learning_rate": 9.055118110236222e-06,
-      "loss": 0.1251,
-      "step": 3360
-    },
-    {
-      "epoch": 19.72,
-      "eval_loss": 0.5608676671981812,
-      "eval_runtime": 4.8884,
-      "eval_samples_per_second": 9.819,
-      "eval_steps_per_second": 1.227,
-      "step": 3360
-    },
-    {
-      "epoch": 19.77,
-      "grad_norm": 1.5234375,
-      "learning_rate": 8.956692913385827e-06,
-      "loss": 0.1299,
-      "step": 3368
-    },
-    {
-      "epoch": 19.77,
-      "eval_loss": 0.5572808980941772,
-      "eval_runtime": 4.8863,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 3368
-    },
-    {
-      "epoch": 19.82,
-      "grad_norm": 1.3828125,
-      "learning_rate": 8.858267716535434e-06,
-      "loss": 0.1281,
-      "step": 3376
-    },
-    {
-      "epoch": 19.82,
-      "eval_loss": 0.5595596432685852,
-      "eval_runtime": 4.8843,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 3376
-    },
-    {
-      "epoch": 19.86,
-      "grad_norm": 1.3671875,
-      "learning_rate": 8.759842519685039e-06,
-      "loss": 0.132,
-      "step": 3384
-    },
-    {
-      "epoch": 19.86,
-      "eval_loss": 0.5580050349235535,
-      "eval_runtime": 4.8864,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 3384
-    },
-    {
-      "epoch": 19.91,
-      "grad_norm": 1.4453125,
-      "learning_rate": 8.661417322834646e-06,
-      "loss": 0.1379,
-      "step": 3392
-    },
-    {
-      "epoch": 19.91,
-      "eval_loss": 0.5603541135787964,
-      "eval_runtime": 4.8836,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 3392
-    },
-    {
-      "epoch": 19.96,
-      "grad_norm": 1.5078125,
-      "learning_rate": 8.562992125984252e-06,
-      "loss": 0.1282,
-      "step": 3400
-    },
-    {
-      "epoch": 19.96,
-      "eval_loss": 0.5588706135749817,
-      "eval_runtime": 4.883,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 3400
-    },
-    {
-      "epoch": 20.0,
-      "grad_norm": 1.328125,
-      "learning_rate": 8.46456692913386e-06,
-      "loss": 0.1384,
-      "step": 3408
-    },
-    {
-      "epoch": 20.0,
-      "eval_loss": 0.5589407086372375,
-      "eval_runtime": 4.8904,
-      "eval_samples_per_second": 9.815,
-      "eval_steps_per_second": 1.227,
-      "step": 3408
-    },
-    {
-      "epoch": 20.05,
-      "grad_norm": 1.4765625,
-      "learning_rate": 8.366141732283466e-06,
-      "loss": 0.1241,
-      "step": 3416
-    },
-    {
-      "epoch": 20.05,
-      "eval_loss": 0.5755249857902527,
-      "eval_runtime": 4.8865,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 3416
-    },
-    {
-      "epoch": 20.1,
-      "grad_norm": 1.3046875,
-      "learning_rate": 8.267716535433071e-06,
-      "loss": 0.1159,
-      "step": 3424
-    },
-    {
-      "epoch": 20.1,
-      "eval_loss": 0.5693344473838806,
-      "eval_runtime": 4.8876,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 3424
-    },
-    {
-      "epoch": 20.14,
-      "grad_norm": 1.640625,
-      "learning_rate": 8.169291338582676e-06,
-      "loss": 0.1315,
-      "step": 3432
-    },
-    {
-      "epoch": 20.14,
-      "eval_loss": 0.566354513168335,
-      "eval_runtime": 4.8833,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 3432
-    },
-    {
-      "epoch": 20.19,
-      "grad_norm": 1.5078125,
-      "learning_rate": 8.070866141732283e-06,
-      "loss": 0.122,
-      "step": 3440
-    },
-    {
-      "epoch": 20.19,
-      "eval_loss": 0.571541428565979,
-      "eval_runtime": 4.8845,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 3440
-    },
-    {
-      "epoch": 20.24,
-      "grad_norm": 1.296875,
-      "learning_rate": 7.97244094488189e-06,
-      "loss": 0.1196,
-      "step": 3448
-    },
-    {
-      "epoch": 20.24,
-      "eval_loss": 0.5679660439491272,
-      "eval_runtime": 4.8818,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 3448
-    },
-    {
-      "epoch": 20.28,
-      "grad_norm": 1.265625,
-      "learning_rate": 7.874015748031496e-06,
-      "loss": 0.1212,
-      "step": 3456
-    },
-    {
-      "epoch": 20.28,
-      "eval_loss": 0.5651406049728394,
-      "eval_runtime": 4.8802,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.229,
-      "step": 3456
-    },
-    {
-      "epoch": 20.33,
-      "grad_norm": 1.34375,
-      "learning_rate": 7.775590551181103e-06,
-      "loss": 0.1253,
-      "step": 3464
-    },
-    {
-      "epoch": 20.33,
-      "eval_loss": 0.5682979822158813,
-      "eval_runtime": 4.8807,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 3464
-    },
-    {
-      "epoch": 20.38,
-      "grad_norm": 1.5234375,
-      "learning_rate": 7.67716535433071e-06,
-      "loss": 0.1247,
-      "step": 3472
-    },
-    {
-      "epoch": 20.38,
-      "eval_loss": 0.5737568736076355,
-      "eval_runtime": 4.8865,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 3472
-    },
-    {
-      "epoch": 20.43,
-      "grad_norm": 1.5625,
-      "learning_rate": 7.578740157480316e-06,
-      "loss": 0.1237,
-      "step": 3480
-    },
-    {
-      "epoch": 20.43,
-      "eval_loss": 0.5683262944221497,
-      "eval_runtime": 4.8826,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 3480
-    },
-    {
-      "epoch": 20.47,
-      "grad_norm": 1.40625,
-      "learning_rate": 7.4803149606299226e-06,
-      "loss": 0.1257,
-      "step": 3488
-    },
-    {
-      "epoch": 20.47,
-      "eval_loss": 0.5667000412940979,
-      "eval_runtime": 4.8835,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 3488
-    },
-    {
-      "epoch": 20.52,
-      "grad_norm": 1.3125,
-      "learning_rate": 7.381889763779528e-06,
-      "loss": 0.1146,
-      "step": 3496
-    },
-    {
-      "epoch": 20.52,
-      "eval_loss": 0.5698798298835754,
-      "eval_runtime": 4.8845,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 3496
-    },
-    {
-      "epoch": 20.57,
-      "grad_norm": 1.4296875,
-      "learning_rate": 7.2834645669291335e-06,
-      "loss": 0.1211,
-      "step": 3504
-    },
-    {
-      "epoch": 20.57,
-      "eval_loss": 0.5667474269866943,
-      "eval_runtime": 4.8839,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.229,
-      "step": 3504
-    },
-    {
-      "epoch": 20.61,
-      "grad_norm": 1.4375,
-      "learning_rate": 7.18503937007874e-06,
-      "loss": 0.133,
-      "step": 3512
-    },
-    {
-      "epoch": 20.61,
-      "eval_loss": 0.5643436312675476,
-      "eval_runtime": 4.8843,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 3512
-    },
-    {
-      "epoch": 20.66,
-      "grad_norm": 1.4921875,
-      "learning_rate": 7.086614173228347e-06,
-      "loss": 0.1253,
-      "step": 3520
-    },
-    {
-      "epoch": 20.66,
-      "eval_loss": 0.5657409429550171,
-      "eval_runtime": 4.8829,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 3520
-    },
-    {
-      "epoch": 20.71,
-      "grad_norm": 1.3203125,
-      "learning_rate": 6.988188976377953e-06,
-      "loss": 0.1217,
-      "step": 3528
-    },
-    {
-      "epoch": 20.71,
-      "eval_loss": 0.5663951635360718,
-      "eval_runtime": 4.8868,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 3528
-    },
-    {
-      "epoch": 20.75,
-      "grad_norm": 1.4921875,
-      "learning_rate": 6.88976377952756e-06,
-      "loss": 0.1385,
-      "step": 3536
-    },
-    {
-      "epoch": 20.75,
-      "eval_loss": 0.5656896233558655,
-      "eval_runtime": 4.8883,
-      "eval_samples_per_second": 9.819,
-      "eval_steps_per_second": 1.227,
-      "step": 3536
-    },
-    {
-      "epoch": 20.8,
-      "grad_norm": 1.34375,
-      "learning_rate": 6.7913385826771665e-06,
-      "loss": 0.1312,
-      "step": 3544
-    },
-    {
-      "epoch": 20.8,
-      "eval_loss": 0.5687183737754822,
-      "eval_runtime": 4.8862,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 3544
-    },
-    {
-      "epoch": 20.85,
-      "grad_norm": 1.5390625,
-      "learning_rate": 6.692913385826772e-06,
-      "loss": 0.1279,
-      "step": 3552
-    },
-    {
-      "epoch": 20.85,
-      "eval_loss": 0.5673196911811829,
-      "eval_runtime": 4.8874,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 3552
-    },
-    {
-      "epoch": 20.9,
-      "grad_norm": 1.3984375,
-      "learning_rate": 6.5944881889763775e-06,
-      "loss": 0.1298,
-      "step": 3560
-    },
-    {
-      "epoch": 20.9,
-      "eval_loss": 0.5674422383308411,
-      "eval_runtime": 4.8854,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 3560
-    },
-    {
-      "epoch": 20.94,
-      "grad_norm": 1.4921875,
-      "learning_rate": 6.496062992125984e-06,
-      "loss": 0.1309,
-      "step": 3568
-    },
-    {
-      "epoch": 20.94,
-      "eval_loss": 0.566185474395752,
-      "eval_runtime": 4.8849,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 3568
-    },
-    {
-      "epoch": 20.99,
-      "grad_norm": 1.4296875,
-      "learning_rate": 6.397637795275591e-06,
-      "loss": 0.125,
-      "step": 3576
-    },
-    {
-      "epoch": 20.99,
-      "eval_loss": 0.5648412704467773,
-      "eval_runtime": 4.895,
-      "eval_samples_per_second": 9.806,
-      "eval_steps_per_second": 1.226,
-      "step": 3576
-    },
-    {
-      "epoch": 21.04,
-      "grad_norm": 1.3046875,
-      "learning_rate": 6.299212598425197e-06,
-      "loss": 0.1252,
-      "step": 3584
-    },
-    {
-      "epoch": 21.04,
-      "eval_loss": 0.5721383690834045,
-      "eval_runtime": 4.8845,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 3584
-    },
-    {
-      "epoch": 21.08,
-      "grad_norm": 1.546875,
-      "learning_rate": 6.200787401574804e-06,
-      "loss": 0.1339,
-      "step": 3592
-    },
-    {
-      "epoch": 21.08,
-      "eval_loss": 0.5767750144004822,
-      "eval_runtime": 4.8849,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 3592
-    },
-    {
-      "epoch": 21.13,
-      "grad_norm": 1.2890625,
-      "learning_rate": 6.10236220472441e-06,
-      "loss": 0.1167,
-      "step": 3600
-    },
-    {
-      "epoch": 21.13,
-      "eval_loss": 0.5756697058677673,
-      "eval_runtime": 4.8824,
-      "eval_samples_per_second": 9.831,
-      "eval_steps_per_second": 1.229,
-      "step": 3600
-    },
-    {
-      "epoch": 21.18,
-      "grad_norm": 1.328125,
-      "learning_rate": 6.0039370078740155e-06,
-      "loss": 0.1307,
-      "step": 3608
-    },
-    {
-      "epoch": 21.18,
-      "eval_loss": 0.5740421414375305,
-      "eval_runtime": 4.8862,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 3608
-    },
-    {
-      "epoch": 21.22,
-      "grad_norm": 1.3984375,
-      "learning_rate": 5.905511811023622e-06,
-      "loss": 0.1229,
-      "step": 3616
-    },
-    {
-      "epoch": 21.22,
-      "eval_loss": 0.5752490162849426,
-      "eval_runtime": 4.8844,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 3616
-    },
-    {
-      "epoch": 21.27,
-      "grad_norm": 1.3046875,
-      "learning_rate": 5.807086614173229e-06,
-      "loss": 0.1177,
-      "step": 3624
-    },
-    {
-      "epoch": 21.27,
-      "eval_loss": 0.5777534246444702,
-      "eval_runtime": 4.8857,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 3624
-    },
-    {
-      "epoch": 21.32,
-      "grad_norm": 1.3515625,
-      "learning_rate": 5.708661417322835e-06,
-      "loss": 0.1255,
-      "step": 3632
-    },
-    {
-      "epoch": 21.32,
-      "eval_loss": 0.5778191685676575,
-      "eval_runtime": 4.888,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.228,
-      "step": 3632
-    },
-    {
-      "epoch": 21.36,
-      "grad_norm": 1.375,
-      "learning_rate": 5.610236220472441e-06,
-      "loss": 0.1263,
-      "step": 3640
-    },
-    {
-      "epoch": 21.36,
-      "eval_loss": 0.5756245255470276,
-      "eval_runtime": 4.8829,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 3640
-    },
-    {
-      "epoch": 21.41,
-      "grad_norm": 1.390625,
-      "learning_rate": 5.511811023622048e-06,
-      "loss": 0.1201,
-      "step": 3648
-    },
-    {
-      "epoch": 21.41,
-      "eval_loss": 0.572148859500885,
-      "eval_runtime": 4.8807,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 3648
-    },
-    {
-      "epoch": 21.46,
-      "grad_norm": 1.3828125,
-      "learning_rate": 5.4133858267716535e-06,
-      "loss": 0.1193,
-      "step": 3656
-    },
-    {
-      "epoch": 21.46,
-      "eval_loss": 0.5735480189323425,
-      "eval_runtime": 4.8918,
-      "eval_samples_per_second": 9.812,
-      "eval_steps_per_second": 1.227,
-      "step": 3656
-    },
-    {
-      "epoch": 21.51,
-      "grad_norm": 1.40625,
-      "learning_rate": 5.31496062992126e-06,
-      "loss": 0.1238,
-      "step": 3664
-    },
-    {
-      "epoch": 21.51,
-      "eval_loss": 0.5760764479637146,
-      "eval_runtime": 4.887,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 3664
-    },
-    {
-      "epoch": 21.55,
-      "grad_norm": 1.4765625,
-      "learning_rate": 5.216535433070866e-06,
-      "loss": 0.1174,
-      "step": 3672
-    },
-    {
-      "epoch": 21.55,
-      "eval_loss": 0.5745072960853577,
-      "eval_runtime": 4.8836,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 3672
-    },
-    {
-      "epoch": 21.6,
-      "grad_norm": 1.328125,
-      "learning_rate": 5.118110236220473e-06,
-      "loss": 0.1207,
-      "step": 3680
-    },
-    {
-      "epoch": 21.6,
-      "eval_loss": 0.5743126273155212,
-      "eval_runtime": 4.8822,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 3680
-    },
-    {
-      "epoch": 21.65,
-      "grad_norm": 1.3046875,
-      "learning_rate": 5.019685039370079e-06,
-      "loss": 0.1276,
-      "step": 3688
-    },
-    {
-      "epoch": 21.65,
-      "eval_loss": 0.5735090374946594,
-      "eval_runtime": 4.8856,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 3688
-    },
-    {
-      "epoch": 21.69,
-      "grad_norm": 1.3125,
-      "learning_rate": 4.921259842519686e-06,
-      "loss": 0.1202,
-      "step": 3696
-    },
-    {
-      "epoch": 21.69,
-      "eval_loss": 0.5732198357582092,
-      "eval_runtime": 4.8809,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
-      "step": 3696
-    },
-    {
-      "epoch": 21.74,
-      "grad_norm": 1.296875,
-      "learning_rate": 4.8228346456692916e-06,
-      "loss": 0.1221,
-      "step": 3704
-    },
-    {
-      "epoch": 21.74,
-      "eval_loss": 0.5724496245384216,
-      "eval_runtime": 4.8876,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 3704
-    },
-    {
-      "epoch": 21.79,
-      "grad_norm": 1.328125,
-      "learning_rate": 4.7244094488188975e-06,
-      "loss": 0.1243,
-      "step": 3712
-    },
-    {
-      "epoch": 21.79,
-      "eval_loss": 0.5729621648788452,
-      "eval_runtime": 4.8869,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 3712
-    },
-    {
-      "epoch": 21.83,
-      "grad_norm": 1.3828125,
-      "learning_rate": 4.625984251968504e-06,
-      "loss": 0.1243,
-      "step": 3720
-    },
-    {
-      "epoch": 21.83,
-      "eval_loss": 0.5742912888526917,
-      "eval_runtime": 4.8842,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.228,
-      "step": 3720
-    },
-    {
-      "epoch": 21.88,
-      "grad_norm": 1.3046875,
-      "learning_rate": 4.527559055118111e-06,
-      "loss": 0.114,
-      "step": 3728
-    },
-    {
-      "epoch": 21.88,
-      "eval_loss": 0.5741546154022217,
-      "eval_runtime": 4.8867,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 3728
-    },
-    {
-      "epoch": 21.93,
-      "grad_norm": 1.3984375,
-      "learning_rate": 4.429133858267717e-06,
-      "loss": 0.1256,
-      "step": 3736
-    },
-    {
-      "epoch": 21.93,
-      "eval_loss": 0.5746021866798401,
-      "eval_runtime": 4.8831,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 3736
-    },
-    {
-      "epoch": 21.98,
-      "grad_norm": 1.28125,
-      "learning_rate": 4.330708661417323e-06,
-      "loss": 0.1167,
-      "step": 3744
-    },
-    {
-      "epoch": 21.98,
-      "eval_loss": 0.5760999321937561,
-      "eval_runtime": 4.8858,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 3744
-    },
-    {
-      "epoch": 22.02,
-      "grad_norm": 1.265625,
-      "learning_rate": 4.23228346456693e-06,
-      "loss": 0.1191,
-      "step": 3752
-    },
-    {
-      "epoch": 22.02,
-      "eval_loss": 0.5751403570175171,
-      "eval_runtime": 4.8805,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 3752
-    },
-    {
-      "epoch": 22.07,
-      "grad_norm": 1.3046875,
-      "learning_rate": 4.1338582677165355e-06,
-      "loss": 0.1213,
-      "step": 3760
-    },
-    {
-      "epoch": 22.07,
-      "eval_loss": 0.5758962631225586,
-      "eval_runtime": 4.8908,
-      "eval_samples_per_second": 9.814,
-      "eval_steps_per_second": 1.227,
-      "step": 3760
-    },
-    {
-      "epoch": 22.12,
-      "grad_norm": 1.3671875,
-      "learning_rate": 4.035433070866141e-06,
-      "loss": 0.117,
-      "step": 3768
-    },
-    {
-      "epoch": 22.12,
-      "eval_loss": 0.5779056549072266,
-      "eval_runtime": 4.8823,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 3768
-    },
-    {
-      "epoch": 22.16,
-      "grad_norm": 1.53125,
-      "learning_rate": 3.937007874015748e-06,
-      "loss": 0.1283,
-      "step": 3776
-    },
-    {
-      "epoch": 22.16,
-      "eval_loss": 0.5779562592506409,
-      "eval_runtime": 4.8838,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.229,
-      "step": 3776
-    },
-    {
-      "epoch": 22.21,
-      "grad_norm": 1.296875,
-      "learning_rate": 3.838582677165355e-06,
-      "loss": 0.1231,
-      "step": 3784
-    },
-    {
-      "epoch": 22.21,
-      "eval_loss": 0.577904999256134,
-      "eval_runtime": 4.8859,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 3784
-    },
-    {
-      "epoch": 22.26,
-      "grad_norm": 1.2890625,
-      "learning_rate": 3.7401574803149613e-06,
-      "loss": 0.1159,
-      "step": 3792
-    },
-    {
-      "epoch": 22.26,
-      "eval_loss": 0.577033519744873,
-      "eval_runtime": 4.8788,
-      "eval_samples_per_second": 9.838,
-      "eval_steps_per_second": 1.23,
-      "step": 3792
-    },
-    {
-      "epoch": 22.3,
-      "grad_norm": 1.4921875,
-      "learning_rate": 3.6417322834645668e-06,
-      "loss": 0.1194,
-      "step": 3800
-    },
-    {
-      "epoch": 22.3,
-      "eval_loss": 0.5773031115531921,
-      "eval_runtime": 4.8861,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 3800
-    },
-    {
-      "epoch": 22.35,
-      "grad_norm": 1.4140625,
-      "learning_rate": 3.5433070866141735e-06,
-      "loss": 0.1258,
-      "step": 3808
-    },
-    {
-      "epoch": 22.35,
-      "eval_loss": 0.5772604942321777,
-      "eval_runtime": 4.8813,
-      "eval_samples_per_second": 9.834,
-      "eval_steps_per_second": 1.229,
-      "step": 3808
-    },
-    {
-      "epoch": 22.4,
-      "grad_norm": 1.203125,
-      "learning_rate": 3.44488188976378e-06,
-      "loss": 0.1194,
-      "step": 3816
-    },
-    {
-      "epoch": 22.4,
-      "eval_loss": 0.576747477054596,
-      "eval_runtime": 4.8816,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 3816
-    },
-    {
-      "epoch": 22.44,
-      "grad_norm": 1.28125,
-      "learning_rate": 3.346456692913386e-06,
-      "loss": 0.117,
-      "step": 3824
-    },
-    {
-      "epoch": 22.44,
-      "eval_loss": 0.577796459197998,
-      "eval_runtime": 4.8802,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.229,
-      "step": 3824
-    },
-    {
-      "epoch": 22.49,
-      "grad_norm": 1.4453125,
-      "learning_rate": 3.248031496062992e-06,
-      "loss": 0.1213,
-      "step": 3832
-    },
-    {
-      "epoch": 22.49,
-      "eval_loss": 0.578058660030365,
-      "eval_runtime": 4.8805,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 3832
-    },
-    {
-      "epoch": 22.54,
-      "grad_norm": 1.4921875,
-      "learning_rate": 3.1496062992125985e-06,
-      "loss": 0.1106,
-      "step": 3840
-    },
-    {
-      "epoch": 22.54,
-      "eval_loss": 0.5779960751533508,
-      "eval_runtime": 4.8851,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 3840
-    },
-    {
-      "epoch": 22.59,
-      "grad_norm": 1.3515625,
-      "learning_rate": 3.051181102362205e-06,
-      "loss": 0.1225,
-      "step": 3848
-    },
-    {
-      "epoch": 22.59,
-      "eval_loss": 0.5778743624687195,
-      "eval_runtime": 4.8853,
-      "eval_samples_per_second": 9.825,
-      "eval_steps_per_second": 1.228,
-      "step": 3848
-    },
-    {
-      "epoch": 22.63,
-      "grad_norm": 1.265625,
-      "learning_rate": 2.952755905511811e-06,
-      "loss": 0.1201,
-      "step": 3856
-    },
-    {
-      "epoch": 22.63,
-      "eval_loss": 0.5778095126152039,
-      "eval_runtime": 4.8862,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 3856
-    },
-    {
-      "epoch": 22.68,
-      "grad_norm": 1.3046875,
-      "learning_rate": 2.8543307086614175e-06,
-      "loss": 0.1123,
-      "step": 3864
-    },
-    {
-      "epoch": 22.68,
-      "eval_loss": 0.576666533946991,
-      "eval_runtime": 4.8875,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 3864
-    },
-    {
-      "epoch": 22.73,
-      "grad_norm": 1.515625,
-      "learning_rate": 2.755905511811024e-06,
-      "loss": 0.1265,
-      "step": 3872
-    },
-    {
-      "epoch": 22.73,
-      "eval_loss": 0.5777731537818909,
-      "eval_runtime": 4.8845,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 3872
-    },
-    {
-      "epoch": 22.77,
-      "grad_norm": 1.390625,
-      "learning_rate": 2.65748031496063e-06,
-      "loss": 0.121,
-      "step": 3880
-    },
-    {
-      "epoch": 22.77,
-      "eval_loss": 0.5775730609893799,
-      "eval_runtime": 4.8883,
-      "eval_samples_per_second": 9.819,
-      "eval_steps_per_second": 1.227,
-      "step": 3880
-    },
-    {
-      "epoch": 22.82,
-      "grad_norm": 1.4375,
-      "learning_rate": 2.5590551181102365e-06,
-      "loss": 0.1233,
-      "step": 3888
-    },
-    {
-      "epoch": 22.82,
-      "eval_loss": 0.5771867036819458,
-      "eval_runtime": 4.8877,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 3888
-    },
-    {
-      "epoch": 22.87,
-      "grad_norm": 1.3046875,
-      "learning_rate": 2.460629921259843e-06,
-      "loss": 0.1191,
-      "step": 3896
-    },
-    {
-      "epoch": 22.87,
-      "eval_loss": 0.5772641897201538,
-      "eval_runtime": 4.8852,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 3896
-    },
-    {
-      "epoch": 22.91,
-      "grad_norm": 1.375,
-      "learning_rate": 2.3622047244094487e-06,
-      "loss": 0.127,
-      "step": 3904
-    },
-    {
-      "epoch": 22.91,
-      "eval_loss": 0.577972412109375,
-      "eval_runtime": 4.8831,
-      "eval_samples_per_second": 9.83,
-      "eval_steps_per_second": 1.229,
-      "step": 3904
-    },
-    {
-      "epoch": 22.96,
-      "grad_norm": 1.2265625,
-      "learning_rate": 2.2637795275590555e-06,
-      "loss": 0.127,
-      "step": 3912
-    },
-    {
-      "epoch": 22.96,
-      "eval_loss": 0.5772601366043091,
-      "eval_runtime": 4.8799,
-      "eval_samples_per_second": 9.836,
-      "eval_steps_per_second": 1.23,
-      "step": 3912
-    },
-    {
-      "epoch": 23.01,
-      "grad_norm": 1.3125,
-      "learning_rate": 2.1653543307086614e-06,
-      "loss": 0.1155,
-      "step": 3920
-    },
-    {
-      "epoch": 23.01,
-      "eval_loss": 0.5777209997177124,
-      "eval_runtime": 4.8863,
-      "eval_samples_per_second": 9.823,
-      "eval_steps_per_second": 1.228,
-      "step": 3920
-    },
-    {
-      "epoch": 23.06,
-      "grad_norm": 1.203125,
-      "learning_rate": 2.0669291338582678e-06,
-      "loss": 0.1133,
-      "step": 3928
-    },
-    {
-      "epoch": 23.06,
-      "eval_loss": 0.5779727697372437,
-      "eval_runtime": 4.8894,
-      "eval_samples_per_second": 9.817,
-      "eval_steps_per_second": 1.227,
-      "step": 3928
-    },
-    {
-      "epoch": 23.1,
-      "grad_norm": 1.3515625,
-      "learning_rate": 1.968503937007874e-06,
-      "loss": 0.115,
-      "step": 3936
-    },
-    {
-      "epoch": 23.1,
-      "eval_loss": 0.5777748823165894,
-      "eval_runtime": 4.8872,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 3936
-    },
-    {
-      "epoch": 23.15,
-      "grad_norm": 1.3203125,
-      "learning_rate": 1.8700787401574806e-06,
-      "loss": 0.1133,
-      "step": 3944
-    },
-    {
-      "epoch": 23.15,
-      "eval_loss": 0.578517735004425,
-      "eval_runtime": 4.8877,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 3944
-    },
-    {
-      "epoch": 23.2,
-      "grad_norm": 1.3828125,
-      "learning_rate": 1.7716535433070868e-06,
-      "loss": 0.1243,
-      "step": 3952
-    },
-    {
-      "epoch": 23.2,
-      "eval_loss": 0.5775905251502991,
-      "eval_runtime": 4.8803,
-      "eval_samples_per_second": 9.835,
-      "eval_steps_per_second": 1.229,
-      "step": 3952
-    },
-    {
-      "epoch": 23.24,
-      "grad_norm": 1.21875,
-      "learning_rate": 1.673228346456693e-06,
-      "loss": 0.1255,
-      "step": 3960
-    },
-    {
-      "epoch": 23.24,
-      "eval_loss": 0.579028308391571,
-      "eval_runtime": 4.8843,
-      "eval_samples_per_second": 9.827,
-      "eval_steps_per_second": 1.228,
-      "step": 3960
-    },
-    {
-      "epoch": 23.29,
-      "grad_norm": 1.3984375,
-      "learning_rate": 1.5748031496062992e-06,
-      "loss": 0.1082,
-      "step": 3968
-    },
-    {
-      "epoch": 23.29,
-      "eval_loss": 0.5788905024528503,
-      "eval_runtime": 4.8872,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 3968
-    },
-    {
-      "epoch": 23.34,
-      "grad_norm": 1.328125,
-      "learning_rate": 1.4763779527559056e-06,
-      "loss": 0.1122,
-      "step": 3976
-    },
-    {
-      "epoch": 23.34,
-      "eval_loss": 0.5791129469871521,
-      "eval_runtime": 4.8841,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.228,
-      "step": 3976
-    },
-    {
-      "epoch": 23.38,
-      "grad_norm": 1.3203125,
-      "learning_rate": 1.377952755905512e-06,
-      "loss": 0.1275,
-      "step": 3984
-    },
-    {
-      "epoch": 23.38,
-      "eval_loss": 0.5788193345069885,
-      "eval_runtime": 4.8848,
-      "eval_samples_per_second": 9.826,
-      "eval_steps_per_second": 1.228,
-      "step": 3984
-    },
-    {
-      "epoch": 23.43,
-      "grad_norm": 1.34375,
-      "learning_rate": 1.2795275590551182e-06,
-      "loss": 0.1234,
-      "step": 3992
-    },
-    {
-      "epoch": 23.43,
-      "eval_loss": 0.5793178081512451,
-      "eval_runtime": 4.8814,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 3992
-    },
-    {
-      "epoch": 23.48,
-      "grad_norm": 1.390625,
-      "learning_rate": 1.1811023622047244e-06,
-      "loss": 0.1239,
-      "step": 4000
-    },
-    {
-      "epoch": 23.48,
-      "eval_loss": 0.5789151787757874,
-      "eval_runtime": 4.8797,
-      "eval_samples_per_second": 9.837,
-      "eval_steps_per_second": 1.23,
-      "step": 4000
-    },
-    {
-      "epoch": 23.52,
-      "grad_norm": 1.4140625,
-      "learning_rate": 1.0826771653543307e-06,
-      "loss": 0.1203,
-      "step": 4008
-    },
-    {
-      "epoch": 23.52,
-      "eval_loss": 0.5790886282920837,
-      "eval_runtime": 4.882,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 4008
-    },
-    {
-      "epoch": 23.57,
-      "grad_norm": 1.3984375,
-      "learning_rate": 9.84251968503937e-07,
-      "loss": 0.1218,
-      "step": 4016
-    },
-    {
-      "epoch": 23.57,
-      "eval_loss": 0.5786433815956116,
-      "eval_runtime": 4.8874,
-      "eval_samples_per_second": 9.821,
-      "eval_steps_per_second": 1.228,
-      "step": 4016
-    },
-    {
-      "epoch": 23.62,
-      "grad_norm": 1.359375,
-      "learning_rate": 8.858267716535434e-07,
-      "loss": 0.13,
-      "step": 4024
-    },
-    {
-      "epoch": 23.62,
-      "eval_loss": 0.5786345601081848,
-      "eval_runtime": 4.8839,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.229,
-      "step": 4024
-    },
-    {
-      "epoch": 23.67,
-      "grad_norm": 1.3203125,
-      "learning_rate": 7.874015748031496e-07,
-      "loss": 0.124,
-      "step": 4032
-    },
-    {
-      "epoch": 23.67,
-      "eval_loss": 0.5792049765586853,
-      "eval_runtime": 4.8834,
-      "eval_samples_per_second": 9.829,
-      "eval_steps_per_second": 1.229,
-      "step": 4032
-    },
-    {
-      "epoch": 23.71,
-      "grad_norm": 1.4375,
-      "learning_rate": 6.88976377952756e-07,
-      "loss": 0.1141,
-      "step": 4040
-    },
-    {
-      "epoch": 23.71,
-      "eval_loss": 0.5789989829063416,
-      "eval_runtime": 4.886,
-      "eval_samples_per_second": 9.824,
-      "eval_steps_per_second": 1.228,
-      "step": 4040
-    },
-    {
-      "epoch": 23.76,
-      "grad_norm": 1.40625,
-      "learning_rate": 5.905511811023622e-07,
-      "loss": 0.118,
-      "step": 4048
-    },
-    {
-      "epoch": 23.76,
-      "eval_loss": 0.5784198641777039,
-      "eval_runtime": 4.8871,
-      "eval_samples_per_second": 9.822,
-      "eval_steps_per_second": 1.228,
-      "step": 4048
-    },
-    {
-      "epoch": 23.81,
-      "grad_norm": 1.390625,
-      "learning_rate": 4.921259842519685e-07,
-      "loss": 0.1276,
-      "step": 4056
-    },
-    {
-      "epoch": 23.81,
-      "eval_loss": 0.57962566614151,
-      "eval_runtime": 4.8775,
-      "eval_samples_per_second": 9.841,
-      "eval_steps_per_second": 1.23,
-      "step": 4056
-    },
-    {
-      "epoch": 23.85,
-      "grad_norm": 1.3515625,
-      "learning_rate": 3.937007874015748e-07,
-      "loss": 0.1152,
-      "step": 4064
-    },
-    {
-      "epoch": 23.85,
-      "eval_loss": 0.5792584419250488,
-      "eval_runtime": 4.884,
-      "eval_samples_per_second": 9.828,
-      "eval_steps_per_second": 1.228,
-      "step": 4064
-    },
-    {
-      "epoch": 23.9,
-      "grad_norm": 1.359375,
-      "learning_rate": 2.952755905511811e-07,
-      "loss": 0.1292,
-      "step": 4072
-    },
-    {
-      "epoch": 23.9,
-      "eval_loss": 0.5795792937278748,
-      "eval_runtime": 4.8815,
-      "eval_samples_per_second": 9.833,
-      "eval_steps_per_second": 1.229,
-      "step": 4072
-    },
-    {
-      "epoch": 23.95,
-      "grad_norm": 1.296875,
-      "learning_rate": 1.968503937007874e-07,
-      "loss": 0.1201,
-      "step": 4080
-    },
-    {
-      "epoch": 23.95,
-      "eval_loss": 0.5792078375816345,
-      "eval_runtime": 4.888,
-      "eval_samples_per_second": 9.82,
-      "eval_steps_per_second": 1.227,
-      "step": 4080
-    },
-    {
-      "epoch": 23.99,
-      "grad_norm": 1.4140625,
-      "learning_rate": 9.84251968503937e-08,
-      "loss": 0.1213,
-      "step": 4088
-    },
-    {
-      "epoch": 23.99,
-      "eval_loss": 0.5791333317756653,
-      "eval_runtime": 4.8822,
-      "eval_samples_per_second": 9.832,
-      "eval_steps_per_second": 1.229,
-      "step": 4088
-    },
-    {
-      "epoch": 24.04,
-      "grad_norm": 1.328125,
-      "learning_rate": 0.0,
-      "loss": 0.1246,
-      "step": 4096
-    },
-    {
-      "epoch": 24.04,
-      "eval_loss": 0.5793049335479736,
-      "eval_runtime": 4.8897,
-      "eval_samples_per_second": 9.817,
-      "eval_steps_per_second": 1.227,
-      "step": 4096
-    },
-    {
-      "epoch": 24.04,
-      "step": 4096,
-      "total_flos": 2.6639402315603804e+18,
-      "train_loss": 0.24291648308280855,
-      "train_runtime": 18372.1559,
-      "train_samples_per_second": 3.567,
-      "train_steps_per_second": 0.223
+      "epoch": 3.0,
+      "step": 512,
+      "total_flos": 9.863515762146509e+16,
+      "train_loss": 0.5746448771096766,
+      "train_runtime": 1124.5772,
+      "train_samples_per_second": 7.285,
+      "train_steps_per_second": 0.455
     }
   ],
   "logging_steps": 8,
-  "max_steps": 4096,
+  "max_steps": 512,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 25,
-  "save_steps": 8,
-  "total_flos": 2.6639402315603804e+18,
-  "train_batch_size": 2,
+  "num_train_epochs": 4,
+  "save_steps": 32,
+  "total_flos": 9.863515762146509e+16,
+  "train_batch_size": 4,
   "trial_name": null,
   "trial_params": null
 }