diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5391 @@
+{
+  "best_metric": 1.1188498735427856,
+  "best_model_checkpoint": "runs/deepseek_lora_20240421-183352/checkpoint-5000",
+  "epoch": 0.17364427234367674,
+  "eval_steps": 500,
+  "global_step": 7500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 8.893424987792969,
+      "learning_rate": 4.0000000000000003e-07,
+      "loss": 2.1665,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 7.90573787689209,
+      "learning_rate": 8.000000000000001e-07,
+      "loss": 2.2972,
+      "step": 20
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 4.421755313873291,
+      "learning_rate": 1.2000000000000002e-06,
+      "loss": 2.0517,
+      "step": 30
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 4.259636402130127,
+      "learning_rate": 1.6000000000000001e-06,
+      "loss": 2.1163,
+      "step": 40
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 4.228359222412109,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 2.2697,
+      "step": 50
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 8.439579963684082,
+      "learning_rate": 2.4000000000000003e-06,
+      "loss": 2.1731,
+      "step": 60
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 2.426865339279175,
+      "learning_rate": 2.8000000000000003e-06,
+      "loss": 2.2052,
+      "step": 70
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 4.410788059234619,
+      "learning_rate": 3.2000000000000003e-06,
+      "loss": 2.0863,
+      "step": 80
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 5.553213596343994,
+      "learning_rate": 3.6000000000000003e-06,
+      "loss": 1.9664,
+      "step": 90
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 6.600622177124023,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 2.0656,
+      "step": 100
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 4.6402153968811035,
+      "learning_rate": 4.4e-06,
+      "loss": 2.0541,
+      "step": 110
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 3.2395169734954834,
+      "learning_rate": 4.800000000000001e-06,
+      "loss": 1.8353,
+      "step": 120
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 2.8458945751190186,
+      "learning_rate": 5.2e-06,
+      "loss": 1.8408,
+      "step": 130
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 5.223249912261963,
+      "learning_rate": 5.600000000000001e-06,
+      "loss": 1.905,
+      "step": 140
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 3.8996775150299072,
+      "learning_rate": 6e-06,
+      "loss": 1.9277,
+      "step": 150
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 3.8909013271331787,
+      "learning_rate": 6.4000000000000006e-06,
+      "loss": 1.777,
+      "step": 160
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 2.76326060295105,
+      "learning_rate": 6.800000000000001e-06,
+      "loss": 1.7483,
+      "step": 170
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 3.6061465740203857,
+      "learning_rate": 7.2000000000000005e-06,
+      "loss": 1.79,
+      "step": 180
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 4.046599388122559,
+      "learning_rate": 7.600000000000001e-06,
+      "loss": 1.7282,
+      "step": 190
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 3.0414559841156006,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 1.8936,
+      "step": 200
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 6.000649452209473,
+      "learning_rate": 8.400000000000001e-06,
+      "loss": 1.7291,
+      "step": 210
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.2610554695129395,
+      "learning_rate": 8.8e-06,
+      "loss": 1.5919,
+      "step": 220
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.346323490142822,
+      "learning_rate": 9.200000000000002e-06,
+      "loss": 1.7002,
+      "step": 230
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.625606060028076,
+      "learning_rate": 9.600000000000001e-06,
+      "loss": 1.7204,
+      "step": 240
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.263577938079834,
+      "learning_rate": 1e-05,
+      "loss": 1.6544,
+      "step": 250
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.526005268096924,
+      "learning_rate": 1.04e-05,
+      "loss": 1.6343,
+      "step": 260
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.9661431312561035,
+      "learning_rate": 1.0800000000000002e-05,
+      "loss": 1.6069,
+      "step": 270
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.983661651611328,
+      "learning_rate": 1.1200000000000001e-05,
+      "loss": 1.718,
+      "step": 280
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.731842517852783,
+      "learning_rate": 1.16e-05,
+      "loss": 1.6998,
+      "step": 290
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.424485683441162,
+      "learning_rate": 1.2e-05,
+      "loss": 1.7128,
+      "step": 300
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.6413464546203613,
+      "learning_rate": 1.2400000000000002e-05,
+      "loss": 1.7114,
+      "step": 310
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.612671375274658,
+      "learning_rate": 1.2800000000000001e-05,
+      "loss": 1.717,
+      "step": 320
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.804788827896118,
+      "learning_rate": 1.3200000000000002e-05,
+      "loss": 1.7916,
+      "step": 330
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.907038688659668,
+      "learning_rate": 1.3600000000000002e-05,
+      "loss": 1.6078,
+      "step": 340
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.003390312194824,
+      "learning_rate": 1.4e-05,
+      "loss": 1.7339,
+      "step": 350
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.2724103927612305,
+      "learning_rate": 1.4400000000000001e-05,
+      "loss": 1.8393,
+      "step": 360
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.7622690200805664,
+      "learning_rate": 1.48e-05,
+      "loss": 1.5091,
+      "step": 370
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.663651943206787,
+      "learning_rate": 1.5200000000000002e-05,
+      "loss": 1.639,
+      "step": 380
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.8204166889190674,
+      "learning_rate": 1.5600000000000003e-05,
+      "loss": 1.4148,
+      "step": 390
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.282427787780762,
+      "learning_rate": 1.6000000000000003e-05,
+      "loss": 1.7065,
+      "step": 400
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 9.151147842407227,
+      "learning_rate": 1.64e-05,
+      "loss": 1.5874,
+      "step": 410
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.359065532684326,
+      "learning_rate": 1.6800000000000002e-05,
+      "loss": 1.4316,
+      "step": 420
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.6506259441375732,
+      "learning_rate": 1.72e-05,
+      "loss": 1.5907,
+      "step": 430
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.2175769805908203,
+      "learning_rate": 1.76e-05,
+      "loss": 1.6386,
+      "step": 440
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.790574550628662,
+      "learning_rate": 1.8e-05,
+      "loss": 1.4903,
+      "step": 450
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.407963514328003,
+      "learning_rate": 1.8400000000000003e-05,
+      "loss": 1.5074,
+      "step": 460
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.586065769195557,
+      "learning_rate": 1.88e-05,
+      "loss": 1.4186,
+      "step": 470
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.2595417499542236,
+      "learning_rate": 1.9200000000000003e-05,
+      "loss": 1.5489,
+      "step": 480
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.8598222732543945,
+      "learning_rate": 1.9600000000000002e-05,
+      "loss": 1.314,
+      "step": 490
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.427089214324951,
+      "learning_rate": 2e-05,
+      "loss": 1.7116,
+      "step": 500
+    },
+    {
+      "epoch": 0.01,
+      "eval_loss": 1.4248766899108887,
+      "eval_runtime": 67.1505,
+      "eval_samples_per_second": 14.892,
+      "eval_steps_per_second": 14.892,
+      "step": 500
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.126067876815796,
+      "learning_rate": 1.999322033898305e-05,
+      "loss": 1.6259,
+      "step": 510
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.224739074707031,
+      "learning_rate": 1.9986440677966104e-05,
+      "loss": 1.6205,
+      "step": 520
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.176899433135986,
+      "learning_rate": 1.9979661016949154e-05,
+      "loss": 1.5216,
+      "step": 530
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.984959840774536,
+      "learning_rate": 1.9972881355932204e-05,
+      "loss": 1.3944,
+      "step": 540
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.6579527854919434,
+      "learning_rate": 1.9966101694915257e-05,
+      "loss": 1.5515,
+      "step": 550
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.6600215435028076,
+      "learning_rate": 1.9959322033898307e-05,
+      "loss": 1.5275,
+      "step": 560
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.994769096374512,
+      "learning_rate": 1.995254237288136e-05,
+      "loss": 1.6558,
+      "step": 570
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.50462532043457,
+      "learning_rate": 1.994576271186441e-05,
+      "loss": 1.641,
+      "step": 580
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.41587233543396,
+      "learning_rate": 1.993898305084746e-05,
+      "loss": 1.461,
+      "step": 590
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.1005308628082275,
+      "learning_rate": 1.9932203389830512e-05,
+      "loss": 1.4009,
+      "step": 600
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.533121347427368,
+      "learning_rate": 1.992542372881356e-05,
+      "loss": 1.7164,
+      "step": 610
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.897855758666992,
+      "learning_rate": 1.991864406779661e-05,
+      "loss": 1.5889,
+      "step": 620
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.890028953552246,
+      "learning_rate": 1.991186440677966e-05,
+      "loss": 1.4549,
+      "step": 630
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.191209316253662,
+      "learning_rate": 1.990508474576271e-05,
+      "loss": 1.4711,
+      "step": 640
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.568540334701538,
+      "learning_rate": 1.9898305084745764e-05,
+      "loss": 1.3507,
+      "step": 650
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.9555094242095947,
+      "learning_rate": 1.9891525423728814e-05,
+      "loss": 1.4569,
+      "step": 660
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.103305339813232,
+      "learning_rate": 1.9884745762711867e-05,
+      "loss": 1.6679,
+      "step": 670
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 5.208203315734863,
+      "learning_rate": 1.9877966101694917e-05,
+      "loss": 1.3169,
+      "step": 680
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.401998996734619,
+      "learning_rate": 1.9871186440677966e-05,
+      "loss": 1.642,
+      "step": 690
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.582605361938477,
+      "learning_rate": 1.986440677966102e-05,
+      "loss": 1.5996,
+      "step": 700
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.554516315460205,
+      "learning_rate": 1.985762711864407e-05,
+      "loss": 1.5291,
+      "step": 710
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.563401460647583,
+      "learning_rate": 1.985084745762712e-05,
+      "loss": 1.4608,
+      "step": 720
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 5.061882972717285,
+      "learning_rate": 1.9844067796610172e-05,
+      "loss": 1.6631,
+      "step": 730
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.9863511323928833,
+      "learning_rate": 1.9837288135593222e-05,
+      "loss": 1.5619,
+      "step": 740
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.3460112810134888,
+      "learning_rate": 1.9830508474576275e-05,
+      "loss": 1.3738,
+      "step": 750
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 5.177835464477539,
+      "learning_rate": 1.9823728813559324e-05,
+      "loss": 1.6424,
+      "step": 760
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.877689242362976,
+      "learning_rate": 1.9816949152542374e-05,
+      "loss": 1.4095,
+      "step": 770
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.7550783157348633,
+      "learning_rate": 1.9810169491525427e-05,
+      "loss": 1.5935,
+      "step": 780
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.4884017705917358,
+      "learning_rate": 1.9803389830508477e-05,
+      "loss": 1.4442,
+      "step": 790
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.309656858444214,
+      "learning_rate": 1.9796610169491527e-05,
+      "loss": 1.5698,
+      "step": 800
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.5358872413635254,
+      "learning_rate": 1.978983050847458e-05,
+      "loss": 1.2542,
+      "step": 810
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.623551368713379,
+      "learning_rate": 1.9783050847457626e-05,
+      "loss": 1.663,
+      "step": 820
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.512385845184326,
+      "learning_rate": 1.977627118644068e-05,
+      "loss": 1.5636,
+      "step": 830
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.330457925796509,
+      "learning_rate": 1.976949152542373e-05,
+      "loss": 1.3357,
+      "step": 840
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.6951920986175537,
+      "learning_rate": 1.9762711864406782e-05,
+      "loss": 1.5766,
+      "step": 850
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.8462748527526855,
+      "learning_rate": 1.9755932203389832e-05,
+      "loss": 1.5554,
+      "step": 860
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.8278895616531372,
+      "learning_rate": 1.974915254237288e-05,
+      "loss": 1.3479,
+      "step": 870
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.1370766162872314,
+      "learning_rate": 1.9742372881355935e-05,
+      "loss": 1.5584,
+      "step": 880
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 8.17900276184082,
+      "learning_rate": 1.9735593220338984e-05,
+      "loss": 1.5172,
+      "step": 890
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.5510640144348145,
+      "learning_rate": 1.9728813559322034e-05,
+      "loss": 1.4574,
+      "step": 900
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.811005115509033,
+      "learning_rate": 1.9722033898305087e-05,
+      "loss": 1.576,
+      "step": 910
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.8449907302856445,
+      "learning_rate": 1.9715254237288137e-05,
+      "loss": 1.4886,
+      "step": 920
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.2932374477386475,
+      "learning_rate": 1.970847457627119e-05,
+      "loss": 1.4845,
+      "step": 930
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.189581871032715,
+      "learning_rate": 1.970169491525424e-05,
+      "loss": 1.4143,
+      "step": 940
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.5806180238723755,
+      "learning_rate": 1.969491525423729e-05,
+      "loss": 1.4859,
+      "step": 950
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 5.293037414550781,
+      "learning_rate": 1.9688135593220342e-05,
+      "loss": 1.4887,
+      "step": 960
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 5.328525543212891,
+      "learning_rate": 1.9681355932203392e-05,
+      "loss": 1.2715,
+      "step": 970
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.8140487670898438,
+      "learning_rate": 1.9674576271186442e-05,
+      "loss": 1.4281,
+      "step": 980
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 5.055774688720703,
+      "learning_rate": 1.9667796610169495e-05,
+      "loss": 1.2676,
+      "step": 990
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.9871106147766113,
+      "learning_rate": 1.9661016949152545e-05,
+      "loss": 1.4242,
+      "step": 1000
+    },
+    {
+      "epoch": 0.02,
+      "eval_loss": 1.2329587936401367,
+      "eval_runtime": 67.1625,
+      "eval_samples_per_second": 14.889,
+      "eval_steps_per_second": 14.889,
+      "step": 1000
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 9.538952827453613,
+      "learning_rate": 1.9654237288135594e-05,
+      "loss": 1.664,
+      "step": 1010
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.894416809082031,
+      "learning_rate": 1.9647457627118644e-05,
+      "loss": 1.5421,
+      "step": 1020
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 15.61988353729248,
+      "learning_rate": 1.9640677966101697e-05,
+      "loss": 1.5037,
+      "step": 1030
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.5587459802627563,
+      "learning_rate": 1.9633898305084747e-05,
+      "loss": 1.4768,
+      "step": 1040
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.0465331077575684,
+      "learning_rate": 1.9627118644067796e-05,
+      "loss": 1.6342,
+      "step": 1050
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.6867032051086426,
+      "learning_rate": 1.962033898305085e-05,
+      "loss": 1.6395,
+      "step": 1060
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.0958399772644043,
+      "learning_rate": 1.96135593220339e-05,
+      "loss": 1.3742,
+      "step": 1070
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.064396858215332,
+      "learning_rate": 1.960677966101695e-05,
+      "loss": 1.258,
+      "step": 1080
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.9529526233673096,
+      "learning_rate": 1.9600000000000002e-05,
+      "loss": 1.6082,
+      "step": 1090
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 8.01111125946045,
+      "learning_rate": 1.9593220338983052e-05,
+      "loss": 1.5338,
+      "step": 1100
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.7759506702423096,
+      "learning_rate": 1.95864406779661e-05,
+      "loss": 1.4728,
+      "step": 1110
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.339688777923584,
+      "learning_rate": 1.9579661016949155e-05,
+      "loss": 1.5947,
+      "step": 1120
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.9827146530151367,
+      "learning_rate": 1.9572881355932204e-05,
+      "loss": 1.4338,
+      "step": 1130
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.902629852294922,
+      "learning_rate": 1.9566101694915257e-05,
+      "loss": 1.5609,
+      "step": 1140
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.317248582839966,
+      "learning_rate": 1.9559322033898307e-05,
+      "loss": 1.5503,
+      "step": 1150
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.522884726524353,
+      "learning_rate": 1.9552542372881357e-05,
+      "loss": 1.3661,
+      "step": 1160
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.1419677734375,
+      "learning_rate": 1.954576271186441e-05,
+      "loss": 1.4839,
+      "step": 1170
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.2150516510009766,
+      "learning_rate": 1.953898305084746e-05,
+      "loss": 1.1754,
+      "step": 1180
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.25466775894165,
+      "learning_rate": 1.953220338983051e-05,
+      "loss": 1.4118,
+      "step": 1190
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.261260986328125,
+      "learning_rate": 1.9525423728813562e-05,
+      "loss": 1.4779,
+      "step": 1200
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.751726150512695,
+      "learning_rate": 1.9518644067796612e-05,
+      "loss": 1.3143,
+      "step": 1210
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.0561487674713135,
+      "learning_rate": 1.9511864406779665e-05,
+      "loss": 1.5102,
+      "step": 1220
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.0936279296875,
+      "learning_rate": 1.950508474576271e-05,
+      "loss": 1.4763,
+      "step": 1230
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.0847556591033936,
+      "learning_rate": 1.9498305084745765e-05,
+      "loss": 1.3039,
+      "step": 1240
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.303075313568115,
+      "learning_rate": 1.9491525423728814e-05,
+      "loss": 1.4491,
+      "step": 1250
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.7520992755889893,
+      "learning_rate": 1.9484745762711864e-05,
+      "loss": 1.3379,
+      "step": 1260
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.233397006988525,
+      "learning_rate": 1.9477966101694917e-05,
+      "loss": 1.5582,
+      "step": 1270
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.988858461380005,
+      "learning_rate": 1.9471186440677967e-05,
+      "loss": 1.4801,
+      "step": 1280
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.172889232635498,
+      "learning_rate": 1.9464406779661017e-05,
+      "loss": 1.5481,
+      "step": 1290
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.5870141983032227,
+      "learning_rate": 1.945762711864407e-05,
+      "loss": 1.4008,
+      "step": 1300
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.538797378540039,
+      "learning_rate": 1.945084745762712e-05,
+      "loss": 1.4145,
+      "step": 1310
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.275447368621826,
+      "learning_rate": 1.9444067796610172e-05,
+      "loss": 1.4524,
+      "step": 1320
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.357238531112671,
+      "learning_rate": 1.9437288135593222e-05,
+      "loss": 1.487,
+      "step": 1330
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.1907289028167725,
+      "learning_rate": 1.9430508474576272e-05,
+      "loss": 1.3633,
+      "step": 1340
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.7265887260437012,
+      "learning_rate": 1.9423728813559325e-05,
+      "loss": 1.3923,
+      "step": 1350
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 10.415843963623047,
+      "learning_rate": 1.9416949152542375e-05,
+      "loss": 1.3302,
+      "step": 1360
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.544619560241699,
+      "learning_rate": 1.9410169491525424e-05,
+      "loss": 1.2906,
+      "step": 1370
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.8395652770996094,
+      "learning_rate": 1.9403389830508477e-05,
+      "loss": 1.4636,
+      "step": 1380
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.81257963180542,
+      "learning_rate": 1.9396610169491527e-05,
+      "loss": 1.408,
+      "step": 1390
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.445089817047119,
+      "learning_rate": 1.938983050847458e-05,
+      "loss": 1.5095,
+      "step": 1400
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.918297290802002,
+      "learning_rate": 1.938305084745763e-05,
+      "loss": 1.154,
+      "step": 1410
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.7477134466171265,
+      "learning_rate": 1.937627118644068e-05,
+      "loss": 1.4866,
+      "step": 1420
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.768603801727295,
+      "learning_rate": 1.9369491525423733e-05,
+      "loss": 1.5642,
+      "step": 1430
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.055950164794922,
+      "learning_rate": 1.936271186440678e-05,
+      "loss": 1.4019,
+      "step": 1440
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 9.180340766906738,
+      "learning_rate": 1.9355932203389832e-05,
+      "loss": 1.509,
+      "step": 1450
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.526889801025391,
+      "learning_rate": 1.9349152542372882e-05,
+      "loss": 1.5197,
+      "step": 1460
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.020207166671753,
+      "learning_rate": 1.934237288135593e-05,
+      "loss": 1.6292,
+      "step": 1470
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.9965012073516846,
+      "learning_rate": 1.9335593220338985e-05,
+      "loss": 1.6703,
+      "step": 1480
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.9574942588806152,
+      "learning_rate": 1.9328813559322034e-05,
+      "loss": 1.3525,
+      "step": 1490
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.8527791500091553,
+      "learning_rate": 1.9322033898305087e-05,
+      "loss": 1.5142,
+      "step": 1500
+    },
+    {
+      "epoch": 0.03,
+      "eval_loss": 1.1832261085510254,
+      "eval_runtime": 67.0686,
+      "eval_samples_per_second": 14.91,
+      "eval_steps_per_second": 14.91,
+      "step": 1500
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.690788745880127,
+      "learning_rate": 1.9315254237288137e-05,
+      "loss": 1.4003,
+      "step": 1510
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.6338605880737305,
+      "learning_rate": 1.9308474576271187e-05,
+      "loss": 1.4324,
+      "step": 1520
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.098454713821411,
+      "learning_rate": 1.930169491525424e-05,
+      "loss": 1.5234,
+      "step": 1530
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.322094440460205,
+      "learning_rate": 1.929491525423729e-05,
+      "loss": 1.2201,
+      "step": 1540
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.089012622833252,
+      "learning_rate": 1.928813559322034e-05,
+      "loss": 1.4444,
+      "step": 1550
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.2624104022979736,
+      "learning_rate": 1.9281355932203392e-05,
+      "loss": 1.4144,
+      "step": 1560
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.4324264526367188,
+      "learning_rate": 1.9274576271186442e-05,
+      "loss": 1.3715,
+      "step": 1570
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.596635341644287,
+      "learning_rate": 1.9267796610169492e-05,
+      "loss": 1.366,
+      "step": 1580
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.6570637226104736,
+      "learning_rate": 1.9261016949152545e-05,
+      "loss": 1.5426,
+      "step": 1590
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.935263156890869,
+      "learning_rate": 1.9254237288135595e-05,
+      "loss": 1.2968,
+      "step": 1600
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.188492298126221,
+      "learning_rate": 1.9247457627118648e-05,
+      "loss": 1.396,
+      "step": 1610
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.421194553375244,
+      "learning_rate": 1.9240677966101698e-05,
+      "loss": 1.4905,
+      "step": 1620
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.735612154006958,
+      "learning_rate": 1.9233898305084747e-05,
+      "loss": 1.4508,
+      "step": 1630
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.989076018333435,
+      "learning_rate": 1.92271186440678e-05,
+      "loss": 1.4195,
+      "step": 1640
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.4910849332809448,
+      "learning_rate": 1.9220338983050847e-05,
+      "loss": 1.374,
+      "step": 1650
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.7809735536575317,
+      "learning_rate": 1.92135593220339e-05,
+      "loss": 1.5008,
+      "step": 1660
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.1804401874542236,
+      "learning_rate": 1.920677966101695e-05,
+      "loss": 1.7274,
+      "step": 1670
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.3297713994979858,
+      "learning_rate": 1.9200000000000003e-05,
+      "loss": 1.3982,
+      "step": 1680
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.2256460189819336,
+      "learning_rate": 1.9193220338983052e-05,
+      "loss": 1.3758,
+      "step": 1690
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.3911705017089844,
+      "learning_rate": 1.9186440677966102e-05,
+      "loss": 1.5054,
+      "step": 1700
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.7374446392059326,
+      "learning_rate": 1.9179661016949155e-05,
+      "loss": 1.3796,
+      "step": 1710
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.8097071647644043,
+      "learning_rate": 1.9172881355932205e-05,
+      "loss": 1.2966,
+      "step": 1720
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.590428352355957,
+      "learning_rate": 1.9166101694915254e-05,
+      "loss": 1.493,
+      "step": 1730
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.458738327026367,
+      "learning_rate": 1.9159322033898308e-05,
+      "loss": 1.3179,
+      "step": 1740
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.689072132110596,
+      "learning_rate": 1.9152542372881357e-05,
+      "loss": 1.2673,
+      "step": 1750
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.915776252746582,
+      "learning_rate": 1.9145762711864407e-05,
+      "loss": 1.5096,
+      "step": 1760
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.790291786193848,
+      "learning_rate": 1.913898305084746e-05,
+      "loss": 1.2601,
+      "step": 1770
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.455094337463379,
+      "learning_rate": 1.913220338983051e-05,
+      "loss": 1.5662,
+      "step": 1780
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.6922054290771484,
+      "learning_rate": 1.9125423728813563e-05,
+      "loss": 1.1819,
+      "step": 1790
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.807013511657715,
+      "learning_rate": 1.9118644067796613e-05,
+      "loss": 1.4572,
+      "step": 1800
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.976742267608643,
+      "learning_rate": 1.9111864406779662e-05,
+      "loss": 1.4976,
+      "step": 1810
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.053407669067383,
+      "learning_rate": 1.9105084745762715e-05,
+      "loss": 1.4944,
+      "step": 1820
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.595452308654785,
+      "learning_rate": 1.9098305084745765e-05,
+      "loss": 1.3392,
+      "step": 1830
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.3358185291290283,
+      "learning_rate": 1.9091525423728815e-05,
+      "loss": 1.3567,
+      "step": 1840
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.6956140995025635,
+      "learning_rate": 1.9084745762711868e-05,
+      "loss": 1.4298,
+      "step": 1850
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.6025402545928955,
+      "learning_rate": 1.9077966101694914e-05,
+      "loss": 1.3674,
+      "step": 1860
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.4478222131729126,
+      "learning_rate": 1.9071186440677967e-05,
+      "loss": 1.5232,
+      "step": 1870
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.7879269123077393,
+      "learning_rate": 1.9064406779661017e-05,
+      "loss": 1.2088,
+      "step": 1880
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.009925365447998,
+      "learning_rate": 1.905762711864407e-05,
+      "loss": 1.306,
+      "step": 1890
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.8789501190185547,
+      "learning_rate": 1.905084745762712e-05,
+      "loss": 1.6352,
+      "step": 1900
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.9612170457839966,
+      "learning_rate": 1.904406779661017e-05,
+      "loss": 1.4313,
+      "step": 1910
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.5845580101013184,
+      "learning_rate": 1.9037288135593223e-05,
+      "loss": 1.464,
+      "step": 1920
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.5218095779418945,
+      "learning_rate": 1.9030508474576272e-05,
+      "loss": 1.3589,
+      "step": 1930
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.3145925998687744,
+      "learning_rate": 1.9023728813559322e-05,
+      "loss": 1.5245,
+      "step": 1940
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.0829366445541382,
+      "learning_rate": 1.9016949152542375e-05,
+      "loss": 1.5265,
+      "step": 1950
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.3161795139312744,
+      "learning_rate": 1.9010169491525425e-05,
+      "loss": 1.3759,
+      "step": 1960
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.037906169891357,
+      "learning_rate": 1.9003389830508478e-05,
+      "loss": 1.576,
+      "step": 1970
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.638181447982788,
+      "learning_rate": 1.8996610169491528e-05,
+      "loss": 1.3606,
+      "step": 1980
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.6348967552185059,
+      "learning_rate": 1.8989830508474577e-05,
+      "loss": 1.4664,
+      "step": 1990
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.778348684310913,
+      "learning_rate": 1.898305084745763e-05,
+      "loss": 1.308,
+      "step": 2000
+    },
+    {
+      "epoch": 0.05,
+      "eval_loss": 1.1971772909164429,
+      "eval_runtime": 67.1893,
+      "eval_samples_per_second": 14.883,
+      "eval_steps_per_second": 14.883,
+      "step": 2000
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.655930757522583,
+      "learning_rate": 1.897627118644068e-05,
+      "loss": 1.5353,
+      "step": 2010
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 7.265784740447998,
+      "learning_rate": 1.896949152542373e-05,
+      "loss": 1.4901,
+      "step": 2020
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.902195930480957,
+      "learning_rate": 1.8962711864406783e-05,
+      "loss": 1.3619,
+      "step": 2030
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.685209035873413,
+      "learning_rate": 1.8955932203389833e-05,
+      "loss": 1.4934,
+      "step": 2040
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.6489458084106445,
+      "learning_rate": 1.8949152542372882e-05,
+      "loss": 1.4807,
+      "step": 2050
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.4234724044799805,
+      "learning_rate": 1.8942372881355932e-05,
+      "loss": 1.4953,
+      "step": 2060
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.339203119277954,
+      "learning_rate": 1.8935593220338985e-05,
+      "loss": 1.4583,
+      "step": 2070
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.0084259510040283,
+      "learning_rate": 1.8928813559322035e-05,
+      "loss": 1.1904,
+      "step": 2080
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.666440963745117,
+      "learning_rate": 1.8922033898305085e-05,
+      "loss": 1.3025,
+      "step": 2090
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.9041775465011597,
+      "learning_rate": 1.8915254237288138e-05,
+      "loss": 1.3873,
+      "step": 2100
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.1416661739349365,
+      "learning_rate": 1.8908474576271187e-05,
+      "loss": 1.3741,
+      "step": 2110
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.8044958114624023,
+      "learning_rate": 1.8901694915254237e-05,
+      "loss": 1.5686,
+      "step": 2120
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.8283112049102783,
+      "learning_rate": 1.889491525423729e-05,
+      "loss": 1.3731,
+      "step": 2130
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.2208967208862305,
+      "learning_rate": 1.888813559322034e-05,
+      "loss": 1.5184,
+      "step": 2140
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.5146480798721313,
+      "learning_rate": 1.8881355932203393e-05,
+      "loss": 1.3791,
+      "step": 2150
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.624804973602295,
+      "learning_rate": 1.8874576271186443e-05,
+      "loss": 1.3399,
+      "step": 2160
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.085561752319336,
+      "learning_rate": 1.8867796610169492e-05,
+      "loss": 1.5298,
+      "step": 2170
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.3589959144592285,
+      "learning_rate": 1.8861016949152545e-05,
+      "loss": 1.2609,
+      "step": 2180
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.165348529815674,
+      "learning_rate": 1.8854237288135595e-05,
+      "loss": 1.5708,
+      "step": 2190
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.508254051208496,
+      "learning_rate": 1.8847457627118645e-05,
+      "loss": 1.5046,
+      "step": 2200
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.811107635498047,
+      "learning_rate": 1.8840677966101698e-05,
+      "loss": 1.5798,
+      "step": 2210
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.915175914764404,
+      "learning_rate": 1.8833898305084748e-05,
+      "loss": 1.3742,
+      "step": 2220
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.212732791900635,
+      "learning_rate": 1.8827118644067797e-05,
+      "loss": 1.4748,
+      "step": 2230
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.4692771434783936,
+      "learning_rate": 1.882033898305085e-05,
+      "loss": 1.1476,
+      "step": 2240
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.057826042175293,
+      "learning_rate": 1.88135593220339e-05,
+      "loss": 1.4235,
+      "step": 2250
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.5246427059173584,
+      "learning_rate": 1.8806779661016953e-05,
+      "loss": 1.4705,
+      "step": 2260
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.8235859870910645,
+      "learning_rate": 1.88e-05,
+      "loss": 1.4017,
+      "step": 2270
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.3737335205078125,
+      "learning_rate": 1.8793220338983053e-05,
+      "loss": 1.2918,
+      "step": 2280
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.6644233465194702,
+      "learning_rate": 1.8786440677966102e-05,
+      "loss": 1.5225,
+      "step": 2290
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.259829044342041,
+      "learning_rate": 1.8779661016949152e-05,
+      "loss": 1.428,
+      "step": 2300
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.356344223022461,
+      "learning_rate": 1.8772881355932205e-05,
+      "loss": 1.4652,
+      "step": 2310
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.862521171569824,
+      "learning_rate": 1.8766101694915255e-05,
+      "loss": 1.3431,
+      "step": 2320
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.1847054958343506,
+      "learning_rate": 1.8759322033898305e-05,
+      "loss": 1.3383,
+      "step": 2330
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.384062767028809,
+      "learning_rate": 1.8752542372881358e-05,
+      "loss": 1.3222,
+      "step": 2340
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.504513740539551,
+      "learning_rate": 1.8745762711864407e-05,
+      "loss": 1.4033,
+      "step": 2350
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.035881757736206,
+      "learning_rate": 1.873898305084746e-05,
+      "loss": 1.4297,
+      "step": 2360
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.0843729972839355,
+      "learning_rate": 1.873220338983051e-05,
+      "loss": 1.5548,
+      "step": 2370
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.0240912437438965,
+      "learning_rate": 1.872542372881356e-05,
+      "loss": 1.5772,
+      "step": 2380
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.927667140960693,
+      "learning_rate": 1.8718644067796613e-05,
+      "loss": 1.471,
+      "step": 2390
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.345820903778076,
+      "learning_rate": 1.8711864406779663e-05,
+      "loss": 1.5337,
+      "step": 2400
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.0480122566223145,
+      "learning_rate": 1.8705084745762712e-05,
+      "loss": 1.5327,
+      "step": 2410
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.2904655933380127,
+      "learning_rate": 1.8698305084745765e-05,
+      "loss": 1.3821,
+      "step": 2420
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.9938225746154785,
+      "learning_rate": 1.8691525423728815e-05,
+      "loss": 1.2987,
+      "step": 2430
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.0605130195617676,
+      "learning_rate": 1.8684745762711868e-05,
+      "loss": 1.7052,
+      "step": 2440
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.4659353494644165,
+      "learning_rate": 1.8677966101694918e-05,
+      "loss": 1.5719,
+      "step": 2450
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.8421217203140259,
+      "learning_rate": 1.8671186440677968e-05,
+      "loss": 1.3397,
+      "step": 2460
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.0998902320861816,
+      "learning_rate": 1.866440677966102e-05,
+      "loss": 1.4649,
+      "step": 2470
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.088279962539673,
+      "learning_rate": 1.8657627118644067e-05,
+      "loss": 1.4717,
+      "step": 2480
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 8.840229034423828,
+      "learning_rate": 1.865084745762712e-05,
+      "loss": 1.4184,
+      "step": 2490
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.5428450107574463,
+      "learning_rate": 1.864406779661017e-05,
+      "loss": 1.5584,
+      "step": 2500
+    },
+    {
+      "epoch": 0.06,
+      "eval_loss": 1.1482951641082764,
+      "eval_runtime": 67.0978,
+      "eval_samples_per_second": 14.904,
+      "eval_steps_per_second": 14.904,
+      "step": 2500
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.455661773681641,
+      "learning_rate": 1.863728813559322e-05,
+      "loss": 1.5017,
+      "step": 2510
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.0317816734313965,
+      "learning_rate": 1.8630508474576273e-05,
+      "loss": 1.5774,
+      "step": 2520
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.15238618850708,
+      "learning_rate": 1.8623728813559322e-05,
+      "loss": 1.3854,
+      "step": 2530
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.9443018436431885,
+      "learning_rate": 1.8616949152542376e-05,
+      "loss": 1.5316,
+      "step": 2540
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.2570126056671143,
+      "learning_rate": 1.8610169491525425e-05,
+      "loss": 1.4141,
+      "step": 2550
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.6843924522399902,
+      "learning_rate": 1.8603389830508475e-05,
+      "loss": 1.4986,
+      "step": 2560
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.697768211364746,
+      "learning_rate": 1.8596610169491528e-05,
+      "loss": 1.3707,
+      "step": 2570
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.394255638122559,
+      "learning_rate": 1.8589830508474578e-05,
+      "loss": 1.4716,
+      "step": 2580
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.255263805389404,
+      "learning_rate": 1.8583050847457627e-05,
+      "loss": 1.5351,
+      "step": 2590
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.587831974029541,
+      "learning_rate": 1.857627118644068e-05,
+      "loss": 1.3261,
+      "step": 2600
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 8.405659675598145,
+      "learning_rate": 1.856949152542373e-05,
+      "loss": 1.4404,
+      "step": 2610
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.0395121574401855,
+      "learning_rate": 1.856271186440678e-05,
+      "loss": 1.368,
+      "step": 2620
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.0600972175598145,
+      "learning_rate": 1.8555932203389833e-05,
+      "loss": 1.4306,
+      "step": 2630
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.951476573944092,
+      "learning_rate": 1.8549152542372883e-05,
+      "loss": 1.4079,
+      "step": 2640
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.6316113471984863,
+      "learning_rate": 1.8542372881355936e-05,
+      "loss": 1.4675,
+      "step": 2650
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.2290899753570557,
+      "learning_rate": 1.8535593220338986e-05,
+      "loss": 1.462,
+      "step": 2660
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.8435947895050049,
+      "learning_rate": 1.8528813559322035e-05,
+      "loss": 1.0381,
+      "step": 2670
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.033623456954956,
+      "learning_rate": 1.852203389830509e-05,
+      "loss": 1.5768,
+      "step": 2680
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.9724245071411133,
+      "learning_rate": 1.8515254237288135e-05,
+      "loss": 1.2659,
+      "step": 2690
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.107775926589966,
+      "learning_rate": 1.8508474576271188e-05,
+      "loss": 1.1751,
+      "step": 2700
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.55193567276001,
+      "learning_rate": 1.8501694915254237e-05,
+      "loss": 1.5424,
+      "step": 2710
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.6639865636825562,
+      "learning_rate": 1.849491525423729e-05,
+      "loss": 1.391,
+      "step": 2720
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.7521861791610718,
+      "learning_rate": 1.848813559322034e-05,
+      "loss": 1.5969,
+      "step": 2730
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.5757704973220825,
+      "learning_rate": 1.848135593220339e-05,
+      "loss": 1.4257,
+      "step": 2740
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.4172585010528564,
+      "learning_rate": 1.8474576271186443e-05,
+      "loss": 1.4596,
+      "step": 2750
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.5575098991394043,
+      "learning_rate": 1.8467796610169493e-05,
+      "loss": 1.5946,
+      "step": 2760
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.385815382003784,
+      "learning_rate": 1.8461016949152542e-05,
+      "loss": 1.3725,
+      "step": 2770
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.108947992324829,
+      "learning_rate": 1.8454237288135596e-05,
+      "loss": 1.3995,
+      "step": 2780
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.180427074432373,
+      "learning_rate": 1.8447457627118645e-05,
+      "loss": 1.7337,
+      "step": 2790
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.628450632095337,
+      "learning_rate": 1.8440677966101695e-05,
+      "loss": 1.3329,
+      "step": 2800
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.4316961765289307,
+      "learning_rate": 1.8433898305084748e-05,
+      "loss": 1.5717,
+      "step": 2810
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.428252100944519,
+      "learning_rate": 1.8427118644067798e-05,
+      "loss": 1.3912,
+      "step": 2820
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.6684863567352295,
+      "learning_rate": 1.842033898305085e-05,
+      "loss": 1.4674,
+      "step": 2830
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.7176380157470703,
+      "learning_rate": 1.84135593220339e-05,
+      "loss": 1.2351,
+      "step": 2840
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.1771388053894043,
+      "learning_rate": 1.840677966101695e-05,
+      "loss": 1.6447,
+      "step": 2850
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.724181652069092,
+      "learning_rate": 1.8400000000000003e-05,
+      "loss": 1.4167,
+      "step": 2860
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.6035029888153076,
+      "learning_rate": 1.8393220338983053e-05,
+      "loss": 1.3391,
+      "step": 2870
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.2763707637786865,
+      "learning_rate": 1.8386440677966103e-05,
+      "loss": 1.5107,
+      "step": 2880
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.8805983066558838,
+      "learning_rate": 1.8379661016949153e-05,
+      "loss": 1.6482,
+      "step": 2890
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.5126351118087769,
+      "learning_rate": 1.8372881355932202e-05,
+      "loss": 1.4343,
+      "step": 2900
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 7.735413074493408,
+      "learning_rate": 1.8366101694915255e-05,
+      "loss": 1.2867,
+      "step": 2910
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.7856109142303467,
+      "learning_rate": 1.8359322033898305e-05,
+      "loss": 1.3506,
+      "step": 2920
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.2521424293518066,
+      "learning_rate": 1.8352542372881358e-05,
+      "loss": 1.3935,
+      "step": 2930
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.2285022735595703,
+      "learning_rate": 1.8345762711864408e-05,
+      "loss": 1.5142,
+      "step": 2940
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.304668664932251,
+      "learning_rate": 1.8338983050847458e-05,
+      "loss": 1.5006,
+      "step": 2950
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.3826477527618408,
+      "learning_rate": 1.833220338983051e-05,
+      "loss": 1.1512,
+      "step": 2960
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.6900218725204468,
+      "learning_rate": 1.832542372881356e-05,
+      "loss": 1.4017,
+      "step": 2970
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.0282979011535645,
+      "learning_rate": 1.831864406779661e-05,
+      "loss": 1.3722,
+      "step": 2980
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.85763692855835,
+      "learning_rate": 1.8311864406779663e-05,
+      "loss": 1.3328,
+      "step": 2990
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.2919251918792725,
+      "learning_rate": 1.8305084745762713e-05,
+      "loss": 1.4177,
+      "step": 3000
+    },
+    {
+      "epoch": 0.07,
+      "eval_loss": 1.2010964155197144,
+      "eval_runtime": 67.1519,
+      "eval_samples_per_second": 14.892,
+      "eval_steps_per_second": 14.892,
+      "step": 3000
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.3012161254882812,
+      "learning_rate": 1.8298305084745766e-05,
+      "loss": 1.2741,
+      "step": 3010
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.728154182434082,
+      "learning_rate": 1.8291525423728816e-05,
+      "loss": 1.495,
+      "step": 3020
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.277280569076538,
+      "learning_rate": 1.8284745762711865e-05,
+      "loss": 1.4877,
+      "step": 3030
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.6222431659698486,
+      "learning_rate": 1.827796610169492e-05,
+      "loss": 1.3664,
+      "step": 3040
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.7156178951263428,
+      "learning_rate": 1.8271186440677968e-05,
+      "loss": 1.3912,
+      "step": 3050
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.9766186475753784,
+      "learning_rate": 1.8264406779661018e-05,
+      "loss": 1.3334,
+      "step": 3060
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 11.627836227416992,
+      "learning_rate": 1.825762711864407e-05,
+      "loss": 1.6583,
+      "step": 3070
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.1804165840148926,
+      "learning_rate": 1.825084745762712e-05,
+      "loss": 1.2931,
+      "step": 3080
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 12.77298641204834,
+      "learning_rate": 1.824406779661017e-05,
+      "loss": 1.3071,
+      "step": 3090
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.189802646636963,
+      "learning_rate": 1.823728813559322e-05,
+      "loss": 1.508,
+      "step": 3100
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.3131775856018066,
+      "learning_rate": 1.8230508474576273e-05,
+      "loss": 1.4604,
+      "step": 3110
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.224339962005615,
+      "learning_rate": 1.8223728813559323e-05,
+      "loss": 1.2131,
+      "step": 3120
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.142259120941162,
+      "learning_rate": 1.8216949152542373e-05,
+      "loss": 1.409,
+      "step": 3130
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.409524917602539,
+      "learning_rate": 1.8210169491525426e-05,
+      "loss": 1.5591,
+      "step": 3140
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.3798465728759766,
+      "learning_rate": 1.8203389830508475e-05,
+      "loss": 1.4358,
+      "step": 3150
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.9663469195365906,
+      "learning_rate": 1.8196610169491525e-05,
+      "loss": 1.4432,
+      "step": 3160
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.4938082695007324,
+      "learning_rate": 1.8189830508474578e-05,
+      "loss": 1.2837,
+      "step": 3170
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.094170093536377,
+      "learning_rate": 1.8183050847457628e-05,
+      "loss": 1.3925,
+      "step": 3180
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.124843120574951,
+      "learning_rate": 1.817627118644068e-05,
+      "loss": 1.5111,
+      "step": 3190
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.090320348739624,
+      "learning_rate": 1.816949152542373e-05,
+      "loss": 1.4229,
+      "step": 3200
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.482816696166992,
+      "learning_rate": 1.816271186440678e-05,
+      "loss": 1.3102,
+      "step": 3210
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.5619702339172363,
+      "learning_rate": 1.8155932203389833e-05,
+      "loss": 1.2051,
+      "step": 3220
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 7.419604301452637,
+      "learning_rate": 1.8149152542372883e-05,
+      "loss": 1.3915,
+      "step": 3230
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.109569072723389,
+      "learning_rate": 1.8142372881355933e-05,
+      "loss": 1.4158,
+      "step": 3240
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.4256246089935303,
+      "learning_rate": 1.8135593220338986e-05,
+      "loss": 1.4494,
+      "step": 3250
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.4669333696365356,
+      "learning_rate": 1.8128813559322036e-05,
+      "loss": 1.6379,
+      "step": 3260
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.800262689590454,
+      "learning_rate": 1.8122033898305085e-05,
+      "loss": 1.5004,
+      "step": 3270
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.3720695972442627,
+      "learning_rate": 1.811525423728814e-05,
+      "loss": 1.425,
+      "step": 3280
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.8338054418563843,
+      "learning_rate": 1.8108474576271188e-05,
+      "loss": 1.4227,
+      "step": 3290
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.969144821166992,
+      "learning_rate": 1.810169491525424e-05,
+      "loss": 1.3806,
+      "step": 3300
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.558366537094116,
+      "learning_rate": 1.8094915254237288e-05,
+      "loss": 1.4254,
+      "step": 3310
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.669677972793579,
+      "learning_rate": 1.808813559322034e-05,
+      "loss": 1.4065,
+      "step": 3320
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.3998602628707886,
+      "learning_rate": 1.808135593220339e-05,
+      "loss": 1.7619,
+      "step": 3330
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.9056684970855713,
+      "learning_rate": 1.807457627118644e-05,
+      "loss": 1.4177,
+      "step": 3340
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.942318916320801,
+      "learning_rate": 1.8067796610169493e-05,
+      "loss": 1.4779,
+      "step": 3350
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.1195051670074463,
+      "learning_rate": 1.8061016949152543e-05,
+      "loss": 1.4865,
+      "step": 3360
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.495617866516113,
+      "learning_rate": 1.8054237288135593e-05,
+      "loss": 1.3127,
+      "step": 3370
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.5328311920166016,
+      "learning_rate": 1.8047457627118646e-05,
+      "loss": 1.4624,
+      "step": 3380
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.0216922760009766,
+      "learning_rate": 1.8040677966101695e-05,
+      "loss": 1.4197,
+      "step": 3390
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.578646183013916,
+      "learning_rate": 1.803389830508475e-05,
+      "loss": 1.2317,
+      "step": 3400
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.8747899532318115,
+      "learning_rate": 1.8027118644067798e-05,
+      "loss": 1.5859,
+      "step": 3410
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.799411773681641,
+      "learning_rate": 1.8020338983050848e-05,
+      "loss": 1.5663,
+      "step": 3420
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.006511688232422,
+      "learning_rate": 1.80135593220339e-05,
+      "loss": 1.4501,
+      "step": 3430
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.5541688203811646,
+      "learning_rate": 1.800677966101695e-05,
+      "loss": 1.6072,
+      "step": 3440
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.2471415996551514,
+      "learning_rate": 1.8e-05,
+      "loss": 1.4034,
+      "step": 3450
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.43544602394104,
+      "learning_rate": 1.7993220338983054e-05,
+      "loss": 1.3181,
+      "step": 3460
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.56860876083374,
+      "learning_rate": 1.7986440677966103e-05,
+      "loss": 1.308,
+      "step": 3470
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.6072529554367065,
+      "learning_rate": 1.7979661016949156e-05,
+      "loss": 1.3989,
+      "step": 3480
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.9302940368652344,
+      "learning_rate": 1.7972881355932206e-05,
+      "loss": 1.5623,
+      "step": 3490
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.61619758605957,
+      "learning_rate": 1.7966101694915256e-05,
+      "loss": 1.4135,
+      "step": 3500
+    },
+    {
+      "epoch": 0.08,
+      "eval_loss": 1.1386640071868896,
+      "eval_runtime": 67.1064,
+      "eval_samples_per_second": 14.902,
+      "eval_steps_per_second": 14.902,
+      "step": 3500
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.315586566925049,
+      "learning_rate": 1.795932203389831e-05,
+      "loss": 1.5134,
+      "step": 3510
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.156167507171631,
+      "learning_rate": 1.7952542372881355e-05,
+      "loss": 1.4425,
+      "step": 3520
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.9159029722213745,
+      "learning_rate": 1.7945762711864408e-05,
+      "loss": 1.3159,
+      "step": 3530
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.3781180381774902,
+      "learning_rate": 1.7938983050847458e-05,
+      "loss": 1.3683,
+      "step": 3540
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.8861943483352661,
+      "learning_rate": 1.7932203389830508e-05,
+      "loss": 1.486,
+      "step": 3550
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.9574244022369385,
+      "learning_rate": 1.792542372881356e-05,
+      "loss": 1.4858,
+      "step": 3560
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.585728168487549,
+      "learning_rate": 1.791864406779661e-05,
+      "loss": 1.4388,
+      "step": 3570
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.258697986602783,
+      "learning_rate": 1.7911864406779664e-05,
+      "loss": 1.5302,
+      "step": 3580
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.292418479919434,
+      "learning_rate": 1.7905084745762713e-05,
+      "loss": 1.3228,
+      "step": 3590
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.342001438140869,
+      "learning_rate": 1.7898305084745763e-05,
+      "loss": 1.1876,
+      "step": 3600
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 7.243982315063477,
+      "learning_rate": 1.7891525423728816e-05,
+      "loss": 1.4241,
+      "step": 3610
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 8.561788558959961,
+      "learning_rate": 1.7884745762711866e-05,
+      "loss": 1.4567,
+      "step": 3620
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.6268651485443115,
+      "learning_rate": 1.7877966101694916e-05,
+      "loss": 1.2928,
+      "step": 3630
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.3656156063079834,
+      "learning_rate": 1.787118644067797e-05,
+      "loss": 1.2128,
+      "step": 3640
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.3748042583465576,
+      "learning_rate": 1.7864406779661018e-05,
+      "loss": 1.472,
+      "step": 3650
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.713076591491699,
+      "learning_rate": 1.785762711864407e-05,
+      "loss": 1.4151,
+      "step": 3660
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.744603157043457,
+      "learning_rate": 1.785084745762712e-05,
+      "loss": 1.3556,
+      "step": 3670
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.440206050872803,
+      "learning_rate": 1.784406779661017e-05,
+      "loss": 1.5536,
+      "step": 3680
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.1328643560409546,
+      "learning_rate": 1.7837288135593224e-05,
+      "loss": 1.3039,
+      "step": 3690
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.336589813232422,
+      "learning_rate": 1.7830508474576274e-05,
+      "loss": 1.4308,
+      "step": 3700
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.2741591930389404,
+      "learning_rate": 1.7823728813559323e-05,
+      "loss": 1.3202,
+      "step": 3710
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.327836990356445,
+      "learning_rate": 1.7816949152542376e-05,
+      "loss": 1.5221,
+      "step": 3720
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.930452823638916,
+      "learning_rate": 1.7810169491525423e-05,
+      "loss": 1.3796,
+      "step": 3730
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.003491401672363,
+      "learning_rate": 1.7803389830508476e-05,
+      "loss": 1.5096,
+      "step": 3740
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.0040380954742432,
+      "learning_rate": 1.7796610169491526e-05,
+      "loss": 1.5853,
+      "step": 3750
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.6716127395629883,
+      "learning_rate": 1.778983050847458e-05,
+      "loss": 1.4564,
+      "step": 3760
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.1395158767700195,
+      "learning_rate": 1.778305084745763e-05,
+      "loss": 1.3919,
+      "step": 3770
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.6375240087509155,
+      "learning_rate": 1.7776271186440678e-05,
+      "loss": 1.2678,
+      "step": 3780
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 11.120715141296387,
+      "learning_rate": 1.776949152542373e-05,
+      "loss": 1.5301,
+      "step": 3790
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.9304801225662231,
+      "learning_rate": 1.776271186440678e-05,
+      "loss": 1.3043,
+      "step": 3800
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.231403112411499,
+      "learning_rate": 1.775593220338983e-05,
+      "loss": 1.3634,
+      "step": 3810
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.287160873413086,
+      "learning_rate": 1.7749152542372884e-05,
+      "loss": 1.306,
+      "step": 3820
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.768348217010498,
+      "learning_rate": 1.7742372881355933e-05,
+      "loss": 1.479,
+      "step": 3830
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.244331121444702,
+      "learning_rate": 1.7735593220338983e-05,
+      "loss": 1.2446,
+      "step": 3840
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.9224987030029297,
+      "learning_rate": 1.7728813559322036e-05,
+      "loss": 1.4043,
+      "step": 3850
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.9674978256225586,
+      "learning_rate": 1.7722033898305086e-05,
+      "loss": 1.3901,
+      "step": 3860
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 7.087489604949951,
+      "learning_rate": 1.771525423728814e-05,
+      "loss": 1.4567,
+      "step": 3870
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.8183858394622803,
+      "learning_rate": 1.770847457627119e-05,
+      "loss": 1.6078,
+      "step": 3880
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.5410759449005127,
+      "learning_rate": 1.770169491525424e-05,
+      "loss": 1.3961,
+      "step": 3890
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.4553983211517334,
+      "learning_rate": 1.769491525423729e-05,
+      "loss": 1.405,
+      "step": 3900
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.181529998779297,
+      "learning_rate": 1.768813559322034e-05,
+      "loss": 1.4944,
+      "step": 3910
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.054699420928955,
+      "learning_rate": 1.768135593220339e-05,
+      "loss": 1.5309,
+      "step": 3920
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.4803507328033447,
+      "learning_rate": 1.767457627118644e-05,
+      "loss": 1.2931,
+      "step": 3930
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.9554686546325684,
+      "learning_rate": 1.7667796610169494e-05,
+      "loss": 1.3651,
+      "step": 3940
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.0756142139434814,
+      "learning_rate": 1.7661016949152543e-05,
+      "loss": 1.4476,
+      "step": 3950
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.552859306335449,
+      "learning_rate": 1.7654237288135593e-05,
+      "loss": 1.4417,
+      "step": 3960
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.7441787719726562,
+      "learning_rate": 1.7647457627118646e-05,
+      "loss": 1.3769,
+      "step": 3970
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.9613826274871826,
+      "learning_rate": 1.7640677966101696e-05,
+      "loss": 1.4381,
+      "step": 3980
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.84116268157959,
+      "learning_rate": 1.7633898305084746e-05,
+      "loss": 1.3539,
+      "step": 3990
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.5078232288360596,
+      "learning_rate": 1.76271186440678e-05,
+      "loss": 1.4277,
+      "step": 4000
+    },
+    {
+      "epoch": 0.09,
+      "eval_loss": 1.158501386642456,
+      "eval_runtime": 67.1002,
+      "eval_samples_per_second": 14.903,
+      "eval_steps_per_second": 14.903,
+      "step": 4000
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.6108415126800537,
+      "learning_rate": 1.762033898305085e-05,
+      "loss": 1.5502,
+      "step": 4010
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.075037479400635,
+      "learning_rate": 1.7613559322033898e-05,
+      "loss": 1.3346,
+      "step": 4020
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.969217300415039,
+      "learning_rate": 1.760677966101695e-05,
+      "loss": 1.3709,
+      "step": 4030
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.885554313659668,
+      "learning_rate": 1.76e-05,
+      "loss": 1.4961,
+      "step": 4040
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.5712426900863647,
+      "learning_rate": 1.7593220338983054e-05,
+      "loss": 1.3091,
+      "step": 4050
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.020577907562256,
+      "learning_rate": 1.7586440677966104e-05,
+      "loss": 1.3308,
+      "step": 4060
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.234301567077637,
+      "learning_rate": 1.7579661016949153e-05,
+      "loss": 1.5322,
+      "step": 4070
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.667493224143982,
+      "learning_rate": 1.7572881355932206e-05,
+      "loss": 1.3986,
+      "step": 4080
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.5255711078643799,
+      "learning_rate": 1.7566101694915256e-05,
+      "loss": 1.1095,
+      "step": 4090
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.342149257659912,
+      "learning_rate": 1.7559322033898306e-05,
+      "loss": 1.468,
+      "step": 4100
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.444767951965332,
+      "learning_rate": 1.755254237288136e-05,
+      "loss": 1.3548,
+      "step": 4110
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.0781197547912598,
+      "learning_rate": 1.754576271186441e-05,
+      "loss": 1.4868,
+      "step": 4120
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.5227110385894775,
+      "learning_rate": 1.753898305084746e-05,
+      "loss": 1.2477,
+      "step": 4130
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.230503797531128,
+      "learning_rate": 1.7532203389830508e-05,
+      "loss": 1.5486,
+      "step": 4140
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.484776496887207,
+      "learning_rate": 1.752542372881356e-05,
+      "loss": 1.4854,
+      "step": 4150
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.312047004699707,
+      "learning_rate": 1.751864406779661e-05,
+      "loss": 1.45,
+      "step": 4160
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.45994758605957,
+      "learning_rate": 1.751186440677966e-05,
+      "loss": 1.2472,
+      "step": 4170
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.4740896224975586,
+      "learning_rate": 1.7505084745762714e-05,
+      "loss": 1.1978,
+      "step": 4180
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.2780067920684814,
+      "learning_rate": 1.7498305084745763e-05,
+      "loss": 1.4335,
+      "step": 4190
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.937509298324585,
+      "learning_rate": 1.7491525423728813e-05,
+      "loss": 1.4036,
+      "step": 4200
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.987236022949219,
+      "learning_rate": 1.7484745762711866e-05,
+      "loss": 1.3751,
+      "step": 4210
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.929296851158142,
+      "learning_rate": 1.7477966101694916e-05,
+      "loss": 1.294,
+      "step": 4220
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.1039040088653564,
+      "learning_rate": 1.747118644067797e-05,
+      "loss": 1.3019,
+      "step": 4230
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.1424946784973145,
+      "learning_rate": 1.746440677966102e-05,
+      "loss": 1.2913,
+      "step": 4240
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.9339830279350281,
+      "learning_rate": 1.745762711864407e-05,
+      "loss": 1.3142,
+      "step": 4250
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.47251296043396,
+      "learning_rate": 1.745084745762712e-05,
+      "loss": 1.4085,
+      "step": 4260
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.0523934364318848,
+      "learning_rate": 1.744406779661017e-05,
+      "loss": 1.3264,
+      "step": 4270
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.680063009262085,
+      "learning_rate": 1.743728813559322e-05,
+      "loss": 1.5524,
+      "step": 4280
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.143898010253906,
+      "learning_rate": 1.7430508474576274e-05,
+      "loss": 1.5795,
+      "step": 4290
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.3456549644470215,
+      "learning_rate": 1.7423728813559324e-05,
+      "loss": 1.3907,
+      "step": 4300
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.776109457015991,
+      "learning_rate": 1.7416949152542373e-05,
+      "loss": 1.2968,
+      "step": 4310
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.241484642028809,
+      "learning_rate": 1.7410169491525427e-05,
+      "loss": 1.305,
+      "step": 4320
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.8671717643737793,
+      "learning_rate": 1.7403389830508476e-05,
+      "loss": 1.5132,
+      "step": 4330
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.0062782764434814,
+      "learning_rate": 1.739661016949153e-05,
+      "loss": 1.2456,
+      "step": 4340
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.044645309448242,
+      "learning_rate": 1.7389830508474576e-05,
+      "loss": 1.4838,
+      "step": 4350
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.291123867034912,
+      "learning_rate": 1.738305084745763e-05,
+      "loss": 1.4495,
+      "step": 4360
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.096938133239746,
+      "learning_rate": 1.737627118644068e-05,
+      "loss": 1.2715,
+      "step": 4370
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.5189833641052246,
+      "learning_rate": 1.7369491525423728e-05,
+      "loss": 1.1358,
+      "step": 4380
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.956369638442993,
+      "learning_rate": 1.736271186440678e-05,
+      "loss": 1.1939,
+      "step": 4390
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 8.631299018859863,
+      "learning_rate": 1.735593220338983e-05,
+      "loss": 1.5566,
+      "step": 4400
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.6200116872787476,
+      "learning_rate": 1.734915254237288e-05,
+      "loss": 1.5092,
+      "step": 4410
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.132010459899902,
+      "learning_rate": 1.7342372881355934e-05,
+      "loss": 1.4241,
+      "step": 4420
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.7074265480041504,
+      "learning_rate": 1.7335593220338983e-05,
+      "loss": 1.4407,
+      "step": 4430
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.0132627487182617,
+      "learning_rate": 1.7328813559322037e-05,
+      "loss": 1.3255,
+      "step": 4440
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.750293254852295,
+      "learning_rate": 1.7322033898305086e-05,
+      "loss": 1.5557,
+      "step": 4450
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.0939888954162598,
+      "learning_rate": 1.7315254237288136e-05,
+      "loss": 1.3632,
+      "step": 4460
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.476027727127075,
+      "learning_rate": 1.730847457627119e-05,
+      "loss": 1.3655,
+      "step": 4470
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.4436848163604736,
+      "learning_rate": 1.730169491525424e-05,
+      "loss": 1.3026,
+      "step": 4480
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.511134147644043,
+      "learning_rate": 1.729491525423729e-05,
+      "loss": 1.5307,
+      "step": 4490
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.5898433923721313,
+      "learning_rate": 1.728813559322034e-05,
+      "loss": 1.4483,
+      "step": 4500
+    },
+    {
+      "epoch": 0.1,
+      "eval_loss": 1.1334296464920044,
+      "eval_runtime": 67.0278,
+      "eval_samples_per_second": 14.919,
+      "eval_steps_per_second": 14.919,
+      "step": 4500
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.7626523971557617,
+      "learning_rate": 1.728135593220339e-05,
+      "loss": 1.3683,
+      "step": 4510
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.1556930541992188,
+      "learning_rate": 1.7274576271186444e-05,
+      "loss": 1.4022,
+      "step": 4520
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.2186577320098877,
+      "learning_rate": 1.7267796610169494e-05,
+      "loss": 1.41,
+      "step": 4530
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.2501068115234375,
+      "learning_rate": 1.7261016949152544e-05,
+      "loss": 1.4536,
+      "step": 4540
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.240054130554199,
+      "learning_rate": 1.7254237288135597e-05,
+      "loss": 1.4073,
+      "step": 4550
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.387373208999634,
+      "learning_rate": 1.7247457627118643e-05,
+      "loss": 1.613,
+      "step": 4560
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.530575752258301,
+      "learning_rate": 1.7240677966101696e-05,
+      "loss": 1.2231,
+      "step": 4570
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.739079236984253,
+      "learning_rate": 1.7233898305084746e-05,
+      "loss": 1.4286,
+      "step": 4580
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.380158424377441,
+      "learning_rate": 1.7227118644067796e-05,
+      "loss": 1.3608,
+      "step": 4590
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.204037189483643,
+      "learning_rate": 1.722033898305085e-05,
+      "loss": 1.2801,
+      "step": 4600
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.6450468301773071,
+      "learning_rate": 1.72135593220339e-05,
+      "loss": 1.29,
+      "step": 4610
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.5391757488250732,
+      "learning_rate": 1.720677966101695e-05,
+      "loss": 1.6189,
+      "step": 4620
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.872121810913086,
+      "learning_rate": 1.72e-05,
+      "loss": 1.2915,
+      "step": 4630
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.8644561767578125,
+      "learning_rate": 1.719322033898305e-05,
+      "loss": 1.5074,
+      "step": 4640
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.774881601333618,
+      "learning_rate": 1.7186440677966104e-05,
+      "loss": 1.3268,
+      "step": 4650
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.5650038719177246,
+      "learning_rate": 1.7179661016949154e-05,
+      "loss": 1.3892,
+      "step": 4660
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.5609863996505737,
+      "learning_rate": 1.7172881355932204e-05,
+      "loss": 1.3675,
+      "step": 4670
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 8.063512802124023,
+      "learning_rate": 1.7166101694915257e-05,
+      "loss": 1.3671,
+      "step": 4680
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.6324501037597656,
+      "learning_rate": 1.7159322033898306e-05,
+      "loss": 1.3284,
+      "step": 4690
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.6894187927246094,
+      "learning_rate": 1.715254237288136e-05,
+      "loss": 1.3032,
+      "step": 4700
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.1557717323303223,
+      "learning_rate": 1.714576271186441e-05,
+      "loss": 1.366,
+      "step": 4710
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.8717806339263916,
+      "learning_rate": 1.713898305084746e-05,
+      "loss": 1.5634,
+      "step": 4720
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.2488980293273926,
+      "learning_rate": 1.7132203389830512e-05,
+      "loss": 1.4814,
+      "step": 4730
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.677203893661499,
+      "learning_rate": 1.712542372881356e-05,
+      "loss": 1.2079,
+      "step": 4740
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.8588718175888062,
+      "learning_rate": 1.711864406779661e-05,
+      "loss": 1.3525,
+      "step": 4750
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.2451605796813965,
+      "learning_rate": 1.711186440677966e-05,
+      "loss": 1.4148,
+      "step": 4760
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.5242979526519775,
+      "learning_rate": 1.710508474576271e-05,
+      "loss": 1.2413,
+      "step": 4770
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.1980478763580322,
+      "learning_rate": 1.7098305084745764e-05,
+      "loss": 1.4558,
+      "step": 4780
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.478370666503906,
+      "learning_rate": 1.7091525423728814e-05,
+      "loss": 1.3639,
+      "step": 4790
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.1923305988311768,
+      "learning_rate": 1.7084745762711867e-05,
+      "loss": 1.5741,
+      "step": 4800
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.293412923812866,
+      "learning_rate": 1.7077966101694916e-05,
+      "loss": 1.2903,
+      "step": 4810
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.4575860500335693,
+      "learning_rate": 1.7071186440677966e-05,
+      "loss": 1.4585,
+      "step": 4820
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.9733614921569824,
+      "learning_rate": 1.706440677966102e-05,
+      "loss": 1.1767,
+      "step": 4830
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.9043350219726562,
+      "learning_rate": 1.705762711864407e-05,
+      "loss": 1.4218,
+      "step": 4840
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.5711846351623535,
+      "learning_rate": 1.705084745762712e-05,
+      "loss": 1.2186,
+      "step": 4850
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.8665926456451416,
+      "learning_rate": 1.704406779661017e-05,
+      "loss": 1.3747,
+      "step": 4860
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.0901689529418945,
+      "learning_rate": 1.703728813559322e-05,
+      "loss": 1.3356,
+      "step": 4870
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.0309970378875732,
+      "learning_rate": 1.703050847457627e-05,
+      "loss": 1.4572,
+      "step": 4880
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.9482085704803467,
+      "learning_rate": 1.7023728813559324e-05,
+      "loss": 1.3365,
+      "step": 4890
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.5880916118621826,
+      "learning_rate": 1.7016949152542374e-05,
+      "loss": 1.4305,
+      "step": 4900
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.12321138381958,
+      "learning_rate": 1.7010169491525427e-05,
+      "loss": 1.5018,
+      "step": 4910
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.508700370788574,
+      "learning_rate": 1.7003389830508477e-05,
+      "loss": 1.237,
+      "step": 4920
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.8924113512039185,
+      "learning_rate": 1.6996610169491526e-05,
+      "loss": 1.3583,
+      "step": 4930
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.551520824432373,
+      "learning_rate": 1.698983050847458e-05,
+      "loss": 1.2687,
+      "step": 4940
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.142587184906006,
+      "learning_rate": 1.698305084745763e-05,
+      "loss": 1.4377,
+      "step": 4950
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.7592850923538208,
+      "learning_rate": 1.697627118644068e-05,
+      "loss": 1.5202,
+      "step": 4960
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.212038040161133,
+      "learning_rate": 1.696949152542373e-05,
+      "loss": 1.3286,
+      "step": 4970
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.211546897888184,
+      "learning_rate": 1.6962711864406782e-05,
+      "loss": 1.5126,
+      "step": 4980
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.877481460571289,
+      "learning_rate": 1.695593220338983e-05,
+      "loss": 1.4129,
+      "step": 4990
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.20015811920166,
+      "learning_rate": 1.694915254237288e-05,
+      "loss": 1.2721,
+      "step": 5000
+    },
+    {
+      "epoch": 0.12,
+      "eval_loss": 1.1188498735427856,
+      "eval_runtime": 67.1151,
+      "eval_samples_per_second": 14.9,
+      "eval_steps_per_second": 14.9,
+      "step": 5000
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.303971290588379,
+      "learning_rate": 1.6942372881355934e-05,
+      "loss": 1.3757,
+      "step": 5010
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.5342512130737305,
+      "learning_rate": 1.6935593220338984e-05,
+      "loss": 1.2597,
+      "step": 5020
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.022207260131836,
+      "learning_rate": 1.6928813559322034e-05,
+      "loss": 1.4939,
+      "step": 5030
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.4725759029388428,
+      "learning_rate": 1.6922033898305087e-05,
+      "loss": 1.2841,
+      "step": 5040
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.944225788116455,
+      "learning_rate": 1.6915254237288136e-05,
+      "loss": 1.3808,
+      "step": 5050
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.0974010229110718,
+      "learning_rate": 1.6908474576271186e-05,
+      "loss": 1.2744,
+      "step": 5060
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.6920721530914307,
+      "learning_rate": 1.690169491525424e-05,
+      "loss": 1.5257,
+      "step": 5070
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.715402126312256,
+      "learning_rate": 1.689491525423729e-05,
+      "loss": 1.5197,
+      "step": 5080
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.5713796615600586,
+      "learning_rate": 1.6888135593220342e-05,
+      "loss": 1.3383,
+      "step": 5090
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.9672067165374756,
+      "learning_rate": 1.6881355932203392e-05,
+      "loss": 1.2505,
+      "step": 5100
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.107706069946289,
+      "learning_rate": 1.687457627118644e-05,
+      "loss": 1.3978,
+      "step": 5110
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.694401979446411,
+      "learning_rate": 1.6867796610169495e-05,
+      "loss": 1.5616,
+      "step": 5120
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.1937147378921509,
+      "learning_rate": 1.6861016949152544e-05,
+      "loss": 1.4247,
+      "step": 5130
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.7874356508255005,
+      "learning_rate": 1.6854237288135594e-05,
+      "loss": 1.1953,
+      "step": 5140
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.081228256225586,
+      "learning_rate": 1.6847457627118647e-05,
+      "loss": 1.3952,
+      "step": 5150
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.792527198791504,
+      "learning_rate": 1.6840677966101697e-05,
+      "loss": 1.3015,
+      "step": 5160
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.482060194015503,
+      "learning_rate": 1.683389830508475e-05,
+      "loss": 1.1953,
+      "step": 5170
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.224519729614258,
+      "learning_rate": 1.6827118644067796e-05,
+      "loss": 1.3729,
+      "step": 5180
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.609736680984497,
+      "learning_rate": 1.682033898305085e-05,
+      "loss": 1.4468,
+      "step": 5190
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.786303997039795,
+      "learning_rate": 1.68135593220339e-05,
+      "loss": 1.4248,
+      "step": 5200
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.584510326385498,
+      "learning_rate": 1.680677966101695e-05,
+      "loss": 1.4896,
+      "step": 5210
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 7.068190574645996,
+      "learning_rate": 1.6800000000000002e-05,
+      "loss": 1.3613,
+      "step": 5220
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.357131242752075,
+      "learning_rate": 1.679322033898305e-05,
+      "loss": 1.1566,
+      "step": 5230
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.547120571136475,
+      "learning_rate": 1.67864406779661e-05,
+      "loss": 1.3428,
+      "step": 5240
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.5816683769226074,
+      "learning_rate": 1.6779661016949154e-05,
+      "loss": 1.3112,
+      "step": 5250
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.817299842834473,
+      "learning_rate": 1.6772881355932204e-05,
+      "loss": 1.4183,
+      "step": 5260
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.906155586242676,
+      "learning_rate": 1.6766101694915257e-05,
+      "loss": 1.4456,
+      "step": 5270
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 8.34229850769043,
+      "learning_rate": 1.6759322033898307e-05,
+      "loss": 1.3624,
+      "step": 5280
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.7926371097564697,
+      "learning_rate": 1.6752542372881357e-05,
+      "loss": 1.4581,
+      "step": 5290
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.89165997505188,
+      "learning_rate": 1.674576271186441e-05,
+      "loss": 1.5302,
+      "step": 5300
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.7644487619400024,
+      "learning_rate": 1.673898305084746e-05,
+      "loss": 1.2788,
+      "step": 5310
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.700753927230835,
+      "learning_rate": 1.673220338983051e-05,
+      "loss": 1.4187,
+      "step": 5320
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.538086175918579,
+      "learning_rate": 1.6725423728813562e-05,
+      "loss": 1.4627,
+      "step": 5330
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.7824482917785645,
+      "learning_rate": 1.6718644067796612e-05,
+      "loss": 1.3147,
+      "step": 5340
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.426088571548462,
+      "learning_rate": 1.671186440677966e-05,
+      "loss": 1.4254,
+      "step": 5350
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.100320339202881,
+      "learning_rate": 1.6705084745762715e-05,
+      "loss": 1.2516,
+      "step": 5360
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.2726573944091797,
+      "learning_rate": 1.6698305084745764e-05,
+      "loss": 1.2826,
+      "step": 5370
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.182670593261719,
+      "learning_rate": 1.6691525423728817e-05,
+      "loss": 1.468,
+      "step": 5380
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.9416656494140625,
+      "learning_rate": 1.6684745762711864e-05,
+      "loss": 1.4455,
+      "step": 5390
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.5433812141418457,
+      "learning_rate": 1.6677966101694917e-05,
+      "loss": 1.4634,
+      "step": 5400
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.191746234893799,
+      "learning_rate": 1.6671186440677967e-05,
+      "loss": 1.3533,
+      "step": 5410
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.96868097782135,
+      "learning_rate": 1.6664406779661016e-05,
+      "loss": 1.3821,
+      "step": 5420
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.689413070678711,
+      "learning_rate": 1.665762711864407e-05,
+      "loss": 1.4411,
+      "step": 5430
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.58811354637146,
+      "learning_rate": 1.665084745762712e-05,
+      "loss": 1.3856,
+      "step": 5440
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.642380952835083,
+      "learning_rate": 1.6644067796610172e-05,
+      "loss": 1.4348,
+      "step": 5450
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.2928035259246826,
+      "learning_rate": 1.6637288135593222e-05,
+      "loss": 1.5305,
+      "step": 5460
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.261194705963135,
+      "learning_rate": 1.663050847457627e-05,
+      "loss": 1.2432,
+      "step": 5470
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.4704355001449585,
+      "learning_rate": 1.6623728813559325e-05,
+      "loss": 1.5543,
+      "step": 5480
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.8535382747650146,
+      "learning_rate": 1.6616949152542374e-05,
+      "loss": 1.3296,
+      "step": 5490
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.3717048168182373,
+      "learning_rate": 1.6610169491525424e-05,
+      "loss": 1.5835,
+      "step": 5500
+    },
+    {
+      "epoch": 0.13,
+      "eval_loss": 1.0649362802505493,
+      "eval_runtime": 66.9134,
+      "eval_samples_per_second": 14.945,
+      "eval_steps_per_second": 14.945,
+      "step": 5500
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.3180487155914307,
+      "learning_rate": 1.6603389830508477e-05,
+      "loss": 1.3602,
+      "step": 5510
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.428536891937256,
+      "learning_rate": 1.6596610169491527e-05,
+      "loss": 1.4902,
+      "step": 5520
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.3657978773117065,
+      "learning_rate": 1.6589830508474577e-05,
+      "loss": 1.3019,
+      "step": 5530
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.775023341178894,
+      "learning_rate": 1.658305084745763e-05,
+      "loss": 1.309,
+      "step": 5540
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.047219753265381,
+      "learning_rate": 1.657627118644068e-05,
+      "loss": 1.463,
+      "step": 5550
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.6937634944915771,
+      "learning_rate": 1.6569491525423732e-05,
+      "loss": 1.4777,
+      "step": 5560
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.0562212467193604,
+      "learning_rate": 1.6562711864406782e-05,
+      "loss": 1.4766,
+      "step": 5570
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.577533483505249,
+      "learning_rate": 1.6555932203389832e-05,
+      "loss": 1.3865,
+      "step": 5580
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.4926679134368896,
+      "learning_rate": 1.654915254237288e-05,
+      "loss": 1.3764,
+      "step": 5590
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.904751777648926,
+      "learning_rate": 1.654237288135593e-05,
+      "loss": 1.2022,
+      "step": 5600
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.692943096160889,
+      "learning_rate": 1.6535593220338984e-05,
+      "loss": 1.2504,
+      "step": 5610
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.7578885555267334,
+      "learning_rate": 1.6528813559322034e-05,
+      "loss": 1.3805,
+      "step": 5620
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.6114630699157715,
+      "learning_rate": 1.6522033898305084e-05,
+      "loss": 1.3391,
+      "step": 5630
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.5723536014556885,
+      "learning_rate": 1.6515254237288137e-05,
+      "loss": 1.2655,
+      "step": 5640
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.141366720199585,
+      "learning_rate": 1.6508474576271187e-05,
+      "loss": 1.4299,
+      "step": 5650
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.970369577407837,
+      "learning_rate": 1.650169491525424e-05,
+      "loss": 1.4145,
+      "step": 5660
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.518489837646484,
+      "learning_rate": 1.649491525423729e-05,
+      "loss": 1.419,
+      "step": 5670
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.4335029125213623,
+      "learning_rate": 1.648813559322034e-05,
+      "loss": 1.3781,
+      "step": 5680
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.025974988937378,
+      "learning_rate": 1.6481355932203392e-05,
+      "loss": 1.456,
+      "step": 5690
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.4364434480667114,
+      "learning_rate": 1.6474576271186442e-05,
+      "loss": 1.4295,
+      "step": 5700
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.17189359664917,
+      "learning_rate": 1.646779661016949e-05,
+      "loss": 1.3155,
+      "step": 5710
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.905094146728516,
+      "learning_rate": 1.6461016949152545e-05,
+      "loss": 1.3161,
+      "step": 5720
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.6364047527313232,
+      "learning_rate": 1.6454237288135594e-05,
+      "loss": 1.378,
+      "step": 5730
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.8336093425750732,
+      "learning_rate": 1.6447457627118648e-05,
+      "loss": 1.2784,
+      "step": 5740
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.631802558898926,
+      "learning_rate": 1.6440677966101697e-05,
+      "loss": 1.2603,
+      "step": 5750
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 9.11829948425293,
+      "learning_rate": 1.6433898305084747e-05,
+      "loss": 1.3563,
+      "step": 5760
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.024040460586548,
+      "learning_rate": 1.64271186440678e-05,
+      "loss": 1.3491,
+      "step": 5770
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.558547019958496,
+      "learning_rate": 1.642033898305085e-05,
+      "loss": 1.3992,
+      "step": 5780
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.277186870574951,
+      "learning_rate": 1.64135593220339e-05,
+      "loss": 1.3636,
+      "step": 5790
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.6262595653533936,
+      "learning_rate": 1.640677966101695e-05,
+      "loss": 1.2908,
+      "step": 5800
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.5076159238815308,
+      "learning_rate": 1.64e-05,
+      "loss": 1.4525,
+      "step": 5810
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.863429069519043,
+      "learning_rate": 1.6393220338983052e-05,
+      "loss": 1.5479,
+      "step": 5820
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.3503921031951904,
+      "learning_rate": 1.63864406779661e-05,
+      "loss": 1.2465,
+      "step": 5830
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.307507276535034,
+      "learning_rate": 1.6379661016949155e-05,
+      "loss": 1.3346,
+      "step": 5840
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.301724910736084,
+      "learning_rate": 1.6372881355932204e-05,
+      "loss": 1.4078,
+      "step": 5850
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.821153163909912,
+      "learning_rate": 1.6366101694915254e-05,
+      "loss": 1.4262,
+      "step": 5860
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.4718036651611328,
+      "learning_rate": 1.6359322033898307e-05,
+      "loss": 1.5426,
+      "step": 5870
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.044488430023193,
+      "learning_rate": 1.6352542372881357e-05,
+      "loss": 1.3703,
+      "step": 5880
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.0791540145874023,
+      "learning_rate": 1.6345762711864407e-05,
+      "loss": 1.3436,
+      "step": 5890
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.2117247581481934,
+      "learning_rate": 1.633898305084746e-05,
+      "loss": 1.428,
+      "step": 5900
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.8183112144470215,
+      "learning_rate": 1.633220338983051e-05,
+      "loss": 1.6166,
+      "step": 5910
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.566955089569092,
+      "learning_rate": 1.6325423728813563e-05,
+      "loss": 1.4127,
+      "step": 5920
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.455868721008301,
+      "learning_rate": 1.6318644067796612e-05,
+      "loss": 1.2684,
+      "step": 5930
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.872295618057251,
+      "learning_rate": 1.6311864406779662e-05,
+      "loss": 1.1753,
+      "step": 5940
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.5006000995635986,
+      "learning_rate": 1.6305084745762715e-05,
+      "loss": 1.1916,
+      "step": 5950
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.872570514678955,
+      "learning_rate": 1.6298305084745765e-05,
+      "loss": 1.1692,
+      "step": 5960
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.5257492065429688,
+      "learning_rate": 1.6291525423728814e-05,
+      "loss": 1.3611,
+      "step": 5970
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.483407497406006,
+      "learning_rate": 1.6284745762711868e-05,
+      "loss": 1.4315,
+      "step": 5980
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.8382426500320435,
+      "learning_rate": 1.6277966101694917e-05,
+      "loss": 1.4698,
+      "step": 5990
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.766387939453125,
+      "learning_rate": 1.6271186440677967e-05,
+      "loss": 1.361,
+      "step": 6000
+    },
+    {
+      "epoch": 0.14,
+      "eval_loss": 1.1337964534759521,
+      "eval_runtime": 67.0656,
+      "eval_samples_per_second": 14.911,
+      "eval_steps_per_second": 14.911,
+      "step": 6000
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.5609889030456543,
+      "learning_rate": 1.6264406779661017e-05,
+      "loss": 1.4736,
+      "step": 6010
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.4243054389953613,
+      "learning_rate": 1.625762711864407e-05,
+      "loss": 1.3073,
+      "step": 6020
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.1757571697235107,
+      "learning_rate": 1.625084745762712e-05,
+      "loss": 1.5282,
+      "step": 6030
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.4322659969329834,
+      "learning_rate": 1.624406779661017e-05,
+      "loss": 1.534,
+      "step": 6040
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.9377753734588623,
+      "learning_rate": 1.6237288135593222e-05,
+      "loss": 1.354,
+      "step": 6050
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.534852981567383,
+      "learning_rate": 1.6230508474576272e-05,
+      "loss": 1.2613,
+      "step": 6060
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.6617101430892944,
+      "learning_rate": 1.6223728813559322e-05,
+      "loss": 1.4414,
+      "step": 6070
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.800033092498779,
+      "learning_rate": 1.6216949152542375e-05,
+      "loss": 1.3286,
+      "step": 6080
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.045539379119873,
+      "learning_rate": 1.6210169491525424e-05,
+      "loss": 1.4305,
+      "step": 6090
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.9078855514526367,
+      "learning_rate": 1.6203389830508474e-05,
+      "loss": 1.3735,
+      "step": 6100
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.0546069145202637,
+      "learning_rate": 1.6196610169491527e-05,
+      "loss": 1.3062,
+      "step": 6110
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.7809054851531982,
+      "learning_rate": 1.6189830508474577e-05,
+      "loss": 1.2676,
+      "step": 6120
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.5306100845336914,
+      "learning_rate": 1.618305084745763e-05,
+      "loss": 1.4725,
+      "step": 6130
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.504068374633789,
+      "learning_rate": 1.617627118644068e-05,
+      "loss": 1.3162,
+      "step": 6140
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.180143356323242,
+      "learning_rate": 1.616949152542373e-05,
+      "loss": 1.3681,
+      "step": 6150
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.692683696746826,
+      "learning_rate": 1.6162711864406783e-05,
+      "loss": 1.2732,
+      "step": 6160
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.6309053897857666,
+      "learning_rate": 1.6155932203389832e-05,
+      "loss": 1.212,
+      "step": 6170
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.247219562530518,
+      "learning_rate": 1.6149152542372882e-05,
+      "loss": 1.3263,
+      "step": 6180
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.420581579208374,
+      "learning_rate": 1.6142372881355935e-05,
+      "loss": 1.2392,
+      "step": 6190
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.2501585483551025,
+      "learning_rate": 1.6135593220338985e-05,
+      "loss": 1.3958,
+      "step": 6200
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.111384868621826,
+      "learning_rate": 1.6128813559322038e-05,
+      "loss": 1.4578,
+      "step": 6210
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.0279738903045654,
+      "learning_rate": 1.6122033898305084e-05,
+      "loss": 1.4083,
+      "step": 6220
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.9996086955070496,
+      "learning_rate": 1.6115254237288137e-05,
+      "loss": 1.2599,
+      "step": 6230
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.6425893306732178,
+      "learning_rate": 1.6108474576271187e-05,
+      "loss": 1.3794,
+      "step": 6240
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.84988260269165,
+      "learning_rate": 1.6101694915254237e-05,
+      "loss": 1.3801,
+      "step": 6250
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 7.806826591491699,
+      "learning_rate": 1.609491525423729e-05,
+      "loss": 1.3006,
+      "step": 6260
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.427366733551025,
+      "learning_rate": 1.608813559322034e-05,
+      "loss": 1.4588,
+      "step": 6270
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.51039719581604,
+      "learning_rate": 1.608135593220339e-05,
+      "loss": 1.4209,
+      "step": 6280
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.638364553451538,
+      "learning_rate": 1.6074576271186442e-05,
+      "loss": 1.3952,
+      "step": 6290
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.9240260124206543,
+      "learning_rate": 1.6067796610169492e-05,
+      "loss": 1.3194,
+      "step": 6300
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.798541784286499,
+      "learning_rate": 1.6061016949152545e-05,
+      "loss": 1.2272,
+      "step": 6310
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.5487680435180664,
+      "learning_rate": 1.6054237288135595e-05,
+      "loss": 1.4795,
+      "step": 6320
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.1405465602874756,
+      "learning_rate": 1.6047457627118645e-05,
+      "loss": 1.3574,
+      "step": 6330
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.4945203065872192,
+      "learning_rate": 1.6040677966101698e-05,
+      "loss": 1.3875,
+      "step": 6340
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.4741933345794678,
+      "learning_rate": 1.6033898305084747e-05,
+      "loss": 1.3757,
+      "step": 6350
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.7438557147979736,
+      "learning_rate": 1.6027118644067797e-05,
+      "loss": 1.4486,
+      "step": 6360
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.901851177215576,
+      "learning_rate": 1.602033898305085e-05,
+      "loss": 1.3954,
+      "step": 6370
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.6728180646896362,
+      "learning_rate": 1.60135593220339e-05,
+      "loss": 1.311,
+      "step": 6380
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.970463752746582,
+      "learning_rate": 1.600677966101695e-05,
+      "loss": 1.4011,
+      "step": 6390
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 12.278678894042969,
+      "learning_rate": 1.6000000000000003e-05,
+      "loss": 1.4062,
+      "step": 6400
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.6090126037597656,
+      "learning_rate": 1.5993220338983052e-05,
+      "loss": 1.3565,
+      "step": 6410
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.103749752044678,
+      "learning_rate": 1.5986440677966105e-05,
+      "loss": 1.4804,
+      "step": 6420
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.822219133377075,
+      "learning_rate": 1.5979661016949152e-05,
+      "loss": 1.4567,
+      "step": 6430
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.869916915893555,
+      "learning_rate": 1.5972881355932205e-05,
+      "loss": 1.3093,
+      "step": 6440
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.4998152256011963,
+      "learning_rate": 1.5966101694915255e-05,
+      "loss": 1.5148,
+      "step": 6450
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.6436197757720947,
+      "learning_rate": 1.5959322033898304e-05,
+      "loss": 1.519,
+      "step": 6460
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.9514076709747314,
+      "learning_rate": 1.5952542372881357e-05,
+      "loss": 1.2673,
+      "step": 6470
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.19160795211792,
+      "learning_rate": 1.5945762711864407e-05,
+      "loss": 1.2964,
+      "step": 6480
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.302166700363159,
+      "learning_rate": 1.593898305084746e-05,
+      "loss": 1.5311,
+      "step": 6490
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.648465156555176,
+      "learning_rate": 1.593220338983051e-05,
+      "loss": 1.5642,
+      "step": 6500
+    },
+    {
+      "epoch": 0.15,
+      "eval_loss": 1.1224900484085083,
+      "eval_runtime": 67.0535,
+      "eval_samples_per_second": 14.913,
+      "eval_steps_per_second": 14.913,
+      "step": 6500
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.6730732917785645,
+      "learning_rate": 1.592542372881356e-05,
+      "loss": 1.3842,
+      "step": 6510
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.7324059009552,
+      "learning_rate": 1.5918644067796613e-05,
+      "loss": 1.4838,
+      "step": 6520
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.5643513202667236,
+      "learning_rate": 1.5911864406779662e-05,
+      "loss": 1.326,
+      "step": 6530
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.981233596801758,
+      "learning_rate": 1.5905084745762712e-05,
+      "loss": 1.1575,
+      "step": 6540
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.4779703617095947,
+      "learning_rate": 1.5898305084745765e-05,
+      "loss": 1.3118,
+      "step": 6550
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.831774711608887,
+      "learning_rate": 1.5891525423728815e-05,
+      "loss": 1.3681,
+      "step": 6560
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.515669345855713,
+      "learning_rate": 1.5884745762711865e-05,
+      "loss": 1.2631,
+      "step": 6570
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.7888094186782837,
+      "learning_rate": 1.5877966101694918e-05,
+      "loss": 1.4389,
+      "step": 6580
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.9246420860290527,
+      "learning_rate": 1.5871186440677967e-05,
+      "loss": 1.2706,
+      "step": 6590
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.629147529602051,
+      "learning_rate": 1.586440677966102e-05,
+      "loss": 1.4099,
+      "step": 6600
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.5322067737579346,
+      "learning_rate": 1.585762711864407e-05,
+      "loss": 1.3107,
+      "step": 6610
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.1020938158035278,
+      "learning_rate": 1.585084745762712e-05,
+      "loss": 1.3415,
+      "step": 6620
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.216763496398926,
+      "learning_rate": 1.584406779661017e-05,
+      "loss": 1.3664,
+      "step": 6630
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.0716798305511475,
+      "learning_rate": 1.583728813559322e-05,
+      "loss": 1.1287,
+      "step": 6640
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.659613609313965,
+      "learning_rate": 1.5830508474576272e-05,
+      "loss": 1.4241,
+      "step": 6650
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 7.536604881286621,
+      "learning_rate": 1.5823728813559322e-05,
+      "loss": 1.426,
+      "step": 6660
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.213017702102661,
+      "learning_rate": 1.5816949152542372e-05,
+      "loss": 1.3564,
+      "step": 6670
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.929147481918335,
+      "learning_rate": 1.5810169491525425e-05,
+      "loss": 1.5176,
+      "step": 6680
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.5247678756713867,
+      "learning_rate": 1.5803389830508475e-05,
+      "loss": 1.4608,
+      "step": 6690
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.313474178314209,
+      "learning_rate": 1.5796610169491528e-05,
+      "loss": 1.411,
+      "step": 6700
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.516223907470703,
+      "learning_rate": 1.5789830508474577e-05,
+      "loss": 1.2478,
+      "step": 6710
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.387552499771118,
+      "learning_rate": 1.5783050847457627e-05,
+      "loss": 1.3225,
+      "step": 6720
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.9858660697937012,
+      "learning_rate": 1.577627118644068e-05,
+      "loss": 1.3289,
+      "step": 6730
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.3014678955078125,
+      "learning_rate": 1.576949152542373e-05,
+      "loss": 1.457,
+      "step": 6740
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 10.245163917541504,
+      "learning_rate": 1.576271186440678e-05,
+      "loss": 1.4311,
+      "step": 6750
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.805487871170044,
+      "learning_rate": 1.5755932203389833e-05,
+      "loss": 1.3857,
+      "step": 6760
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.2822433710098267,
+      "learning_rate": 1.5749152542372882e-05,
+      "loss": 1.4489,
+      "step": 6770
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.254396915435791,
+      "learning_rate": 1.5742372881355936e-05,
+      "loss": 1.3914,
+      "step": 6780
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.3311165571212769,
+      "learning_rate": 1.5735593220338985e-05,
+      "loss": 1.243,
+      "step": 6790
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.7225605249404907,
+      "learning_rate": 1.5728813559322035e-05,
+      "loss": 1.425,
+      "step": 6800
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.585598945617676,
+      "learning_rate": 1.5722033898305088e-05,
+      "loss": 1.414,
+      "step": 6810
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.3213350772857666,
+      "learning_rate": 1.5715254237288138e-05,
+      "loss": 1.3233,
+      "step": 6820
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.5480399131774902,
+      "learning_rate": 1.5708474576271187e-05,
+      "loss": 1.3096,
+      "step": 6830
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.9461381435394287,
+      "learning_rate": 1.5701694915254237e-05,
+      "loss": 1.2209,
+      "step": 6840
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.8784675598144531,
+      "learning_rate": 1.5694915254237287e-05,
+      "loss": 1.2154,
+      "step": 6850
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.964296817779541,
+      "learning_rate": 1.568813559322034e-05,
+      "loss": 1.3913,
+      "step": 6860
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.013964891433716,
+      "learning_rate": 1.568135593220339e-05,
+      "loss": 1.4206,
+      "step": 6870
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.359316110610962,
+      "learning_rate": 1.5674576271186443e-05,
+      "loss": 1.3806,
+      "step": 6880
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.8272175788879395,
+      "learning_rate": 1.5667796610169492e-05,
+      "loss": 1.2084,
+      "step": 6890
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.1895947456359863,
+      "learning_rate": 1.5661016949152542e-05,
+      "loss": 1.5326,
+      "step": 6900
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.105100631713867,
+      "learning_rate": 1.5654237288135595e-05,
+      "loss": 1.3586,
+      "step": 6910
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.031907081604004,
+      "learning_rate": 1.5647457627118645e-05,
+      "loss": 1.0995,
+      "step": 6920
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 15.673041343688965,
+      "learning_rate": 1.5640677966101695e-05,
+      "loss": 1.5304,
+      "step": 6930
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.542800426483154,
+      "learning_rate": 1.5633898305084748e-05,
+      "loss": 1.4612,
+      "step": 6940
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.4466352462768555,
+      "learning_rate": 1.5627118644067798e-05,
+      "loss": 1.4641,
+      "step": 6950
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.0358104705810547,
+      "learning_rate": 1.562033898305085e-05,
+      "loss": 1.5169,
+      "step": 6960
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.3914594650268555,
+      "learning_rate": 1.56135593220339e-05,
+      "loss": 1.3483,
+      "step": 6970
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.626600503921509,
+      "learning_rate": 1.560677966101695e-05,
+      "loss": 1.3114,
+      "step": 6980
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.424988746643066,
+      "learning_rate": 1.5600000000000003e-05,
+      "loss": 1.2768,
+      "step": 6990
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 7.747529029846191,
+      "learning_rate": 1.5593220338983053e-05,
+      "loss": 1.1494,
+      "step": 7000
+    },
+    {
+      "epoch": 0.16,
+      "eval_loss": 1.0665773153305054,
+      "eval_runtime": 67.1469,
+      "eval_samples_per_second": 14.893,
+      "eval_steps_per_second": 14.893,
+      "step": 7000
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.9942423701286316,
+      "learning_rate": 1.5586440677966103e-05,
+      "loss": 1.3625,
+      "step": 7010
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.6136345863342285,
+      "learning_rate": 1.5579661016949156e-05,
+      "loss": 1.2317,
+      "step": 7020
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.3277318477630615,
+      "learning_rate": 1.5572881355932205e-05,
+      "loss": 1.5961,
+      "step": 7030
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.660468578338623,
+      "learning_rate": 1.5566101694915255e-05,
+      "loss": 1.3753,
+      "step": 7040
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.3660237789154053,
+      "learning_rate": 1.5559322033898305e-05,
+      "loss": 1.311,
+      "step": 7050
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.831724643707275,
+      "learning_rate": 1.5552542372881358e-05,
+      "loss": 1.3895,
+      "step": 7060
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.309049606323242,
+      "learning_rate": 1.5545762711864408e-05,
+      "loss": 1.325,
+      "step": 7070
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.5130770206451416,
+      "learning_rate": 1.5538983050847457e-05,
+      "loss": 1.3108,
+      "step": 7080
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 11.385069847106934,
+      "learning_rate": 1.553220338983051e-05,
+      "loss": 1.3483,
+      "step": 7090
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.305030107498169,
+      "learning_rate": 1.552542372881356e-05,
+      "loss": 1.12,
+      "step": 7100
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.7694568634033203,
+      "learning_rate": 1.551864406779661e-05,
+      "loss": 1.4892,
+      "step": 7110
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.4342284202575684,
+      "learning_rate": 1.5511864406779663e-05,
+      "loss": 1.3841,
+      "step": 7120
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 8.617453575134277,
+      "learning_rate": 1.5505084745762713e-05,
+      "loss": 1.2856,
+      "step": 7130
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.346908092498779,
+      "learning_rate": 1.5498305084745762e-05,
+      "loss": 1.2906,
+      "step": 7140
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 3.7418603897094727,
+      "learning_rate": 1.5491525423728815e-05,
+      "loss": 1.5261,
+      "step": 7150
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.1197525262832642,
+      "learning_rate": 1.5484745762711865e-05,
+      "loss": 1.2972,
+      "step": 7160
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 3.5833466053009033,
+      "learning_rate": 1.5477966101694918e-05,
+      "loss": 1.3393,
+      "step": 7170
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 3.046830892562866,
+      "learning_rate": 1.5471186440677968e-05,
+      "loss": 1.4726,
+      "step": 7180
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 10.847610473632812,
+      "learning_rate": 1.5464406779661018e-05,
+      "loss": 1.4872,
+      "step": 7190
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.379472255706787,
+      "learning_rate": 1.545762711864407e-05,
+      "loss": 1.3873,
+      "step": 7200
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.1769838333129883,
+      "learning_rate": 1.545084745762712e-05,
+      "loss": 1.432,
+      "step": 7210
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.547250747680664,
+      "learning_rate": 1.544406779661017e-05,
+      "loss": 1.5696,
+      "step": 7220
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 8.801671981811523,
+      "learning_rate": 1.5437288135593223e-05,
+      "loss": 1.4942,
+      "step": 7230
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.674645185470581,
+      "learning_rate": 1.5430508474576273e-05,
+      "loss": 1.2567,
+      "step": 7240
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 3.047281503677368,
+      "learning_rate": 1.5423728813559326e-05,
+      "loss": 1.1286,
+      "step": 7250
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.422094464302063,
+      "learning_rate": 1.5416949152542372e-05,
+      "loss": 1.4765,
+      "step": 7260
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 3.8606622219085693,
+      "learning_rate": 1.5410169491525425e-05,
+      "loss": 1.2492,
+      "step": 7270
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.5654449462890625,
+      "learning_rate": 1.5403389830508475e-05,
+      "loss": 1.2715,
+      "step": 7280
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.5264976024627686,
+      "learning_rate": 1.5396610169491525e-05,
+      "loss": 1.4896,
+      "step": 7290
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.585784435272217,
+      "learning_rate": 1.5389830508474578e-05,
+      "loss": 1.3929,
+      "step": 7300
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.7968121767044067,
+      "learning_rate": 1.5383050847457628e-05,
+      "loss": 1.3018,
+      "step": 7310
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.1171720027923584,
+      "learning_rate": 1.5376271186440677e-05,
+      "loss": 1.4814,
+      "step": 7320
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.140902519226074,
+      "learning_rate": 1.536949152542373e-05,
+      "loss": 1.454,
+      "step": 7330
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 8.208806037902832,
+      "learning_rate": 1.536271186440678e-05,
+      "loss": 1.3176,
+      "step": 7340
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.932653903961182,
+      "learning_rate": 1.5355932203389833e-05,
+      "loss": 1.3674,
+      "step": 7350
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.9559627175331116,
+      "learning_rate": 1.5349152542372883e-05,
+      "loss": 1.3701,
+      "step": 7360
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.199779033660889,
+      "learning_rate": 1.5342372881355933e-05,
+      "loss": 1.2293,
+      "step": 7370
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.100563049316406,
+      "learning_rate": 1.5335593220338986e-05,
+      "loss": 1.3448,
+      "step": 7380
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.8607914447784424,
+      "learning_rate": 1.5328813559322035e-05,
+      "loss": 1.4457,
+      "step": 7390
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.619777202606201,
+      "learning_rate": 1.5322033898305085e-05,
+      "loss": 1.5492,
+      "step": 7400
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.614234447479248,
+      "learning_rate": 1.5315254237288138e-05,
+      "loss": 1.2983,
+      "step": 7410
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.3404347896575928,
+      "learning_rate": 1.5308474576271188e-05,
+      "loss": 1.3885,
+      "step": 7420
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.3918747901916504,
+      "learning_rate": 1.530169491525424e-05,
+      "loss": 1.2908,
+      "step": 7430
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.5815324783325195,
+      "learning_rate": 1.529491525423729e-05,
+      "loss": 1.2632,
+      "step": 7440
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.126127243041992,
+      "learning_rate": 1.528813559322034e-05,
+      "loss": 1.6003,
+      "step": 7450
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.3783955574035645,
+      "learning_rate": 1.528135593220339e-05,
+      "loss": 1.3901,
+      "step": 7460
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.654667615890503,
+      "learning_rate": 1.527457627118644e-05,
+      "loss": 1.2822,
+      "step": 7470
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 7.821676731109619,
+      "learning_rate": 1.5267796610169493e-05,
+      "loss": 1.396,
+      "step": 7480
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.0116024017333984,
+      "learning_rate": 1.5261016949152543e-05,
+      "loss": 1.4361,
+      "step": 7490
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.1018917560577393,
+      "learning_rate": 1.5254237288135594e-05,
+      "loss": 1.3302,
+      "step": 7500
+    },
+    {
+      "epoch": 0.17,
+      "eval_loss": 1.1425343751907349,
+      "eval_runtime": 67.2314,
+      "eval_samples_per_second": 14.874,
+      "eval_steps_per_second": 14.874,
+      "step": 7500
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 30000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 2500,
+  "total_flos": 1.2076594495488e+17,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}