diff --git "a/checkpoint-80000/trainer_state.json" "b/checkpoint-80000/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-80000/trainer_state.json"
@@ -0,0 +1,6274 @@
+{
+  "best_global_step": 80000,
+  "best_metric": 2.1577999591827393,
+  "best_model_checkpoint": "models/plausibert/checkpoint-80000",
+  "epoch": 75.47180938900684,
+  "eval_steps": 1000,
+  "global_step": 80000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.09436187780136825,
+      "grad_norm": 3.896272659301758,
+      "learning_rate": 9.9e-07,
+      "loss": 10.1635,
+      "step": 100
+    },
+    {
+      "epoch": 0.1887237556027365,
+      "grad_norm": 2.5898056030273438,
+      "learning_rate": 1.99e-06,
+      "loss": 9.1407,
+      "step": 200
+    },
+    {
+      "epoch": 0.28308563340410475,
+      "grad_norm": 2.2082550525665283,
+      "learning_rate": 2.99e-06,
+      "loss": 8.6837,
+      "step": 300
+    },
+    {
+      "epoch": 0.377447511205473,
+      "grad_norm": 2.3000612258911133,
+      "learning_rate": 3.99e-06,
+      "loss": 8.2813,
+      "step": 400
+    },
+    {
+      "epoch": 0.4718093890068412,
+      "grad_norm": 1.8331196308135986,
+      "learning_rate": 4.9900000000000005e-06,
+      "loss": 7.8228,
+      "step": 500
+    },
+    {
+      "epoch": 0.5661712668082095,
+      "grad_norm": 1.509745717048645,
+      "learning_rate": 5.99e-06,
+      "loss": 7.3049,
+      "step": 600
+    },
+    {
+      "epoch": 0.6605331446095777,
+      "grad_norm": 1.0882461071014404,
+      "learning_rate": 6.990000000000001e-06,
+      "loss": 6.8271,
+      "step": 700
+    },
+    {
+      "epoch": 0.754895022410946,
+      "grad_norm": 0.9488278031349182,
+      "learning_rate": 7.99e-06,
+      "loss": 6.4755,
+      "step": 800
+    },
+    {
+      "epoch": 0.8492569002123143,
+      "grad_norm": 0.7304624319076538,
+      "learning_rate": 8.99e-06,
+      "loss": 6.2671,
+      "step": 900
+    },
+    {
+      "epoch": 0.9436187780136824,
+      "grad_norm": 0.7234325408935547,
+      "learning_rate": 9.990000000000001e-06,
+      "loss": 6.1688,
+      "step": 1000
+    },
+    {
+      "epoch": 0.9436187780136824,
+      "eval_loss": 6.0837483406066895,
+      "eval_runtime": 74.3964,
+      "eval_samples_per_second": 202.577,
+      "eval_steps_per_second": 6.331,
+      "step": 1000
+    },
+    {
+      "epoch": 1.0377447511205473,
+      "grad_norm": 0.5878735780715942,
+      "learning_rate": 1.099e-05,
+      "loss": 6.0804,
+      "step": 1100
+    },
+    {
+      "epoch": 1.1321066289219155,
+      "grad_norm": 1.0044981241226196,
+      "learning_rate": 1.199e-05,
+      "loss": 6.0195,
+      "step": 1200
+    },
+    {
+      "epoch": 1.226468506723284,
+      "grad_norm": 0.6754953265190125,
+      "learning_rate": 1.299e-05,
+      "loss": 5.9978,
+      "step": 1300
+    },
+    {
+      "epoch": 1.320830384524652,
+      "grad_norm": 0.7222411036491394,
+      "learning_rate": 1.399e-05,
+      "loss": 5.9548,
+      "step": 1400
+    },
+    {
+      "epoch": 1.4151922623260202,
+      "grad_norm": 0.6526289582252502,
+      "learning_rate": 1.499e-05,
+      "loss": 5.9321,
+      "step": 1500
+    },
+    {
+      "epoch": 1.5095541401273884,
+      "grad_norm": 0.7330458164215088,
+      "learning_rate": 1.599e-05,
+      "loss": 5.8904,
+      "step": 1600
+    },
+    {
+      "epoch": 1.6039160179287568,
+      "grad_norm": 0.7171738743782043,
+      "learning_rate": 1.699e-05,
+      "loss": 5.887,
+      "step": 1700
+    },
+    {
+      "epoch": 1.698277895730125,
+      "grad_norm": 0.6907537579536438,
+      "learning_rate": 1.7990000000000002e-05,
+      "loss": 5.8503,
+      "step": 1800
+    },
+    {
+      "epoch": 1.7926397735314934,
+      "grad_norm": 1.0631508827209473,
+      "learning_rate": 1.8990000000000003e-05,
+      "loss": 5.8309,
+      "step": 1900
+    },
+    {
+      "epoch": 1.8870016513328616,
+      "grad_norm": 0.5577560663223267,
+      "learning_rate": 1.999e-05,
+      "loss": 5.8133,
+      "step": 2000
+    },
+    {
+      "epoch": 1.8870016513328616,
+      "eval_loss": 5.77741003036499,
+      "eval_runtime": 74.4232,
+      "eval_samples_per_second": 202.504,
+      "eval_steps_per_second": 6.329,
+      "step": 2000
+    },
+    {
+      "epoch": 1.9813635291342298,
+      "grad_norm": 0.878653347492218,
+      "learning_rate": 2.099e-05,
+      "loss": 5.7944,
+      "step": 2100
+    },
+    {
+      "epoch": 2.0754895022410946,
+      "grad_norm": 0.5987277626991272,
+      "learning_rate": 2.199e-05,
+      "loss": 5.7819,
+      "step": 2200
+    },
+    {
+      "epoch": 2.169851380042463,
+      "grad_norm": 0.6282809972763062,
+      "learning_rate": 2.2990000000000002e-05,
+      "loss": 5.7676,
+      "step": 2300
+    },
+    {
+      "epoch": 2.264213257843831,
+      "grad_norm": 0.7545515298843384,
+      "learning_rate": 2.3990000000000002e-05,
+      "loss": 5.7626,
+      "step": 2400
+    },
+    {
+      "epoch": 2.358575135645199,
+      "grad_norm": 0.9032400846481323,
+      "learning_rate": 2.4990000000000003e-05,
+      "loss": 5.7559,
+      "step": 2500
+    },
+    {
+      "epoch": 2.452937013446568,
+      "grad_norm": 0.6327532529830933,
+      "learning_rate": 2.5990000000000004e-05,
+      "loss": 5.7318,
+      "step": 2600
+    },
+    {
+      "epoch": 2.547298891247936,
+      "grad_norm": 0.6848271489143372,
+      "learning_rate": 2.6989999999999997e-05,
+      "loss": 5.7179,
+      "step": 2700
+    },
+    {
+      "epoch": 2.641660769049304,
+      "grad_norm": 0.5881490707397461,
+      "learning_rate": 2.7989999999999998e-05,
+      "loss": 5.7122,
+      "step": 2800
+    },
+    {
+      "epoch": 2.7360226468506723,
+      "grad_norm": 1.0944511890411377,
+      "learning_rate": 2.8990000000000002e-05,
+      "loss": 5.7164,
+      "step": 2900
+    },
+    {
+      "epoch": 2.8303845246520405,
+      "grad_norm": 0.9962751865386963,
+      "learning_rate": 2.9990000000000003e-05,
+      "loss": 5.6835,
+      "step": 3000
+    },
+    {
+      "epoch": 2.8303845246520405,
+      "eval_loss": 5.670976161956787,
+      "eval_runtime": 74.4573,
+      "eval_samples_per_second": 202.411,
+      "eval_steps_per_second": 6.326,
+      "step": 3000
+    },
+    {
+      "epoch": 2.9247464024534087,
+      "grad_norm": 0.8781231641769409,
+      "learning_rate": 3.099e-05,
+      "loss": 5.6835,
+      "step": 3100
+    },
+    {
+      "epoch": 3.0188723755602735,
+      "grad_norm": 0.806686282157898,
+      "learning_rate": 3.1990000000000004e-05,
+      "loss": 5.6689,
+      "step": 3200
+    },
+    {
+      "epoch": 3.1132342533616417,
+      "grad_norm": 0.8316852450370789,
+      "learning_rate": 3.299e-05,
+      "loss": 5.6723,
+      "step": 3300
+    },
+    {
+      "epoch": 3.2075961311630103,
+      "grad_norm": 0.82398521900177,
+      "learning_rate": 3.399e-05,
+      "loss": 5.6536,
+      "step": 3400
+    },
+    {
+      "epoch": 3.3019580089643785,
+      "grad_norm": 0.9566921591758728,
+      "learning_rate": 3.499e-05,
+      "loss": 5.6482,
+      "step": 3500
+    },
+    {
+      "epoch": 3.3963198867657467,
+      "grad_norm": 0.8480503559112549,
+      "learning_rate": 3.599e-05,
+      "loss": 5.6511,
+      "step": 3600
+    },
+    {
+      "epoch": 3.490681764567115,
+      "grad_norm": 0.9649183750152588,
+      "learning_rate": 3.699e-05,
+      "loss": 5.644,
+      "step": 3700
+    },
+    {
+      "epoch": 3.585043642368483,
+      "grad_norm": 0.870224118232727,
+      "learning_rate": 3.799e-05,
+      "loss": 5.6256,
+      "step": 3800
+    },
+    {
+      "epoch": 3.6794055201698512,
+      "grad_norm": 0.6461514830589294,
+      "learning_rate": 3.8990000000000004e-05,
+      "loss": 5.6282,
+      "step": 3900
+    },
+    {
+      "epoch": 3.77376739797122,
+      "grad_norm": 0.7242473363876343,
+      "learning_rate": 3.999e-05,
+      "loss": 5.627,
+      "step": 4000
+    },
+    {
+      "epoch": 3.77376739797122,
+      "eval_loss": 5.605582237243652,
+      "eval_runtime": 74.4202,
+      "eval_samples_per_second": 202.512,
+      "eval_steps_per_second": 6.329,
+      "step": 4000
+    },
+    {
+      "epoch": 3.868129275772588,
+      "grad_norm": 0.6954628825187683,
+      "learning_rate": 4.099e-05,
+      "loss": 5.6102,
+      "step": 4100
+    },
+    {
+      "epoch": 3.962491153573956,
+      "grad_norm": 0.676533579826355,
+      "learning_rate": 4.199e-05,
+      "loss": 5.6129,
+      "step": 4200
+    },
+    {
+      "epoch": 4.056617126680821,
+      "grad_norm": 0.7012454271316528,
+      "learning_rate": 4.299e-05,
+      "loss": 5.6066,
+      "step": 4300
+    },
+    {
+      "epoch": 4.150979004482189,
+      "grad_norm": 1.0515376329421997,
+      "learning_rate": 4.3990000000000004e-05,
+      "loss": 5.5951,
+      "step": 4400
+    },
+    {
+      "epoch": 4.245340882283557,
+      "grad_norm": 0.7566595077514648,
+      "learning_rate": 4.499e-05,
+      "loss": 5.5894,
+      "step": 4500
+    },
+    {
+      "epoch": 4.339702760084926,
+      "grad_norm": 0.8826500177383423,
+      "learning_rate": 4.599e-05,
+      "loss": 5.5889,
+      "step": 4600
+    },
+    {
+      "epoch": 4.434064637886294,
+      "grad_norm": 1.1490343809127808,
+      "learning_rate": 4.699e-05,
+      "loss": 5.5867,
+      "step": 4700
+    },
+    {
+      "epoch": 4.528426515687662,
+      "grad_norm": 0.8356620669364929,
+      "learning_rate": 4.799e-05,
+      "loss": 5.571,
+      "step": 4800
+    },
+    {
+      "epoch": 4.622788393489031,
+      "grad_norm": 0.7320058345794678,
+      "learning_rate": 4.8990000000000004e-05,
+      "loss": 5.5733,
+      "step": 4900
+    },
+    {
+      "epoch": 4.717150271290398,
+      "grad_norm": 0.9290647506713867,
+      "learning_rate": 4.999e-05,
+      "loss": 5.5623,
+      "step": 5000
+    },
+    {
+      "epoch": 4.717150271290398,
+      "eval_loss": 5.556262016296387,
+      "eval_runtime": 74.4326,
+      "eval_samples_per_second": 202.479,
+      "eval_steps_per_second": 6.328,
+      "step": 5000
+    },
+    {
+      "epoch": 4.811512149091767,
+      "grad_norm": 0.733899712562561,
+      "learning_rate": 5.0990000000000005e-05,
+      "loss": 5.5642,
+      "step": 5100
+    },
+    {
+      "epoch": 4.905874026893136,
+      "grad_norm": 0.7362884879112244,
+      "learning_rate": 5.199000000000001e-05,
+      "loss": 5.5571,
+      "step": 5200
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 0.9703426957130432,
+      "learning_rate": 5.2990000000000006e-05,
+      "loss": 5.5584,
+      "step": 5300
+    },
+    {
+      "epoch": 5.094361877801369,
+      "grad_norm": 0.8319281339645386,
+      "learning_rate": 5.399000000000001e-05,
+      "loss": 5.5516,
+      "step": 5400
+    },
+    {
+      "epoch": 5.188723755602736,
+      "grad_norm": 0.9939006567001343,
+      "learning_rate": 5.499000000000001e-05,
+      "loss": 5.535,
+      "step": 5500
+    },
+    {
+      "epoch": 5.283085633404105,
+      "grad_norm": 1.1493581533432007,
+      "learning_rate": 5.599e-05,
+      "loss": 5.5439,
+      "step": 5600
+    },
+    {
+      "epoch": 5.377447511205473,
+      "grad_norm": 0.9831592440605164,
+      "learning_rate": 5.699e-05,
+      "loss": 5.5357,
+      "step": 5700
+    },
+    {
+      "epoch": 5.471809389006841,
+      "grad_norm": 0.8419109582901001,
+      "learning_rate": 5.799e-05,
+      "loss": 5.5192,
+      "step": 5800
+    },
+    {
+      "epoch": 5.56617126680821,
+      "grad_norm": 0.803837239742279,
+      "learning_rate": 5.899e-05,
+      "loss": 5.5352,
+      "step": 5900
+    },
+    {
+      "epoch": 5.660533144609578,
+      "grad_norm": 0.8247868418693542,
+      "learning_rate": 5.999e-05,
+      "loss": 5.5243,
+      "step": 6000
+    },
+    {
+      "epoch": 5.660533144609578,
+      "eval_loss": 5.519050598144531,
+      "eval_runtime": 74.4404,
+      "eval_samples_per_second": 202.457,
+      "eval_steps_per_second": 6.327,
+      "step": 6000
+    },
+    {
+      "epoch": 5.754895022410946,
+      "grad_norm": 0.8315047025680542,
+      "learning_rate": 6.0990000000000004e-05,
+      "loss": 5.5147,
+      "step": 6100
+    },
+    {
+      "epoch": 5.849256900212314,
+      "grad_norm": 0.7833096385002136,
+      "learning_rate": 6.199000000000001e-05,
+      "loss": 5.523,
+      "step": 6200
+    },
+    {
+      "epoch": 5.943618778013683,
+      "grad_norm": 0.8719848990440369,
+      "learning_rate": 6.299e-05,
+      "loss": 5.5146,
+      "step": 6300
+    },
+    {
+      "epoch": 6.037744751120547,
+      "grad_norm": 0.8079792857170105,
+      "learning_rate": 6.399e-05,
+      "loss": 5.5031,
+      "step": 6400
+    },
+    {
+      "epoch": 6.132106628921916,
+      "grad_norm": 0.9179410338401794,
+      "learning_rate": 6.499000000000001e-05,
+      "loss": 5.5075,
+      "step": 6500
+    },
+    {
+      "epoch": 6.2264685067232834,
+      "grad_norm": 0.9099860191345215,
+      "learning_rate": 6.599000000000001e-05,
+      "loss": 5.4979,
+      "step": 6600
+    },
+    {
+      "epoch": 6.320830384524652,
+      "grad_norm": 0.7887560129165649,
+      "learning_rate": 6.699000000000001e-05,
+      "loss": 5.4848,
+      "step": 6700
+    },
+    {
+      "epoch": 6.415192262326021,
+      "grad_norm": 0.9273598194122314,
+      "learning_rate": 6.799e-05,
+      "loss": 5.4865,
+      "step": 6800
+    },
+    {
+      "epoch": 6.509554140127388,
+      "grad_norm": 0.8765608072280884,
+      "learning_rate": 6.899e-05,
+      "loss": 5.4878,
+      "step": 6900
+    },
+    {
+      "epoch": 6.603916017928757,
+      "grad_norm": 0.8114970922470093,
+      "learning_rate": 6.999e-05,
+      "loss": 5.4844,
+      "step": 7000
+    },
+    {
+      "epoch": 6.603916017928757,
+      "eval_loss": 5.4852294921875,
+      "eval_runtime": 74.3954,
+      "eval_samples_per_second": 202.58,
+      "eval_steps_per_second": 6.331,
+      "step": 7000
+    },
+    {
+      "epoch": 6.698277895730125,
+      "grad_norm": 0.8651418089866638,
+      "learning_rate": 7.099e-05,
+      "loss": 5.4928,
+      "step": 7100
+    },
+    {
+      "epoch": 6.792639773531493,
+      "grad_norm": 1.0564604997634888,
+      "learning_rate": 7.199000000000001e-05,
+      "loss": 5.4697,
+      "step": 7200
+    },
+    {
+      "epoch": 6.887001651332861,
+      "grad_norm": 0.6973365545272827,
+      "learning_rate": 7.299e-05,
+      "loss": 5.477,
+      "step": 7300
+    },
+    {
+      "epoch": 6.98136352913423,
+      "grad_norm": 0.9218101501464844,
+      "learning_rate": 7.399e-05,
+      "loss": 5.482,
+      "step": 7400
+    },
+    {
+      "epoch": 7.075489502241094,
+      "grad_norm": 0.8261534571647644,
+      "learning_rate": 7.499e-05,
+      "loss": 5.4594,
+      "step": 7500
+    },
+    {
+      "epoch": 7.169851380042463,
+      "grad_norm": 0.9930619597434998,
+      "learning_rate": 7.599000000000001e-05,
+      "loss": 5.4673,
+      "step": 7600
+    },
+    {
+      "epoch": 7.264213257843831,
+      "grad_norm": 0.9928425550460815,
+      "learning_rate": 7.699e-05,
+      "loss": 5.4517,
+      "step": 7700
+    },
+    {
+      "epoch": 7.358575135645199,
+      "grad_norm": 0.9189549088478088,
+      "learning_rate": 7.799e-05,
+      "loss": 5.4587,
+      "step": 7800
+    },
+    {
+      "epoch": 7.452937013446568,
+      "grad_norm": 0.8727918267250061,
+      "learning_rate": 7.899000000000001e-05,
+      "loss": 5.4462,
+      "step": 7900
+    },
+    {
+      "epoch": 7.5472988912479355,
+      "grad_norm": 0.87702476978302,
+      "learning_rate": 7.999000000000001e-05,
+      "loss": 5.4608,
+      "step": 8000
+    },
+    {
+      "epoch": 7.5472988912479355,
+      "eval_loss": 5.4536590576171875,
+      "eval_runtime": 74.428,
+      "eval_samples_per_second": 202.491,
+      "eval_steps_per_second": 6.328,
+      "step": 8000
+    },
+    {
+      "epoch": 7.641660769049304,
+      "grad_norm": 0.9891802668571472,
+      "learning_rate": 8.099e-05,
+      "loss": 5.4534,
+      "step": 8100
+    },
+    {
+      "epoch": 7.736022646850673,
+      "grad_norm": 1.0268616676330566,
+      "learning_rate": 8.199e-05,
+      "loss": 5.4537,
+      "step": 8200
+    },
+    {
+      "epoch": 7.8303845246520405,
+      "grad_norm": 0.7097076177597046,
+      "learning_rate": 8.299e-05,
+      "loss": 5.4557,
+      "step": 8300
+    },
+    {
+      "epoch": 7.924746402453409,
+      "grad_norm": 0.955984354019165,
+      "learning_rate": 8.399e-05,
+      "loss": 5.4478,
+      "step": 8400
+    },
+    {
+      "epoch": 8.018872375560274,
+      "grad_norm": 1.1078083515167236,
+      "learning_rate": 8.499e-05,
+      "loss": 5.4363,
+      "step": 8500
+    },
+    {
+      "epoch": 8.113234253361641,
+      "grad_norm": 0.8345474004745483,
+      "learning_rate": 8.599000000000001e-05,
+      "loss": 5.4326,
+      "step": 8600
+    },
+    {
+      "epoch": 8.20759613116301,
+      "grad_norm": 0.8543062210083008,
+      "learning_rate": 8.699e-05,
+      "loss": 5.4211,
+      "step": 8700
+    },
+    {
+      "epoch": 8.301958008964379,
+      "grad_norm": 0.9149999022483826,
+      "learning_rate": 8.799e-05,
+      "loss": 5.4342,
+      "step": 8800
+    },
+    {
+      "epoch": 8.396319886765747,
+      "grad_norm": 0.8864068984985352,
+      "learning_rate": 8.899e-05,
+      "loss": 5.4274,
+      "step": 8900
+    },
+    {
+      "epoch": 8.490681764567114,
+      "grad_norm": 0.8433997631072998,
+      "learning_rate": 8.999000000000001e-05,
+      "loss": 5.4176,
+      "step": 9000
+    },
+    {
+      "epoch": 8.490681764567114,
+      "eval_loss": 5.430926322937012,
+      "eval_runtime": 74.4314,
+      "eval_samples_per_second": 202.482,
+      "eval_steps_per_second": 6.328,
+      "step": 9000
+    },
+    {
+      "epoch": 8.585043642368483,
+      "grad_norm": 0.9321457147598267,
+      "learning_rate": 9.099000000000001e-05,
+      "loss": 5.4309,
+      "step": 9100
+    },
+    {
+      "epoch": 8.679405520169851,
+      "grad_norm": 0.9454299211502075,
+      "learning_rate": 9.199e-05,
+      "loss": 5.4222,
+      "step": 9200
+    },
+    {
+      "epoch": 8.77376739797122,
+      "grad_norm": 0.9724736213684082,
+      "learning_rate": 9.299e-05,
+      "loss": 5.4294,
+      "step": 9300
+    },
+    {
+      "epoch": 8.868129275772588,
+      "grad_norm": 0.8900998830795288,
+      "learning_rate": 9.399e-05,
+      "loss": 5.4178,
+      "step": 9400
+    },
+    {
+      "epoch": 8.962491153573955,
+      "grad_norm": 0.8408654928207397,
+      "learning_rate": 9.499e-05,
+      "loss": 5.4098,
+      "step": 9500
+    },
+    {
+      "epoch": 9.056617126680822,
+      "grad_norm": 0.9535470008850098,
+      "learning_rate": 9.599000000000001e-05,
+      "loss": 5.4211,
+      "step": 9600
+    },
+    {
+      "epoch": 9.150979004482188,
+      "grad_norm": 0.9209726452827454,
+      "learning_rate": 9.699e-05,
+      "loss": 5.406,
+      "step": 9700
+    },
+    {
+      "epoch": 9.245340882283557,
+      "grad_norm": 0.9139648079872131,
+      "learning_rate": 9.799e-05,
+      "loss": 5.4004,
+      "step": 9800
+    },
+    {
+      "epoch": 9.339702760084926,
+      "grad_norm": 0.9717345833778381,
+      "learning_rate": 9.899e-05,
+      "loss": 5.4054,
+      "step": 9900
+    },
+    {
+      "epoch": 9.434064637886294,
+      "grad_norm": 0.9255397319793701,
+      "learning_rate": 9.999000000000001e-05,
+      "loss": 5.4013,
+      "step": 10000
+    },
+    {
+      "epoch": 9.434064637886294,
+      "eval_loss": 5.412718296051025,
+      "eval_runtime": 74.3953,
+      "eval_samples_per_second": 202.58,
+      "eval_steps_per_second": 6.331,
+      "step": 10000
+    },
+    {
+      "epoch": 9.528426515687663,
+      "grad_norm": 1.028997540473938,
+      "learning_rate": 9.999970144476398e-05,
+      "loss": 5.4009,
+      "step": 10100
+    },
+    {
+      "epoch": 9.62278839348903,
+      "grad_norm": 0.9851202964782715,
+      "learning_rate": 9.999879368940656e-05,
+      "loss": 5.4014,
+      "step": 10200
+    },
+    {
+      "epoch": 9.717150271290398,
+      "grad_norm": 0.8858544230461121,
+      "learning_rate": 9.999727671452668e-05,
+      "loss": 5.3941,
+      "step": 10300
+    },
+    {
+      "epoch": 9.811512149091767,
+      "grad_norm": 1.0239341259002686,
+      "learning_rate": 9.999515053860821e-05,
+      "loss": 5.4031,
+      "step": 10400
+    },
+    {
+      "epoch": 9.905874026893136,
+      "grad_norm": 1.0142488479614258,
+      "learning_rate": 9.999241518755793e-05,
+      "loss": 5.395,
+      "step": 10500
+    },
+    {
+      "epoch": 10.0,
+      "grad_norm": 0.9302240014076233,
+      "learning_rate": 9.998907069470524e-05,
+      "loss": 5.3871,
+      "step": 10600
+    },
+    {
+      "epoch": 10.094361877801369,
+      "grad_norm": 0.9354774951934814,
+      "learning_rate": 9.998511710080171e-05,
+      "loss": 5.3863,
+      "step": 10700
+    },
+    {
+      "epoch": 10.188723755602737,
+      "grad_norm": 1.126571774482727,
+      "learning_rate": 9.998055445402067e-05,
+      "loss": 5.377,
+      "step": 10800
+    },
+    {
+      "epoch": 10.283085633404104,
+      "grad_norm": 0.9135729670524597,
+      "learning_rate": 9.997538280995651e-05,
+      "loss": 5.3809,
+      "step": 10900
+    },
+    {
+      "epoch": 10.377447511205473,
+      "grad_norm": 0.8609702587127686,
+      "learning_rate": 9.996960223162406e-05,
+      "loss": 5.3784,
+      "step": 11000
+    },
+    {
+      "epoch": 10.377447511205473,
+      "eval_loss": 5.395565986633301,
+      "eval_runtime": 74.4304,
+      "eval_samples_per_second": 202.484,
+      "eval_steps_per_second": 6.328,
+      "step": 11000
+    },
+    {
+      "epoch": 10.471809389006841,
+      "grad_norm": 1.0586097240447998,
+      "learning_rate": 9.996321278945788e-05,
+      "loss": 5.3636,
+      "step": 11100
+    },
+    {
+      "epoch": 10.56617126680821,
+      "grad_norm": 0.8425832390785217,
+      "learning_rate": 9.995621456131128e-05,
+      "loss": 5.3767,
+      "step": 11200
+    },
+    {
+      "epoch": 10.660533144609577,
+      "grad_norm": 0.9150360822677612,
+      "learning_rate": 9.994860763245549e-05,
+      "loss": 5.3744,
+      "step": 11300
+    },
+    {
+      "epoch": 10.754895022410945,
+      "grad_norm": 0.9344813227653503,
+      "learning_rate": 9.99403920955785e-05,
+      "loss": 5.3904,
+      "step": 11400
+    },
+    {
+      "epoch": 10.849256900212314,
+      "grad_norm": 1.030868411064148,
+      "learning_rate": 9.993156805078405e-05,
+      "loss": 5.3683,
+      "step": 11500
+    },
+    {
+      "epoch": 10.943618778013683,
+      "grad_norm": 0.9607158899307251,
+      "learning_rate": 9.992213560559034e-05,
+      "loss": 5.3726,
+      "step": 11600
+    },
+    {
+      "epoch": 11.037744751120547,
+      "grad_norm": 0.9345623254776001,
+      "learning_rate": 9.991209487492876e-05,
+      "loss": 5.3647,
+      "step": 11700
+    },
+    {
+      "epoch": 11.132106628921916,
+      "grad_norm": 0.8988705277442932,
+      "learning_rate": 9.990144598114242e-05,
+      "loss": 5.3618,
+      "step": 11800
+    },
+    {
+      "epoch": 11.226468506723284,
+      "grad_norm": 1.0196607112884521,
+      "learning_rate": 9.989018905398473e-05,
+      "loss": 5.364,
+      "step": 11900
+    },
+    {
+      "epoch": 11.320830384524651,
+      "grad_norm": 0.8753230571746826,
+      "learning_rate": 9.98783242306178e-05,
+      "loss": 5.3542,
+      "step": 12000
+    },
+    {
+      "epoch": 11.320830384524651,
+      "eval_loss": 5.378592491149902,
+      "eval_runtime": 74.4306,
+      "eval_samples_per_second": 202.484,
+      "eval_steps_per_second": 6.328,
+      "step": 12000
+    },
+    {
+      "epoch": 11.41519226232602,
+      "grad_norm": 0.8527357578277588,
+      "learning_rate": 9.986585165561076e-05,
+      "loss": 5.3541,
+      "step": 12100
+    },
+    {
+      "epoch": 11.509554140127388,
+      "grad_norm": 0.8356689214706421,
+      "learning_rate": 9.9852771480938e-05,
+      "loss": 5.3532,
+      "step": 12200
+    },
+    {
+      "epoch": 11.603916017928757,
+      "grad_norm": 0.8830206990242004,
+      "learning_rate": 9.983908386597732e-05,
+      "loss": 5.3614,
+      "step": 12300
+    },
+    {
+      "epoch": 11.698277895730126,
+      "grad_norm": 0.937129557132721,
+      "learning_rate": 9.9824788977508e-05,
+      "loss": 5.3556,
+      "step": 12400
+    },
+    {
+      "epoch": 11.792639773531493,
+      "grad_norm": 0.8641858100891113,
+      "learning_rate": 9.980988698970872e-05,
+      "loss": 5.3481,
+      "step": 12500
+    },
+    {
+      "epoch": 11.887001651332861,
+      "grad_norm": 0.9982609152793884,
+      "learning_rate": 9.979437808415552e-05,
+      "loss": 5.348,
+      "step": 12600
+    },
+    {
+      "epoch": 11.98136352913423,
+      "grad_norm": 0.9662423133850098,
+      "learning_rate": 9.977826244981952e-05,
+      "loss": 5.3448,
+      "step": 12700
+    },
+    {
+      "epoch": 12.075489502241094,
+      "grad_norm": 0.991277277469635,
+      "learning_rate": 9.976154028306461e-05,
+      "loss": 5.3463,
+      "step": 12800
+    },
+    {
+      "epoch": 12.169851380042463,
+      "grad_norm": 0.841627299785614,
+      "learning_rate": 9.974421178764515e-05,
+      "loss": 5.3306,
+      "step": 12900
+    },
+    {
+      "epoch": 12.264213257843831,
+      "grad_norm": 0.886860191822052,
+      "learning_rate": 9.972627717470337e-05,
+      "loss": 5.3562,
+      "step": 13000
+    },
+    {
+      "epoch": 12.264213257843831,
+      "eval_loss": 5.368251800537109,
+      "eval_runtime": 74.4164,
+      "eval_samples_per_second": 202.522,
+      "eval_steps_per_second": 6.329,
+      "step": 13000
+    },
+    {
+      "epoch": 12.3585751356452,
+      "grad_norm": 0.9396846294403076,
+      "learning_rate": 9.970773666276686e-05,
+      "loss": 5.3461,
+      "step": 13100
+    },
+    {
+      "epoch": 12.452937013446567,
+      "grad_norm": 0.9581524729728699,
+      "learning_rate": 9.968859047774595e-05,
+      "loss": 5.3421,
+      "step": 13200
+    },
+    {
+      "epoch": 12.547298891247936,
+      "grad_norm": 0.9131600856781006,
+      "learning_rate": 9.966883885293081e-05,
+      "loss": 5.3472,
+      "step": 13300
+    },
+    {
+      "epoch": 12.641660769049304,
+      "grad_norm": 0.9555962085723877,
+      "learning_rate": 9.964848202898879e-05,
+      "loss": 5.3453,
+      "step": 13400
+    },
+    {
+      "epoch": 12.736022646850673,
+      "grad_norm": 0.8874040842056274,
+      "learning_rate": 9.962752025396133e-05,
+      "loss": 5.3367,
+      "step": 13500
+    },
+    {
+      "epoch": 12.830384524652041,
+      "grad_norm": 1.0667670965194702,
+      "learning_rate": 9.96059537832611e-05,
+      "loss": 5.3221,
+      "step": 13600
+    },
+    {
+      "epoch": 12.924746402453408,
+      "grad_norm": 1.00221848487854,
+      "learning_rate": 9.958378287966868e-05,
+      "loss": 5.3402,
+      "step": 13700
+    },
+    {
+      "epoch": 13.018872375560274,
+      "grad_norm": 1.027253270149231,
+      "learning_rate": 9.956100781332958e-05,
+      "loss": 5.3324,
+      "step": 13800
+    },
+    {
+      "epoch": 13.113234253361641,
+      "grad_norm": 1.002406120300293,
+      "learning_rate": 9.953762886175075e-05,
+      "loss": 5.326,
+      "step": 13900
+    },
+    {
+      "epoch": 13.20759613116301,
+      "grad_norm": 0.8681198954582214,
+      "learning_rate": 9.951364630979738e-05,
+      "loss": 5.3311,
+      "step": 14000
+    },
+    {
+      "epoch": 13.20759613116301,
+      "eval_loss": 5.359684944152832,
+      "eval_runtime": 74.4319,
+      "eval_samples_per_second": 202.48,
+      "eval_steps_per_second": 6.328,
+      "step": 14000
+    },
+    {
+      "epoch": 13.301958008964379,
+      "grad_norm": 0.8745918273925781,
+      "learning_rate": 9.948906044968926e-05,
+      "loss": 5.3234,
+      "step": 14100
+    },
+    {
+      "epoch": 13.396319886765747,
+      "grad_norm": 0.8706562519073486,
+      "learning_rate": 9.946387158099738e-05,
+      "loss": 5.327,
+      "step": 14200
+    },
+    {
+      "epoch": 13.490681764567114,
+      "grad_norm": 0.9299585819244385,
+      "learning_rate": 9.943808001064013e-05,
+      "loss": 5.3347,
+      "step": 14300
+    },
+    {
+      "epoch": 13.585043642368483,
+      "grad_norm": 0.9840534925460815,
+      "learning_rate": 9.941168605287965e-05,
+      "loss": 5.3222,
+      "step": 14400
+    },
+    {
+      "epoch": 13.679405520169851,
+      "grad_norm": 0.8881776332855225,
+      "learning_rate": 9.938469002931798e-05,
+      "loss": 5.3211,
+      "step": 14500
+    },
+    {
+      "epoch": 13.77376739797122,
+      "grad_norm": 1.0282058715820312,
+      "learning_rate": 9.935709226889319e-05,
+      "loss": 5.3328,
+      "step": 14600
+    },
+    {
+      "epoch": 13.868129275772588,
+      "grad_norm": 0.9646194577217102,
+      "learning_rate": 9.932889310787522e-05,
+      "loss": 5.3127,
+      "step": 14700
+    },
+    {
+      "epoch": 13.962491153573955,
+      "grad_norm": 0.9002189636230469,
+      "learning_rate": 9.9300092889862e-05,
+      "loss": 5.3208,
+      "step": 14800
+    },
+    {
+      "epoch": 14.056617126680822,
+      "grad_norm": 0.9314839839935303,
+      "learning_rate": 9.927069196577507e-05,
+      "loss": 5.3141,
+      "step": 14900
+    },
+    {
+      "epoch": 14.150979004482188,
+      "grad_norm": 0.9567928314208984,
+      "learning_rate": 9.924069069385543e-05,
+      "loss": 5.3125,
+      "step": 15000
+    },
+    {
+      "epoch": 14.150979004482188,
+      "eval_loss": 5.350492477416992,
+      "eval_runtime": 74.4863,
+      "eval_samples_per_second": 202.333,
+      "eval_steps_per_second": 6.323,
+      "step": 15000
+    },
+    {
+      "epoch": 14.245340882283557,
+      "grad_norm": 0.9932417273521423,
+      "learning_rate": 9.921008943965908e-05,
+      "loss": 5.3151,
+      "step": 15100
+    },
+    {
+      "epoch": 14.339702760084926,
+      "grad_norm": 1.0583364963531494,
+      "learning_rate": 9.917888857605268e-05,
+      "loss": 5.3138,
+      "step": 15200
+    },
+    {
+      "epoch": 14.434064637886294,
+      "grad_norm": 0.9595634341239929,
+      "learning_rate": 9.91470884832089e-05,
+      "loss": 5.2995,
+      "step": 15300
+    },
+    {
+      "epoch": 14.528426515687663,
+      "grad_norm": 1.0111640691757202,
+      "learning_rate": 9.911468954860181e-05,
+      "loss": 5.3113,
+      "step": 15400
+    },
+    {
+      "epoch": 14.62278839348903,
+      "grad_norm": 0.872369647026062,
+      "learning_rate": 9.908169216700223e-05,
+      "loss": 5.3142,
+      "step": 15500
+    },
+    {
+      "epoch": 14.717150271290398,
+      "grad_norm": 1.0529162883758545,
+      "learning_rate": 9.904809674047284e-05,
+      "loss": 5.325,
+      "step": 15600
+    },
+    {
+      "epoch": 14.811512149091767,
+      "grad_norm": 0.9173269271850586,
+      "learning_rate": 9.90139036783633e-05,
+      "loss": 5.307,
+      "step": 15700
+    },
+    {
+      "epoch": 14.905874026893136,
+      "grad_norm": 0.8593319058418274,
+      "learning_rate": 9.897911339730527e-05,
+      "loss": 5.3098,
+      "step": 15800
+    },
+    {
+      "epoch": 15.0,
+      "grad_norm": 1.0840786695480347,
+      "learning_rate": 9.894372632120738e-05,
+      "loss": 5.3069,
+      "step": 15900
+    },
+    {
+      "epoch": 15.094361877801369,
+      "grad_norm": 0.9976385235786438,
+      "learning_rate": 9.890774288124996e-05,
+      "loss": 5.301,
+      "step": 16000
+    },
+    {
+      "epoch": 15.094361877801369,
+      "eval_loss": 5.339163303375244,
+      "eval_runtime": 74.4349,
+      "eval_samples_per_second": 202.472,
+      "eval_steps_per_second": 6.328,
+      "step": 16000
+    },
+    {
+      "epoch": 15.188723755602737,
+      "grad_norm": 0.8718565702438354,
+      "learning_rate": 9.887116351587985e-05,
+      "loss": 5.3039,
+      "step": 16100
+    },
+    {
+      "epoch": 15.283085633404104,
+      "grad_norm": 0.942323625087738,
+      "learning_rate": 9.883398867080513e-05,
+      "loss": 5.2966,
+      "step": 16200
+    },
+    {
+      "epoch": 15.377447511205473,
+      "grad_norm": 0.9328644871711731,
+      "learning_rate": 9.87962187989895e-05,
+      "loss": 5.2976,
+      "step": 16300
+    },
+    {
+      "epoch": 15.471809389006841,
+      "grad_norm": 0.9764389395713806,
+      "learning_rate": 9.875785436064697e-05,
+      "loss": 5.3005,
+      "step": 16400
+    },
+    {
+      "epoch": 15.56617126680821,
+      "grad_norm": 0.951077401638031,
+      "learning_rate": 9.871889582323609e-05,
+      "loss": 5.2958,
+      "step": 16500
+    },
+    {
+      "epoch": 15.660533144609577,
+      "grad_norm": 0.8876622915267944,
+      "learning_rate": 9.867934366145435e-05,
+      "loss": 5.3006,
+      "step": 16600
+    },
+    {
+      "epoch": 15.754895022410945,
+      "grad_norm": 1.1068483591079712,
+      "learning_rate": 9.863919835723236e-05,
+      "loss": 5.3,
+      "step": 16700
+    },
+    {
+      "epoch": 15.849256900212314,
+      "grad_norm": 0.9473031163215637,
+      "learning_rate": 9.859846039972798e-05,
+      "loss": 5.2966,
+      "step": 16800
+    },
+    {
+      "epoch": 15.943618778013683,
+      "grad_norm": 1.0127873420715332,
+      "learning_rate": 9.855713028532036e-05,
+      "loss": 5.2963,
+      "step": 16900
+    },
+    {
+      "epoch": 16.03774475112055,
+      "grad_norm": 0.8792197108268738,
+      "learning_rate": 9.851520851760394e-05,
+      "loss": 5.2898,
+      "step": 17000
+    },
+    {
+      "epoch": 16.03774475112055,
+      "eval_loss": 5.332190036773682,
+      "eval_runtime": 74.4273,
+      "eval_samples_per_second": 202.493,
+      "eval_steps_per_second": 6.328,
+      "step": 17000
+    },
+    {
+      "epoch": 16.132106628921914,
+      "grad_norm": 0.9135002493858337,
+      "learning_rate": 9.847269560738218e-05,
+      "loss": 5.2841,
+      "step": 17100
+    },
+    {
+      "epoch": 16.226468506723283,
+      "grad_norm": 0.8058662414550781,
+      "learning_rate": 9.842959207266149e-05,
+      "loss": 5.2849,
+      "step": 17200
+    },
+    {
+      "epoch": 16.32083038452465,
+      "grad_norm": 0.919384777545929,
+      "learning_rate": 9.838589843864484e-05,
+      "loss": 5.2938,
+      "step": 17300
+    },
+    {
+      "epoch": 16.41519226232602,
+      "grad_norm": 0.9282925128936768,
+      "learning_rate": 9.834161523772539e-05,
+      "loss": 5.2806,
+      "step": 17400
+    },
+    {
+      "epoch": 16.50955414012739,
+      "grad_norm": 0.8830996155738831,
+      "learning_rate": 9.829674300947993e-05,
+      "loss": 5.2875,
+      "step": 17500
+    },
+    {
+      "epoch": 16.603916017928757,
+      "grad_norm": 0.8510077595710754,
+      "learning_rate": 9.825128230066244e-05,
+      "loss": 5.296,
+      "step": 17600
+    },
+    {
+      "epoch": 16.698277895730126,
+      "grad_norm": 0.9782845377922058,
+      "learning_rate": 9.82052336651973e-05,
+      "loss": 5.2783,
+      "step": 17700
+    },
+    {
+      "epoch": 16.792639773531494,
+      "grad_norm": 0.9722960591316223,
+      "learning_rate": 9.815859766417257e-05,
+      "loss": 5.2933,
+      "step": 17800
+    },
+    {
+      "epoch": 16.887001651332863,
+      "grad_norm": 0.9551956653594971,
+      "learning_rate": 9.811137486583324e-05,
+      "loss": 5.2802,
+      "step": 17900
+    },
+    {
+      "epoch": 16.981363529134228,
+      "grad_norm": 0.9742000699043274,
+      "learning_rate": 9.806356584557419e-05,
+      "loss": 5.296,
+      "step": 18000
+    },
+    {
+      "epoch": 16.981363529134228,
+      "eval_loss": 5.326197147369385,
+      "eval_runtime": 74.3924,
+      "eval_samples_per_second": 202.588,
+      "eval_steps_per_second": 6.331,
+      "step": 18000
+    },
+    {
+      "epoch": 17.075489502241094,
+      "grad_norm": 0.8250226974487305,
+      "learning_rate": 9.801517118593327e-05,
+      "loss": 5.283,
+      "step": 18100
+    },
+    {
+      "epoch": 17.169851380042463,
+      "grad_norm": 0.9234561324119568,
+      "learning_rate": 9.796619147658408e-05,
+      "loss": 5.2766,
+      "step": 18200
+    },
+    {
+      "epoch": 17.26421325784383,
+      "grad_norm": 0.920927107334137,
+      "learning_rate": 9.791662731432898e-05,
+      "loss": 5.2737,
+      "step": 18300
+    },
+    {
+      "epoch": 17.3585751356452,
+      "grad_norm": 0.9160899519920349,
+      "learning_rate": 9.78664793030916e-05,
+      "loss": 5.2762,
+      "step": 18400
+    },
+    {
+      "epoch": 17.45293701344657,
+      "grad_norm": 0.9943878054618835,
+      "learning_rate": 9.781574805390967e-05,
+      "loss": 5.2694,
+      "step": 18500
+    },
+    {
+      "epoch": 17.547298891247937,
+      "grad_norm": 0.9411502480506897,
+      "learning_rate": 9.776443418492744e-05,
+      "loss": 5.2783,
+      "step": 18600
+    },
+    {
+      "epoch": 17.641660769049302,
+      "grad_norm": 0.9028970003128052,
+      "learning_rate": 9.771253832138819e-05,
+      "loss": 5.2812,
+      "step": 18700
+    },
+    {
+      "epoch": 17.73602264685067,
+      "grad_norm": 0.930030345916748,
+      "learning_rate": 9.766006109562664e-05,
+      "loss": 5.279,
+      "step": 18800
+    },
+    {
+      "epoch": 17.83038452465204,
+      "grad_norm": 1.0116451978683472,
+      "learning_rate": 9.760700314706125e-05,
+      "loss": 5.2862,
+      "step": 18900
+    },
+    {
+      "epoch": 17.924746402453408,
+      "grad_norm": 0.8852808475494385,
+      "learning_rate": 9.755336512218638e-05,
+      "loss": 5.2797,
+      "step": 19000
+    },
+    {
+      "epoch": 17.924746402453408,
+      "eval_loss": 5.320762634277344,
+      "eval_runtime": 74.4346,
+      "eval_samples_per_second": 202.473,
+      "eval_steps_per_second": 6.328,
+      "step": 19000
+    },
+    {
+      "epoch": 18.018872375560274,
+      "grad_norm": 0.9398846626281738,
+      "learning_rate": 9.749914767456441e-05,
+      "loss": 5.2747,
+      "step": 19100
+    },
+    {
+      "epoch": 18.113234253361643,
+      "grad_norm": 0.8749862909317017,
+      "learning_rate": 9.744435146481785e-05,
+      "loss": 5.2593,
+      "step": 19200
+    },
+    {
+      "epoch": 18.20759613116301,
+      "grad_norm": 0.812777578830719,
+      "learning_rate": 9.738897716062121e-05,
+      "loss": 5.2673,
+      "step": 19300
+    },
+    {
+      "epoch": 18.301958008964377,
+      "grad_norm": 0.9763257503509521,
+      "learning_rate": 9.733302543669291e-05,
+      "loss": 5.2755,
+      "step": 19400
+    },
+    {
+      "epoch": 18.396319886765745,
+      "grad_norm": 0.9894685745239258,
+      "learning_rate": 9.727649697478708e-05,
+      "loss": 5.2701,
+      "step": 19500
+    },
+    {
+      "epoch": 18.490681764567114,
+      "grad_norm": 1.0243197679519653,
+      "learning_rate": 9.721939246368515e-05,
+      "loss": 5.2652,
+      "step": 19600
+    },
+    {
+      "epoch": 18.585043642368483,
+      "grad_norm": 0.9538172483444214,
+      "learning_rate": 9.716171259918758e-05,
+      "loss": 5.2685,
+      "step": 19700
+    },
+    {
+      "epoch": 18.67940552016985,
+      "grad_norm": 1.0511356592178345,
+      "learning_rate": 9.710345808410532e-05,
+      "loss": 5.2751,
+      "step": 19800
+    },
+    {
+      "epoch": 18.77376739797122,
+      "grad_norm": 0.8636530637741089,
+      "learning_rate": 9.704462962825124e-05,
+      "loss": 5.2657,
+      "step": 19900
+    },
+    {
+      "epoch": 18.86812927577259,
+      "grad_norm": 0.820347785949707,
+      "learning_rate": 9.698522794843154e-05,
+      "loss": 5.2805,
+      "step": 20000
+    },
+    {
+      "epoch": 18.86812927577259,
+      "eval_loss": 5.320885181427002,
+      "eval_runtime": 74.4058,
+      "eval_samples_per_second": 202.551,
+      "eval_steps_per_second": 6.33,
+      "step": 20000
+    },
+    {
+      "epoch": 18.962491153573957,
+      "grad_norm": 0.9779840707778931,
+      "learning_rate": 9.692525376843691e-05,
+      "loss": 5.2646,
+      "step": 20100
+    },
+    {
+      "epoch": 19.05661712668082,
+      "grad_norm": 0.883474588394165,
+      "learning_rate": 9.686470781903383e-05,
+      "loss": 5.2645,
+      "step": 20200
+    },
+    {
+      "epoch": 19.15097900448219,
+      "grad_norm": 1.031437873840332,
+      "learning_rate": 9.680359083795557e-05,
+      "loss": 5.2595,
+      "step": 20300
+    },
+    {
+      "epoch": 19.245340882283557,
+      "grad_norm": 0.8232062458992004,
+      "learning_rate": 9.674190356989325e-05,
+      "loss": 5.2606,
+      "step": 20400
+    },
+    {
+      "epoch": 19.339702760084926,
+      "grad_norm": 0.8686699271202087,
+      "learning_rate": 9.66796467664868e-05,
+      "loss": 5.2575,
+      "step": 20500
+    },
+    {
+      "epoch": 19.434064637886294,
+      "grad_norm": 0.8924238085746765,
+      "learning_rate": 9.661682118631568e-05,
+      "loss": 5.2488,
+      "step": 20600
+    },
+    {
+      "epoch": 19.528426515687663,
+      "grad_norm": 0.9776726365089417,
+      "learning_rate": 9.655342759488979e-05,
+      "loss": 5.264,
+      "step": 20700
+    },
+    {
+      "epoch": 19.62278839348903,
+      "grad_norm": 0.9295623302459717,
+      "learning_rate": 9.648946676464002e-05,
+      "loss": 5.2589,
+      "step": 20800
+    },
+    {
+      "epoch": 19.7171502712904,
+      "grad_norm": 0.9422546625137329,
+      "learning_rate": 9.642493947490889e-05,
+      "loss": 5.2616,
+      "step": 20900
+    },
+    {
+      "epoch": 19.81151214909177,
+      "grad_norm": 1.0146008729934692,
+      "learning_rate": 9.635984651194109e-05,
+      "loss": 5.2631,
+      "step": 21000
+    },
+    {
+      "epoch": 19.81151214909177,
+      "eval_loss": 5.316196918487549,
+      "eval_runtime": 74.4315,
+      "eval_samples_per_second": 202.481,
+      "eval_steps_per_second": 6.328,
+      "step": 21000
+    },
+    {
+      "epoch": 19.905874026893134,
+      "grad_norm": 0.8792014718055725,
+      "learning_rate": 9.629418866887381e-05,
+      "loss": 5.2685,
+      "step": 21100
+    },
+    {
+      "epoch": 20.0,
+      "grad_norm": 1.1166847944259644,
+      "learning_rate": 9.622796674572716e-05,
+      "loss": 5.2622,
+      "step": 21200
+    },
+    {
+      "epoch": 20.09436187780137,
+      "grad_norm": 0.888293445110321,
+      "learning_rate": 9.616118154939436e-05,
+      "loss": 5.2506,
+      "step": 21300
+    },
+    {
+      "epoch": 20.188723755602737,
+      "grad_norm": 0.9711266756057739,
+      "learning_rate": 9.609383389363198e-05,
+      "loss": 5.2503,
+      "step": 21400
+    },
+    {
+      "epoch": 20.283085633404106,
+      "grad_norm": 0.9572346806526184,
+      "learning_rate": 9.602592459904993e-05,
+      "loss": 5.2483,
+      "step": 21500
+    },
+    {
+      "epoch": 20.377447511205474,
+      "grad_norm": 0.8501202464103699,
+      "learning_rate": 9.595745449310152e-05,
+      "loss": 5.2526,
+      "step": 21600
+    },
+    {
+      "epoch": 20.47180938900684,
+      "grad_norm": 0.9165418744087219,
+      "learning_rate": 9.588842441007342e-05,
+      "loss": 5.2609,
+      "step": 21700
+    },
+    {
+      "epoch": 20.566171266808208,
+      "grad_norm": 0.8789685964584351,
+      "learning_rate": 9.581883519107538e-05,
+      "loss": 5.2498,
+      "step": 21800
+    },
+    {
+      "epoch": 20.660533144609577,
+      "grad_norm": 0.9526987671852112,
+      "learning_rate": 9.574868768403007e-05,
+      "loss": 5.2476,
+      "step": 21900
+    },
+    {
+      "epoch": 20.754895022410945,
+      "grad_norm": 0.8855116963386536,
+      "learning_rate": 9.567798274366273e-05,
+      "loss": 5.2582,
+      "step": 22000
+    },
+    {
+      "epoch": 20.754895022410945,
+      "eval_loss": 5.307859897613525,
+      "eval_runtime": 74.4524,
+      "eval_samples_per_second": 202.425,
+      "eval_steps_per_second": 6.326,
+      "step": 22000
+    },
+    {
+      "epoch": 20.849256900212314,
+      "grad_norm": 0.821835994720459,
+      "learning_rate": 9.560672123149077e-05,
+      "loss": 5.2586,
+      "step": 22100
+    },
+    {
+      "epoch": 20.943618778013683,
+      "grad_norm": 1.0579211711883545,
+      "learning_rate": 9.55349040158132e-05,
+      "loss": 5.2541,
+      "step": 22200
+    },
+    {
+      "epoch": 21.03774475112055,
+      "grad_norm": 1.004662275314331,
+      "learning_rate": 9.546253197170015e-05,
+      "loss": 5.2489,
+      "step": 22300
+    },
+    {
+      "epoch": 21.132106628921914,
+      "grad_norm": 0.9038196802139282,
+      "learning_rate": 9.538960598098211e-05,
+      "loss": 5.2426,
+      "step": 22400
+    },
+    {
+      "epoch": 21.226468506723283,
+      "grad_norm": 0.8407565951347351,
+      "learning_rate": 9.531612693223928e-05,
+      "loss": 5.2479,
+      "step": 22500
+    },
+    {
+      "epoch": 21.32083038452465,
+      "grad_norm": 1.034454107284546,
+      "learning_rate": 9.524209572079068e-05,
+      "loss": 5.2437,
+      "step": 22600
+    },
+    {
+      "epoch": 21.41519226232602,
+      "grad_norm": 0.9391712546348572,
+      "learning_rate": 9.516751324868326e-05,
+      "loss": 5.2527,
+      "step": 22700
+    },
+    {
+      "epoch": 21.50955414012739,
+      "grad_norm": 0.9005853533744812,
+      "learning_rate": 9.509238042468091e-05,
+      "loss": 5.2437,
+      "step": 22800
+    },
+    {
+      "epoch": 21.603916017928757,
+      "grad_norm": 0.8823351860046387,
+      "learning_rate": 9.501669816425337e-05,
+      "loss": 5.2555,
+      "step": 22900
+    },
+    {
+      "epoch": 21.698277895730126,
+      "grad_norm": 0.9329714179039001,
+      "learning_rate": 9.494046738956508e-05,
+      "loss": 5.237,
+      "step": 23000
+    },
+    {
+      "epoch": 21.698277895730126,
+      "eval_loss": 5.306467056274414,
+      "eval_runtime": 74.4426,
+      "eval_samples_per_second": 202.451,
+      "eval_steps_per_second": 6.327,
+      "step": 23000
+    },
+    {
+      "epoch": 21.792639773531494,
+      "grad_norm": 0.8710659742355347,
+      "learning_rate": 9.486368902946402e-05,
+      "loss": 5.2476,
+      "step": 23100
+    },
+    {
+      "epoch": 21.887001651332863,
+      "grad_norm": 0.9512269496917725,
+      "learning_rate": 9.478636401947026e-05,
+      "loss": 5.2428,
+      "step": 23200
+    },
+    {
+      "epoch": 21.981363529134228,
+      "grad_norm": 1.0027644634246826,
+      "learning_rate": 9.47084933017646e-05,
+      "loss": 5.2451,
+      "step": 23300
+    },
+    {
+      "epoch": 22.075489502241094,
+      "grad_norm": 0.9528675079345703,
+      "learning_rate": 9.463007782517723e-05,
+      "loss": 5.2369,
+      "step": 23400
+    },
+    {
+      "epoch": 22.169851380042463,
+      "grad_norm": 0.9792280197143555,
+      "learning_rate": 9.455111854517595e-05,
+      "loss": 5.229,
+      "step": 23500
+    },
+    {
+      "epoch": 22.26421325784383,
+      "grad_norm": 0.8722760081291199,
+      "learning_rate": 9.447161642385467e-05,
+      "loss": 5.2364,
+      "step": 23600
+    },
+    {
+      "epoch": 22.3585751356452,
+      "grad_norm": 0.9419755935668945,
+      "learning_rate": 9.439157242992164e-05,
+      "loss": 5.2363,
+      "step": 23700
+    },
+    {
+      "epoch": 22.45293701344657,
+      "grad_norm": 0.8856399059295654,
+      "learning_rate": 9.43109875386877e-05,
+      "loss": 5.2399,
+      "step": 23800
+    },
+    {
+      "epoch": 22.547298891247937,
+      "grad_norm": 0.9981026649475098,
+      "learning_rate": 9.422986273205429e-05,
+      "loss": 5.2433,
+      "step": 23900
+    },
+    {
+      "epoch": 22.641660769049302,
+      "grad_norm": 0.9505060911178589,
+      "learning_rate": 9.414819899850158e-05,
+      "loss": 5.2481,
+      "step": 24000
+    },
+    {
+      "epoch": 22.641660769049302,
+      "eval_loss": 5.302011489868164,
+      "eval_runtime": 74.4717,
+      "eval_samples_per_second": 202.372,
+      "eval_steps_per_second": 6.325,
+      "step": 24000
+    },
+    {
+      "epoch": 22.73602264685067,
+      "grad_norm": 0.8508577942848206,
+      "learning_rate": 9.40659973330764e-05,
+      "loss": 5.2348,
+      "step": 24100
+    },
+    {
+      "epoch": 22.83038452465204,
+      "grad_norm": 0.9357001781463623,
+      "learning_rate": 9.398325873738007e-05,
+      "loss": 5.2411,
+      "step": 24200
+    },
+    {
+      "epoch": 22.924746402453408,
+      "grad_norm": 1.0099997520446777,
+      "learning_rate": 9.389998421955632e-05,
+      "loss": 5.2383,
+      "step": 24300
+    },
+    {
+      "epoch": 23.018872375560274,
+      "grad_norm": 0.9384352564811707,
+      "learning_rate": 9.381617479427885e-05,
+      "loss": 5.2366,
+      "step": 24400
+    },
+    {
+      "epoch": 23.113234253361643,
+      "grad_norm": 1.0836660861968994,
+      "learning_rate": 9.373183148273905e-05,
+      "loss": 5.2278,
+      "step": 24500
+    },
+    {
+      "epoch": 23.20759613116301,
+      "grad_norm": 0.9002297520637512,
+      "learning_rate": 9.364695531263354e-05,
+      "loss": 5.229,
+      "step": 24600
+    },
+    {
+      "epoch": 23.301958008964377,
+      "grad_norm": 0.9582403302192688,
+      "learning_rate": 9.35615473181517e-05,
+      "loss": 5.2378,
+      "step": 24700
+    },
+    {
+      "epoch": 23.396319886765745,
+      "grad_norm": 0.9208775162696838,
+      "learning_rate": 9.347560853996298e-05,
+      "loss": 5.2292,
+      "step": 24800
+    },
+    {
+      "epoch": 23.490681764567114,
+      "grad_norm": 0.9435613751411438,
+      "learning_rate": 9.338914002520426e-05,
+      "loss": 5.2313,
+      "step": 24900
+    },
+    {
+      "epoch": 23.585043642368483,
+      "grad_norm": 0.8621221780776978,
+      "learning_rate": 9.330214282746712e-05,
+      "loss": 5.2405,
+      "step": 25000
+    },
+    {
+      "epoch": 23.585043642368483,
+      "eval_loss": 5.301399230957031,
+      "eval_runtime": 74.4481,
+      "eval_samples_per_second": 202.436,
+      "eval_steps_per_second": 6.327,
+      "step": 25000
+    },
+    {
+      "epoch": 23.67940552016985,
+      "grad_norm": 0.9942018389701843,
+      "learning_rate": 9.321461800678494e-05,
+      "loss": 5.2424,
+      "step": 25100
+    },
+    {
+      "epoch": 23.77376739797122,
+      "grad_norm": 0.8850314021110535,
+      "learning_rate": 9.312656662962004e-05,
+      "loss": 5.2341,
+      "step": 25200
+    },
+    {
+      "epoch": 23.86812927577259,
+      "grad_norm": 0.8494738936424255,
+      "learning_rate": 9.30379897688507e-05,
+      "loss": 5.2369,
+      "step": 25300
+    },
+    {
+      "epoch": 23.962491153573957,
+      "grad_norm": 0.8826471567153931,
+      "learning_rate": 9.294888850375796e-05,
+      "loss": 5.2394,
+      "step": 25400
+    },
+    {
+      "epoch": 24.05661712668082,
+      "grad_norm": 0.9835422039031982,
+      "learning_rate": 9.285926392001265e-05,
+      "loss": 5.2274,
+      "step": 25500
+    },
+    {
+      "epoch": 24.15097900448219,
+      "grad_norm": 0.9036953449249268,
+      "learning_rate": 9.276911710966205e-05,
+      "loss": 5.2244,
+      "step": 25600
+    },
+    {
+      "epoch": 24.245340882283557,
+      "grad_norm": 0.916057288646698,
+      "learning_rate": 9.267844917111657e-05,
+      "loss": 5.2275,
+      "step": 25700
+    },
+    {
+      "epoch": 24.339702760084926,
+      "grad_norm": 0.8915982246398926,
+      "learning_rate": 9.258726120913643e-05,
+      "loss": 5.232,
+      "step": 25800
+    },
+    {
+      "epoch": 24.434064637886294,
+      "grad_norm": 1.0098341703414917,
+      "learning_rate": 9.249555433481819e-05,
+      "loss": 5.2243,
+      "step": 25900
+    },
+    {
+      "epoch": 24.528426515687663,
+      "grad_norm": 0.9247532486915588,
+      "learning_rate": 9.240332966558116e-05,
+      "loss": 5.2289,
+      "step": 26000
+    },
+    {
+      "epoch": 24.528426515687663,
+      "eval_loss": 5.294754505157471,
+      "eval_runtime": 74.4828,
+      "eval_samples_per_second": 202.342,
+      "eval_steps_per_second": 6.324,
+      "step": 26000
+    },
+    {
+      "epoch": 24.62278839348903,
+      "grad_norm": 0.9268798828125,
+      "learning_rate": 9.231058832515383e-05,
+      "loss": 5.2264,
+      "step": 26100
+    },
+    {
+      "epoch": 24.7171502712904,
+      "grad_norm": 0.9089873433113098,
+      "learning_rate": 9.221733144356015e-05,
+      "loss": 5.2231,
+      "step": 26200
+    },
+    {
+      "epoch": 24.81151214909177,
+      "grad_norm": 0.9429314136505127,
+      "learning_rate": 9.212356015710581e-05,
+      "loss": 5.2244,
+      "step": 26300
+    },
+    {
+      "epoch": 24.905874026893134,
+      "grad_norm": 0.9047618508338928,
+      "learning_rate": 9.202927560836436e-05,
+      "loss": 5.2291,
+      "step": 26400
+    },
+    {
+      "epoch": 25.0,
+      "grad_norm": 1.1686391830444336,
+      "learning_rate": 9.193447894616324e-05,
+      "loss": 5.2308,
+      "step": 26500
+    },
+    {
+      "epoch": 25.09436187780137,
+      "grad_norm": 0.9237032532691956,
+      "learning_rate": 9.183917132556987e-05,
+      "loss": 5.2191,
+      "step": 26600
+    },
+    {
+      "epoch": 25.188723755602737,
+      "grad_norm": 0.8873036503791809,
+      "learning_rate": 9.174335390787754e-05,
+      "loss": 5.2141,
+      "step": 26700
+    },
+    {
+      "epoch": 25.283085633404106,
+      "grad_norm": 0.964063286781311,
+      "learning_rate": 9.164702786059125e-05,
+      "loss": 5.2186,
+      "step": 26800
+    },
+    {
+      "epoch": 25.377447511205474,
+      "grad_norm": 0.8931275010108948,
+      "learning_rate": 9.155019435741348e-05,
+      "loss": 5.2149,
+      "step": 26900
+    },
+    {
+      "epoch": 25.47180938900684,
+      "grad_norm": 0.9907437562942505,
+      "learning_rate": 9.14528545782299e-05,
+      "loss": 5.2133,
+      "step": 27000
+    },
+    {
+      "epoch": 25.47180938900684,
+      "eval_loss": 5.286930084228516,
+      "eval_runtime": 74.443,
+      "eval_samples_per_second": 202.45,
+      "eval_steps_per_second": 6.327,
+      "step": 27000
+    },
+    {
+      "epoch": 25.566171266808208,
+      "grad_norm": 0.9528295397758484,
+      "learning_rate": 9.135500970909501e-05,
+      "loss": 5.2296,
+      "step": 27100
+    },
+    {
+      "epoch": 25.660533144609577,
+      "grad_norm": 0.939414918422699,
+      "learning_rate": 9.125666094221766e-05,
+      "loss": 5.2269,
+      "step": 27200
+    },
+    {
+      "epoch": 25.754895022410945,
+      "grad_norm": 0.9641194343566895,
+      "learning_rate": 9.115780947594654e-05,
+      "loss": 5.2203,
+      "step": 27300
+    },
+    {
+      "epoch": 25.849256900212314,
+      "grad_norm": 0.9802380204200745,
+      "learning_rate": 9.105845651475556e-05,
+      "loss": 5.2267,
+      "step": 27400
+    },
+    {
+      "epoch": 25.943618778013683,
+      "grad_norm": 0.8933775424957275,
+      "learning_rate": 9.09586032692292e-05,
+      "loss": 5.2196,
+      "step": 27500
+    },
+    {
+      "epoch": 26.03774475112055,
+      "grad_norm": 0.846247136592865,
+      "learning_rate": 9.085825095604778e-05,
+      "loss": 5.2226,
+      "step": 27600
+    },
+    {
+      "epoch": 26.132106628921914,
+      "grad_norm": 0.9873145222663879,
+      "learning_rate": 9.075740079797253e-05,
+      "loss": 5.2118,
+      "step": 27700
+    },
+    {
+      "epoch": 26.226468506723283,
+      "grad_norm": 0.9664213061332703,
+      "learning_rate": 9.06560540238308e-05,
+      "loss": 5.2127,
+      "step": 27800
+    },
+    {
+      "epoch": 26.32083038452465,
+      "grad_norm": 0.902673065662384,
+      "learning_rate": 9.055421186850104e-05,
+      "loss": 5.222,
+      "step": 27900
+    },
+    {
+      "epoch": 26.41519226232602,
+      "grad_norm": 1.0140800476074219,
+      "learning_rate": 9.045187557289783e-05,
+      "loss": 5.2122,
+      "step": 28000
+    },
+    {
+      "epoch": 26.41519226232602,
+      "eval_loss": 5.291759490966797,
+      "eval_runtime": 74.4092,
+      "eval_samples_per_second": 202.542,
+      "eval_steps_per_second": 6.33,
+      "step": 28000
+    },
+    {
+      "epoch": 26.50955414012739,
+      "grad_norm": 0.9105413556098938,
+      "learning_rate": 9.034904638395656e-05,
+      "loss": 5.2119,
+      "step": 28100
+    },
+    {
+      "epoch": 26.603916017928757,
+      "grad_norm": 0.8379244208335876,
+      "learning_rate": 9.024572555461852e-05,
+      "loss": 5.2099,
+      "step": 28200
+    },
+    {
+      "epoch": 26.698277895730126,
+      "grad_norm": 0.9082247614860535,
+      "learning_rate": 9.014191434381535e-05,
+      "loss": 5.2044,
+      "step": 28300
+    },
+    {
+      "epoch": 26.792639773531494,
+      "grad_norm": 0.9054832458496094,
+      "learning_rate": 9.003761401645393e-05,
+      "loss": 5.2102,
+      "step": 28400
+    },
+    {
+      "epoch": 26.887001651332863,
+      "grad_norm": 0.8595414757728577,
+      "learning_rate": 8.99328258434008e-05,
+      "loss": 5.2244,
+      "step": 28500
+    },
+    {
+      "epoch": 26.981363529134228,
+      "grad_norm": 0.9551483988761902,
+      "learning_rate": 8.982755110146681e-05,
+      "loss": 5.2267,
+      "step": 28600
+    },
+    {
+      "epoch": 27.075489502241094,
+      "grad_norm": 0.9670704007148743,
+      "learning_rate": 8.972179107339148e-05,
+      "loss": 5.2109,
+      "step": 28700
+    },
+    {
+      "epoch": 27.169851380042463,
+      "grad_norm": 0.9536374807357788,
+      "learning_rate": 8.961554704782731e-05,
+      "loss": 5.197,
+      "step": 28800
+    },
+    {
+      "epoch": 27.26421325784383,
+      "grad_norm": 0.9553530812263489,
+      "learning_rate": 8.95088203193243e-05,
+      "loss": 5.2085,
+      "step": 28900
+    },
+    {
+      "epoch": 27.3585751356452,
+      "grad_norm": 0.9752506017684937,
+      "learning_rate": 8.940161218831391e-05,
+      "loss": 5.2078,
+      "step": 29000
+    },
+    {
+      "epoch": 27.3585751356452,
+      "eval_loss": 5.286397457122803,
+      "eval_runtime": 74.4089,
+      "eval_samples_per_second": 202.543,
+      "eval_steps_per_second": 6.33,
+      "step": 29000
+    },
+    {
+      "epoch": 27.45293701344657,
+      "grad_norm": 0.9288462400436401,
+      "learning_rate": 8.929392396109341e-05,
+      "loss": 5.2063,
+      "step": 29100
+    },
+    {
+      "epoch": 27.547298891247937,
+      "grad_norm": 0.9132038950920105,
+      "learning_rate": 8.918575694980983e-05,
+      "loss": 5.217,
+      "step": 29200
+    },
+    {
+      "epoch": 27.641660769049302,
+      "grad_norm": 1.0119094848632812,
+      "learning_rate": 8.90771124724441e-05,
+      "loss": 5.2139,
+      "step": 29300
+    },
+    {
+      "epoch": 27.73602264685067,
+      "grad_norm": 0.9567475318908691,
+      "learning_rate": 8.896799185279487e-05,
+      "loss": 5.2077,
+      "step": 29400
+    },
+    {
+      "epoch": 27.83038452465204,
+      "grad_norm": 0.8996191620826721,
+      "learning_rate": 8.885839642046249e-05,
+      "loss": 5.2139,
+      "step": 29500
+    },
+    {
+      "epoch": 27.924746402453408,
+      "grad_norm": 0.942906379699707,
+      "learning_rate": 8.874832751083266e-05,
+      "loss": 5.219,
+      "step": 29600
+    },
+    {
+      "epoch": 28.018872375560274,
+      "grad_norm": 0.8692095279693604,
+      "learning_rate": 8.863778646506035e-05,
+      "loss": 5.2142,
+      "step": 29700
+    },
+    {
+      "epoch": 28.113234253361643,
+      "grad_norm": 0.9690754413604736,
+      "learning_rate": 8.85267746300533e-05,
+      "loss": 5.2075,
+      "step": 29800
+    },
+    {
+      "epoch": 28.20759613116301,
+      "grad_norm": 0.9451592564582825,
+      "learning_rate": 8.841529335845569e-05,
+      "loss": 5.2045,
+      "step": 29900
+    },
+    {
+      "epoch": 28.301958008964377,
+      "grad_norm": 0.8922234177589417,
+      "learning_rate": 8.830334400863164e-05,
+      "loss": 5.2029,
+      "step": 30000
+    },
+    {
+      "epoch": 28.301958008964377,
+      "eval_loss": 5.287634372711182,
+      "eval_runtime": 74.4172,
+      "eval_samples_per_second": 202.52,
+      "eval_steps_per_second": 6.329,
+      "step": 30000
+    },
+    {
+      "epoch": 28.396319886765745,
+      "grad_norm": 0.9044151306152344,
+      "learning_rate": 8.819092794464863e-05,
+      "loss": 5.2027,
+      "step": 30100
+    },
+    {
+      "epoch": 28.490681764567114,
+      "grad_norm": 0.8109731078147888,
+      "learning_rate": 8.807804653626095e-05,
+      "loss": 5.2031,
+      "step": 30200
+    },
+    {
+      "epoch": 28.585043642368483,
+      "grad_norm": 0.9353645443916321,
+      "learning_rate": 8.796470115889292e-05,
+      "loss": 5.208,
+      "step": 30300
+    },
+    {
+      "epoch": 28.67940552016985,
+      "grad_norm": 0.956208348274231,
+      "learning_rate": 8.785089319362221e-05,
+      "loss": 5.2047,
+      "step": 30400
+    },
+    {
+      "epoch": 28.77376739797122,
+      "grad_norm": 0.799325704574585,
+      "learning_rate": 8.773662402716294e-05,
+      "loss": 5.2049,
+      "step": 30500
+    },
+    {
+      "epoch": 28.86812927577259,
+      "grad_norm": 0.8981509804725647,
+      "learning_rate": 8.762189505184885e-05,
+      "loss": 5.2081,
+      "step": 30600
+    },
+    {
+      "epoch": 28.962491153573957,
+      "grad_norm": 0.9017074704170227,
+      "learning_rate": 8.75067076656163e-05,
+      "loss": 5.1975,
+      "step": 30700
+    },
+    {
+      "epoch": 29.05661712668082,
+      "grad_norm": 0.888663649559021,
+      "learning_rate": 8.739106327198724e-05,
+      "loss": 5.1941,
+      "step": 30800
+    },
+    {
+      "epoch": 29.15097900448219,
+      "grad_norm": 0.900802731513977,
+      "learning_rate": 8.727496328005211e-05,
+      "loss": 5.192,
+      "step": 30900
+    },
+    {
+      "epoch": 29.245340882283557,
+      "grad_norm": 0.9100425839424133,
+      "learning_rate": 8.715840910445267e-05,
+      "loss": 5.2038,
+      "step": 31000
+    },
+    {
+      "epoch": 29.245340882283557,
+      "eval_loss": 5.285486221313477,
+      "eval_runtime": 74.4352,
+      "eval_samples_per_second": 202.471,
+      "eval_steps_per_second": 6.328,
+      "step": 31000
+    },
+    {
+      "epoch": 29.339702760084926,
+      "grad_norm": 0.8852345943450928,
+      "learning_rate": 8.704140216536478e-05,
+      "loss": 5.2091,
+      "step": 31100
+    },
+    {
+      "epoch": 29.434064637886294,
+      "grad_norm": 0.9929203391075134,
+      "learning_rate": 8.692394388848107e-05,
+      "loss": 5.2047,
+      "step": 31200
+    },
+    {
+      "epoch": 29.528426515687663,
+      "grad_norm": 0.9344567656517029,
+      "learning_rate": 8.680603570499354e-05,
+      "loss": 5.1932,
+      "step": 31300
+    },
+    {
+      "epoch": 29.62278839348903,
+      "grad_norm": 0.8371131420135498,
+      "learning_rate": 8.668767905157625e-05,
+      "loss": 5.199,
+      "step": 31400
+    },
+    {
+      "epoch": 29.7171502712904,
+      "grad_norm": 0.9597654342651367,
+      "learning_rate": 8.656887537036762e-05,
+      "loss": 5.1964,
+      "step": 31500
+    },
+    {
+      "epoch": 29.81151214909177,
+      "grad_norm": 0.9672077894210815,
+      "learning_rate": 8.644962610895303e-05,
+      "loss": 5.2033,
+      "step": 31600
+    },
+    {
+      "epoch": 29.905874026893134,
+      "grad_norm": 0.9107534289360046,
+      "learning_rate": 8.63299327203471e-05,
+      "loss": 5.198,
+      "step": 31700
+    },
+    {
+      "epoch": 30.0,
+      "grad_norm": 1.1500047445297241,
+      "learning_rate": 8.620979666297603e-05,
+      "loss": 5.2054,
+      "step": 31800
+    },
+    {
+      "epoch": 30.09436187780137,
+      "grad_norm": 0.8798314929008484,
+      "learning_rate": 8.608921940065973e-05,
+      "loss": 5.1939,
+      "step": 31900
+    },
+    {
+      "epoch": 30.188723755602737,
+      "grad_norm": 0.9284983277320862,
+      "learning_rate": 8.596820240259408e-05,
+      "loss": 5.1908,
+      "step": 32000
+    },
+    {
+      "epoch": 30.188723755602737,
+      "eval_loss": 5.2796783447265625,
+      "eval_runtime": 74.4825,
+      "eval_samples_per_second": 202.343,
+      "eval_steps_per_second": 6.324,
+      "step": 32000
+    },
+    {
+      "epoch": 30.283085633404106,
+      "grad_norm": 0.9129772782325745,
+      "learning_rate": 8.584674714333303e-05,
+      "loss": 5.1838,
+      "step": 32100
+    },
+    {
+      "epoch": 30.377447511205474,
+      "grad_norm": 0.902010977268219,
+      "learning_rate": 8.57248551027706e-05,
+      "loss": 5.197,
+      "step": 32200
+    },
+    {
+      "epoch": 30.47180938900684,
+      "grad_norm": 0.877648651599884,
+      "learning_rate": 8.56025277661228e-05,
+      "loss": 5.1797,
+      "step": 32300
+    },
+    {
+      "epoch": 30.566171266808208,
+      "grad_norm": 0.9197627305984497,
+      "learning_rate": 8.547976662390964e-05,
+      "loss": 5.1965,
+      "step": 32400
+    },
+    {
+      "epoch": 30.660533144609577,
+      "grad_norm": 0.9505943655967712,
+      "learning_rate": 8.535657317193692e-05,
+      "loss": 5.2051,
+      "step": 32500
+    },
+    {
+      "epoch": 30.754895022410945,
+      "grad_norm": 0.9619107246398926,
+      "learning_rate": 8.523294891127794e-05,
+      "loss": 5.1884,
+      "step": 32600
+    },
+    {
+      "epoch": 30.849256900212314,
+      "grad_norm": 0.9347026944160461,
+      "learning_rate": 8.510889534825532e-05,
+      "loss": 5.1951,
+      "step": 32700
+    },
+    {
+      "epoch": 30.943618778013683,
+      "grad_norm": 0.9318591952323914,
+      "learning_rate": 8.498441399442258e-05,
+      "loss": 5.2003,
+      "step": 32800
+    },
+    {
+      "epoch": 31.03774475112055,
+      "grad_norm": 0.9306088089942932,
+      "learning_rate": 8.485950636654572e-05,
+      "loss": 5.1937,
+      "step": 32900
+    },
+    {
+      "epoch": 31.132106628921914,
+      "grad_norm": 0.9033887386322021,
+      "learning_rate": 8.473417398658476e-05,
+      "loss": 5.1805,
+      "step": 33000
+    },
+    {
+      "epoch": 31.132106628921914,
+      "eval_loss": 5.279585361480713,
+      "eval_runtime": 74.456,
+      "eval_samples_per_second": 202.415,
+      "eval_steps_per_second": 6.326,
+      "step": 33000
+    },
+    {
+      "epoch": 31.226468506723283,
+      "grad_norm": 0.873508095741272,
+      "learning_rate": 8.460841838167523e-05,
+      "loss": 5.1898,
+      "step": 33100
+    },
+    {
+      "epoch": 31.32083038452465,
+      "grad_norm": 0.9387214183807373,
+      "learning_rate": 8.448224108410947e-05,
+      "loss": 5.186,
+      "step": 33200
+    },
+    {
+      "epoch": 31.41519226232602,
+      "grad_norm": 0.9126655459403992,
+      "learning_rate": 8.435564363131803e-05,
+      "loss": 5.1895,
+      "step": 33300
+    },
+    {
+      "epoch": 31.50955414012739,
+      "grad_norm": 0.9194200038909912,
+      "learning_rate": 8.422862756585091e-05,
+      "loss": 5.1815,
+      "step": 33400
+    },
+    {
+      "epoch": 31.603916017928757,
+      "grad_norm": 0.9167635440826416,
+      "learning_rate": 8.41011944353588e-05,
+      "loss": 5.1931,
+      "step": 33500
+    },
+    {
+      "epoch": 31.698277895730126,
+      "grad_norm": 0.9253891110420227,
+      "learning_rate": 8.397334579257418e-05,
+      "loss": 5.1931,
+      "step": 33600
+    },
+    {
+      "epoch": 31.792639773531494,
+      "grad_norm": 0.890615701675415,
+      "learning_rate": 8.384508319529242e-05,
+      "loss": 5.1991,
+      "step": 33700
+    },
+    {
+      "epoch": 31.887001651332863,
+      "grad_norm": 0.9657667875289917,
+      "learning_rate": 8.371640820635278e-05,
+      "loss": 5.1965,
+      "step": 33800
+    },
+    {
+      "epoch": 31.981363529134228,
+      "grad_norm": 1.0598945617675781,
+      "learning_rate": 8.358732239361938e-05,
+      "loss": 5.1899,
+      "step": 33900
+    },
+    {
+      "epoch": 32.0754895022411,
+      "grad_norm": 1.0309218168258667,
+      "learning_rate": 8.345782732996215e-05,
+      "loss": 5.1919,
+      "step": 34000
+    },
+    {
+      "epoch": 32.0754895022411,
+      "eval_loss": 5.279961109161377,
+      "eval_runtime": 74.46,
+      "eval_samples_per_second": 202.404,
+      "eval_steps_per_second": 6.326,
+      "step": 34000
+    },
+    {
+      "epoch": 32.169851380042466,
+      "grad_norm": 0.9428073167800903,
+      "learning_rate": 8.332792459323753e-05,
+      "loss": 5.1871,
+      "step": 34100
+    },
+    {
+      "epoch": 32.26421325784383,
+      "grad_norm": 0.8995642066001892,
+      "learning_rate": 8.31976157662694e-05,
+      "loss": 5.1898,
+      "step": 34200
+    },
+    {
+      "epoch": 32.3585751356452,
+      "grad_norm": 0.9593998193740845,
+      "learning_rate": 8.30669024368297e-05,
+      "loss": 5.1826,
+      "step": 34300
+    },
+    {
+      "epoch": 32.452937013446565,
+      "grad_norm": 0.9802656769752502,
+      "learning_rate": 8.293578619761906e-05,
+      "loss": 5.1928,
+      "step": 34400
+    },
+    {
+      "epoch": 32.547298891247934,
+      "grad_norm": 0.9958524703979492,
+      "learning_rate": 8.280426864624753e-05,
+      "loss": 5.1782,
+      "step": 34500
+    },
+    {
+      "epoch": 32.6416607690493,
+      "grad_norm": 0.9810296297073364,
+      "learning_rate": 8.267235138521492e-05,
+      "loss": 5.1867,
+      "step": 34600
+    },
+    {
+      "epoch": 32.73602264685067,
+      "grad_norm": 0.9276301264762878,
+      "learning_rate": 8.254003602189146e-05,
+      "loss": 5.1786,
+      "step": 34700
+    },
+    {
+      "epoch": 32.83038452465204,
+      "grad_norm": 0.913356363773346,
+      "learning_rate": 8.240732416849807e-05,
+      "loss": 5.1913,
+      "step": 34800
+    },
+    {
+      "epoch": 32.92474640245341,
+      "grad_norm": 0.9506176114082336,
+      "learning_rate": 8.227421744208683e-05,
+      "loss": 5.1852,
+      "step": 34900
+    },
+    {
+      "epoch": 33.01887237556027,
+      "grad_norm": 0.9290357232093811,
+      "learning_rate": 8.214071746452117e-05,
+      "loss": 5.1822,
+      "step": 35000
+    },
+    {
+      "epoch": 33.01887237556027,
+      "eval_loss": 5.276256084442139,
+      "eval_runtime": 74.4707,
+      "eval_samples_per_second": 202.375,
+      "eval_steps_per_second": 6.325,
+      "step": 35000
+    },
+    {
+      "epoch": 33.11323425336164,
+      "grad_norm": 0.9290668368339539,
+      "learning_rate": 8.200682586245621e-05,
+      "loss": 5.1844,
+      "step": 35100
+    },
+    {
+      "epoch": 33.20759613116301,
+      "grad_norm": 0.9466826319694519,
+      "learning_rate": 8.187254426731884e-05,
+      "loss": 5.1794,
+      "step": 35200
+    },
+    {
+      "epoch": 33.30195800896438,
+      "grad_norm": 1.0010716915130615,
+      "learning_rate": 8.173787431528794e-05,
+      "loss": 5.1817,
+      "step": 35300
+    },
+    {
+      "epoch": 33.396319886765745,
+      "grad_norm": 0.9706283807754517,
+      "learning_rate": 8.160281764727436e-05,
+      "loss": 5.1868,
+      "step": 35400
+    },
+    {
+      "epoch": 33.490681764567114,
+      "grad_norm": 0.9433771371841431,
+      "learning_rate": 8.146737590890101e-05,
+      "loss": 5.1779,
+      "step": 35500
+    },
+    {
+      "epoch": 33.58504364236848,
+      "grad_norm": 0.9575541615486145,
+      "learning_rate": 8.133155075048269e-05,
+      "loss": 5.1879,
+      "step": 35600
+    },
+    {
+      "epoch": 33.67940552016985,
+      "grad_norm": 0.9401003122329712,
+      "learning_rate": 8.119534382700613e-05,
+      "loss": 5.1768,
+      "step": 35700
+    },
+    {
+      "epoch": 33.77376739797122,
+      "grad_norm": 0.9027467370033264,
+      "learning_rate": 8.105875679810968e-05,
+      "loss": 5.1786,
+      "step": 35800
+    },
+    {
+      "epoch": 33.86812927577259,
+      "grad_norm": 0.9518372416496277,
+      "learning_rate": 8.092179132806317e-05,
+      "loss": 5.1776,
+      "step": 35900
+    },
+    {
+      "epoch": 33.96249115357396,
+      "grad_norm": 0.9268783926963806,
+      "learning_rate": 8.078444908574767e-05,
+      "loss": 5.1755,
+      "step": 36000
+    },
+    {
+      "epoch": 33.96249115357396,
+      "eval_loss": 5.270759105682373,
+      "eval_runtime": 74.4429,
+      "eval_samples_per_second": 202.45,
+      "eval_steps_per_second": 6.327,
+      "step": 36000
+    },
+    {
+      "epoch": 34.05661712668082,
+      "grad_norm": 0.8898890614509583,
+      "learning_rate": 8.064673174463505e-05,
+      "loss": 5.1774,
+      "step": 36100
+    },
+    {
+      "epoch": 34.15097900448219,
+      "grad_norm": 0.9448688626289368,
+      "learning_rate": 8.050864098276762e-05,
+      "loss": 5.1719,
+      "step": 36200
+    },
+    {
+      "epoch": 34.24534088228356,
+      "grad_norm": 0.9344834685325623,
+      "learning_rate": 8.037017848273776e-05,
+      "loss": 5.1745,
+      "step": 36300
+    },
+    {
+      "epoch": 34.339702760084926,
+      "grad_norm": 1.0767810344696045,
+      "learning_rate": 8.023134593166734e-05,
+      "loss": 5.155,
+      "step": 36400
+    },
+    {
+      "epoch": 34.434064637886294,
+      "grad_norm": 1.1843194961547852,
+      "learning_rate": 8.009214502118718e-05,
+      "loss": 5.108,
+      "step": 36500
+    },
+    {
+      "epoch": 34.52842651568766,
+      "grad_norm": 1.1245286464691162,
+      "learning_rate": 7.995257744741642e-05,
+      "loss": 5.0859,
+      "step": 36600
+    },
+    {
+      "epoch": 34.62278839348903,
+      "grad_norm": 1.4160962104797363,
+      "learning_rate": 7.981264491094192e-05,
+      "loss": 5.068,
+      "step": 36700
+    },
+    {
+      "epoch": 34.7171502712904,
+      "grad_norm": 1.0435899496078491,
+      "learning_rate": 7.967234911679749e-05,
+      "loss": 5.0546,
+      "step": 36800
+    },
+    {
+      "epoch": 34.81151214909177,
+      "grad_norm": 1.2017834186553955,
+      "learning_rate": 7.953169177444309e-05,
+      "loss": 5.0384,
+      "step": 36900
+    },
+    {
+      "epoch": 34.90587402689314,
+      "grad_norm": 1.2122361660003662,
+      "learning_rate": 7.939067459774405e-05,
+      "loss": 5.0225,
+      "step": 37000
+    },
+    {
+      "epoch": 34.90587402689314,
+      "eval_loss": 5.060994625091553,
+      "eval_runtime": 74.4648,
+      "eval_samples_per_second": 202.391,
+      "eval_steps_per_second": 6.325,
+      "step": 37000
+    },
+    {
+      "epoch": 35.0,
+      "grad_norm": 1.66022527217865,
+      "learning_rate": 7.924929930495018e-05,
+      "loss": 4.9977,
+      "step": 37100
+    },
+    {
+      "epoch": 35.09436187780137,
+      "grad_norm": 1.222360610961914,
+      "learning_rate": 7.910756761867479e-05,
+      "loss": 4.9734,
+      "step": 37200
+    },
+    {
+      "epoch": 35.18872375560274,
+      "grad_norm": 1.9777710437774658,
+      "learning_rate": 7.896548126587374e-05,
+      "loss": 4.9506,
+      "step": 37300
+    },
+    {
+      "epoch": 35.283085633404106,
+      "grad_norm": 1.3450212478637695,
+      "learning_rate": 7.882304197782443e-05,
+      "loss": 4.9493,
+      "step": 37400
+    },
+    {
+      "epoch": 35.377447511205474,
+      "grad_norm": 1.3580466508865356,
+      "learning_rate": 7.86802514901046e-05,
+      "loss": 4.9329,
+      "step": 37500
+    },
+    {
+      "epoch": 35.47180938900684,
+      "grad_norm": 1.2852448225021362,
+      "learning_rate": 7.853711154257133e-05,
+      "loss": 4.9236,
+      "step": 37600
+    },
+    {
+      "epoch": 35.56617126680821,
+      "grad_norm": 1.6216447353363037,
+      "learning_rate": 7.839362387933965e-05,
+      "loss": 4.9067,
+      "step": 37700
+    },
+    {
+      "epoch": 35.66053314460958,
+      "grad_norm": 1.6207915544509888,
+      "learning_rate": 7.824979024876149e-05,
+      "loss": 4.875,
+      "step": 37800
+    },
+    {
+      "epoch": 35.75489502241095,
+      "grad_norm": 1.4836673736572266,
+      "learning_rate": 7.810561240340424e-05,
+      "loss": 4.8447,
+      "step": 37900
+    },
+    {
+      "epoch": 35.84925690021232,
+      "grad_norm": 1.5666450262069702,
+      "learning_rate": 7.796109210002945e-05,
+      "loss": 4.7756,
+      "step": 38000
+    },
+    {
+      "epoch": 35.84925690021232,
+      "eval_loss": 4.731563091278076,
+      "eval_runtime": 74.4706,
+      "eval_samples_per_second": 202.375,
+      "eval_steps_per_second": 6.325,
+      "step": 38000
+    },
+    {
+      "epoch": 35.94361877801368,
+      "grad_norm": 2.0269269943237305,
+      "learning_rate": 7.781623109957139e-05,
+      "loss": 4.698,
+      "step": 38100
+    },
+    {
+      "epoch": 36.03774475112055,
+      "grad_norm": 3.2447714805603027,
+      "learning_rate": 7.767103116711566e-05,
+      "loss": 4.6137,
+      "step": 38200
+    },
+    {
+      "epoch": 36.13210662892192,
+      "grad_norm": 2.3450400829315186,
+      "learning_rate": 7.752549407187761e-05,
+      "loss": 4.5311,
+      "step": 38300
+    },
+    {
+      "epoch": 36.226468506723286,
+      "grad_norm": 2.9792938232421875,
+      "learning_rate": 7.73796215871808e-05,
+      "loss": 4.467,
+      "step": 38400
+    },
+    {
+      "epoch": 36.320830384524655,
+      "grad_norm": 2.1335771083831787,
+      "learning_rate": 7.723341549043543e-05,
+      "loss": 4.4239,
+      "step": 38500
+    },
+    {
+      "epoch": 36.41519226232602,
+      "grad_norm": 2.37923526763916,
+      "learning_rate": 7.708687756311666e-05,
+      "loss": 4.3596,
+      "step": 38600
+    },
+    {
+      "epoch": 36.50955414012739,
+      "grad_norm": 2.716700553894043,
+      "learning_rate": 7.694000959074288e-05,
+      "loss": 4.3063,
+      "step": 38700
+    },
+    {
+      "epoch": 36.60391601792875,
+      "grad_norm": 1.805467963218689,
+      "learning_rate": 7.679281336285398e-05,
+      "loss": 4.2623,
+      "step": 38800
+    },
+    {
+      "epoch": 36.69827789573012,
+      "grad_norm": 2.4338560104370117,
+      "learning_rate": 7.664529067298954e-05,
+      "loss": 4.2091,
+      "step": 38900
+    },
+    {
+      "epoch": 36.79263977353149,
+      "grad_norm": 1.896016001701355,
+      "learning_rate": 7.649744331866702e-05,
+      "loss": 4.1527,
+      "step": 39000
+    },
+    {
+      "epoch": 36.79263977353149,
+      "eval_loss": 4.014511585235596,
+      "eval_runtime": 74.4548,
+      "eval_samples_per_second": 202.418,
+      "eval_steps_per_second": 6.326,
+      "step": 39000
+    },
+    {
+      "epoch": 36.88700165133286,
+      "grad_norm": 2.2716317176818848,
+      "learning_rate": 7.634927310135972e-05,
+      "loss": 4.0514,
+      "step": 39100
+    },
+    {
+      "epoch": 36.98136352913423,
+      "grad_norm": 1.6702080965042114,
+      "learning_rate": 7.620078182647502e-05,
+      "loss": 3.9428,
+      "step": 39200
+    },
+    {
+      "epoch": 37.0754895022411,
+      "grad_norm": 1.7880933284759521,
+      "learning_rate": 7.605197130333222e-05,
+      "loss": 3.8424,
+      "step": 39300
+    },
+    {
+      "epoch": 37.169851380042466,
+      "grad_norm": 1.788912296295166,
+      "learning_rate": 7.590284334514057e-05,
+      "loss": 3.732,
+      "step": 39400
+    },
+    {
+      "epoch": 37.26421325784383,
+      "grad_norm": 1.85847806930542,
+      "learning_rate": 7.575339976897722e-05,
+      "loss": 3.6599,
+      "step": 39500
+    },
+    {
+      "epoch": 37.3585751356452,
+      "grad_norm": 1.744305968284607,
+      "learning_rate": 7.560364239576496e-05,
+      "loss": 3.5681,
+      "step": 39600
+    },
+    {
+      "epoch": 37.452937013446565,
+      "grad_norm": 1.8111028671264648,
+      "learning_rate": 7.545357305025013e-05,
+      "loss": 3.5125,
+      "step": 39700
+    },
+    {
+      "epoch": 37.547298891247934,
+      "grad_norm": 1.7670716047286987,
+      "learning_rate": 7.530319356098033e-05,
+      "loss": 3.4513,
+      "step": 39800
+    },
+    {
+      "epoch": 37.6416607690493,
+      "grad_norm": 1.5341713428497314,
+      "learning_rate": 7.51525057602822e-05,
+      "loss": 3.4147,
+      "step": 39900
+    },
+    {
+      "epoch": 37.73602264685067,
+      "grad_norm": 1.5770525932312012,
+      "learning_rate": 7.500151148423902e-05,
+      "loss": 3.3725,
+      "step": 40000
+    },
+    {
+      "epoch": 37.73602264685067,
+      "eval_loss": 3.2892656326293945,
+      "eval_runtime": 74.4068,
+      "eval_samples_per_second": 202.549,
+      "eval_steps_per_second": 6.33,
+      "step": 40000
+    },
+    {
+      "epoch": 37.83038452465204,
+      "grad_norm": 1.4793765544891357,
+      "learning_rate": 7.485021257266841e-05,
+      "loss": 3.3299,
+      "step": 40100
+    },
+    {
+      "epoch": 37.92474640245341,
+      "grad_norm": 1.519507646560669,
+      "learning_rate": 7.469861086909983e-05,
+      "loss": 3.3008,
+      "step": 40200
+    },
+    {
+      "epoch": 38.01887237556027,
+      "grad_norm": 1.6102982759475708,
+      "learning_rate": 7.454670822075225e-05,
+      "loss": 3.2762,
+      "step": 40300
+    },
+    {
+      "epoch": 38.11323425336164,
+      "grad_norm": 1.501295804977417,
+      "learning_rate": 7.439450647851145e-05,
+      "loss": 3.2414,
+      "step": 40400
+    },
+    {
+      "epoch": 38.20759613116301,
+      "grad_norm": 1.3700292110443115,
+      "learning_rate": 7.424200749690763e-05,
+      "loss": 3.2046,
+      "step": 40500
+    },
+    {
+      "epoch": 38.30195800896438,
+      "grad_norm": 1.5433059930801392,
+      "learning_rate": 7.40892131340928e-05,
+      "loss": 3.1777,
+      "step": 40600
+    },
+    {
+      "epoch": 38.396319886765745,
+      "grad_norm": 1.410462498664856,
+      "learning_rate": 7.393612525181801e-05,
+      "loss": 3.1619,
+      "step": 40700
+    },
+    {
+      "epoch": 38.490681764567114,
+      "grad_norm": 1.526921272277832,
+      "learning_rate": 7.37827457154108e-05,
+      "loss": 3.1474,
+      "step": 40800
+    },
+    {
+      "epoch": 38.58504364236848,
+      "grad_norm": 1.4830299615859985,
+      "learning_rate": 7.362907639375244e-05,
+      "loss": 3.1138,
+      "step": 40900
+    },
+    {
+      "epoch": 38.67940552016985,
+      "grad_norm": 1.3642972707748413,
+      "learning_rate": 7.347511915925512e-05,
+      "loss": 3.0895,
+      "step": 41000
+    },
+    {
+      "epoch": 38.67940552016985,
+      "eval_loss": 3.0351099967956543,
+      "eval_runtime": 74.4301,
+      "eval_samples_per_second": 202.485,
+      "eval_steps_per_second": 6.328,
+      "step": 41000
+    },
+    {
+      "epoch": 38.77376739797122,
+      "grad_norm": 1.26517653465271,
+      "learning_rate": 7.33208758878391e-05,
+      "loss": 3.0726,
+      "step": 41100
+    },
+    {
+      "epoch": 38.86812927577259,
+      "grad_norm": 1.4588021039962769,
+      "learning_rate": 7.316634845891003e-05,
+      "loss": 3.0682,
+      "step": 41200
+    },
+    {
+      "epoch": 38.96249115357396,
+      "grad_norm": 1.3476386070251465,
+      "learning_rate": 7.301153875533583e-05,
+      "loss": 3.0388,
+      "step": 41300
+    },
+    {
+      "epoch": 39.05661712668082,
+      "grad_norm": 1.5515798330307007,
+      "learning_rate": 7.28564486634239e-05,
+      "loss": 3.0181,
+      "step": 41400
+    },
+    {
+      "epoch": 39.15097900448219,
+      "grad_norm": 1.422370195388794,
+      "learning_rate": 7.270108007289807e-05,
+      "loss": 2.9994,
+      "step": 41500
+    },
+    {
+      "epoch": 39.24534088228356,
+      "grad_norm": 1.347652792930603,
+      "learning_rate": 7.254543487687558e-05,
+      "loss": 2.9834,
+      "step": 41600
+    },
+    {
+      "epoch": 39.339702760084926,
+      "grad_norm": 1.3476914167404175,
+      "learning_rate": 7.2389514971844e-05,
+      "loss": 2.9631,
+      "step": 41700
+    },
+    {
+      "epoch": 39.434064637886294,
+      "grad_norm": 1.3662831783294678,
+      "learning_rate": 7.22333222576382e-05,
+      "loss": 2.9611,
+      "step": 41800
+    },
+    {
+      "epoch": 39.52842651568766,
+      "grad_norm": 1.3495389223098755,
+      "learning_rate": 7.207685863741711e-05,
+      "loss": 2.9458,
+      "step": 41900
+    },
+    {
+      "epoch": 39.62278839348903,
+      "grad_norm": 1.4348385334014893,
+      "learning_rate": 7.192012601764053e-05,
+      "loss": 2.9351,
+      "step": 42000
+    },
+    {
+      "epoch": 39.62278839348903,
+      "eval_loss": 2.878948450088501,
+      "eval_runtime": 74.419,
+      "eval_samples_per_second": 202.516,
+      "eval_steps_per_second": 6.329,
+      "step": 42000
+    },
+    {
+      "epoch": 39.7171502712904,
+      "grad_norm": 1.299336314201355,
+      "learning_rate": 7.1763126308046e-05,
+      "loss": 2.9141,
+      "step": 42100
+    },
+    {
+      "epoch": 39.81151214909177,
+      "grad_norm": 1.3337122201919556,
+      "learning_rate": 7.160586142162544e-05,
+      "loss": 2.8945,
+      "step": 42200
+    },
+    {
+      "epoch": 39.90587402689314,
+      "grad_norm": 1.3183374404907227,
+      "learning_rate": 7.144833327460186e-05,
+      "loss": 2.8872,
+      "step": 42300
+    },
+    {
+      "epoch": 40.0,
+      "grad_norm": 1.5197392702102661,
+      "learning_rate": 7.129054378640599e-05,
+      "loss": 2.8863,
+      "step": 42400
+    },
+    {
+      "epoch": 40.09436187780137,
+      "grad_norm": 1.358708381652832,
+      "learning_rate": 7.1132494879653e-05,
+      "loss": 2.854,
+      "step": 42500
+    },
+    {
+      "epoch": 40.18872375560274,
+      "grad_norm": 1.3579105138778687,
+      "learning_rate": 7.097418848011888e-05,
+      "loss": 2.8579,
+      "step": 42600
+    },
+    {
+      "epoch": 40.283085633404106,
+      "grad_norm": 1.2787528038024902,
+      "learning_rate": 7.081562651671719e-05,
+      "loss": 2.8426,
+      "step": 42700
+    },
+    {
+      "epoch": 40.377447511205474,
+      "grad_norm": 1.3120508193969727,
+      "learning_rate": 7.065681092147542e-05,
+      "loss": 2.8356,
+      "step": 42800
+    },
+    {
+      "epoch": 40.47180938900684,
+      "grad_norm": 1.29078209400177,
+      "learning_rate": 7.049774362951144e-05,
+      "loss": 2.8274,
+      "step": 42900
+    },
+    {
+      "epoch": 40.56617126680821,
+      "grad_norm": 1.2766653299331665,
+      "learning_rate": 7.033842657901005e-05,
+      "loss": 2.8066,
+      "step": 43000
+    },
+    {
+      "epoch": 40.56617126680821,
+      "eval_loss": 2.7730472087860107,
+      "eval_runtime": 74.4694,
+      "eval_samples_per_second": 202.378,
+      "eval_steps_per_second": 6.325,
+      "step": 43000
+    },
+    {
+      "epoch": 40.66053314460958,
+      "grad_norm": 1.3247416019439697,
+      "learning_rate": 7.017886171119917e-05,
+      "loss": 2.7981,
+      "step": 43100
+    },
+    {
+      "epoch": 40.75489502241095,
+      "grad_norm": 1.271553635597229,
+      "learning_rate": 7.001905097032644e-05,
+      "loss": 2.794,
+      "step": 43200
+    },
+    {
+      "epoch": 40.84925690021232,
+      "grad_norm": 1.228617787361145,
+      "learning_rate": 6.985899630363526e-05,
+      "loss": 2.7911,
+      "step": 43300
+    },
+    {
+      "epoch": 40.94361877801368,
+      "grad_norm": 1.2349623441696167,
+      "learning_rate": 6.969869966134123e-05,
+      "loss": 2.7901,
+      "step": 43400
+    },
+    {
+      "epoch": 41.03774475112055,
+      "grad_norm": 1.2398724555969238,
+      "learning_rate": 6.953816299660834e-05,
+      "loss": 2.7727,
+      "step": 43500
+    },
+    {
+      "epoch": 41.13210662892192,
+      "grad_norm": 1.2596232891082764,
+      "learning_rate": 6.937738826552524e-05,
+      "loss": 2.7543,
+      "step": 43600
+    },
+    {
+      "epoch": 41.226468506723286,
+      "grad_norm": 1.349109172821045,
+      "learning_rate": 6.921637742708123e-05,
+      "loss": 2.7419,
+      "step": 43700
+    },
+    {
+      "epoch": 41.320830384524655,
+      "grad_norm": 1.1863901615142822,
+      "learning_rate": 6.905513244314259e-05,
+      "loss": 2.7433,
+      "step": 43800
+    },
+    {
+      "epoch": 41.41519226232602,
+      "grad_norm": 1.2365715503692627,
+      "learning_rate": 6.889365527842857e-05,
+      "loss": 2.7391,
+      "step": 43900
+    },
+    {
+      "epoch": 41.50955414012739,
+      "grad_norm": 1.2725337743759155,
+      "learning_rate": 6.873194790048746e-05,
+      "loss": 2.7429,
+      "step": 44000
+    },
+    {
+      "epoch": 41.50955414012739,
+      "eval_loss": 2.692702054977417,
+      "eval_runtime": 74.4255,
+      "eval_samples_per_second": 202.498,
+      "eval_steps_per_second": 6.328,
+      "step": 44000
+    },
+    {
+      "epoch": 41.60391601792875,
+      "grad_norm": 1.2004797458648682,
+      "learning_rate": 6.857001227967263e-05,
+      "loss": 2.727,
+      "step": 44100
+    },
+    {
+      "epoch": 41.69827789573012,
+      "grad_norm": 1.2816728353500366,
+      "learning_rate": 6.84078503891185e-05,
+      "loss": 2.7233,
+      "step": 44200
+    },
+    {
+      "epoch": 41.79263977353149,
+      "grad_norm": 1.2110925912857056,
+      "learning_rate": 6.824546420471653e-05,
+      "loss": 2.7166,
+      "step": 44300
+    },
+    {
+      "epoch": 41.88700165133286,
+      "grad_norm": 1.2935187816619873,
+      "learning_rate": 6.808285570509117e-05,
+      "loss": 2.7073,
+      "step": 44400
+    },
+    {
+      "epoch": 41.98136352913423,
+      "grad_norm": 1.2628765106201172,
+      "learning_rate": 6.792002687157564e-05,
+      "loss": 2.6934,
+      "step": 44500
+    },
+    {
+      "epoch": 42.0754895022411,
+      "grad_norm": 1.2200449705123901,
+      "learning_rate": 6.775697968818788e-05,
+      "loss": 2.6895,
+      "step": 44600
+    },
+    {
+      "epoch": 42.169851380042466,
+      "grad_norm": 1.1676024198532104,
+      "learning_rate": 6.759371614160639e-05,
+      "loss": 2.6743,
+      "step": 44700
+    },
+    {
+      "epoch": 42.26421325784383,
+      "grad_norm": 1.1828370094299316,
+      "learning_rate": 6.743023822114596e-05,
+      "loss": 2.6713,
+      "step": 44800
+    },
+    {
+      "epoch": 42.3585751356452,
+      "grad_norm": 1.2179129123687744,
+      "learning_rate": 6.726654791873343e-05,
+      "loss": 2.6621,
+      "step": 44900
+    },
+    {
+      "epoch": 42.452937013446565,
+      "grad_norm": 1.232735514640808,
+      "learning_rate": 6.710264722888352e-05,
+      "loss": 2.6762,
+      "step": 45000
+    },
+    {
+      "epoch": 42.452937013446565,
+      "eval_loss": 2.6310536861419678,
+      "eval_runtime": 74.4881,
+      "eval_samples_per_second": 202.328,
+      "eval_steps_per_second": 6.323,
+      "step": 45000
+    },
+    {
+      "epoch": 42.547298891247934,
+      "grad_norm": 1.1474517583847046,
+      "learning_rate": 6.693853814867439e-05,
+      "loss": 2.6642,
+      "step": 45100
+    },
+    {
+      "epoch": 42.6416607690493,
+      "grad_norm": 1.2410264015197754,
+      "learning_rate": 6.677422267772338e-05,
+      "loss": 2.6602,
+      "step": 45200
+    },
+    {
+      "epoch": 42.73602264685067,
+      "grad_norm": 1.3303461074829102,
+      "learning_rate": 6.660970281816269e-05,
+      "loss": 2.665,
+      "step": 45300
+    },
+    {
+      "epoch": 42.83038452465204,
+      "grad_norm": 1.1773780584335327,
+      "learning_rate": 6.644498057461485e-05,
+      "loss": 2.6539,
+      "step": 45400
+    },
+    {
+      "epoch": 42.92474640245341,
+      "grad_norm": 1.2169513702392578,
+      "learning_rate": 6.628005795416842e-05,
+      "loss": 2.6352,
+      "step": 45500
+    },
+    {
+      "epoch": 43.01887237556027,
+      "grad_norm": 1.1680723428726196,
+      "learning_rate": 6.611493696635351e-05,
+      "loss": 2.6335,
+      "step": 45600
+    },
+    {
+      "epoch": 43.11323425336164,
+      "grad_norm": 1.14376699924469,
+      "learning_rate": 6.594961962311722e-05,
+      "loss": 2.614,
+      "step": 45700
+    },
+    {
+      "epoch": 43.20759613116301,
+      "grad_norm": 1.1640697717666626,
+      "learning_rate": 6.578410793879921e-05,
+      "loss": 2.6253,
+      "step": 45800
+    },
+    {
+      "epoch": 43.30195800896438,
+      "grad_norm": 1.2426387071609497,
+      "learning_rate": 6.561840393010713e-05,
+      "loss": 2.6163,
+      "step": 45900
+    },
+    {
+      "epoch": 43.396319886765745,
+      "grad_norm": 1.1884350776672363,
+      "learning_rate": 6.545250961609202e-05,
+      "loss": 2.6201,
+      "step": 46000
+    },
+    {
+      "epoch": 43.396319886765745,
+      "eval_loss": 2.5862319469451904,
+      "eval_runtime": 74.4528,
+      "eval_samples_per_second": 202.423,
+      "eval_steps_per_second": 6.326,
+      "step": 46000
+    },
+    {
+      "epoch": 43.490681764567114,
+      "grad_norm": 1.2242307662963867,
+      "learning_rate": 6.528642701812378e-05,
+      "loss": 2.6032,
+      "step": 46100
+    },
+    {
+      "epoch": 43.58504364236848,
+      "grad_norm": 1.1581670045852661,
+      "learning_rate": 6.51201581598664e-05,
+      "loss": 2.6036,
+      "step": 46200
+    },
+    {
+      "epoch": 43.67940552016985,
+      "grad_norm": 1.1338080167770386,
+      "learning_rate": 6.49537050672535e-05,
+      "loss": 2.6141,
+      "step": 46300
+    },
+    {
+      "epoch": 43.77376739797122,
+      "grad_norm": 1.2403849363327026,
+      "learning_rate": 6.478706976846344e-05,
+      "loss": 2.5999,
+      "step": 46400
+    },
+    {
+      "epoch": 43.86812927577259,
+      "grad_norm": 1.2618863582611084,
+      "learning_rate": 6.462025429389475e-05,
+      "loss": 2.5949,
+      "step": 46500
+    },
+    {
+      "epoch": 43.96249115357396,
+      "grad_norm": 1.1581882238388062,
+      "learning_rate": 6.445326067614139e-05,
+      "loss": 2.5872,
+      "step": 46600
+    },
+    {
+      "epoch": 44.05661712668082,
+      "grad_norm": 1.1863130331039429,
+      "learning_rate": 6.428609094996785e-05,
+      "loss": 2.5819,
+      "step": 46700
+    },
+    {
+      "epoch": 44.15097900448219,
+      "grad_norm": 1.1343599557876587,
+      "learning_rate": 6.411874715228447e-05,
+      "loss": 2.5694,
+      "step": 46800
+    },
+    {
+      "epoch": 44.24534088228356,
+      "grad_norm": 1.1974676847457886,
+      "learning_rate": 6.395123132212268e-05,
+      "loss": 2.5648,
+      "step": 46900
+    },
+    {
+      "epoch": 44.339702760084926,
+      "grad_norm": 1.151482343673706,
+      "learning_rate": 6.378354550060997e-05,
+      "loss": 2.5672,
+      "step": 47000
+    },
+    {
+      "epoch": 44.339702760084926,
+      "eval_loss": 2.544923782348633,
+      "eval_runtime": 74.4495,
+      "eval_samples_per_second": 202.432,
+      "eval_steps_per_second": 6.326,
+      "step": 47000
+    },
+    {
+      "epoch": 44.434064637886294,
+      "grad_norm": 1.144623875617981,
+      "learning_rate": 6.361569173094515e-05,
+      "loss": 2.5668,
+      "step": 47100
+    },
+    {
+      "epoch": 44.52842651568766,
+      "grad_norm": 1.1448100805282593,
+      "learning_rate": 6.344767205837345e-05,
+      "loss": 2.5664,
+      "step": 47200
+    },
+    {
+      "epoch": 44.62278839348903,
+      "grad_norm": 1.1526908874511719,
+      "learning_rate": 6.327948853016153e-05,
+      "loss": 2.5536,
+      "step": 47300
+    },
+    {
+      "epoch": 44.7171502712904,
+      "grad_norm": 1.1646764278411865,
+      "learning_rate": 6.311114319557261e-05,
+      "loss": 2.5634,
+      "step": 47400
+    },
+    {
+      "epoch": 44.81151214909177,
+      "grad_norm": 1.1626538038253784,
+      "learning_rate": 6.29426381058415e-05,
+      "loss": 2.5495,
+      "step": 47500
+    },
+    {
+      "epoch": 44.90587402689314,
+      "grad_norm": 1.1385385990142822,
+      "learning_rate": 6.277397531414951e-05,
+      "loss": 2.5466,
+      "step": 47600
+    },
+    {
+      "epoch": 45.0,
+      "grad_norm": 1.4365875720977783,
+      "learning_rate": 6.260515687559953e-05,
+      "loss": 2.5465,
+      "step": 47700
+    },
+    {
+      "epoch": 45.09436187780137,
+      "grad_norm": 1.1712549924850464,
+      "learning_rate": 6.243618484719098e-05,
+      "loss": 2.534,
+      "step": 47800
+    },
+    {
+      "epoch": 45.18872375560274,
+      "grad_norm": 1.1397508382797241,
+      "learning_rate": 6.226706128779468e-05,
+      "loss": 2.5308,
+      "step": 47900
+    },
+    {
+      "epoch": 45.283085633404106,
+      "grad_norm": 1.0902845859527588,
+      "learning_rate": 6.209778825812784e-05,
+      "loss": 2.5285,
+      "step": 48000
+    },
+    {
+      "epoch": 45.283085633404106,
+      "eval_loss": 2.5119264125823975,
+      "eval_runtime": 74.4894,
+      "eval_samples_per_second": 202.324,
+      "eval_steps_per_second": 6.323,
+      "step": 48000
+    },
+    {
+      "epoch": 45.377447511205474,
+      "grad_norm": 1.174816370010376,
+      "learning_rate": 6.19283678207289e-05,
+      "loss": 2.5255,
+      "step": 48100
+    },
+    {
+      "epoch": 45.47180938900684,
+      "grad_norm": 1.1556349992752075,
+      "learning_rate": 6.175880203993243e-05,
+      "loss": 2.5208,
+      "step": 48200
+    },
+    {
+      "epoch": 45.56617126680821,
+      "grad_norm": 1.1801139116287231,
+      "learning_rate": 6.158909298184395e-05,
+      "loss": 2.5132,
+      "step": 48300
+    },
+    {
+      "epoch": 45.66053314460958,
+      "grad_norm": 1.0904967784881592,
+      "learning_rate": 6.14192427143148e-05,
+      "loss": 2.531,
+      "step": 48400
+    },
+    {
+      "epoch": 45.75489502241095,
+      "grad_norm": 1.1581918001174927,
+      "learning_rate": 6.124925330691685e-05,
+      "loss": 2.5225,
+      "step": 48500
+    },
+    {
+      "epoch": 45.84925690021232,
+      "grad_norm": 1.130543828010559,
+      "learning_rate": 6.107912683091741e-05,
+      "loss": 2.501,
+      "step": 48600
+    },
+    {
+      "epoch": 45.94361877801368,
+      "grad_norm": 1.114476203918457,
+      "learning_rate": 6.0908865359253886e-05,
+      "loss": 2.5157,
+      "step": 48700
+    },
+    {
+      "epoch": 46.03774475112055,
+      "grad_norm": 1.150086760520935,
+      "learning_rate": 6.07384709665086e-05,
+      "loss": 2.501,
+      "step": 48800
+    },
+    {
+      "epoch": 46.13210662892192,
+      "grad_norm": 1.1558949947357178,
+      "learning_rate": 6.0567945728883435e-05,
+      "loss": 2.4929,
+      "step": 48900
+    },
+    {
+      "epoch": 46.226468506723286,
+      "grad_norm": 1.1270654201507568,
+      "learning_rate": 6.03972917241746e-05,
+      "loss": 2.4942,
+      "step": 49000
+    },
+    {
+      "epoch": 46.226468506723286,
+      "eval_loss": 2.4788951873779297,
+      "eval_runtime": 74.4909,
+      "eval_samples_per_second": 202.32,
+      "eval_steps_per_second": 6.323,
+      "step": 49000
+    },
+    {
+      "epoch": 46.320830384524655,
+      "grad_norm": 1.1517997980117798,
+      "learning_rate": 6.02265110317473e-05,
+      "loss": 2.5006,
+      "step": 49100
+    },
+    {
+      "epoch": 46.41519226232602,
+      "grad_norm": 1.0966933965682983,
+      "learning_rate": 6.005560573251037e-05,
+      "loss": 2.4881,
+      "step": 49200
+    },
+    {
+      "epoch": 46.50955414012739,
+      "grad_norm": 1.147829532623291,
+      "learning_rate": 5.9884577908890926e-05,
+      "loss": 2.4907,
+      "step": 49300
+    },
+    {
+      "epoch": 46.60391601792875,
+      "grad_norm": 1.1360352039337158,
+      "learning_rate": 5.971342964480906e-05,
+      "loss": 2.4895,
+      "step": 49400
+    },
+    {
+      "epoch": 46.69827789573012,
+      "grad_norm": 1.1289290189743042,
+      "learning_rate": 5.954216302565235e-05,
+      "loss": 2.4808,
+      "step": 49500
+    },
+    {
+      "epoch": 46.79263977353149,
+      "grad_norm": 1.0954254865646362,
+      "learning_rate": 5.9370780138250484e-05,
+      "loss": 2.4772,
+      "step": 49600
+    },
+    {
+      "epoch": 46.88700165133286,
+      "grad_norm": 1.141011118888855,
+      "learning_rate": 5.9199283070849875e-05,
+      "loss": 2.4766,
+      "step": 49700
+    },
+    {
+      "epoch": 46.98136352913423,
+      "grad_norm": 1.0591996908187866,
+      "learning_rate": 5.9027673913088165e-05,
+      "loss": 2.4805,
+      "step": 49800
+    },
+    {
+      "epoch": 47.0754895022411,
+      "grad_norm": 1.0702204704284668,
+      "learning_rate": 5.885595475596878e-05,
+      "loss": 2.4638,
+      "step": 49900
+    },
+    {
+      "epoch": 47.169851380042466,
+      "grad_norm": 1.1920737028121948,
+      "learning_rate": 5.868412769183547e-05,
+      "loss": 2.4721,
+      "step": 50000
+    },
+    {
+      "epoch": 47.169851380042466,
+      "eval_loss": 2.450434923171997,
+      "eval_runtime": 74.4123,
+      "eval_samples_per_second": 202.534,
+      "eval_steps_per_second": 6.33,
+      "step": 50000
+    },
+    {
+      "epoch": 47.26421325784383,
+      "grad_norm": 1.0812413692474365,
+      "learning_rate": 5.8512194814346775e-05,
+      "loss": 2.4477,
+      "step": 50100
+    },
+    {
+      "epoch": 47.3585751356452,
+      "grad_norm": 1.0759973526000977,
+      "learning_rate": 5.8340158218450555e-05,
+      "loss": 2.454,
+      "step": 50200
+    },
+    {
+      "epoch": 47.452937013446565,
+      "grad_norm": 1.1221734285354614,
+      "learning_rate": 5.8168020000358435e-05,
+      "loss": 2.4554,
+      "step": 50300
+    },
+    {
+      "epoch": 47.547298891247934,
+      "grad_norm": 1.0968127250671387,
+      "learning_rate": 5.799578225752028e-05,
+      "loss": 2.4586,
+      "step": 50400
+    },
+    {
+      "epoch": 47.6416607690493,
+      "grad_norm": 1.1258801221847534,
+      "learning_rate": 5.7823447088598624e-05,
+      "loss": 2.462,
+      "step": 50500
+    },
+    {
+      "epoch": 47.73602264685067,
+      "grad_norm": 1.1324305534362793,
+      "learning_rate": 5.765101659344313e-05,
+      "loss": 2.4518,
+      "step": 50600
+    },
+    {
+      "epoch": 47.83038452465204,
+      "grad_norm": 1.1313503980636597,
+      "learning_rate": 5.747849287306496e-05,
+      "loss": 2.4556,
+      "step": 50700
+    },
+    {
+      "epoch": 47.92474640245341,
+      "grad_norm": 1.088899850845337,
+      "learning_rate": 5.730587802961119e-05,
+      "loss": 2.4553,
+      "step": 50800
+    },
+    {
+      "epoch": 48.01887237556027,
+      "grad_norm": 1.1050726175308228,
+      "learning_rate": 5.7133174166339245e-05,
+      "loss": 2.4449,
+      "step": 50900
+    },
+    {
+      "epoch": 48.11323425336164,
+      "grad_norm": 1.2336230278015137,
+      "learning_rate": 5.696038338759117e-05,
+      "loss": 2.4343,
+      "step": 51000
+    },
+    {
+      "epoch": 48.11323425336164,
+      "eval_loss": 2.423373222351074,
+      "eval_runtime": 74.4359,
+      "eval_samples_per_second": 202.47,
+      "eval_steps_per_second": 6.328,
+      "step": 51000
+    },
+    {
+      "epoch": 48.20759613116301,
+      "grad_norm": 1.10746169090271,
+      "learning_rate": 5.678750779876807e-05,
+      "loss": 2.4329,
+      "step": 51100
+    },
+    {
+      "epoch": 48.30195800896438,
+      "grad_norm": 1.1919947862625122,
+      "learning_rate": 5.661454950630445e-05,
+      "loss": 2.429,
+      "step": 51200
+    },
+    {
+      "epoch": 48.396319886765745,
+      "grad_norm": 1.0968834161758423,
+      "learning_rate": 5.6441510617642526e-05,
+      "loss": 2.4384,
+      "step": 51300
+    },
+    {
+      "epoch": 48.490681764567114,
+      "grad_norm": 1.112357497215271,
+      "learning_rate": 5.626839324120654e-05,
+      "loss": 2.4317,
+      "step": 51400
+    },
+    {
+      "epoch": 48.58504364236848,
+      "grad_norm": 1.0847004652023315,
+      "learning_rate": 5.609519948637708e-05,
+      "loss": 2.4362,
+      "step": 51500
+    },
+    {
+      "epoch": 48.67940552016985,
+      "grad_norm": 1.1191656589508057,
+      "learning_rate": 5.592193146346543e-05,
+      "loss": 2.4201,
+      "step": 51600
+    },
+    {
+      "epoch": 48.77376739797122,
+      "grad_norm": 1.1359890699386597,
+      "learning_rate": 5.5748591283687725e-05,
+      "loss": 2.4272,
+      "step": 51700
+    },
+    {
+      "epoch": 48.86812927577259,
+      "grad_norm": 1.1025972366333008,
+      "learning_rate": 5.557518105913939e-05,
+      "loss": 2.4205,
+      "step": 51800
+    },
+    {
+      "epoch": 48.96249115357396,
+      "grad_norm": 1.1198244094848633,
+      "learning_rate": 5.540170290276927e-05,
+      "loss": 2.4185,
+      "step": 51900
+    },
+    {
+      "epoch": 49.05661712668082,
+      "grad_norm": 1.0897448062896729,
+      "learning_rate": 5.5228158928353944e-05,
+      "loss": 2.4059,
+      "step": 52000
+    },
+    {
+      "epoch": 49.05661712668082,
+      "eval_loss": 2.4050960540771484,
+      "eval_runtime": 74.4452,
+      "eval_samples_per_second": 202.444,
+      "eval_steps_per_second": 6.327,
+      "step": 52000
+    },
+    {
+      "epoch": 49.15097900448219,
+      "grad_norm": 1.1507387161254883,
+      "learning_rate": 5.5054551250471985e-05,
+      "loss": 2.4074,
+      "step": 52100
+    },
+    {
+      "epoch": 49.24534088228356,
+      "grad_norm": 1.1095491647720337,
+      "learning_rate": 5.488088198447816e-05,
+      "loss": 2.4071,
+      "step": 52200
+    },
+    {
+      "epoch": 49.339702760084926,
+      "grad_norm": 1.0501704216003418,
+      "learning_rate": 5.470715324647766e-05,
+      "loss": 2.4103,
+      "step": 52300
+    },
+    {
+      "epoch": 49.434064637886294,
+      "grad_norm": 1.1576809883117676,
+      "learning_rate": 5.453336715330034e-05,
+      "loss": 2.4132,
+      "step": 52400
+    },
+    {
+      "epoch": 49.52842651568766,
+      "grad_norm": 1.136675238609314,
+      "learning_rate": 5.43595258224749e-05,
+      "loss": 2.4065,
+      "step": 52500
+    },
+    {
+      "epoch": 49.62278839348903,
+      "grad_norm": 1.1476448774337769,
+      "learning_rate": 5.4185631372203106e-05,
+      "loss": 2.4027,
+      "step": 52600
+    },
+    {
+      "epoch": 49.7171502712904,
+      "grad_norm": 1.0723971128463745,
+      "learning_rate": 5.401168592133394e-05,
+      "loss": 2.3952,
+      "step": 52700
+    },
+    {
+      "epoch": 49.81151214909177,
+      "grad_norm": 1.1071858406066895,
+      "learning_rate": 5.3837691589337833e-05,
+      "loss": 2.3957,
+      "step": 52800
+    },
+    {
+      "epoch": 49.90587402689314,
+      "grad_norm": 1.0979355573654175,
+      "learning_rate": 5.3663650496280814e-05,
+      "loss": 2.4061,
+      "step": 52900
+    },
+    {
+      "epoch": 50.0,
+      "grad_norm": 1.390066385269165,
+      "learning_rate": 5.348956476279867e-05,
+      "loss": 2.3988,
+      "step": 53000
+    },
+    {
+      "epoch": 50.0,
+      "eval_loss": 2.382521152496338,
+      "eval_runtime": 74.4633,
+      "eval_samples_per_second": 202.395,
+      "eval_steps_per_second": 6.325,
+      "step": 53000
+    },
+    {
+      "epoch": 50.09436187780137,
+      "grad_norm": 1.0886965990066528,
+      "learning_rate": 5.331543651007114e-05,
+      "loss": 2.3889,
+      "step": 53100
+    },
+    {
+      "epoch": 50.18872375560274,
+      "grad_norm": 1.1360448598861694,
+      "learning_rate": 5.314126785979601e-05,
+      "loss": 2.373,
+      "step": 53200
+    },
+    {
+      "epoch": 50.283085633404106,
+      "grad_norm": 1.1651926040649414,
+      "learning_rate": 5.296706093416334e-05,
+      "loss": 2.3825,
+      "step": 53300
+    },
+    {
+      "epoch": 50.377447511205474,
+      "grad_norm": 1.0949503183364868,
+      "learning_rate": 5.2792817855829534e-05,
+      "loss": 2.3856,
+      "step": 53400
+    },
+    {
+      "epoch": 50.47180938900684,
+      "grad_norm": 1.0986069440841675,
+      "learning_rate": 5.261854074789151e-05,
+      "loss": 2.3846,
+      "step": 53500
+    },
+    {
+      "epoch": 50.56617126680821,
+      "grad_norm": 1.0695897340774536,
+      "learning_rate": 5.244423173386084e-05,
+      "loss": 2.385,
+      "step": 53600
+    },
+    {
+      "epoch": 50.66053314460958,
+      "grad_norm": 1.084670066833496,
+      "learning_rate": 5.226989293763784e-05,
+      "loss": 2.3788,
+      "step": 53700
+    },
+    {
+      "epoch": 50.75489502241095,
+      "grad_norm": 1.0809028148651123,
+      "learning_rate": 5.2095526483485736e-05,
+      "loss": 2.381,
+      "step": 53800
+    },
+    {
+      "epoch": 50.84925690021232,
+      "grad_norm": 1.0939772129058838,
+      "learning_rate": 5.192113449600473e-05,
+      "loss": 2.3798,
+      "step": 53900
+    },
+    {
+      "epoch": 50.94361877801368,
+      "grad_norm": 1.149915099143982,
+      "learning_rate": 5.1746719100106164e-05,
+      "loss": 2.3664,
+      "step": 54000
+    },
+    {
+      "epoch": 50.94361877801368,
+      "eval_loss": 2.366403102874756,
+      "eval_runtime": 74.4788,
+      "eval_samples_per_second": 202.353,
+      "eval_steps_per_second": 6.324,
+      "step": 54000
+    },
+    {
+      "epoch": 51.03774475112055,
+      "grad_norm": 1.0868983268737793,
+      "learning_rate": 5.1572282420986615e-05,
+      "loss": 2.3738,
+      "step": 54100
+    },
+    {
+      "epoch": 51.13210662892192,
+      "grad_norm": 1.0766066312789917,
+      "learning_rate": 5.139782658410193e-05,
+      "loss": 2.3648,
+      "step": 54200
+    },
+    {
+      "epoch": 51.226468506723286,
+      "grad_norm": 1.1454092264175415,
+      "learning_rate": 5.122335371514144e-05,
+      "loss": 2.3695,
+      "step": 54300
+    },
+    {
+      "epoch": 51.320830384524655,
+      "grad_norm": 1.0294060707092285,
+      "learning_rate": 5.1048865940002e-05,
+      "loss": 2.3615,
+      "step": 54400
+    },
+    {
+      "epoch": 51.41519226232602,
+      "grad_norm": 1.1249427795410156,
+      "learning_rate": 5.0874365384762093e-05,
+      "loss": 2.3652,
+      "step": 54500
+    },
+    {
+      "epoch": 51.50955414012739,
+      "grad_norm": 1.0959460735321045,
+      "learning_rate": 5.069985417565589e-05,
+      "loss": 2.3635,
+      "step": 54600
+    },
+    {
+      "epoch": 51.60391601792875,
+      "grad_norm": 1.0896023511886597,
+      "learning_rate": 5.0525334439047435e-05,
+      "loss": 2.3554,
+      "step": 54700
+    },
+    {
+      "epoch": 51.69827789573012,
+      "grad_norm": 1.1385838985443115,
+      "learning_rate": 5.035080830140462e-05,
+      "loss": 2.3591,
+      "step": 54800
+    },
+    {
+      "epoch": 51.79263977353149,
+      "grad_norm": 1.1226471662521362,
+      "learning_rate": 5.017627788927336e-05,
+      "loss": 2.3575,
+      "step": 54900
+    },
+    {
+      "epoch": 51.88700165133286,
+      "grad_norm": 1.1267681121826172,
+      "learning_rate": 5.000174532925165e-05,
+      "loss": 2.3521,
+      "step": 55000
+    },
+    {
+      "epoch": 51.88700165133286,
+      "eval_loss": 2.3515806198120117,
+      "eval_runtime": 74.4619,
+      "eval_samples_per_second": 202.399,
+      "eval_steps_per_second": 6.325,
+      "step": 55000
+    },
+    {
+      "epoch": 51.98136352913423,
+      "grad_norm": 1.100253701210022,
+      "learning_rate": 4.982721274796365e-05,
+      "loss": 2.3561,
+      "step": 55100
+    },
+    {
+      "epoch": 52.0754895022411,
+      "grad_norm": 1.0981378555297852,
+      "learning_rate": 4.9652682272033776e-05,
+      "loss": 2.3454,
+      "step": 55200
+    },
+    {
+      "epoch": 52.169851380042466,
+      "grad_norm": 1.14932382106781,
+      "learning_rate": 4.947815602806083e-05,
+      "loss": 2.3457,
+      "step": 55300
+    },
+    {
+      "epoch": 52.26421325784383,
+      "grad_norm": 1.112889051437378,
+      "learning_rate": 4.9303636142592005e-05,
+      "loss": 2.3483,
+      "step": 55400
+    },
+    {
+      "epoch": 52.3585751356452,
+      "grad_norm": 1.1395831108093262,
+      "learning_rate": 4.912912474209699e-05,
+      "loss": 2.3499,
+      "step": 55500
+    },
+    {
+      "epoch": 52.452937013446565,
+      "grad_norm": 1.0740474462509155,
+      "learning_rate": 4.8954623952942196e-05,
+      "loss": 2.337,
+      "step": 55600
+    },
+    {
+      "epoch": 52.547298891247934,
+      "grad_norm": 1.0660241842269897,
+      "learning_rate": 4.878013590136461e-05,
+      "loss": 2.3385,
+      "step": 55700
+    },
+    {
+      "epoch": 52.6416607690493,
+      "grad_norm": 1.138458490371704,
+      "learning_rate": 4.860566271344612e-05,
+      "loss": 2.3486,
+      "step": 55800
+    },
+    {
+      "epoch": 52.73602264685067,
+      "grad_norm": 1.1012473106384277,
+      "learning_rate": 4.8431206515087425e-05,
+      "loss": 2.3311,
+      "step": 55900
+    },
+    {
+      "epoch": 52.83038452465204,
+      "grad_norm": 1.0752781629562378,
+      "learning_rate": 4.825676943198228e-05,
+      "loss": 2.3411,
+      "step": 56000
+    },
+    {
+      "epoch": 52.83038452465204,
+      "eval_loss": 2.3332431316375732,
+      "eval_runtime": 74.4827,
+      "eval_samples_per_second": 202.342,
+      "eval_steps_per_second": 6.324,
+      "step": 56000
+    },
+    {
+      "epoch": 52.92474640245341,
+      "grad_norm": 1.1414703130722046,
+      "learning_rate": 4.808235358959146e-05,
+      "loss": 2.3436,
+      "step": 56100
+    },
+    {
+      "epoch": 53.01887237556027,
+      "grad_norm": 1.0539082288742065,
+      "learning_rate": 4.790796111311697e-05,
+      "loss": 2.3465,
+      "step": 56200
+    },
+    {
+      "epoch": 53.11323425336164,
+      "grad_norm": 1.176776647567749,
+      "learning_rate": 4.773359412747614e-05,
+      "loss": 2.3285,
+      "step": 56300
+    },
+    {
+      "epoch": 53.20759613116301,
+      "grad_norm": 1.0964272022247314,
+      "learning_rate": 4.75592547572756e-05,
+      "loss": 2.3213,
+      "step": 56400
+    },
+    {
+      "epoch": 53.30195800896438,
+      "grad_norm": 1.0305556058883667,
+      "learning_rate": 4.738494512678562e-05,
+      "loss": 2.3247,
+      "step": 56500
+    },
+    {
+      "epoch": 53.396319886765745,
+      "grad_norm": 1.0632230043411255,
+      "learning_rate": 4.7210667359913984e-05,
+      "loss": 2.3313,
+      "step": 56600
+    },
+    {
+      "epoch": 53.490681764567114,
+      "grad_norm": 1.098626732826233,
+      "learning_rate": 4.7036423580180325e-05,
+      "loss": 2.3334,
+      "step": 56700
+    },
+    {
+      "epoch": 53.58504364236848,
+      "grad_norm": 1.1028692722320557,
+      "learning_rate": 4.6862215910690103e-05,
+      "loss": 2.3131,
+      "step": 56800
+    },
+    {
+      "epoch": 53.67940552016985,
+      "grad_norm": 1.1090211868286133,
+      "learning_rate": 4.668804647410876e-05,
+      "loss": 2.3232,
+      "step": 56900
+    },
+    {
+      "epoch": 53.77376739797122,
+      "grad_norm": 1.062287449836731,
+      "learning_rate": 4.6513917392635945e-05,
+      "loss": 2.3135,
+      "step": 57000
+    },
+    {
+      "epoch": 53.77376739797122,
+      "eval_loss": 2.322282314300537,
+      "eval_runtime": 74.4526,
+      "eval_samples_per_second": 202.424,
+      "eval_steps_per_second": 6.326,
+      "step": 57000
+    },
+    {
+      "epoch": 53.86812927577259,
+      "grad_norm": 1.0684916973114014,
+      "learning_rate": 4.6339830787979574e-05,
+      "loss": 2.3177,
+      "step": 57100
+    },
+    {
+      "epoch": 53.96249115357396,
+      "grad_norm": 1.158130407333374,
+      "learning_rate": 4.616578878132996e-05,
+      "loss": 2.3185,
+      "step": 57200
+    },
+    {
+      "epoch": 54.05661712668082,
+      "grad_norm": 1.0510168075561523,
+      "learning_rate": 4.5991793493334035e-05,
+      "loss": 2.3252,
+      "step": 57300
+    },
+    {
+      "epoch": 54.15097900448219,
+      "grad_norm": 1.1067076921463013,
+      "learning_rate": 4.58178470440695e-05,
+      "loss": 2.3158,
+      "step": 57400
+    },
+    {
+      "epoch": 54.24534088228356,
+      "grad_norm": 1.0520684719085693,
+      "learning_rate": 4.564395155301891e-05,
+      "loss": 2.3024,
+      "step": 57500
+    },
+    {
+      "epoch": 54.339702760084926,
+      "grad_norm": 1.1140958070755005,
+      "learning_rate": 4.5470109139043984e-05,
+      "loss": 2.3164,
+      "step": 57600
+    },
+    {
+      "epoch": 54.434064637886294,
+      "grad_norm": 1.0808565616607666,
+      "learning_rate": 4.529632192035965e-05,
+      "loss": 2.3067,
+      "step": 57700
+    },
+    {
+      "epoch": 54.52842651568766,
+      "grad_norm": 1.1175216436386108,
+      "learning_rate": 4.51225920145083e-05,
+      "loss": 2.3154,
+      "step": 57800
+    },
+    {
+      "epoch": 54.62278839348903,
+      "grad_norm": 1.0674301385879517,
+      "learning_rate": 4.494892153833406e-05,
+      "loss": 2.3054,
+      "step": 57900
+    },
+    {
+      "epoch": 54.7171502712904,
+      "grad_norm": 1.0364117622375488,
+      "learning_rate": 4.477531260795683e-05,
+      "loss": 2.3042,
+      "step": 58000
+    },
+    {
+      "epoch": 54.7171502712904,
+      "eval_loss": 2.3081283569335938,
+      "eval_runtime": 74.4358,
+      "eval_samples_per_second": 202.47,
+      "eval_steps_per_second": 6.328,
+      "step": 58000
+    },
+    {
+      "epoch": 54.81151214909177,
+      "grad_norm": 1.1012537479400635,
+      "learning_rate": 4.460176733874668e-05,
+      "loss": 2.2988,
+      "step": 58100
+    },
+    {
+      "epoch": 54.90587402689314,
+      "grad_norm": 1.0269622802734375,
+      "learning_rate": 4.442828784529791e-05,
+      "loss": 2.3111,
+      "step": 58200
+    },
+    {
+      "epoch": 55.0,
+      "grad_norm": 1.3486961126327515,
+      "learning_rate": 4.4254876241403444e-05,
+      "loss": 2.3058,
+      "step": 58300
+    },
+    {
+      "epoch": 55.09436187780137,
+      "grad_norm": 1.0656137466430664,
+      "learning_rate": 4.4081534640028924e-05,
+      "loss": 2.2969,
+      "step": 58400
+    },
+    {
+      "epoch": 55.18872375560274,
+      "grad_norm": 1.0997986793518066,
+      "learning_rate": 4.390826515328704e-05,
+      "loss": 2.2923,
+      "step": 58500
+    },
+    {
+      "epoch": 55.283085633404106,
+      "grad_norm": 1.061912178993225,
+      "learning_rate": 4.373506989241186e-05,
+      "loss": 2.2893,
+      "step": 58600
+    },
+    {
+      "epoch": 55.377447511205474,
+      "grad_norm": 1.12472403049469,
+      "learning_rate": 4.356195096773292e-05,
+      "loss": 2.2975,
+      "step": 58700
+    },
+    {
+      "epoch": 55.47180938900684,
+      "grad_norm": 1.0722994804382324,
+      "learning_rate": 4.338891048864973e-05,
+      "loss": 2.2996,
+      "step": 58800
+    },
+    {
+      "epoch": 55.56617126680821,
+      "grad_norm": 1.0319920778274536,
+      "learning_rate": 4.321595056360589e-05,
+      "loss": 2.2927,
+      "step": 58900
+    },
+    {
+      "epoch": 55.66053314460958,
+      "grad_norm": 1.0819764137268066,
+      "learning_rate": 4.304307330006352e-05,
+      "loss": 2.2922,
+      "step": 59000
+    },
+    {
+      "epoch": 55.66053314460958,
+      "eval_loss": 2.297449827194214,
+      "eval_runtime": 74.4766,
+      "eval_samples_per_second": 202.359,
+      "eval_steps_per_second": 6.324,
+      "step": 59000
+    },
+    {
+      "epoch": 55.75489502241095,
+      "grad_norm": 1.0915340185165405,
+      "learning_rate": 4.2870280804477525e-05,
+      "loss": 2.2955,
+      "step": 59100
+    },
+    {
+      "epoch": 55.84925690021232,
+      "grad_norm": 1.072871208190918,
+      "learning_rate": 4.2697575182269924e-05,
+      "loss": 2.2896,
+      "step": 59200
+    },
+    {
+      "epoch": 55.94361877801368,
+      "grad_norm": 1.1387864351272583,
+      "learning_rate": 4.2524958537804226e-05,
+      "loss": 2.2905,
+      "step": 59300
+    },
+    {
+      "epoch": 56.03774475112055,
+      "grad_norm": 1.0957666635513306,
+      "learning_rate": 4.235243297435975e-05,
+      "loss": 2.2877,
+      "step": 59400
+    },
+    {
+      "epoch": 56.13210662892192,
+      "grad_norm": 1.1053986549377441,
+      "learning_rate": 4.2180000594106076e-05,
+      "loss": 2.2804,
+      "step": 59500
+    },
+    {
+      "epoch": 56.226468506723286,
+      "grad_norm": 1.094224214553833,
+      "learning_rate": 4.200766349807731e-05,
+      "loss": 2.2813,
+      "step": 59600
+    },
+    {
+      "epoch": 56.320830384524655,
+      "grad_norm": 1.0825510025024414,
+      "learning_rate": 4.18354237861466e-05,
+      "loss": 2.2922,
+      "step": 59700
+    },
+    {
+      "epoch": 56.41519226232602,
+      "grad_norm": 1.0940465927124023,
+      "learning_rate": 4.1663283557000455e-05,
+      "loss": 2.2865,
+      "step": 59800
+    },
+    {
+      "epoch": 56.50955414012739,
+      "grad_norm": 1.053289771080017,
+      "learning_rate": 4.1491244908113266e-05,
+      "loss": 2.2806,
+      "step": 59900
+    },
+    {
+      "epoch": 56.60391601792875,
+      "grad_norm": 1.0980818271636963,
+      "learning_rate": 4.1319309935721695e-05,
+      "loss": 2.2792,
+      "step": 60000
+    },
+    {
+      "epoch": 56.60391601792875,
+      "eval_loss": 2.285236120223999,
+      "eval_runtime": 74.4453,
+      "eval_samples_per_second": 202.444,
+      "eval_steps_per_second": 6.327,
+      "step": 60000
+    },
+    {
+      "epoch": 56.69827789573012,
+      "grad_norm": 1.0499736070632935,
+      "learning_rate": 4.114748073479907e-05,
+      "loss": 2.2796,
+      "step": 60100
+    },
+    {
+      "epoch": 56.79263977353149,
+      "grad_norm": 1.0926116704940796,
+      "learning_rate": 4.097575939903003e-05,
+      "loss": 2.2816,
+      "step": 60200
+    },
+    {
+      "epoch": 56.88700165133286,
+      "grad_norm": 1.1019150018692017,
+      "learning_rate": 4.080414802078481e-05,
+      "loss": 2.2815,
+      "step": 60300
+    },
+    {
+      "epoch": 56.98136352913423,
+      "grad_norm": 1.0841056108474731,
+      "learning_rate": 4.063264869109395e-05,
+      "loss": 2.277,
+      "step": 60400
+    },
+    {
+      "epoch": 57.0754895022411,
+      "grad_norm": 1.0328353643417358,
+      "learning_rate": 4.046126349962261e-05,
+      "loss": 2.2802,
+      "step": 60500
+    },
+    {
+      "epoch": 57.169851380042466,
+      "grad_norm": 1.1018388271331787,
+      "learning_rate": 4.0289994534645305e-05,
+      "loss": 2.2737,
+      "step": 60600
+    },
+    {
+      "epoch": 57.26421325784383,
+      "grad_norm": 1.0932639837265015,
+      "learning_rate": 4.01188438830203e-05,
+      "loss": 2.2687,
+      "step": 60700
+    },
+    {
+      "epoch": 57.3585751356452,
+      "grad_norm": 1.0744473934173584,
+      "learning_rate": 3.994781363016427e-05,
+      "loss": 2.2597,
+      "step": 60800
+    },
+    {
+      "epoch": 57.452937013446565,
+      "grad_norm": 1.1107820272445679,
+      "learning_rate": 3.977690586002688e-05,
+      "loss": 2.2658,
+      "step": 60900
+    },
+    {
+      "epoch": 57.547298891247934,
+      "grad_norm": 1.0874978303909302,
+      "learning_rate": 3.9606122655065365e-05,
+      "loss": 2.2689,
+      "step": 61000
+    },
+    {
+      "epoch": 57.547298891247934,
+      "eval_loss": 2.2746315002441406,
+      "eval_runtime": 74.4452,
+      "eval_samples_per_second": 202.444,
+      "eval_steps_per_second": 6.327,
+      "step": 61000
+    },
+    {
+      "epoch": 57.6416607690493,
+      "grad_norm": 1.073294758796692,
+      "learning_rate": 3.943546609621921e-05,
+      "loss": 2.2597,
+      "step": 61100
+    },
+    {
+      "epoch": 57.73602264685067,
+      "grad_norm": 1.0983116626739502,
+      "learning_rate": 3.926493826288469e-05,
+      "loss": 2.2643,
+      "step": 61200
+    },
+    {
+      "epoch": 57.83038452465204,
+      "grad_norm": 1.0410650968551636,
+      "learning_rate": 3.909454123288968e-05,
+      "loss": 2.2739,
+      "step": 61300
+    },
+    {
+      "epoch": 57.92474640245341,
+      "grad_norm": 1.1241108179092407,
+      "learning_rate": 3.892427708246818e-05,
+      "loss": 2.2623,
+      "step": 61400
+    },
+    {
+      "epoch": 58.01887237556027,
+      "grad_norm": 1.1001108884811401,
+      "learning_rate": 3.875414788623515e-05,
+      "loss": 2.2569,
+      "step": 61500
+    },
+    {
+      "epoch": 58.11323425336164,
+      "grad_norm": 1.0980916023254395,
+      "learning_rate": 3.858415571716116e-05,
+      "loss": 2.2532,
+      "step": 61600
+    },
+    {
+      "epoch": 58.20759613116301,
+      "grad_norm": 1.1283693313598633,
+      "learning_rate": 3.8414302646547114e-05,
+      "loss": 2.2601,
+      "step": 61700
+    },
+    {
+      "epoch": 58.30195800896438,
+      "grad_norm": 1.0919461250305176,
+      "learning_rate": 3.824459074399911e-05,
+      "loss": 2.2589,
+      "step": 61800
+    },
+    {
+      "epoch": 58.396319886765745,
+      "grad_norm": 1.0695630311965942,
+      "learning_rate": 3.8075022077403095e-05,
+      "loss": 2.2421,
+      "step": 61900
+    },
+    {
+      "epoch": 58.490681764567114,
+      "grad_norm": 1.106446385383606,
+      "learning_rate": 3.790559871289979e-05,
+      "loss": 2.2574,
+      "step": 62000
+    },
+    {
+      "epoch": 58.490681764567114,
+      "eval_loss": 2.263988494873047,
+      "eval_runtime": 74.4531,
+      "eval_samples_per_second": 202.423,
+      "eval_steps_per_second": 6.326,
+      "step": 62000
+    },
+    {
+      "epoch": 58.58504364236848,
+      "grad_norm": 1.0683228969573975,
+      "learning_rate": 3.77363227148594e-05,
+      "loss": 2.2527,
+      "step": 62100
+    },
+    {
+      "epoch": 58.67940552016985,
+      "grad_norm": 1.1563806533813477,
+      "learning_rate": 3.756719614585656e-05,
+      "loss": 2.2533,
+      "step": 62200
+    },
+    {
+      "epoch": 58.77376739797122,
+      "grad_norm": 1.0460565090179443,
+      "learning_rate": 3.739822106664513e-05,
+      "loss": 2.2522,
+      "step": 62300
+    },
+    {
+      "epoch": 58.86812927577259,
+      "grad_norm": 1.110073208808899,
+      "learning_rate": 3.7229399536133106e-05,
+      "loss": 2.2516,
+      "step": 62400
+    },
+    {
+      "epoch": 58.96249115357396,
+      "grad_norm": 1.0516654253005981,
+      "learning_rate": 3.706073361135759e-05,
+      "loss": 2.2577,
+      "step": 62500
+    },
+    {
+      "epoch": 59.05661712668082,
+      "grad_norm": 1.1236464977264404,
+      "learning_rate": 3.6892225347459624e-05,
+      "loss": 2.2567,
+      "step": 62600
+    },
+    {
+      "epoch": 59.15097900448219,
+      "grad_norm": 1.089566946029663,
+      "learning_rate": 3.672387679765925e-05,
+      "loss": 2.2364,
+      "step": 62700
+    },
+    {
+      "epoch": 59.24534088228356,
+      "grad_norm": 1.051033854484558,
+      "learning_rate": 3.65556900132304e-05,
+      "loss": 2.2515,
+      "step": 62800
+    },
+    {
+      "epoch": 59.339702760084926,
+      "grad_norm": 1.058039903640747,
+      "learning_rate": 3.638766704347598e-05,
+      "loss": 2.2509,
+      "step": 62900
+    },
+    {
+      "epoch": 59.434064637886294,
+      "grad_norm": 1.0932092666625977,
+      "learning_rate": 3.621980993570283e-05,
+      "loss": 2.2361,
+      "step": 63000
+    },
+    {
+      "epoch": 59.434064637886294,
+      "eval_loss": 2.256521701812744,
+      "eval_runtime": 74.4604,
+      "eval_samples_per_second": 202.403,
+      "eval_steps_per_second": 6.326,
+      "step": 63000
+    },
+    {
+      "epoch": 59.52842651568766,
+      "grad_norm": 1.0966417789459229,
+      "learning_rate": 3.605212073519687e-05,
+      "loss": 2.2486,
+      "step": 63100
+    },
+    {
+      "epoch": 59.62278839348903,
+      "grad_norm": 1.0995506048202515,
+      "learning_rate": 3.588460148519808e-05,
+      "loss": 2.2425,
+      "step": 63200
+    },
+    {
+      "epoch": 59.7171502712904,
+      "grad_norm": 1.0373094081878662,
+      "learning_rate": 3.5717254226875605e-05,
+      "loss": 2.2398,
+      "step": 63300
+    },
+    {
+      "epoch": 59.81151214909177,
+      "grad_norm": 1.1059927940368652,
+      "learning_rate": 3.555008099930305e-05,
+      "loss": 2.2465,
+      "step": 63400
+    },
+    {
+      "epoch": 59.90587402689314,
+      "grad_norm": 1.104933261871338,
+      "learning_rate": 3.5383083839433385e-05,
+      "loss": 2.2386,
+      "step": 63500
+    },
+    {
+      "epoch": 60.0,
+      "grad_norm": 1.4146462678909302,
+      "learning_rate": 3.521626478207432e-05,
+      "loss": 2.2446,
+      "step": 63600
+    },
+    {
+      "epoch": 60.09436187780137,
+      "grad_norm": 1.073795199394226,
+      "learning_rate": 3.504962585986342e-05,
+      "loss": 2.2379,
+      "step": 63700
+    },
+    {
+      "epoch": 60.18872375560274,
+      "grad_norm": 1.1478886604309082,
+      "learning_rate": 3.488316910324338e-05,
+      "loss": 2.2373,
+      "step": 63800
+    },
+    {
+      "epoch": 60.283085633404106,
+      "grad_norm": 1.112461805343628,
+      "learning_rate": 3.471689654043724e-05,
+      "loss": 2.2395,
+      "step": 63900
+    },
+    {
+      "epoch": 60.377447511205474,
+      "grad_norm": 1.0776383876800537,
+      "learning_rate": 3.455081019742368e-05,
+      "loss": 2.2296,
+      "step": 64000
+    },
+    {
+      "epoch": 60.377447511205474,
+      "eval_loss": 2.2467448711395264,
+      "eval_runtime": 74.4598,
+      "eval_samples_per_second": 202.404,
+      "eval_steps_per_second": 6.326,
+      "step": 64000
+    },
+    {
+      "epoch": 60.47180938900684,
+      "grad_norm": 1.0833407640457153,
+      "learning_rate": 3.438491209791242e-05,
+      "loss": 2.2313,
+      "step": 64100
+    },
+    {
+      "epoch": 60.56617126680821,
+      "grad_norm": 1.1021618843078613,
+      "learning_rate": 3.42192042633194e-05,
+      "loss": 2.2326,
+      "step": 64200
+    },
+    {
+      "epoch": 60.66053314460958,
+      "grad_norm": 1.0688246488571167,
+      "learning_rate": 3.405368871274234e-05,
+      "loss": 2.2187,
+      "step": 64300
+    },
+    {
+      "epoch": 60.75489502241095,
+      "grad_norm": 1.0851497650146484,
+      "learning_rate": 3.3888367462935946e-05,
+      "loss": 2.2369,
+      "step": 64400
+    },
+    {
+      "epoch": 60.84925690021232,
+      "grad_norm": 1.0779963731765747,
+      "learning_rate": 3.3723242528287515e-05,
+      "loss": 2.2317,
+      "step": 64500
+    },
+    {
+      "epoch": 60.94361877801368,
+      "grad_norm": 1.113876461982727,
+      "learning_rate": 3.355831592079223e-05,
+      "loss": 2.2348,
+      "step": 64600
+    },
+    {
+      "epoch": 61.03774475112055,
+      "grad_norm": 1.0656490325927734,
+      "learning_rate": 3.3393589650028766e-05,
+      "loss": 2.2316,
+      "step": 64700
+    },
+    {
+      "epoch": 61.13210662892192,
+      "grad_norm": 1.1197242736816406,
+      "learning_rate": 3.322906572313477e-05,
+      "loss": 2.2145,
+      "step": 64800
+    },
+    {
+      "epoch": 61.226468506723286,
+      "grad_norm": 1.1261602640151978,
+      "learning_rate": 3.306474614478234e-05,
+      "loss": 2.2289,
+      "step": 64900
+    },
+    {
+      "epoch": 61.320830384524655,
+      "grad_norm": 1.1310150623321533,
+      "learning_rate": 3.2900632917153705e-05,
+      "loss": 2.2245,
+      "step": 65000
+    },
+    {
+      "epoch": 61.320830384524655,
+      "eval_loss": 2.2367091178894043,
+      "eval_runtime": 74.4387,
+      "eval_samples_per_second": 202.462,
+      "eval_steps_per_second": 6.327,
+      "step": 65000
+    },
+    {
+      "epoch": 61.41519226232602,
+      "grad_norm": 1.089492678642273,
+      "learning_rate": 3.273672803991673e-05,
+      "loss": 2.2191,
+      "step": 65100
+    },
+    {
+      "epoch": 61.50955414012739,
+      "grad_norm": 1.0696125030517578,
+      "learning_rate": 3.257303351020066e-05,
+      "loss": 2.2212,
+      "step": 65200
+    },
+    {
+      "epoch": 61.60391601792875,
+      "grad_norm": 1.116721749305725,
+      "learning_rate": 3.240955132257162e-05,
+      "loss": 2.2255,
+      "step": 65300
+    },
+    {
+      "epoch": 61.69827789573012,
+      "grad_norm": 1.1005743741989136,
+      "learning_rate": 3.224628346900853e-05,
+      "loss": 2.2215,
+      "step": 65400
+    },
+    {
+      "epoch": 61.79263977353149,
+      "grad_norm": 1.0481823682785034,
+      "learning_rate": 3.208323193887863e-05,
+      "loss": 2.2267,
+      "step": 65500
+    },
+    {
+      "epoch": 61.88700165133286,
+      "grad_norm": 1.0808483362197876,
+      "learning_rate": 3.192039871891336e-05,
+      "loss": 2.2148,
+      "step": 65600
+    },
+    {
+      "epoch": 61.98136352913423,
+      "grad_norm": 1.1039003133773804,
+      "learning_rate": 3.1757785793184144e-05,
+      "loss": 2.2253,
+      "step": 65700
+    },
+    {
+      "epoch": 62.0754895022411,
+      "grad_norm": 1.0520274639129639,
+      "learning_rate": 3.159539514307812e-05,
+      "loss": 2.2174,
+      "step": 65800
+    },
+    {
+      "epoch": 62.169851380042466,
+      "grad_norm": 1.1201003789901733,
+      "learning_rate": 3.143322874727417e-05,
+      "loss": 2.2051,
+      "step": 65900
+    },
+    {
+      "epoch": 62.26421325784383,
+      "grad_norm": 1.1086193323135376,
+      "learning_rate": 3.1271288581718586e-05,
+      "loss": 2.2222,
+      "step": 66000
+    },
+    {
+      "epoch": 62.26421325784383,
+      "eval_loss": 2.226095676422119,
+      "eval_runtime": 74.4774,
+      "eval_samples_per_second": 202.357,
+      "eval_steps_per_second": 6.324,
+      "step": 66000
+    },
+    {
+      "epoch": 62.3585751356452,
+      "grad_norm": 1.087605357170105,
+      "learning_rate": 3.1109576619601245e-05,
+      "loss": 2.2157,
+      "step": 66100
+    },
+    {
+      "epoch": 62.452937013446565,
+      "grad_norm": 1.1039416790008545,
+      "learning_rate": 3.0948094831331334e-05,
+      "loss": 2.2156,
+      "step": 66200
+    },
+    {
+      "epoch": 62.547298891247934,
+      "grad_norm": 1.0668971538543701,
+      "learning_rate": 3.078684518451346e-05,
+      "loss": 2.2123,
+      "step": 66300
+    },
+    {
+      "epoch": 62.6416607690493,
+      "grad_norm": 1.1140378713607788,
+      "learning_rate": 3.062582964392373e-05,
+      "loss": 2.2119,
+      "step": 66400
+    },
+    {
+      "epoch": 62.73602264685067,
+      "grad_norm": 1.0850133895874023,
+      "learning_rate": 3.0465050171485677e-05,
+      "loss": 2.2167,
+      "step": 66500
+    },
+    {
+      "epoch": 62.83038452465204,
+      "grad_norm": 1.1161259412765503,
+      "learning_rate": 3.0304508726246428e-05,
+      "loss": 2.2059,
+      "step": 66600
+    },
+    {
+      "epoch": 62.92474640245341,
+      "grad_norm": 1.0978820323944092,
+      "learning_rate": 3.0144207264352814e-05,
+      "loss": 2.2089,
+      "step": 66700
+    },
+    {
+      "epoch": 63.01887237556027,
+      "grad_norm": 1.107197880744934,
+      "learning_rate": 2.99841477390276e-05,
+      "loss": 2.2076,
+      "step": 66800
+    },
+    {
+      "epoch": 63.11323425336164,
+      "grad_norm": 1.0904182195663452,
+      "learning_rate": 2.982433210054557e-05,
+      "loss": 2.1997,
+      "step": 66900
+    },
+    {
+      "epoch": 63.20759613116301,
+      "grad_norm": 1.082038402557373,
+      "learning_rate": 2.9664762296209824e-05,
+      "loss": 2.2159,
+      "step": 67000
+    },
+    {
+      "epoch": 63.20759613116301,
+      "eval_loss": 2.2268171310424805,
+      "eval_runtime": 74.4765,
+      "eval_samples_per_second": 202.359,
+      "eval_steps_per_second": 6.324,
+      "step": 67000
+    },
+    {
+      "epoch": 63.30195800896438,
+      "grad_norm": 1.0698933601379395,
+      "learning_rate": 2.9505440270328112e-05,
+      "loss": 2.2073,
+      "step": 67100
+    },
+    {
+      "epoch": 63.396319886765745,
+      "grad_norm": 1.1355483531951904,
+      "learning_rate": 2.9346367964188992e-05,
+      "loss": 2.2007,
+      "step": 67200
+    },
+    {
+      "epoch": 63.490681764567114,
+      "grad_norm": 1.0896204710006714,
+      "learning_rate": 2.918754731603835e-05,
+      "loss": 2.2074,
+      "step": 67300
+    },
+    {
+      "epoch": 63.58504364236848,
+      "grad_norm": 1.0855309963226318,
+      "learning_rate": 2.9028980261055637e-05,
+      "loss": 2.2171,
+      "step": 67400
+    },
+    {
+      "epoch": 63.67940552016985,
+      "grad_norm": 1.0763812065124512,
+      "learning_rate": 2.8870668731330426e-05,
+      "loss": 2.2012,
+      "step": 67500
+    },
+    {
+      "epoch": 63.77376739797122,
+      "grad_norm": 1.0958815813064575,
+      "learning_rate": 2.8712614655838683e-05,
+      "loss": 2.2026,
+      "step": 67600
+    },
+    {
+      "epoch": 63.86812927577259,
+      "grad_norm": 1.1061776876449585,
+      "learning_rate": 2.8554819960419493e-05,
+      "loss": 2.1959,
+      "step": 67700
+    },
+    {
+      "epoch": 63.96249115357396,
+      "grad_norm": 1.0832457542419434,
+      "learning_rate": 2.8397286567751397e-05,
+      "loss": 2.1905,
+      "step": 67800
+    },
+    {
+      "epoch": 64.05661712668082,
+      "grad_norm": 1.0960924625396729,
+      "learning_rate": 2.824001639732905e-05,
+      "loss": 2.1935,
+      "step": 67900
+    },
+    {
+      "epoch": 64.1509790044822,
+      "grad_norm": 1.039017677307129,
+      "learning_rate": 2.8083011365439892e-05,
+      "loss": 2.1978,
+      "step": 68000
+    },
+    {
+      "epoch": 64.1509790044822,
+      "eval_loss": 2.213329315185547,
+      "eval_runtime": 74.4287,
+      "eval_samples_per_second": 202.489,
+      "eval_steps_per_second": 6.328,
+      "step": 68000
+    },
+    {
+      "epoch": 64.24534088228356,
+      "grad_norm": 1.052176594734192,
+      "learning_rate": 2.792627338514065e-05,
+      "loss": 2.1899,
+      "step": 68100
+    },
+    {
+      "epoch": 64.33970276008493,
+      "grad_norm": 1.1135153770446777,
+      "learning_rate": 2.7769804366234187e-05,
+      "loss": 2.1966,
+      "step": 68200
+    },
+    {
+      "epoch": 64.4340646378863,
+      "grad_norm": 1.0758830308914185,
+      "learning_rate": 2.7613606215246067e-05,
+      "loss": 2.2014,
+      "step": 68300
+    },
+    {
+      "epoch": 64.52842651568766,
+      "grad_norm": 1.0544888973236084,
+      "learning_rate": 2.7457680835401533e-05,
+      "loss": 2.1979,
+      "step": 68400
+    },
+    {
+      "epoch": 64.62278839348903,
+      "grad_norm": 1.0929769277572632,
+      "learning_rate": 2.730203012660209e-05,
+      "loss": 2.1956,
+      "step": 68500
+    },
+    {
+      "epoch": 64.7171502712904,
+      "grad_norm": 1.10369074344635,
+      "learning_rate": 2.714665598540249e-05,
+      "loss": 2.1912,
+      "step": 68600
+    },
+    {
+      "epoch": 64.81151214909177,
+      "grad_norm": 1.0878170728683472,
+      "learning_rate": 2.699156030498764e-05,
+      "loss": 2.1846,
+      "step": 68700
+    },
+    {
+      "epoch": 64.90587402689313,
+      "grad_norm": 1.1195570230484009,
+      "learning_rate": 2.6836744975149463e-05,
+      "loss": 2.1976,
+      "step": 68800
+    },
+    {
+      "epoch": 65.0,
+      "grad_norm": 1.416730523109436,
+      "learning_rate": 2.6682211882263873e-05,
+      "loss": 2.2005,
+      "step": 68900
+    },
+    {
+      "epoch": 65.09436187780136,
+      "grad_norm": 1.0857688188552856,
+      "learning_rate": 2.6527962909267856e-05,
+      "loss": 2.1879,
+      "step": 69000
+    },
+    {
+      "epoch": 65.09436187780136,
+      "eval_loss": 2.2124412059783936,
+      "eval_runtime": 74.4579,
+      "eval_samples_per_second": 202.41,
+      "eval_steps_per_second": 6.326,
+      "step": 69000
+    },
+    {
+      "epoch": 65.18872375560274,
+      "grad_norm": 1.096017837524414,
+      "learning_rate": 2.637399993563648e-05,
+      "loss": 2.1833,
+      "step": 69100
+    },
+    {
+      "epoch": 65.2830856334041,
+      "grad_norm": 1.0832054615020752,
+      "learning_rate": 2.6220324837359956e-05,
+      "loss": 2.193,
+      "step": 69200
+    },
+    {
+      "epoch": 65.37744751120547,
+      "grad_norm": 1.1171544790267944,
+      "learning_rate": 2.6066939486920904e-05,
+      "loss": 2.187,
+      "step": 69300
+    },
+    {
+      "epoch": 65.47180938900684,
+      "grad_norm": 1.0824764966964722,
+      "learning_rate": 2.5913845753271393e-05,
+      "loss": 2.1909,
+      "step": 69400
+    },
+    {
+      "epoch": 65.56617126680821,
+      "grad_norm": 1.1256800889968872,
+      "learning_rate": 2.5761045501810222e-05,
+      "loss": 2.1933,
+      "step": 69500
+    },
+    {
+      "epoch": 65.66053314460957,
+      "grad_norm": 1.117341160774231,
+      "learning_rate": 2.560854059436029e-05,
+      "loss": 2.189,
+      "step": 69600
+    },
+    {
+      "epoch": 65.75489502241095,
+      "grad_norm": 1.0918456315994263,
+      "learning_rate": 2.5456332889145718e-05,
+      "loss": 2.1832,
+      "step": 69700
+    },
+    {
+      "epoch": 65.84925690021231,
+      "grad_norm": 1.0791367292404175,
+      "learning_rate": 2.530442424076941e-05,
+      "loss": 2.1835,
+      "step": 69800
+    },
+    {
+      "epoch": 65.94361877801369,
+      "grad_norm": 1.0746275186538696,
+      "learning_rate": 2.5152816500190253e-05,
+      "loss": 2.1803,
+      "step": 69900
+    },
+    {
+      "epoch": 66.03774475112054,
+      "grad_norm": 1.0764819383621216,
+      "learning_rate": 2.500151151470077e-05,
+      "loss": 2.1937,
+      "step": 70000
+    },
+    {
+      "epoch": 66.03774475112054,
+      "eval_loss": 2.2022976875305176,
+      "eval_runtime": 74.4488,
+      "eval_samples_per_second": 202.434,
+      "eval_steps_per_second": 6.326,
+      "step": 70000
+    },
+    {
+      "epoch": 66.13210662892192,
+      "grad_norm": 1.081490397453308,
+      "learning_rate": 2.4850511127904437e-05,
+      "loss": 2.1802,
+      "step": 70100
+    },
+    {
+      "epoch": 66.22646850672328,
+      "grad_norm": 1.111242413520813,
+      "learning_rate": 2.469981717969329e-05,
+      "loss": 2.1761,
+      "step": 70200
+    },
+    {
+      "epoch": 66.32083038452465,
+      "grad_norm": 1.1172429323196411,
+      "learning_rate": 2.4549431506225586e-05,
+      "loss": 2.1866,
+      "step": 70300
+    },
+    {
+      "epoch": 66.41519226232602,
+      "grad_norm": 1.1029917001724243,
+      "learning_rate": 2.4399355939903245e-05,
+      "loss": 2.1849,
+      "step": 70400
+    },
+    {
+      "epoch": 66.50955414012739,
+      "grad_norm": 1.1302073001861572,
+      "learning_rate": 2.4249592309349728e-05,
+      "loss": 2.1863,
+      "step": 70500
+    },
+    {
+      "epoch": 66.60391601792875,
+      "grad_norm": 1.1342121362686157,
+      "learning_rate": 2.410014243938757e-05,
+      "loss": 2.1741,
+      "step": 70600
+    },
+    {
+      "epoch": 66.69827789573013,
+      "grad_norm": 1.1101330518722534,
+      "learning_rate": 2.3951008151016285e-05,
+      "loss": 2.1791,
+      "step": 70700
+    },
+    {
+      "epoch": 66.79263977353149,
+      "grad_norm": 1.1350260972976685,
+      "learning_rate": 2.380219126139014e-05,
+      "loss": 2.1803,
+      "step": 70800
+    },
+    {
+      "epoch": 66.88700165133287,
+      "grad_norm": 1.0897001028060913,
+      "learning_rate": 2.3653693583795932e-05,
+      "loss": 2.171,
+      "step": 70900
+    },
+    {
+      "epoch": 66.98136352913423,
+      "grad_norm": 1.1115797758102417,
+      "learning_rate": 2.3505516927631037e-05,
+      "loss": 2.1757,
+      "step": 71000
+    },
+    {
+      "epoch": 66.98136352913423,
+      "eval_loss": 2.197711706161499,
+      "eval_runtime": 74.4988,
+      "eval_samples_per_second": 202.299,
+      "eval_steps_per_second": 6.322,
+      "step": 71000
+    },
+    {
+      "epoch": 67.0754895022411,
+      "grad_norm": 1.1194738149642944,
+      "learning_rate": 2.3357663098381217e-05,
+      "loss": 2.1772,
+      "step": 71100
+    },
+    {
+      "epoch": 67.16985138004246,
+      "grad_norm": 1.0791974067687988,
+      "learning_rate": 2.3210133897598744e-05,
+      "loss": 2.1742,
+      "step": 71200
+    },
+    {
+      "epoch": 67.26421325784383,
+      "grad_norm": 1.1039748191833496,
+      "learning_rate": 2.3062931122880348e-05,
+      "loss": 2.17,
+      "step": 71300
+    },
+    {
+      "epoch": 67.3585751356452,
+      "grad_norm": 1.0685378313064575,
+      "learning_rate": 2.2916056567845418e-05,
+      "loss": 2.1741,
+      "step": 71400
+    },
+    {
+      "epoch": 67.45293701344657,
+      "grad_norm": 1.154982328414917,
+      "learning_rate": 2.276951202211402e-05,
+      "loss": 2.1699,
+      "step": 71500
+    },
+    {
+      "epoch": 67.54729889124793,
+      "grad_norm": 1.0707484483718872,
+      "learning_rate": 2.262329927128523e-05,
+      "loss": 2.1746,
+      "step": 71600
+    },
+    {
+      "epoch": 67.64166076904931,
+      "grad_norm": 1.0761550664901733,
+      "learning_rate": 2.2477420096915257e-05,
+      "loss": 2.1778,
+      "step": 71700
+    },
+    {
+      "epoch": 67.73602264685067,
+      "grad_norm": 1.1259441375732422,
+      "learning_rate": 2.2331876276495796e-05,
+      "loss": 2.1711,
+      "step": 71800
+    },
+    {
+      "epoch": 67.83038452465205,
+      "grad_norm": 1.1506456136703491,
+      "learning_rate": 2.218666958343239e-05,
+      "loss": 2.1731,
+      "step": 71900
+    },
+    {
+      "epoch": 67.92474640245341,
+      "grad_norm": 1.1114802360534668,
+      "learning_rate": 2.2041801787022742e-05,
+      "loss": 2.1747,
+      "step": 72000
+    },
+    {
+      "epoch": 67.92474640245341,
+      "eval_loss": 2.1910886764526367,
+      "eval_runtime": 74.4667,
+      "eval_samples_per_second": 202.386,
+      "eval_steps_per_second": 6.325,
+      "step": 72000
+    },
+    {
+      "epoch": 68.01887237556028,
+      "grad_norm": 1.0863537788391113,
+      "learning_rate": 2.189727465243527e-05,
+      "loss": 2.1734,
+      "step": 72100
+    },
+    {
+      "epoch": 68.11323425336164,
+      "grad_norm": 1.1258056163787842,
+      "learning_rate": 2.1753089940687456e-05,
+      "loss": 2.1707,
+      "step": 72200
+    },
+    {
+      "epoch": 68.20759613116302,
+      "grad_norm": 1.1622432470321655,
+      "learning_rate": 2.1609249408624547e-05,
+      "loss": 2.1651,
+      "step": 72300
+    },
+    {
+      "epoch": 68.30195800896438,
+      "grad_norm": 1.0975794792175293,
+      "learning_rate": 2.1465754808898007e-05,
+      "loss": 2.1642,
+      "step": 72400
+    },
+    {
+      "epoch": 68.39631988676575,
+      "grad_norm": 1.0919430255889893,
+      "learning_rate": 2.132260788994428e-05,
+      "loss": 2.1662,
+      "step": 72500
+    },
+    {
+      "epoch": 68.49068176456711,
+      "grad_norm": 1.0746418237686157,
+      "learning_rate": 2.1179810395963363e-05,
+      "loss": 2.1791,
+      "step": 72600
+    },
+    {
+      "epoch": 68.58504364236849,
+      "grad_norm": 1.0876717567443848,
+      "learning_rate": 2.1037364066897674e-05,
+      "loss": 2.1635,
+      "step": 72700
+    },
+    {
+      "epoch": 68.67940552016985,
+      "grad_norm": 1.1063474416732788,
+      "learning_rate": 2.0895270638410797e-05,
+      "loss": 2.1609,
+      "step": 72800
+    },
+    {
+      "epoch": 68.77376739797121,
+      "grad_norm": 1.1409770250320435,
+      "learning_rate": 2.0753531841866282e-05,
+      "loss": 2.1673,
+      "step": 72900
+    },
+    {
+      "epoch": 68.86812927577259,
+      "grad_norm": 1.0966746807098389,
+      "learning_rate": 2.0612149404306664e-05,
+      "loss": 2.1635,
+      "step": 73000
+    },
+    {
+      "epoch": 68.86812927577259,
+      "eval_loss": 2.1918320655822754,
+      "eval_runtime": 74.4341,
+      "eval_samples_per_second": 202.475,
+      "eval_steps_per_second": 6.328,
+      "step": 73000
+    },
+    {
+      "epoch": 68.96249115357395,
+      "grad_norm": 1.1267166137695312,
+      "learning_rate": 2.0471125048432317e-05,
+      "loss": 2.1717,
+      "step": 73100
+    },
+    {
+      "epoch": 69.05661712668082,
+      "grad_norm": 1.0646253824234009,
+      "learning_rate": 2.033046049258049e-05,
+      "loss": 2.1562,
+      "step": 73200
+    },
+    {
+      "epoch": 69.1509790044822,
+      "grad_norm": 1.1580528020858765,
+      "learning_rate": 2.0190157450704443e-05,
+      "loss": 2.1547,
+      "step": 73300
+    },
+    {
+      "epoch": 69.24534088228356,
+      "grad_norm": 1.0343353748321533,
+      "learning_rate": 2.005021763235243e-05,
+      "loss": 2.1636,
+      "step": 73400
+    },
+    {
+      "epoch": 69.33970276008493,
+      "grad_norm": 1.0552024841308594,
+      "learning_rate": 1.9910642742647013e-05,
+      "loss": 2.1634,
+      "step": 73500
+    },
+    {
+      "epoch": 69.4340646378863,
+      "grad_norm": 1.071376919746399,
+      "learning_rate": 1.9771434482264132e-05,
+      "loss": 2.1661,
+      "step": 73600
+    },
+    {
+      "epoch": 69.52842651568766,
+      "grad_norm": 1.0911625623703003,
+      "learning_rate": 1.963259454741255e-05,
+      "loss": 2.1656,
+      "step": 73700
+    },
+    {
+      "epoch": 69.62278839348903,
+      "grad_norm": 1.0990372896194458,
+      "learning_rate": 1.949412462981302e-05,
+      "loss": 2.1524,
+      "step": 73800
+    },
+    {
+      "epoch": 69.7171502712904,
+      "grad_norm": 1.093256950378418,
+      "learning_rate": 1.935602641667783e-05,
+      "loss": 2.1594,
+      "step": 73900
+    },
+    {
+      "epoch": 69.81151214909177,
+      "grad_norm": 1.104551911354065,
+      "learning_rate": 1.9218301590690103e-05,
+      "loss": 2.1617,
+      "step": 74000
+    },
+    {
+      "epoch": 69.81151214909177,
+      "eval_loss": 2.179980993270874,
+      "eval_runtime": 74.4861,
+      "eval_samples_per_second": 202.333,
+      "eval_steps_per_second": 6.323,
+      "step": 74000
+    },
+    {
+      "epoch": 69.90587402689313,
+      "grad_norm": 1.1245603561401367,
+      "learning_rate": 1.9080951829983358e-05,
+      "loss": 2.1602,
+      "step": 74100
+    },
+    {
+      "epoch": 70.0,
+      "grad_norm": 1.3696141242980957,
+      "learning_rate": 1.894397880812113e-05,
+      "loss": 2.1683,
+      "step": 74200
+    },
+    {
+      "epoch": 70.09436187780136,
+      "grad_norm": 1.111162543296814,
+      "learning_rate": 1.8807384194076426e-05,
+      "loss": 2.1507,
+      "step": 74300
+    },
+    {
+      "epoch": 70.18872375560274,
+      "grad_norm": 1.0995087623596191,
+      "learning_rate": 1.8671169652211524e-05,
+      "loss": 2.1524,
+      "step": 74400
+    },
+    {
+      "epoch": 70.2830856334041,
+      "grad_norm": 1.139926791191101,
+      "learning_rate": 1.8535336842257657e-05,
+      "loss": 2.1656,
+      "step": 74500
+    },
+    {
+      "epoch": 70.37744751120547,
+      "grad_norm": 1.114806056022644,
+      "learning_rate": 1.8399887419294696e-05,
+      "loss": 2.1443,
+      "step": 74600
+    },
+    {
+      "epoch": 70.47180938900684,
+      "grad_norm": 1.0854475498199463,
+      "learning_rate": 1.826482303373117e-05,
+      "loss": 2.1584,
+      "step": 74700
+    },
+    {
+      "epoch": 70.56617126680821,
+      "grad_norm": 1.087457299232483,
+      "learning_rate": 1.8130145331283948e-05,
+      "loss": 2.1424,
+      "step": 74800
+    },
+    {
+      "epoch": 70.66053314460957,
+      "grad_norm": 1.1035183668136597,
+      "learning_rate": 1.799585595295837e-05,
+      "loss": 2.1533,
+      "step": 74900
+    },
+    {
+      "epoch": 70.75489502241095,
+      "grad_norm": 1.1080502271652222,
+      "learning_rate": 1.786195653502809e-05,
+      "loss": 2.1583,
+      "step": 75000
+    },
+    {
+      "epoch": 70.75489502241095,
+      "eval_loss": 2.1788103580474854,
+      "eval_runtime": 74.4505,
+      "eval_samples_per_second": 202.43,
+      "eval_steps_per_second": 6.326,
+      "step": 75000
+    },
+    {
+      "epoch": 70.84925690021231,
+      "grad_norm": 1.12478506565094,
+      "learning_rate": 1.7728448709015304e-05,
+      "loss": 2.1595,
+      "step": 75100
+    },
+    {
+      "epoch": 70.94361877801369,
+      "grad_norm": 1.0864694118499756,
+      "learning_rate": 1.7595334101670703e-05,
+      "loss": 2.1529,
+      "step": 75200
+    },
+    {
+      "epoch": 71.03774475112054,
+      "grad_norm": 1.1255881786346436,
+      "learning_rate": 1.7462614334953798e-05,
+      "loss": 2.162,
+      "step": 75300
+    },
+    {
+      "epoch": 71.13210662892192,
+      "grad_norm": 1.1375105381011963,
+      "learning_rate": 1.733029102601303e-05,
+      "loss": 2.1481,
+      "step": 75400
+    },
+    {
+      "epoch": 71.22646850672328,
+      "grad_norm": 1.1133302450180054,
+      "learning_rate": 1.71983657871662e-05,
+      "loss": 2.1476,
+      "step": 75500
+    },
+    {
+      "epoch": 71.32083038452465,
+      "grad_norm": 1.090040683746338,
+      "learning_rate": 1.706684022588068e-05,
+      "loss": 2.153,
+      "step": 75600
+    },
+    {
+      "epoch": 71.41519226232602,
+      "grad_norm": 1.110629916191101,
+      "learning_rate": 1.6935715944753928e-05,
+      "loss": 2.147,
+      "step": 75700
+    },
+    {
+      "epoch": 71.50955414012739,
+      "grad_norm": 1.1086583137512207,
+      "learning_rate": 1.6804994541493953e-05,
+      "loss": 2.1422,
+      "step": 75800
+    },
+    {
+      "epoch": 71.60391601792875,
+      "grad_norm": 1.1097800731658936,
+      "learning_rate": 1.6674677608899763e-05,
+      "loss": 2.1469,
+      "step": 75900
+    },
+    {
+      "epoch": 71.69827789573013,
+      "grad_norm": 1.1248265504837036,
+      "learning_rate": 1.6544766734842093e-05,
+      "loss": 2.153,
+      "step": 76000
+    },
+    {
+      "epoch": 71.69827789573013,
+      "eval_loss": 2.1701321601867676,
+      "eval_runtime": 74.4907,
+      "eval_samples_per_second": 202.321,
+      "eval_steps_per_second": 6.323,
+      "step": 76000
+    },
+    {
+      "epoch": 71.79263977353149,
+      "grad_norm": 1.0759069919586182,
+      "learning_rate": 1.641526350224392e-05,
+      "loss": 2.1415,
+      "step": 76100
+    },
+    {
+      "epoch": 71.88700165133287,
+      "grad_norm": 1.1383440494537354,
+      "learning_rate": 1.628616948906129e-05,
+      "loss": 2.146,
+      "step": 76200
+    },
+    {
+      "epoch": 71.98136352913423,
+      "grad_norm": 1.1579523086547852,
+      "learning_rate": 1.615748626826398e-05,
+      "loss": 2.1467,
+      "step": 76300
+    },
+    {
+      "epoch": 72.0754895022411,
+      "grad_norm": 1.0908124446868896,
+      "learning_rate": 1.602921540781645e-05,
+      "loss": 2.1503,
+      "step": 76400
+    },
+    {
+      "epoch": 72.16985138004246,
+      "grad_norm": 1.1176483631134033,
+      "learning_rate": 1.5901358470658667e-05,
+      "loss": 2.1398,
+      "step": 76500
+    },
+    {
+      "epoch": 72.26421325784383,
+      "grad_norm": 1.0756431818008423,
+      "learning_rate": 1.5773917014687024e-05,
+      "loss": 2.1512,
+      "step": 76600
+    },
+    {
+      "epoch": 72.3585751356452,
+      "grad_norm": 1.0899015665054321,
+      "learning_rate": 1.5646892592735478e-05,
+      "loss": 2.1345,
+      "step": 76700
+    },
+    {
+      "epoch": 72.45293701344657,
+      "grad_norm": 1.118424892425537,
+      "learning_rate": 1.55202867525565e-05,
+      "loss": 2.1433,
+      "step": 76800
+    },
+    {
+      "epoch": 72.54729889124793,
+      "grad_norm": 1.1094263792037964,
+      "learning_rate": 1.5394101036802316e-05,
+      "loss": 2.1485,
+      "step": 76900
+    },
+    {
+      "epoch": 72.64166076904931,
+      "grad_norm": 1.0856695175170898,
+      "learning_rate": 1.5268336983006048e-05,
+      "loss": 2.1478,
+      "step": 77000
+    },
+    {
+      "epoch": 72.64166076904931,
+      "eval_loss": 2.1706361770629883,
+      "eval_runtime": 74.46,
+      "eval_samples_per_second": 202.404,
+      "eval_steps_per_second": 6.326,
+      "step": 77000
+    },
+    {
+      "epoch": 72.73602264685067,
+      "grad_norm": 1.095223069190979,
+      "learning_rate": 1.514299612356298e-05,
+      "loss": 2.1441,
+      "step": 77100
+    },
+    {
+      "epoch": 72.83038452465205,
+      "grad_norm": 1.1260662078857422,
+      "learning_rate": 1.5018079985711963e-05,
+      "loss": 2.1505,
+      "step": 77200
+    },
+    {
+      "epoch": 72.92474640245341,
+      "grad_norm": 1.1021870374679565,
+      "learning_rate": 1.4893590091516686e-05,
+      "loss": 2.1412,
+      "step": 77300
+    },
+    {
+      "epoch": 73.01887237556028,
+      "grad_norm": 1.0860086679458618,
+      "learning_rate": 1.4769527957847246e-05,
+      "loss": 2.1347,
+      "step": 77400
+    },
+    {
+      "epoch": 73.11323425336164,
+      "grad_norm": 1.0829050540924072,
+      "learning_rate": 1.4645895096361568e-05,
+      "loss": 2.135,
+      "step": 77500
+    },
+    {
+      "epoch": 73.20759613116302,
+      "grad_norm": 1.0621412992477417,
+      "learning_rate": 1.4522693013487077e-05,
+      "loss": 2.1357,
+      "step": 77600
+    },
+    {
+      "epoch": 73.30195800896438,
+      "grad_norm": 1.0788235664367676,
+      "learning_rate": 1.439992321040225e-05,
+      "loss": 2.143,
+      "step": 77700
+    },
+    {
+      "epoch": 73.39631988676575,
+      "grad_norm": 1.1159855127334595,
+      "learning_rate": 1.427758718301842e-05,
+      "loss": 2.1329,
+      "step": 77800
+    },
+    {
+      "epoch": 73.49068176456711,
+      "grad_norm": 1.1019457578659058,
+      "learning_rate": 1.4155686421961456e-05,
+      "loss": 2.1427,
+      "step": 77900
+    },
+    {
+      "epoch": 73.58504364236849,
+      "grad_norm": 1.1145848035812378,
+      "learning_rate": 1.4034222412553655e-05,
+      "loss": 2.1413,
+      "step": 78000
+    },
+    {
+      "epoch": 73.58504364236849,
+      "eval_loss": 2.165809154510498,
+      "eval_runtime": 74.4896,
+      "eval_samples_per_second": 202.324,
+      "eval_steps_per_second": 6.323,
+      "step": 78000
+    },
+    {
+      "epoch": 73.67940552016985,
+      "grad_norm": 1.0918891429901123,
+      "learning_rate": 1.3913196634795644e-05,
+      "loss": 2.1466,
+      "step": 78100
+    },
+    {
+      "epoch": 73.77376739797121,
+      "grad_norm": 1.0411089658737183,
+      "learning_rate": 1.3792610563348352e-05,
+      "loss": 2.1351,
+      "step": 78200
+    },
+    {
+      "epoch": 73.86812927577259,
+      "grad_norm": 1.089095950126648,
+      "learning_rate": 1.3672465667514977e-05,
+      "loss": 2.1422,
+      "step": 78300
+    },
+    {
+      "epoch": 73.96249115357395,
+      "grad_norm": 1.1013556718826294,
+      "learning_rate": 1.3552763411223173e-05,
+      "loss": 2.1326,
+      "step": 78400
+    },
+    {
+      "epoch": 74.05661712668082,
+      "grad_norm": 1.0978729724884033,
+      "learning_rate": 1.3433505253007172e-05,
+      "loss": 2.1358,
+      "step": 78500
+    },
+    {
+      "epoch": 74.1509790044822,
+      "grad_norm": 1.1121299266815186,
+      "learning_rate": 1.3314692645989978e-05,
+      "loss": 2.1314,
+      "step": 78600
+    },
+    {
+      "epoch": 74.24534088228356,
+      "grad_norm": 1.087158441543579,
+      "learning_rate": 1.3196327037865701e-05,
+      "loss": 2.1449,
+      "step": 78700
+    },
+    {
+      "epoch": 74.33970276008493,
+      "grad_norm": 1.1194955110549927,
+      "learning_rate": 1.3078409870881952e-05,
+      "loss": 2.1358,
+      "step": 78800
+    },
+    {
+      "epoch": 74.4340646378863,
+      "grad_norm": 1.1199134588241577,
+      "learning_rate": 1.2960942581822166e-05,
+      "loss": 2.1389,
+      "step": 78900
+    },
+    {
+      "epoch": 74.52842651568766,
+      "grad_norm": 1.0960466861724854,
+      "learning_rate": 1.2843926601988227e-05,
+      "loss": 2.1328,
+      "step": 79000
+    },
+    {
+      "epoch": 74.52842651568766,
+      "eval_loss": 2.165520429611206,
+      "eval_runtime": 74.4987,
+      "eval_samples_per_second": 202.299,
+      "eval_steps_per_second": 6.322,
+      "step": 79000
+    },
+    {
+      "epoch": 74.62278839348903,
+      "grad_norm": 1.1289753913879395,
+      "learning_rate": 1.272736335718288e-05,
+      "loss": 2.1309,
+      "step": 79100
+    },
+    {
+      "epoch": 74.7171502712904,
+      "grad_norm": 1.109093427658081,
+      "learning_rate": 1.2611254267692518e-05,
+      "loss": 2.1349,
+      "step": 79200
+    },
+    {
+      "epoch": 74.81151214909177,
+      "grad_norm": 1.1008985042572021,
+      "learning_rate": 1.2495600748269732e-05,
+      "loss": 2.1366,
+      "step": 79300
+    },
+    {
+      "epoch": 74.90587402689313,
+      "grad_norm": 1.0962027311325073,
+      "learning_rate": 1.2380404208116148e-05,
+      "loss": 2.1223,
+      "step": 79400
+    },
+    {
+      "epoch": 75.0,
+      "grad_norm": 1.3603086471557617,
+      "learning_rate": 1.2265666050865283e-05,
+      "loss": 2.1387,
+      "step": 79500
+    },
+    {
+      "epoch": 75.09436187780136,
+      "grad_norm": 1.0967055559158325,
+      "learning_rate": 1.215138767456534e-05,
+      "loss": 2.1333,
+      "step": 79600
+    },
+    {
+      "epoch": 75.18872375560274,
+      "grad_norm": 1.1414368152618408,
+      "learning_rate": 1.2037570471662307e-05,
+      "loss": 2.1335,
+      "step": 79700
+    },
+    {
+      "epoch": 75.2830856334041,
+      "grad_norm": 1.0993640422821045,
+      "learning_rate": 1.1924215828982842e-05,
+      "loss": 2.1304,
+      "step": 79800
+    },
+    {
+      "epoch": 75.37744751120547,
+      "grad_norm": 1.0678730010986328,
+      "learning_rate": 1.1811325127717544e-05,
+      "loss": 2.1344,
+      "step": 79900
+    },
+    {
+      "epoch": 75.47180938900684,
+      "grad_norm": 1.0797874927520752,
+      "learning_rate": 1.169889974340393e-05,
+      "loss": 2.1234,
+      "step": 80000
+    },
+    {
+      "epoch": 75.47180938900684,
+      "eval_loss": 2.1577999591827393,
+      "eval_runtime": 74.4676,
+      "eval_samples_per_second": 202.383,
+      "eval_steps_per_second": 6.325,
+      "step": 80000
+    }
+  ],
+  "logging_steps": 100,
+  "max_steps": 100000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 95,
+  "save_steps": 5000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.6943289668288e+18,
+  "train_batch_size": 32,
+  "trial_name": null,
+  "trial_params": null
+}