diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,7286 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 30.914898651313194,
+  "eval_steps": 1000,
+  "global_step": 49000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.031548229355627413,
+      "grad_norm": 32.0,
+      "learning_rate": 4.9e-07,
+      "loss": 17.4073,
+      "step": 50
+    },
+    {
+      "epoch": 0.06309645871125483,
+      "grad_norm": 30.625,
+      "learning_rate": 9.9e-07,
+      "loss": 17.6876,
+      "step": 100
+    },
+    {
+      "epoch": 0.09464468806688224,
+      "grad_norm": 32.25,
+      "learning_rate": 1.4900000000000001e-06,
+      "loss": 17.5782,
+      "step": 150
+    },
+    {
+      "epoch": 0.12619291742250965,
+      "grad_norm": 34.5,
+      "learning_rate": 1.99e-06,
+      "loss": 17.7185,
+      "step": 200
+    },
+    {
+      "epoch": 0.15774114677813708,
+      "grad_norm": 31.125,
+      "learning_rate": 2.49e-06,
+      "loss": 17.5788,
+      "step": 250
+    },
+    {
+      "epoch": 0.18928937613376448,
+      "grad_norm": 31.875,
+      "learning_rate": 2.99e-06,
+      "loss": 17.3918,
+      "step": 300
+    },
+    {
+      "epoch": 0.2208376054893919,
+      "grad_norm": 36.5,
+      "learning_rate": 3.49e-06,
+      "loss": 17.5019,
+      "step": 350
+    },
+    {
+      "epoch": 0.2523858348450193,
+      "grad_norm": 31.5,
+      "learning_rate": 3.99e-06,
+      "loss": 17.1858,
+      "step": 400
+    },
+    {
+      "epoch": 0.28393406420064676,
+      "grad_norm": 36.5,
+      "learning_rate": 4.49e-06,
+      "loss": 17.0212,
+      "step": 450
+    },
+    {
+      "epoch": 0.31548229355627416,
+      "grad_norm": 33.75,
+      "learning_rate": 4.9900000000000005e-06,
+      "loss": 16.7389,
+      "step": 500
+    },
+    {
+      "epoch": 0.34703052291190156,
+      "grad_norm": 35.5,
+      "learning_rate": 5.49e-06,
+      "loss": 16.4118,
+      "step": 550
+    },
+    {
+      "epoch": 0.37857875226752896,
+      "grad_norm": 31.625,
+      "learning_rate": 5.99e-06,
+      "loss": 16.4162,
+      "step": 600
+    },
+    {
+      "epoch": 0.4101269816231564,
+      "grad_norm": 63.5,
+      "learning_rate": 6.4900000000000005e-06,
+      "loss": 15.9327,
+      "step": 650
+    },
+    {
+      "epoch": 0.4416752109787838,
+      "grad_norm": 30.375,
+      "learning_rate": 6.990000000000001e-06,
+      "loss": 15.402,
+      "step": 700
+    },
+    {
+      "epoch": 0.4732234403344112,
+      "grad_norm": 27.625,
+      "learning_rate": 7.4899999999999994e-06,
+      "loss": 15.0267,
+      "step": 750
+    },
+    {
+      "epoch": 0.5047716696900386,
+      "grad_norm": 32.0,
+      "learning_rate": 7.99e-06,
+      "loss": 14.6909,
+      "step": 800
+    },
+    {
+      "epoch": 0.536319899045666,
+      "grad_norm": 34.0,
+      "learning_rate": 8.49e-06,
+      "loss": 14.1192,
+      "step": 850
+    },
+    {
+      "epoch": 0.5678681284012935,
+      "grad_norm": 30.25,
+      "learning_rate": 8.99e-06,
+      "loss": 13.2533,
+      "step": 900
+    },
+    {
+      "epoch": 0.5994163577569209,
+      "grad_norm": 30.625,
+      "learning_rate": 9.49e-06,
+      "loss": 12.4446,
+      "step": 950
+    },
+    {
+      "epoch": 0.6309645871125483,
+      "grad_norm": 28.5,
+      "learning_rate": 9.990000000000001e-06,
+      "loss": 11.4839,
+      "step": 1000
+    },
+    {
+      "epoch": 0.6309645871125483,
+      "eval_loss": 12.30927848815918,
+      "eval_runtime": 267.4428,
+      "eval_samples_per_second": 94.813,
+      "eval_steps_per_second": 5.927,
+      "step": 1000
+    },
+    {
+      "epoch": 0.6625128164681757,
+      "grad_norm": 23.25,
+      "learning_rate": 1.049e-05,
+      "loss": 10.7284,
+      "step": 1050
+    },
+    {
+      "epoch": 0.6940610458238031,
+      "grad_norm": 26.5,
+      "learning_rate": 1.099e-05,
+      "loss": 10.0224,
+      "step": 1100
+    },
+    {
+      "epoch": 0.7256092751794305,
+      "grad_norm": 21.25,
+      "learning_rate": 1.149e-05,
+      "loss": 9.6291,
+      "step": 1150
+    },
+    {
+      "epoch": 0.7571575045350579,
+      "grad_norm": 18.75,
+      "learning_rate": 1.199e-05,
+      "loss": 8.6819,
+      "step": 1200
+    },
+    {
+      "epoch": 0.7887057338906854,
+      "grad_norm": 19.5,
+      "learning_rate": 1.249e-05,
+      "loss": 8.564,
+      "step": 1250
+    },
+    {
+      "epoch": 0.8202539632463128,
+      "grad_norm": 19.125,
+      "learning_rate": 1.299e-05,
+      "loss": 7.7527,
+      "step": 1300
+    },
+    {
+      "epoch": 0.8518021926019402,
+      "grad_norm": 15.9375,
+      "learning_rate": 1.349e-05,
+      "loss": 7.3225,
+      "step": 1350
+    },
+    {
+      "epoch": 0.8833504219575676,
+      "grad_norm": 14.875,
+      "learning_rate": 1.399e-05,
+      "loss": 6.9508,
+      "step": 1400
+    },
+    {
+      "epoch": 0.914898651313195,
+      "grad_norm": 15.0625,
+      "learning_rate": 1.449e-05,
+      "loss": 6.516,
+      "step": 1450
+    },
+    {
+      "epoch": 0.9464468806688224,
+      "grad_norm": 13.6875,
+      "learning_rate": 1.499e-05,
+      "loss": 6.0257,
+      "step": 1500
+    },
+    {
+      "epoch": 0.9779951100244498,
+      "grad_norm": 11.0,
+      "learning_rate": 1.5490000000000002e-05,
+      "loss": 5.6735,
+      "step": 1550
+    },
+    {
+      "epoch": 1.0094644688066883,
+      "grad_norm": 15.75,
+      "learning_rate": 1.599e-05,
+      "loss": 5.0756,
+      "step": 1600
+    },
+    {
+      "epoch": 1.0410126981623156,
+      "grad_norm": 8.125,
+      "learning_rate": 1.649e-05,
+      "loss": 5.0824,
+      "step": 1650
+    },
+    {
+      "epoch": 1.072560927517943,
+      "grad_norm": 7.25,
+      "learning_rate": 1.699e-05,
+      "loss": 4.5511,
+      "step": 1700
+    },
+    {
+      "epoch": 1.1041091568735704,
+      "grad_norm": 9.0625,
+      "learning_rate": 1.749e-05,
+      "loss": 4.2476,
+      "step": 1750
+    },
+    {
+      "epoch": 1.1356573862291979,
+      "grad_norm": 4.125,
+      "learning_rate": 1.7990000000000002e-05,
+      "loss": 3.9672,
+      "step": 1800
+    },
+    {
+      "epoch": 1.1672056155848254,
+      "grad_norm": 5.25,
+      "learning_rate": 1.849e-05,
+      "loss": 3.8173,
+      "step": 1850
+    },
+    {
+      "epoch": 1.1987538449404527,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.8990000000000003e-05,
+      "loss": 3.5025,
+      "step": 1900
+    },
+    {
+      "epoch": 1.2303020742960802,
+      "grad_norm": 30.375,
+      "learning_rate": 1.949e-05,
+      "loss": 3.3593,
+      "step": 1950
+    },
+    {
+      "epoch": 1.2618503036517075,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.999e-05,
+      "loss": 3.2297,
+      "step": 2000
+    },
+    {
+      "epoch": 1.2618503036517075,
+      "eval_loss": 3.776749849319458,
+      "eval_runtime": 267.0502,
+      "eval_samples_per_second": 94.952,
+      "eval_steps_per_second": 5.935,
+      "step": 2000
+    },
+    {
+      "epoch": 1.293398533007335,
+      "grad_norm": 5.5625,
+      "learning_rate": 2.0490000000000002e-05,
+      "loss": 3.288,
+      "step": 2050
+    },
+    {
+      "epoch": 1.3249467623629623,
+      "grad_norm": 7.5625,
+      "learning_rate": 2.099e-05,
+      "loss": 2.9833,
+      "step": 2100
+    },
+    {
+      "epoch": 1.3564949917185898,
+      "grad_norm": 4.3125,
+      "learning_rate": 2.1490000000000003e-05,
+      "loss": 3.0733,
+      "step": 2150
+    },
+    {
+      "epoch": 1.3880432210742173,
+      "grad_norm": 3.796875,
+      "learning_rate": 2.199e-05,
+      "loss": 3.0319,
+      "step": 2200
+    },
+    {
+      "epoch": 1.4195914504298446,
+      "grad_norm": 3.78125,
+      "learning_rate": 2.249e-05,
+      "loss": 2.9367,
+      "step": 2250
+    },
+    {
+      "epoch": 1.451139679785472,
+      "grad_norm": 3.0,
+      "learning_rate": 2.2990000000000002e-05,
+      "loss": 2.7882,
+      "step": 2300
+    },
+    {
+      "epoch": 1.4826879091410994,
+      "grad_norm": 4.8125,
+      "learning_rate": 2.349e-05,
+      "loss": 2.7725,
+      "step": 2350
+    },
+    {
+      "epoch": 1.514236138496727,
+      "grad_norm": 3.21875,
+      "learning_rate": 2.3990000000000002e-05,
+      "loss": 2.619,
+      "step": 2400
+    },
+    {
+      "epoch": 1.5457843678523542,
+      "grad_norm": 2.515625,
+      "learning_rate": 2.449e-05,
+      "loss": 2.6745,
+      "step": 2450
+    },
+    {
+      "epoch": 1.5773325972079817,
+      "grad_norm": 3.109375,
+      "learning_rate": 2.4990000000000003e-05,
+      "loss": 2.5296,
+      "step": 2500
+    },
+    {
+      "epoch": 1.6088808265636092,
+      "grad_norm": 1.984375,
+      "learning_rate": 2.549e-05,
+      "loss": 2.6223,
+      "step": 2550
+    },
+    {
+      "epoch": 1.6404290559192365,
+      "grad_norm": 2.90625,
+      "learning_rate": 2.5990000000000004e-05,
+      "loss": 2.4497,
+      "step": 2600
+    },
+    {
+      "epoch": 1.671977285274864,
+      "grad_norm": 3.09375,
+      "learning_rate": 2.6490000000000002e-05,
+      "loss": 2.5528,
+      "step": 2650
+    },
+    {
+      "epoch": 1.7035255146304915,
+      "grad_norm": 4.53125,
+      "learning_rate": 2.6989999999999997e-05,
+      "loss": 2.4371,
+      "step": 2700
+    },
+    {
+      "epoch": 1.7350737439861188,
+      "grad_norm": 2.515625,
+      "learning_rate": 2.749e-05,
+      "loss": 2.3439,
+      "step": 2750
+    },
+    {
+      "epoch": 1.766621973341746,
+      "grad_norm": 2.703125,
+      "learning_rate": 2.7989999999999998e-05,
+      "loss": 2.3008,
+      "step": 2800
+    },
+    {
+      "epoch": 1.7981702026973736,
+      "grad_norm": 3.5625,
+      "learning_rate": 2.849e-05,
+      "loss": 2.429,
+      "step": 2850
+    },
+    {
+      "epoch": 1.829718432053001,
+      "grad_norm": 3.109375,
+      "learning_rate": 2.8990000000000002e-05,
+      "loss": 2.2902,
+      "step": 2900
+    },
+    {
+      "epoch": 1.8612666614086284,
+      "grad_norm": 2.578125,
+      "learning_rate": 2.949e-05,
+      "loss": 2.3473,
+      "step": 2950
+    },
+    {
+      "epoch": 1.892814890764256,
+      "grad_norm": 3.34375,
+      "learning_rate": 2.9990000000000003e-05,
+      "loss": 2.3278,
+      "step": 3000
+    },
+    {
+      "epoch": 1.892814890764256,
+      "eval_loss": 2.846893787384033,
+      "eval_runtime": 267.2695,
+      "eval_samples_per_second": 94.874,
+      "eval_steps_per_second": 5.93,
+      "step": 3000
+    },
+    {
+      "epoch": 1.9243631201198834,
+      "grad_norm": 1.9140625,
+      "learning_rate": 3.049e-05,
+      "loss": 2.2279,
+      "step": 3050
+    },
+    {
+      "epoch": 1.9559113494755107,
+      "grad_norm": 3.0625,
+      "learning_rate": 3.099e-05,
+      "loss": 2.1692,
+      "step": 3100
+    },
+    {
+      "epoch": 1.987459578831138,
+      "grad_norm": 2.703125,
+      "learning_rate": 3.1490000000000005e-05,
+      "loss": 2.1651,
+      "step": 3150
+    },
+    {
+      "epoch": 2.0189289376133766,
+      "grad_norm": 2.609375,
+      "learning_rate": 3.1990000000000004e-05,
+      "loss": 2.3521,
+      "step": 3200
+    },
+    {
+      "epoch": 2.050477166969004,
+      "grad_norm": 2.8125,
+      "learning_rate": 3.249e-05,
+      "loss": 2.1494,
+      "step": 3250
+    },
+    {
+      "epoch": 2.082025396324631,
+      "grad_norm": 2.3125,
+      "learning_rate": 3.299e-05,
+      "loss": 2.1754,
+      "step": 3300
+    },
+    {
+      "epoch": 2.113573625680259,
+      "grad_norm": 2.15625,
+      "learning_rate": 3.349e-05,
+      "loss": 2.1549,
+      "step": 3350
+    },
+    {
+      "epoch": 2.145121855035886,
+      "grad_norm": 2.09375,
+      "learning_rate": 3.399e-05,
+      "loss": 2.1574,
+      "step": 3400
+    },
+    {
+      "epoch": 2.1766700843915134,
+      "grad_norm": 2.421875,
+      "learning_rate": 3.449e-05,
+      "loss": 2.075,
+      "step": 3450
+    },
+    {
+      "epoch": 2.2082183137471407,
+      "grad_norm": 3.140625,
+      "learning_rate": 3.499e-05,
+      "loss": 2.1074,
+      "step": 3500
+    },
+    {
+      "epoch": 2.2397665431027685,
+      "grad_norm": 6.34375,
+      "learning_rate": 3.549e-05,
+      "loss": 2.083,
+      "step": 3550
+    },
+    {
+      "epoch": 2.2713147724583957,
+      "grad_norm": 2.15625,
+      "learning_rate": 3.599e-05,
+      "loss": 2.1259,
+      "step": 3600
+    },
+    {
+      "epoch": 2.302863001814023,
+      "grad_norm": 2.03125,
+      "learning_rate": 3.6490000000000005e-05,
+      "loss": 2.0767,
+      "step": 3650
+    },
+    {
+      "epoch": 2.3344112311696508,
+      "grad_norm": 2.609375,
+      "learning_rate": 3.699e-05,
+      "loss": 2.027,
+      "step": 3700
+    },
+    {
+      "epoch": 2.365959460525278,
+      "grad_norm": 2.140625,
+      "learning_rate": 3.749e-05,
+      "loss": 2.1235,
+      "step": 3750
+    },
+    {
+      "epoch": 2.3975076898809053,
+      "grad_norm": 2.203125,
+      "learning_rate": 3.799e-05,
+      "loss": 1.9496,
+      "step": 3800
+    },
+    {
+      "epoch": 2.429055919236533,
+      "grad_norm": 1.8828125,
+      "learning_rate": 3.8490000000000006e-05,
+      "loss": 1.8565,
+      "step": 3850
+    },
+    {
+      "epoch": 2.4606041485921604,
+      "grad_norm": 2.46875,
+      "learning_rate": 3.8990000000000004e-05,
+      "loss": 1.9914,
+      "step": 3900
+    },
+    {
+      "epoch": 2.4921523779477877,
+      "grad_norm": 1.5078125,
+      "learning_rate": 3.9489999999999996e-05,
+      "loss": 1.9249,
+      "step": 3950
+    },
+    {
+      "epoch": 2.523700607303415,
+      "grad_norm": 2.671875,
+      "learning_rate": 3.999e-05,
+      "loss": 2.0286,
+      "step": 4000
+    },
+    {
+      "epoch": 2.523700607303415,
+      "eval_loss": 2.5390400886535645,
+      "eval_runtime": 268.2567,
+      "eval_samples_per_second": 94.525,
+      "eval_steps_per_second": 5.909,
+      "step": 4000
+    },
+    {
+      "epoch": 2.5552488366590427,
+      "grad_norm": 2.03125,
+      "learning_rate": 4.049e-05,
+      "loss": 1.895,
+      "step": 4050
+    },
+    {
+      "epoch": 2.58679706601467,
+      "grad_norm": 2.015625,
+      "learning_rate": 4.099e-05,
+      "loss": 2.037,
+      "step": 4100
+    },
+    {
+      "epoch": 2.6183452953702973,
+      "grad_norm": 2.40625,
+      "learning_rate": 4.1490000000000004e-05,
+      "loss": 1.7669,
+      "step": 4150
+    },
+    {
+      "epoch": 2.6498935247259245,
+      "grad_norm": 2.4375,
+      "learning_rate": 4.199e-05,
+      "loss": 1.8471,
+      "step": 4200
+    },
+    {
+      "epoch": 2.6814417540815523,
+      "grad_norm": 1.7578125,
+      "learning_rate": 4.249e-05,
+      "loss": 2.0354,
+      "step": 4250
+    },
+    {
+      "epoch": 2.7129899834371796,
+      "grad_norm": 2.390625,
+      "learning_rate": 4.299e-05,
+      "loss": 1.9273,
+      "step": 4300
+    },
+    {
+      "epoch": 2.744538212792807,
+      "grad_norm": 1.75,
+      "learning_rate": 4.3490000000000005e-05,
+      "loss": 1.951,
+      "step": 4350
+    },
+    {
+      "epoch": 2.7760864421484346,
+      "grad_norm": 1.6171875,
+      "learning_rate": 4.3990000000000004e-05,
+      "loss": 1.8838,
+      "step": 4400
+    },
+    {
+      "epoch": 2.807634671504062,
+      "grad_norm": 1.625,
+      "learning_rate": 4.449e-05,
+      "loss": 1.9785,
+      "step": 4450
+    },
+    {
+      "epoch": 2.839182900859689,
+      "grad_norm": 1.796875,
+      "learning_rate": 4.499e-05,
+      "loss": 1.8728,
+      "step": 4500
+    },
+    {
+      "epoch": 2.870731130215317,
+      "grad_norm": 1.6015625,
+      "learning_rate": 4.549000000000001e-05,
+      "loss": 1.914,
+      "step": 4550
+    },
+    {
+      "epoch": 2.902279359570944,
+      "grad_norm": 2.8125,
+      "learning_rate": 4.599e-05,
+      "loss": 1.8725,
+      "step": 4600
+    },
+    {
+      "epoch": 2.9338275889265715,
+      "grad_norm": 2.265625,
+      "learning_rate": 4.649e-05,
+      "loss": 1.9947,
+      "step": 4650
+    },
+    {
+      "epoch": 2.9653758182821988,
+      "grad_norm": 2.03125,
+      "learning_rate": 4.699e-05,
+      "loss": 1.7939,
+      "step": 4700
+    },
+    {
+      "epoch": 2.996924047637826,
+      "grad_norm": 1.953125,
+      "learning_rate": 4.749e-05,
+      "loss": 1.7749,
+      "step": 4750
+    },
+    {
+      "epoch": 3.0283934064200646,
+      "grad_norm": 1.6015625,
+      "learning_rate": 4.799e-05,
+      "loss": 1.9298,
+      "step": 4800
+    },
+    {
+      "epoch": 3.059941635775692,
+      "grad_norm": 2.28125,
+      "learning_rate": 4.8490000000000005e-05,
+      "loss": 1.816,
+      "step": 4850
+    },
+    {
+      "epoch": 3.0914898651313196,
+      "grad_norm": 3.078125,
+      "learning_rate": 4.8990000000000004e-05,
+      "loss": 1.7261,
+      "step": 4900
+    },
+    {
+      "epoch": 3.123038094486947,
+      "grad_norm": 1.5234375,
+      "learning_rate": 4.949e-05,
+      "loss": 1.7376,
+      "step": 4950
+    },
+    {
+      "epoch": 3.154586323842574,
+      "grad_norm": 1.6328125,
+      "learning_rate": 4.999e-05,
+      "loss": 1.8473,
+      "step": 5000
+    },
+    {
+      "epoch": 3.154586323842574,
+      "eval_loss": 2.3419270515441895,
+      "eval_runtime": 267.4485,
+      "eval_samples_per_second": 94.811,
+      "eval_steps_per_second": 5.926,
+      "step": 5000
+    },
+    {
+      "epoch": 3.186134553198202,
+      "grad_norm": 1.25,
+      "learning_rate": 5.0490000000000006e-05,
+      "loss": 1.7936,
+      "step": 5050
+    },
+    {
+      "epoch": 3.2176827825538292,
+      "grad_norm": 2.171875,
+      "learning_rate": 5.0990000000000005e-05,
+      "loss": 1.8275,
+      "step": 5100
+    },
+    {
+      "epoch": 3.2492310119094565,
+      "grad_norm": 2.90625,
+      "learning_rate": 5.149e-05,
+      "loss": 1.8632,
+      "step": 5150
+    },
+    {
+      "epoch": 3.2807792412650842,
+      "grad_norm": 1.8515625,
+      "learning_rate": 5.199000000000001e-05,
+      "loss": 1.5972,
+      "step": 5200
+    },
+    {
+      "epoch": 3.3123274706207115,
+      "grad_norm": 1.546875,
+      "learning_rate": 5.249000000000001e-05,
+      "loss": 1.7239,
+      "step": 5250
+    },
+    {
+      "epoch": 3.343875699976339,
+      "grad_norm": 2.84375,
+      "learning_rate": 5.2990000000000006e-05,
+      "loss": 1.8685,
+      "step": 5300
+    },
+    {
+      "epoch": 3.375423929331966,
+      "grad_norm": 1.703125,
+      "learning_rate": 5.3490000000000005e-05,
+      "loss": 1.7002,
+      "step": 5350
+    },
+    {
+      "epoch": 3.406972158687594,
+      "grad_norm": 1.296875,
+      "learning_rate": 5.399000000000001e-05,
+      "loss": 1.8572,
+      "step": 5400
+    },
+    {
+      "epoch": 3.438520388043221,
+      "grad_norm": 1.5859375,
+      "learning_rate": 5.449000000000001e-05,
+      "loss": 1.6519,
+      "step": 5450
+    },
+    {
+      "epoch": 3.4700686173988484,
+      "grad_norm": 1.703125,
+      "learning_rate": 5.499000000000001e-05,
+      "loss": 1.7819,
+      "step": 5500
+    },
+    {
+      "epoch": 3.5016168467544757,
+      "grad_norm": 2.421875,
+      "learning_rate": 5.549e-05,
+      "loss": 1.6989,
+      "step": 5550
+    },
+    {
+      "epoch": 3.5331650761101034,
+      "grad_norm": 2.15625,
+      "learning_rate": 5.599e-05,
+      "loss": 1.6876,
+      "step": 5600
+    },
+    {
+      "epoch": 3.5647133054657307,
+      "grad_norm": 1.6328125,
+      "learning_rate": 5.6489999999999996e-05,
+      "loss": 1.6419,
+      "step": 5650
+    },
+    {
+      "epoch": 3.596261534821358,
+      "grad_norm": 1.53125,
+      "learning_rate": 5.699e-05,
+      "loss": 1.7303,
+      "step": 5700
+    },
+    {
+      "epoch": 3.6278097641769858,
+      "grad_norm": 2.109375,
+      "learning_rate": 5.749e-05,
+      "loss": 1.6267,
+      "step": 5750
+    },
+    {
+      "epoch": 3.659357993532613,
+      "grad_norm": 1.9765625,
+      "learning_rate": 5.799e-05,
+      "loss": 1.7364,
+      "step": 5800
+    },
+    {
+      "epoch": 3.6909062228882403,
+      "grad_norm": 2.140625,
+      "learning_rate": 5.849e-05,
+      "loss": 1.8087,
+      "step": 5850
+    },
+    {
+      "epoch": 3.722454452243868,
+      "grad_norm": 1.859375,
+      "learning_rate": 5.899e-05,
+      "loss": 1.532,
+      "step": 5900
+    },
+    {
+      "epoch": 3.7540026815994954,
+      "grad_norm": 1.6015625,
+      "learning_rate": 5.949e-05,
+      "loss": 1.7565,
+      "step": 5950
+    },
+    {
+      "epoch": 3.7855509109551226,
+      "grad_norm": 1.5703125,
+      "learning_rate": 5.999e-05,
+      "loss": 1.5554,
+      "step": 6000
+    },
+    {
+      "epoch": 3.7855509109551226,
+      "eval_loss": 2.193551778793335,
+      "eval_runtime": 267.7225,
+      "eval_samples_per_second": 94.714,
+      "eval_steps_per_second": 5.92,
+      "step": 6000
+    },
+    {
+      "epoch": 3.81709914031075,
+      "grad_norm": 1.78125,
+      "learning_rate": 6.0490000000000005e-05,
+      "loss": 1.6697,
+      "step": 6050
+    },
+    {
+      "epoch": 3.848647369666377,
+      "grad_norm": 1.6171875,
+      "learning_rate": 6.0990000000000004e-05,
+      "loss": 1.6499,
+      "step": 6100
+    },
+    {
+      "epoch": 3.880195599022005,
+      "grad_norm": 1.796875,
+      "learning_rate": 6.149000000000001e-05,
+      "loss": 1.6098,
+      "step": 6150
+    },
+    {
+      "epoch": 3.9117438283776322,
+      "grad_norm": 1.5078125,
+      "learning_rate": 6.199000000000001e-05,
+      "loss": 1.6471,
+      "step": 6200
+    },
+    {
+      "epoch": 3.9432920577332595,
+      "grad_norm": 1.9296875,
+      "learning_rate": 6.249e-05,
+      "loss": 1.6636,
+      "step": 6250
+    },
+    {
+      "epoch": 3.9748402870888873,
+      "grad_norm": 1.5859375,
+      "learning_rate": 6.299e-05,
+      "loss": 1.68,
+      "step": 6300
+    },
+    {
+      "epoch": 4.006309645871125,
+      "grad_norm": 1.453125,
+      "learning_rate": 6.349e-05,
+      "loss": 1.6271,
+      "step": 6350
+    },
+    {
+      "epoch": 4.037857875226753,
+      "grad_norm": 1.3515625,
+      "learning_rate": 6.399e-05,
+      "loss": 1.708,
+      "step": 6400
+    },
+    {
+      "epoch": 4.06940610458238,
+      "grad_norm": 1.625,
+      "learning_rate": 6.449e-05,
+      "loss": 1.5844,
+      "step": 6450
+    },
+    {
+      "epoch": 4.100954333938008,
+      "grad_norm": 1.546875,
+      "learning_rate": 6.499000000000001e-05,
+      "loss": 1.7148,
+      "step": 6500
+    },
+    {
+      "epoch": 4.132502563293635,
+      "grad_norm": 1.640625,
+      "learning_rate": 6.549000000000001e-05,
+      "loss": 1.657,
+      "step": 6550
+    },
+    {
+      "epoch": 4.164050792649262,
+      "grad_norm": 7.09375,
+      "learning_rate": 6.599000000000001e-05,
+      "loss": 1.5734,
+      "step": 6600
+    },
+    {
+      "epoch": 4.19559902200489,
+      "grad_norm": 1.7265625,
+      "learning_rate": 6.649000000000001e-05,
+      "loss": 1.6004,
+      "step": 6650
+    },
+    {
+      "epoch": 4.227147251360518,
+      "grad_norm": 2.125,
+      "learning_rate": 6.699000000000001e-05,
+      "loss": 1.5555,
+      "step": 6700
+    },
+    {
+      "epoch": 4.258695480716145,
+      "grad_norm": 1.5859375,
+      "learning_rate": 6.749e-05,
+      "loss": 1.5842,
+      "step": 6750
+    },
+    {
+      "epoch": 4.290243710071772,
+      "grad_norm": 1.8203125,
+      "learning_rate": 6.799e-05,
+      "loss": 1.4458,
+      "step": 6800
+    },
+    {
+      "epoch": 4.3217919394274,
+      "grad_norm": 1.8515625,
+      "learning_rate": 6.849e-05,
+      "loss": 1.5308,
+      "step": 6850
+    },
+    {
+      "epoch": 4.353340168783027,
+      "grad_norm": 2.890625,
+      "learning_rate": 6.899e-05,
+      "loss": 1.6275,
+      "step": 6900
+    },
+    {
+      "epoch": 4.384888398138655,
+      "grad_norm": 1.5,
+      "learning_rate": 6.949e-05,
+      "loss": 1.6239,
+      "step": 6950
+    },
+    {
+      "epoch": 4.4164366274942815,
+      "grad_norm": 2.703125,
+      "learning_rate": 6.999e-05,
+      "loss": 1.6261,
+      "step": 7000
+    },
+    {
+      "epoch": 4.4164366274942815,
+      "eval_loss": 2.089057207107544,
+      "eval_runtime": 266.8117,
+      "eval_samples_per_second": 95.037,
+      "eval_steps_per_second": 5.941,
+      "step": 7000
+    },
+    {
+      "epoch": 4.447984856849909,
+      "grad_norm": 1.25,
+      "learning_rate": 7.049e-05,
+      "loss": 1.6696,
+      "step": 7050
+    },
+    {
+      "epoch": 4.479533086205537,
+      "grad_norm": 2.546875,
+      "learning_rate": 7.099e-05,
+      "loss": 1.5402,
+      "step": 7100
+    },
+    {
+      "epoch": 4.511081315561164,
+      "grad_norm": 1.703125,
+      "learning_rate": 7.149e-05,
+      "loss": 1.5104,
+      "step": 7150
+    },
+    {
+      "epoch": 4.5426295449167915,
+      "grad_norm": 5.3125,
+      "learning_rate": 7.199000000000001e-05,
+      "loss": 1.5624,
+      "step": 7200
+    },
+    {
+      "epoch": 4.574177774272419,
+      "grad_norm": 1.2109375,
+      "learning_rate": 7.249e-05,
+      "loss": 1.4863,
+      "step": 7250
+    },
+    {
+      "epoch": 4.605726003628046,
+      "grad_norm": 1.78125,
+      "learning_rate": 7.299e-05,
+      "loss": 1.5396,
+      "step": 7300
+    },
+    {
+      "epoch": 4.637274232983674,
+      "grad_norm": 1.9765625,
+      "learning_rate": 7.349e-05,
+      "loss": 1.5752,
+      "step": 7350
+    },
+    {
+      "epoch": 4.6688224623393015,
+      "grad_norm": 1.3203125,
+      "learning_rate": 7.399e-05,
+      "loss": 1.596,
+      "step": 7400
+    },
+    {
+      "epoch": 4.700370691694928,
+      "grad_norm": 1.609375,
+      "learning_rate": 7.449e-05,
+      "loss": 1.5409,
+      "step": 7450
+    },
+    {
+      "epoch": 4.731918921050556,
+      "grad_norm": 1.75,
+      "learning_rate": 7.499e-05,
+      "loss": 1.4706,
+      "step": 7500
+    },
+    {
+      "epoch": 4.763467150406184,
+      "grad_norm": 1.4921875,
+      "learning_rate": 7.549000000000001e-05,
+      "loss": 1.6109,
+      "step": 7550
+    },
+    {
+      "epoch": 4.795015379761811,
+      "grad_norm": 1.2578125,
+      "learning_rate": 7.599000000000001e-05,
+      "loss": 1.4949,
+      "step": 7600
+    },
+    {
+      "epoch": 4.826563609117438,
+      "grad_norm": 2.5,
+      "learning_rate": 7.649000000000001e-05,
+      "loss": 1.5716,
+      "step": 7650
+    },
+    {
+      "epoch": 4.858111838473066,
+      "grad_norm": 2.21875,
+      "learning_rate": 7.699e-05,
+      "loss": 1.4256,
+      "step": 7700
+    },
+    {
+      "epoch": 4.889660067828693,
+      "grad_norm": 1.7109375,
+      "learning_rate": 7.749e-05,
+      "loss": 1.4907,
+      "step": 7750
+    },
+    {
+      "epoch": 4.921208297184321,
+      "grad_norm": 1.328125,
+      "learning_rate": 7.799e-05,
+      "loss": 1.5904,
+      "step": 7800
+    },
+    {
+      "epoch": 4.952756526539948,
+      "grad_norm": 1.46875,
+      "learning_rate": 7.849e-05,
+      "loss": 1.3684,
+      "step": 7850
+    },
+    {
+      "epoch": 4.984304755895575,
+      "grad_norm": 2.09375,
+      "learning_rate": 7.899000000000001e-05,
+      "loss": 1.4927,
+      "step": 7900
+    },
+    {
+      "epoch": 5.015774114677813,
+      "grad_norm": 1.46875,
+      "learning_rate": 7.949000000000001e-05,
+      "loss": 1.5083,
+      "step": 7950
+    },
+    {
+      "epoch": 5.047322344033441,
+      "grad_norm": 1.3671875,
+      "learning_rate": 7.999000000000001e-05,
+      "loss": 1.5748,
+      "step": 8000
+    },
+    {
+      "epoch": 5.047322344033441,
+      "eval_loss": 1.9995193481445312,
+      "eval_runtime": 267.072,
+      "eval_samples_per_second": 94.944,
+      "eval_steps_per_second": 5.935,
+      "step": 8000
+    },
+    {
+      "epoch": 5.078870573389069,
+      "grad_norm": 1.2578125,
+      "learning_rate": 8.049e-05,
+      "loss": 1.4866,
+      "step": 8050
+    },
+    {
+      "epoch": 5.110418802744696,
+      "grad_norm": 1.1796875,
+      "learning_rate": 8.099e-05,
+      "loss": 1.4125,
+      "step": 8100
+    },
+    {
+      "epoch": 5.1419670321003235,
+      "grad_norm": 1.1640625,
+      "learning_rate": 8.149e-05,
+      "loss": 1.5001,
+      "step": 8150
+    },
+    {
+      "epoch": 5.173515261455951,
+      "grad_norm": 4.0625,
+      "learning_rate": 8.199e-05,
+      "loss": 1.4564,
+      "step": 8200
+    },
+    {
+      "epoch": 5.205063490811578,
+      "grad_norm": 1.3359375,
+      "learning_rate": 8.249e-05,
+      "loss": 1.512,
+      "step": 8250
+    },
+    {
+      "epoch": 5.236611720167206,
+      "grad_norm": 1.4375,
+      "learning_rate": 8.299e-05,
+      "loss": 1.409,
+      "step": 8300
+    },
+    {
+      "epoch": 5.268159949522833,
+      "grad_norm": 1.3046875,
+      "learning_rate": 8.349e-05,
+      "loss": 1.5431,
+      "step": 8350
+    },
+    {
+      "epoch": 5.29970817887846,
+      "grad_norm": 3.640625,
+      "learning_rate": 8.399e-05,
+      "loss": 1.518,
+      "step": 8400
+    },
+    {
+      "epoch": 5.331256408234088,
+      "grad_norm": 1.5859375,
+      "learning_rate": 8.449e-05,
+      "loss": 1.4523,
+      "step": 8450
+    },
+    {
+      "epoch": 5.362804637589715,
+      "grad_norm": 2.3125,
+      "learning_rate": 8.499e-05,
+      "loss": 1.4272,
+      "step": 8500
+    },
+    {
+      "epoch": 5.394352866945343,
+      "grad_norm": 1.453125,
+      "learning_rate": 8.549000000000001e-05,
+      "loss": 1.3483,
+      "step": 8550
+    },
+    {
+      "epoch": 5.42590109630097,
+      "grad_norm": 1.5078125,
+      "learning_rate": 8.599000000000001e-05,
+      "loss": 1.528,
+      "step": 8600
+    },
+    {
+      "epoch": 5.457449325656597,
+      "grad_norm": 1.3046875,
+      "learning_rate": 8.649000000000001e-05,
+      "loss": 1.3666,
+      "step": 8650
+    },
+    {
+      "epoch": 5.488997555012225,
+      "grad_norm": 1.1484375,
+      "learning_rate": 8.699e-05,
+      "loss": 1.4751,
+      "step": 8700
+    },
+    {
+      "epoch": 5.520545784367853,
+      "grad_norm": 1.328125,
+      "learning_rate": 8.749e-05,
+      "loss": 1.503,
+      "step": 8750
+    },
+    {
+      "epoch": 5.5520940137234795,
+      "grad_norm": 18.625,
+      "learning_rate": 8.799e-05,
+      "loss": 1.4122,
+      "step": 8800
+    },
+    {
+      "epoch": 5.583642243079107,
+      "grad_norm": 79.5,
+      "learning_rate": 8.849e-05,
+      "loss": 1.4222,
+      "step": 8850
+    },
+    {
+      "epoch": 5.615190472434735,
+      "grad_norm": 2.125,
+      "learning_rate": 8.899e-05,
+      "loss": 1.5393,
+      "step": 8900
+    },
+    {
+      "epoch": 5.646738701790362,
+      "grad_norm": 1.0078125,
+      "learning_rate": 8.949000000000001e-05,
+      "loss": 1.3101,
+      "step": 8950
+    },
+    {
+      "epoch": 5.67828693114599,
+      "grad_norm": 1.25,
+      "learning_rate": 8.999000000000001e-05,
+      "loss": 1.3791,
+      "step": 9000
+    },
+    {
+      "epoch": 5.67828693114599,
+      "eval_loss": 1.9208731651306152,
+      "eval_runtime": 266.9196,
+      "eval_samples_per_second": 94.999,
+      "eval_steps_per_second": 5.938,
+      "step": 9000
+    },
+    {
+      "epoch": 5.709835160501617,
+      "grad_norm": 1.34375,
+      "learning_rate": 9.049000000000001e-05,
+      "loss": 1.3748,
+      "step": 9050
+    },
+    {
+      "epoch": 5.741383389857244,
+      "grad_norm": 1.1875,
+      "learning_rate": 9.099000000000001e-05,
+      "loss": 1.456,
+      "step": 9100
+    },
+    {
+      "epoch": 5.772931619212872,
+      "grad_norm": 1.171875,
+      "learning_rate": 9.149e-05,
+      "loss": 1.3566,
+      "step": 9150
+    },
+    {
+      "epoch": 5.804479848568499,
+      "grad_norm": 1.2578125,
+      "learning_rate": 9.199e-05,
+      "loss": 1.4585,
+      "step": 9200
+    },
+    {
+      "epoch": 5.8360280779241265,
+      "grad_norm": 1.1875,
+      "learning_rate": 9.249e-05,
+      "loss": 1.4235,
+      "step": 9250
+    },
+    {
+      "epoch": 5.867576307279754,
+      "grad_norm": 0.86328125,
+      "learning_rate": 9.299e-05,
+      "loss": 1.4555,
+      "step": 9300
+    },
+    {
+      "epoch": 5.899124536635381,
+      "grad_norm": 1.390625,
+      "learning_rate": 9.349e-05,
+      "loss": 1.365,
+      "step": 9350
+    },
+    {
+      "epoch": 5.930672765991009,
+      "grad_norm": 1.609375,
+      "learning_rate": 9.399e-05,
+      "loss": 1.4072,
+      "step": 9400
+    },
+    {
+      "epoch": 5.9622209953466365,
+      "grad_norm": 1.1640625,
+      "learning_rate": 9.449e-05,
+      "loss": 1.3952,
+      "step": 9450
+    },
+    {
+      "epoch": 5.993769224702263,
+      "grad_norm": 1.2421875,
+      "learning_rate": 9.499e-05,
+      "loss": 1.4411,
+      "step": 9500
+    },
+    {
+      "epoch": 6.025238583484502,
+      "grad_norm": 1.1875,
+      "learning_rate": 9.549e-05,
+      "loss": 1.4389,
+      "step": 9550
+    },
+    {
+      "epoch": 6.056786812840129,
+      "grad_norm": 1.1640625,
+      "learning_rate": 9.599000000000001e-05,
+      "loss": 1.4392,
+      "step": 9600
+    },
+    {
+      "epoch": 6.088335042195757,
+      "grad_norm": 1.4140625,
+      "learning_rate": 9.649e-05,
+      "loss": 1.3544,
+      "step": 9650
+    },
+    {
+      "epoch": 6.119883271551384,
+      "grad_norm": 1.203125,
+      "learning_rate": 9.699e-05,
+      "loss": 1.335,
+      "step": 9700
+    },
+    {
+      "epoch": 6.1514315009070115,
+      "grad_norm": 1.609375,
+      "learning_rate": 9.749e-05,
+      "loss": 1.3299,
+      "step": 9750
+    },
+    {
+      "epoch": 6.182979730262639,
+      "grad_norm": 1.3515625,
+      "learning_rate": 9.799e-05,
+      "loss": 1.4326,
+      "step": 9800
+    },
+    {
+      "epoch": 6.214527959618266,
+      "grad_norm": 1.1484375,
+      "learning_rate": 9.849e-05,
+      "loss": 1.3592,
+      "step": 9850
+    },
+    {
+      "epoch": 6.246076188973894,
+      "grad_norm": 1.5703125,
+      "learning_rate": 9.899e-05,
+      "loss": 1.4231,
+      "step": 9900
+    },
+    {
+      "epoch": 6.277624418329522,
+      "grad_norm": 1.1171875,
+      "learning_rate": 9.949000000000001e-05,
+      "loss": 1.4076,
+      "step": 9950
+    },
+    {
+      "epoch": 6.309172647685148,
+      "grad_norm": 1.34375,
+      "learning_rate": 9.999000000000001e-05,
+      "loss": 1.3761,
+      "step": 10000
+    },
+    {
+      "epoch": 6.309172647685148,
+      "eval_loss": 1.8519904613494873,
+      "eval_runtime": 267.0769,
+      "eval_samples_per_second": 94.943,
+      "eval_steps_per_second": 5.935,
+      "step": 10000
+    },
+    {
+      "epoch": 6.340720877040776,
+      "grad_norm": 1.359375,
+      "learning_rate": 9.99742105263158e-05,
+      "loss": 1.4797,
+      "step": 10050
+    },
+    {
+      "epoch": 6.372269106396404,
+      "grad_norm": 1.734375,
+      "learning_rate": 9.994789473684211e-05,
+      "loss": 1.3483,
+      "step": 10100
+    },
+    {
+      "epoch": 6.403817335752031,
+      "grad_norm": 1.34375,
+      "learning_rate": 9.992157894736842e-05,
+      "loss": 1.4807,
+      "step": 10150
+    },
+    {
+      "epoch": 6.4353655651076584,
+      "grad_norm": 1.4921875,
+      "learning_rate": 9.989526315789473e-05,
+      "loss": 1.3523,
+      "step": 10200
+    },
+    {
+      "epoch": 6.466913794463286,
+      "grad_norm": 1.46875,
+      "learning_rate": 9.986894736842106e-05,
+      "loss": 1.3903,
+      "step": 10250
+    },
+    {
+      "epoch": 6.498462023818913,
+      "grad_norm": 1.328125,
+      "learning_rate": 9.984263157894738e-05,
+      "loss": 1.3053,
+      "step": 10300
+    },
+    {
+      "epoch": 6.530010253174541,
+      "grad_norm": 1.40625,
+      "learning_rate": 9.98163157894737e-05,
+      "loss": 1.3541,
+      "step": 10350
+    },
+    {
+      "epoch": 6.5615584825301685,
+      "grad_norm": 1.3125,
+      "learning_rate": 9.979e-05,
+      "loss": 1.316,
+      "step": 10400
+    },
+    {
+      "epoch": 6.593106711885795,
+      "grad_norm": 1.28125,
+      "learning_rate": 9.976368421052632e-05,
+      "loss": 1.3258,
+      "step": 10450
+    },
+    {
+      "epoch": 6.624654941241423,
+      "grad_norm": 1.2734375,
+      "learning_rate": 9.973736842105263e-05,
+      "loss": 1.2155,
+      "step": 10500
+    },
+    {
+      "epoch": 6.65620317059705,
+      "grad_norm": 1.140625,
+      "learning_rate": 9.971105263157895e-05,
+      "loss": 1.3149,
+      "step": 10550
+    },
+    {
+      "epoch": 6.687751399952678,
+      "grad_norm": 1.2109375,
+      "learning_rate": 9.968473684210526e-05,
+      "loss": 1.4807,
+      "step": 10600
+    },
+    {
+      "epoch": 6.719299629308305,
+      "grad_norm": 1.5,
+      "learning_rate": 9.965842105263158e-05,
+      "loss": 1.3725,
+      "step": 10650
+    },
+    {
+      "epoch": 6.750847858663932,
+      "grad_norm": 2.15625,
+      "learning_rate": 9.96321052631579e-05,
+      "loss": 1.3688,
+      "step": 10700
+    },
+    {
+      "epoch": 6.78239608801956,
+      "grad_norm": 1.6875,
+      "learning_rate": 9.960578947368421e-05,
+      "loss": 1.4244,
+      "step": 10750
+    },
+    {
+      "epoch": 6.813944317375188,
+      "grad_norm": 2.6875,
+      "learning_rate": 9.957947368421054e-05,
+      "loss": 1.2727,
+      "step": 10800
+    },
+    {
+      "epoch": 6.8454925467308145,
+      "grad_norm": 1.21875,
+      "learning_rate": 9.955315789473685e-05,
+      "loss": 1.3025,
+      "step": 10850
+    },
+    {
+      "epoch": 6.877040776086442,
+      "grad_norm": 1.5546875,
+      "learning_rate": 9.952684210526316e-05,
+      "loss": 1.3731,
+      "step": 10900
+    },
+    {
+      "epoch": 6.908589005442069,
+      "grad_norm": 1.3515625,
+      "learning_rate": 9.950052631578947e-05,
+      "loss": 1.364,
+      "step": 10950
+    },
+    {
+      "epoch": 6.940137234797697,
+      "grad_norm": 1.640625,
+      "learning_rate": 9.94742105263158e-05,
+      "loss": 1.4627,
+      "step": 11000
+    },
+    {
+      "epoch": 6.940137234797697,
+      "eval_loss": 1.7950236797332764,
+      "eval_runtime": 267.3284,
+      "eval_samples_per_second": 94.853,
+      "eval_steps_per_second": 5.929,
+      "step": 11000
+    },
+    {
+      "epoch": 6.971685464153325,
+      "grad_norm": 1.1953125,
+      "learning_rate": 9.944789473684211e-05,
+      "loss": 1.2709,
+      "step": 11050
+    },
+    {
+      "epoch": 7.003154822935563,
+      "grad_norm": 1.3046875,
+      "learning_rate": 9.942157894736842e-05,
+      "loss": 1.2428,
+      "step": 11100
+    },
+    {
+      "epoch": 7.03470305229119,
+      "grad_norm": 1.46875,
+      "learning_rate": 9.939526315789475e-05,
+      "loss": 1.3802,
+      "step": 11150
+    },
+    {
+      "epoch": 7.066251281646817,
+      "grad_norm": 1.5859375,
+      "learning_rate": 9.936894736842106e-05,
+      "loss": 1.416,
+      "step": 11200
+    },
+    {
+      "epoch": 7.097799511002445,
+      "grad_norm": 1.203125,
+      "learning_rate": 9.934263157894737e-05,
+      "loss": 1.2825,
+      "step": 11250
+    },
+    {
+      "epoch": 7.129347740358073,
+      "grad_norm": 1.421875,
+      "learning_rate": 9.93163157894737e-05,
+      "loss": 1.3154,
+      "step": 11300
+    },
+    {
+      "epoch": 7.1608959697137,
+      "grad_norm": 1.640625,
+      "learning_rate": 9.929e-05,
+      "loss": 1.271,
+      "step": 11350
+    },
+    {
+      "epoch": 7.192444199069327,
+      "grad_norm": 1.0703125,
+      "learning_rate": 9.926368421052632e-05,
+      "loss": 1.3287,
+      "step": 11400
+    },
+    {
+      "epoch": 7.223992428424955,
+      "grad_norm": 1.1953125,
+      "learning_rate": 9.923736842105263e-05,
+      "loss": 1.3063,
+      "step": 11450
+    },
+    {
+      "epoch": 7.255540657780582,
+      "grad_norm": 1.2734375,
+      "learning_rate": 9.921105263157895e-05,
+      "loss": 1.2098,
+      "step": 11500
+    },
+    {
+      "epoch": 7.28708888713621,
+      "grad_norm": 1.1640625,
+      "learning_rate": 9.918473684210527e-05,
+      "loss": 1.4312,
+      "step": 11550
+    },
+    {
+      "epoch": 7.318637116491837,
+      "grad_norm": 1.421875,
+      "learning_rate": 9.915842105263158e-05,
+      "loss": 1.3148,
+      "step": 11600
+    },
+    {
+      "epoch": 7.350185345847464,
+      "grad_norm": 0.98828125,
+      "learning_rate": 9.91321052631579e-05,
+      "loss": 1.2584,
+      "step": 11650
+    },
+    {
+      "epoch": 7.381733575203092,
+      "grad_norm": 1.296875,
+      "learning_rate": 9.910578947368421e-05,
+      "loss": 1.3112,
+      "step": 11700
+    },
+    {
+      "epoch": 7.413281804558719,
+      "grad_norm": 1.4609375,
+      "learning_rate": 9.907947368421054e-05,
+      "loss": 1.2949,
+      "step": 11750
+    },
+    {
+      "epoch": 7.4448300339143465,
+      "grad_norm": 1.015625,
+      "learning_rate": 9.905315789473685e-05,
+      "loss": 1.2393,
+      "step": 11800
+    },
+    {
+      "epoch": 7.476378263269974,
+      "grad_norm": 1.8046875,
+      "learning_rate": 9.902684210526316e-05,
+      "loss": 1.4107,
+      "step": 11850
+    },
+    {
+      "epoch": 7.507926492625601,
+      "grad_norm": 3.296875,
+      "learning_rate": 9.900052631578947e-05,
+      "loss": 1.1609,
+      "step": 11900
+    },
+    {
+      "epoch": 7.539474721981229,
+      "grad_norm": 2.25,
+      "learning_rate": 9.897421052631579e-05,
+      "loss": 1.2993,
+      "step": 11950
+    },
+    {
+      "epoch": 7.5710229513368565,
+      "grad_norm": 1.203125,
+      "learning_rate": 9.894789473684211e-05,
+      "loss": 1.3573,
+      "step": 12000
+    },
+    {
+      "epoch": 7.5710229513368565,
+      "eval_loss": 1.7552872896194458,
+      "eval_runtime": 267.1568,
+      "eval_samples_per_second": 94.914,
+      "eval_steps_per_second": 5.933,
+      "step": 12000
+    },
+    {
+      "epoch": 7.602571180692483,
+      "grad_norm": 1.2578125,
+      "learning_rate": 9.892157894736842e-05,
+      "loss": 1.1903,
+      "step": 12050
+    },
+    {
+      "epoch": 7.634119410048111,
+      "grad_norm": 1.3671875,
+      "learning_rate": 9.889526315789475e-05,
+      "loss": 1.3354,
+      "step": 12100
+    },
+    {
+      "epoch": 7.665667639403739,
+      "grad_norm": 1.6796875,
+      "learning_rate": 9.886894736842106e-05,
+      "loss": 1.298,
+      "step": 12150
+    },
+    {
+      "epoch": 7.697215868759366,
+      "grad_norm": 1.265625,
+      "learning_rate": 9.884263157894737e-05,
+      "loss": 1.3834,
+      "step": 12200
+    },
+    {
+      "epoch": 7.728764098114993,
+      "grad_norm": 1.3046875,
+      "learning_rate": 9.88163157894737e-05,
+      "loss": 1.3978,
+      "step": 12250
+    },
+    {
+      "epoch": 7.76031232747062,
+      "grad_norm": 1.4140625,
+      "learning_rate": 9.879000000000001e-05,
+      "loss": 1.2518,
+      "step": 12300
+    },
+    {
+      "epoch": 7.791860556826248,
+      "grad_norm": 1.1484375,
+      "learning_rate": 9.876368421052632e-05,
+      "loss": 1.3163,
+      "step": 12350
+    },
+    {
+      "epoch": 7.823408786181876,
+      "grad_norm": 1.3515625,
+      "learning_rate": 9.873736842105263e-05,
+      "loss": 1.3215,
+      "step": 12400
+    },
+    {
+      "epoch": 7.854957015537503,
+      "grad_norm": 1.2109375,
+      "learning_rate": 9.871105263157894e-05,
+      "loss": 1.2938,
+      "step": 12450
+    },
+    {
+      "epoch": 7.88650524489313,
+      "grad_norm": 0.875,
+      "learning_rate": 9.868473684210527e-05,
+      "loss": 1.2684,
+      "step": 12500
+    },
+    {
+      "epoch": 7.918053474248758,
+      "grad_norm": 1.234375,
+      "learning_rate": 9.865842105263159e-05,
+      "loss": 1.2393,
+      "step": 12550
+    },
+    {
+      "epoch": 7.949601703604385,
+      "grad_norm": 1.3984375,
+      "learning_rate": 9.86321052631579e-05,
+      "loss": 1.3817,
+      "step": 12600
+    },
+    {
+      "epoch": 7.981149932960013,
+      "grad_norm": 1.15625,
+      "learning_rate": 9.860578947368422e-05,
+      "loss": 1.3158,
+      "step": 12650
+    },
+    {
+      "epoch": 8.01261929174225,
+      "grad_norm": 1.296875,
+      "learning_rate": 9.857947368421053e-05,
+      "loss": 1.3428,
+      "step": 12700
+    },
+    {
+      "epoch": 8.044167521097878,
+      "grad_norm": 2.375,
+      "learning_rate": 9.855315789473685e-05,
+      "loss": 1.2863,
+      "step": 12750
+    },
+    {
+      "epoch": 8.075715750453506,
+      "grad_norm": 1.0859375,
+      "learning_rate": 9.852684210526316e-05,
+      "loss": 1.2989,
+      "step": 12800
+    },
+    {
+      "epoch": 8.107263979809133,
+      "grad_norm": 1.203125,
+      "learning_rate": 9.850052631578948e-05,
+      "loss": 1.2974,
+      "step": 12850
+    },
+    {
+      "epoch": 8.13881220916476,
+      "grad_norm": 1.25,
+      "learning_rate": 9.847421052631579e-05,
+      "loss": 1.2585,
+      "step": 12900
+    },
+    {
+      "epoch": 8.170360438520389,
+      "grad_norm": 1.1015625,
+      "learning_rate": 9.84478947368421e-05,
+      "loss": 1.1535,
+      "step": 12950
+    },
+    {
+      "epoch": 8.201908667876015,
+      "grad_norm": 1.09375,
+      "learning_rate": 9.842157894736842e-05,
+      "loss": 1.2961,
+      "step": 13000
+    },
+    {
+      "epoch": 8.201908667876015,
+      "eval_loss": 1.72089684009552,
+      "eval_runtime": 267.2684,
+      "eval_samples_per_second": 94.875,
+      "eval_steps_per_second": 5.93,
+      "step": 13000
+    },
+    {
+      "epoch": 8.233456897231642,
+      "grad_norm": 1.1953125,
+      "learning_rate": 9.839526315789475e-05,
+      "loss": 1.1665,
+      "step": 13050
+    },
+    {
+      "epoch": 8.26500512658727,
+      "grad_norm": 1.0625,
+      "learning_rate": 9.836894736842106e-05,
+      "loss": 1.3976,
+      "step": 13100
+    },
+    {
+      "epoch": 8.296553355942898,
+      "grad_norm": 1.640625,
+      "learning_rate": 9.834263157894737e-05,
+      "loss": 1.2169,
+      "step": 13150
+    },
+    {
+      "epoch": 8.328101585298525,
+      "grad_norm": 0.97265625,
+      "learning_rate": 9.831631578947368e-05,
+      "loss": 1.3242,
+      "step": 13200
+    },
+    {
+      "epoch": 8.359649814654153,
+      "grad_norm": 1.4765625,
+      "learning_rate": 9.829000000000001e-05,
+      "loss": 1.2378,
+      "step": 13250
+    },
+    {
+      "epoch": 8.39119804400978,
+      "grad_norm": 1.4453125,
+      "learning_rate": 9.826368421052632e-05,
+      "loss": 1.3139,
+      "step": 13300
+    },
+    {
+      "epoch": 8.422746273365407,
+      "grad_norm": 1.296875,
+      "learning_rate": 9.823736842105263e-05,
+      "loss": 1.2975,
+      "step": 13350
+    },
+    {
+      "epoch": 8.454294502721035,
+      "grad_norm": 1.25,
+      "learning_rate": 9.821105263157894e-05,
+      "loss": 1.2578,
+      "step": 13400
+    },
+    {
+      "epoch": 8.485842732076662,
+      "grad_norm": 2.796875,
+      "learning_rate": 9.818473684210527e-05,
+      "loss": 1.2879,
+      "step": 13450
+    },
+    {
+      "epoch": 8.51739096143229,
+      "grad_norm": 1.0234375,
+      "learning_rate": 9.81584210526316e-05,
+      "loss": 1.34,
+      "step": 13500
+    },
+    {
+      "epoch": 8.548939190787918,
+      "grad_norm": 1.1484375,
+      "learning_rate": 9.81321052631579e-05,
+      "loss": 1.248,
+      "step": 13550
+    },
+    {
+      "epoch": 8.580487420143545,
+      "grad_norm": 4.40625,
+      "learning_rate": 9.810578947368422e-05,
+      "loss": 1.3071,
+      "step": 13600
+    },
+    {
+      "epoch": 8.612035649499171,
+      "grad_norm": 1.1328125,
+      "learning_rate": 9.807947368421053e-05,
+      "loss": 1.2628,
+      "step": 13650
+    },
+    {
+      "epoch": 8.6435838788548,
+      "grad_norm": 1.046875,
+      "learning_rate": 9.805315789473684e-05,
+      "loss": 1.211,
+      "step": 13700
+    },
+    {
+      "epoch": 8.675132108210427,
+      "grad_norm": 0.76953125,
+      "learning_rate": 9.802684210526317e-05,
+      "loss": 1.27,
+      "step": 13750
+    },
+    {
+      "epoch": 8.706680337566054,
+      "grad_norm": 1.6171875,
+      "learning_rate": 9.800052631578948e-05,
+      "loss": 1.383,
+      "step": 13800
+    },
+    {
+      "epoch": 8.73822856692168,
+      "grad_norm": 1.5625,
+      "learning_rate": 9.797421052631579e-05,
+      "loss": 1.2795,
+      "step": 13850
+    },
+    {
+      "epoch": 8.76977679627731,
+      "grad_norm": 1.53125,
+      "learning_rate": 9.79478947368421e-05,
+      "loss": 1.278,
+      "step": 13900
+    },
+    {
+      "epoch": 8.801325025632936,
+      "grad_norm": 1.1640625,
+      "learning_rate": 9.792157894736843e-05,
+      "loss": 1.1991,
+      "step": 13950
+    },
+    {
+      "epoch": 8.832873254988563,
+      "grad_norm": 1.375,
+      "learning_rate": 9.789526315789475e-05,
+      "loss": 1.2975,
+      "step": 14000
+    },
+    {
+      "epoch": 8.832873254988563,
+      "eval_loss": 1.693904995918274,
+      "eval_runtime": 266.8978,
+      "eval_samples_per_second": 95.006,
+      "eval_steps_per_second": 5.939,
+      "step": 14000
+    },
+    {
+      "epoch": 8.864421484344192,
+      "grad_norm": 1.3359375,
+      "learning_rate": 9.786894736842106e-05,
+      "loss": 1.2153,
+      "step": 14050
+    },
+    {
+      "epoch": 8.895969713699818,
+      "grad_norm": 1.4140625,
+      "learning_rate": 9.784263157894737e-05,
+      "loss": 1.2027,
+      "step": 14100
+    },
+    {
+      "epoch": 8.927517943055445,
+      "grad_norm": 1.265625,
+      "learning_rate": 9.781631578947369e-05,
+      "loss": 1.3394,
+      "step": 14150
+    },
+    {
+      "epoch": 8.959066172411074,
+      "grad_norm": 1.3984375,
+      "learning_rate": 9.779e-05,
+      "loss": 1.3347,
+      "step": 14200
+    },
+    {
+      "epoch": 8.9906144017667,
+      "grad_norm": 1.1015625,
+      "learning_rate": 9.776368421052632e-05,
+      "loss": 1.2417,
+      "step": 14250
+    },
+    {
+      "epoch": 9.02208376054894,
+      "grad_norm": 1.6015625,
+      "learning_rate": 9.773736842105263e-05,
+      "loss": 1.3005,
+      "step": 14300
+    },
+    {
+      "epoch": 9.053631989904567,
+      "grad_norm": 1.2265625,
+      "learning_rate": 9.771105263157895e-05,
+      "loss": 1.141,
+      "step": 14350
+    },
+    {
+      "epoch": 9.085180219260193,
+      "grad_norm": 1.265625,
+      "learning_rate": 9.768473684210527e-05,
+      "loss": 1.3553,
+      "step": 14400
+    },
+    {
+      "epoch": 9.116728448615822,
+      "grad_norm": 0.87890625,
+      "learning_rate": 9.765842105263158e-05,
+      "loss": 1.3469,
+      "step": 14450
+    },
+    {
+      "epoch": 9.148276677971449,
+      "grad_norm": 0.81640625,
+      "learning_rate": 9.763210526315791e-05,
+      "loss": 1.3001,
+      "step": 14500
+    },
+    {
+      "epoch": 9.179824907327076,
+      "grad_norm": 0.89453125,
+      "learning_rate": 9.760578947368422e-05,
+      "loss": 1.2651,
+      "step": 14550
+    },
+    {
+      "epoch": 9.211373136682704,
+      "grad_norm": 1.109375,
+      "learning_rate": 9.757947368421053e-05,
+      "loss": 1.2318,
+      "step": 14600
+    },
+    {
+      "epoch": 9.242921366038331,
+      "grad_norm": 1.4140625,
+      "learning_rate": 9.755315789473684e-05,
+      "loss": 1.2397,
+      "step": 14650
+    },
+    {
+      "epoch": 9.274469595393958,
+      "grad_norm": 2.234375,
+      "learning_rate": 9.752684210526317e-05,
+      "loss": 1.2617,
+      "step": 14700
+    },
+    {
+      "epoch": 9.306017824749587,
+      "grad_norm": 1.203125,
+      "learning_rate": 9.750052631578948e-05,
+      "loss": 1.308,
+      "step": 14750
+    },
+    {
+      "epoch": 9.337566054105213,
+      "grad_norm": 1.3359375,
+      "learning_rate": 9.747421052631579e-05,
+      "loss": 1.2499,
+      "step": 14800
+    },
+    {
+      "epoch": 9.36911428346084,
+      "grad_norm": 1.2109375,
+      "learning_rate": 9.74478947368421e-05,
+      "loss": 1.2713,
+      "step": 14850
+    },
+    {
+      "epoch": 9.400662512816469,
+      "grad_norm": 1.0,
+      "learning_rate": 9.742157894736843e-05,
+      "loss": 1.16,
+      "step": 14900
+    },
+    {
+      "epoch": 9.432210742172096,
+      "grad_norm": 1.71875,
+      "learning_rate": 9.739526315789474e-05,
+      "loss": 1.2691,
+      "step": 14950
+    },
+    {
+      "epoch": 9.463758971527723,
+      "grad_norm": 1.59375,
+      "learning_rate": 9.736894736842106e-05,
+      "loss": 1.2895,
+      "step": 15000
+    },
+    {
+      "epoch": 9.463758971527723,
+      "eval_loss": 1.6718602180480957,
+      "eval_runtime": 267.6306,
+      "eval_samples_per_second": 94.746,
+      "eval_steps_per_second": 5.922,
+      "step": 15000
+    },
+    {
+      "epoch": 9.495307200883351,
+      "grad_norm": 1.421875,
+      "learning_rate": 9.734263157894738e-05,
+      "loss": 1.2325,
+      "step": 15050
+    },
+    {
+      "epoch": 9.526855430238978,
+      "grad_norm": 1.96875,
+      "learning_rate": 9.731631578947369e-05,
+      "loss": 1.211,
+      "step": 15100
+    },
+    {
+      "epoch": 9.558403659594605,
+      "grad_norm": 3.078125,
+      "learning_rate": 9.729e-05,
+      "loss": 1.2128,
+      "step": 15150
+    },
+    {
+      "epoch": 9.589951888950232,
+      "grad_norm": 1.265625,
+      "learning_rate": 9.726368421052632e-05,
+      "loss": 1.3211,
+      "step": 15200
+    },
+    {
+      "epoch": 9.62150011830586,
+      "grad_norm": 1.0546875,
+      "learning_rate": 9.723736842105264e-05,
+      "loss": 1.1707,
+      "step": 15250
+    },
+    {
+      "epoch": 9.653048347661487,
+      "grad_norm": 1.2265625,
+      "learning_rate": 9.721105263157895e-05,
+      "loss": 1.1723,
+      "step": 15300
+    },
+    {
+      "epoch": 9.684596577017114,
+      "grad_norm": 1.0703125,
+      "learning_rate": 9.718473684210527e-05,
+      "loss": 1.1736,
+      "step": 15350
+    },
+    {
+      "epoch": 9.716144806372743,
+      "grad_norm": 1.2109375,
+      "learning_rate": 9.715842105263158e-05,
+      "loss": 1.2063,
+      "step": 15400
+    },
+    {
+      "epoch": 9.74769303572837,
+      "grad_norm": 0.97265625,
+      "learning_rate": 9.713210526315791e-05,
+      "loss": 1.2337,
+      "step": 15450
+    },
+    {
+      "epoch": 9.779241265083996,
+      "grad_norm": 1.1640625,
+      "learning_rate": 9.710578947368422e-05,
+      "loss": 1.1708,
+      "step": 15500
+    },
+    {
+      "epoch": 9.810789494439625,
+      "grad_norm": 1.1171875,
+      "learning_rate": 9.707947368421053e-05,
+      "loss": 1.303,
+      "step": 15550
+    },
+    {
+      "epoch": 9.842337723795252,
+      "grad_norm": 1.140625,
+      "learning_rate": 9.705315789473684e-05,
+      "loss": 1.1924,
+      "step": 15600
+    },
+    {
+      "epoch": 9.873885953150879,
+      "grad_norm": 1.203125,
+      "learning_rate": 9.702684210526316e-05,
+      "loss": 1.2857,
+      "step": 15650
+    },
+    {
+      "epoch": 9.905434182506507,
+      "grad_norm": 1.703125,
+      "learning_rate": 9.700052631578948e-05,
+      "loss": 1.2931,
+      "step": 15700
+    },
+    {
+      "epoch": 9.936982411862134,
+      "grad_norm": 1.1015625,
+      "learning_rate": 9.697421052631579e-05,
+      "loss": 1.2534,
+      "step": 15750
+    },
+    {
+      "epoch": 9.968530641217761,
+      "grad_norm": 1.0,
+      "learning_rate": 9.694789473684212e-05,
+      "loss": 1.2154,
+      "step": 15800
+    },
+    {
+      "epoch": 10.0,
+      "grad_norm": 1.2109375,
+      "learning_rate": 9.692157894736843e-05,
+      "loss": 1.1951,
+      "step": 15850
+    },
+    {
+      "epoch": 10.031548229355627,
+      "grad_norm": 1.0859375,
+      "learning_rate": 9.689526315789474e-05,
+      "loss": 1.1643,
+      "step": 15900
+    },
+    {
+      "epoch": 10.063096458711255,
+      "grad_norm": 1.734375,
+      "learning_rate": 9.686894736842107e-05,
+      "loss": 1.2679,
+      "step": 15950
+    },
+    {
+      "epoch": 10.094644688066882,
+      "grad_norm": 1.0859375,
+      "learning_rate": 9.684263157894738e-05,
+      "loss": 1.2818,
+      "step": 16000
+    },
+    {
+      "epoch": 10.094644688066882,
+      "eval_loss": 1.653988242149353,
+      "eval_runtime": 267.0729,
+      "eval_samples_per_second": 94.944,
+      "eval_steps_per_second": 5.935,
+      "step": 16000
+    },
+    {
+      "epoch": 10.12619291742251,
+      "grad_norm": 1.453125,
+      "learning_rate": 9.681631578947369e-05,
+      "loss": 1.0915,
+      "step": 16050
+    },
+    {
+      "epoch": 10.157741146778138,
+      "grad_norm": 0.99609375,
+      "learning_rate": 9.679e-05,
+      "loss": 1.2581,
+      "step": 16100
+    },
+    {
+      "epoch": 10.189289376133765,
+      "grad_norm": 1.6484375,
+      "learning_rate": 9.676368421052631e-05,
+      "loss": 1.3474,
+      "step": 16150
+    },
+    {
+      "epoch": 10.220837605489391,
+      "grad_norm": 1.25,
+      "learning_rate": 9.673736842105264e-05,
+      "loss": 1.2432,
+      "step": 16200
+    },
+    {
+      "epoch": 10.25238583484502,
+      "grad_norm": 0.921875,
+      "learning_rate": 9.671105263157895e-05,
+      "loss": 1.1038,
+      "step": 16250
+    },
+    {
+      "epoch": 10.283934064200647,
+      "grad_norm": 1.1875,
+      "learning_rate": 9.668473684210527e-05,
+      "loss": 1.2775,
+      "step": 16300
+    },
+    {
+      "epoch": 10.315482293556274,
+      "grad_norm": 1.2890625,
+      "learning_rate": 9.665842105263158e-05,
+      "loss": 1.2662,
+      "step": 16350
+    },
+    {
+      "epoch": 10.347030522911902,
+      "grad_norm": 1.8984375,
+      "learning_rate": 9.66321052631579e-05,
+      "loss": 1.1715,
+      "step": 16400
+    },
+    {
+      "epoch": 10.37857875226753,
+      "grad_norm": 1.0078125,
+      "learning_rate": 9.660578947368422e-05,
+      "loss": 1.2537,
+      "step": 16450
+    },
+    {
+      "epoch": 10.410126981623156,
+      "grad_norm": 1.3125,
+      "learning_rate": 9.657947368421053e-05,
+      "loss": 1.3039,
+      "step": 16500
+    },
+    {
+      "epoch": 10.441675210978783,
+      "grad_norm": 1.2421875,
+      "learning_rate": 9.655315789473684e-05,
+      "loss": 1.188,
+      "step": 16550
+    },
+    {
+      "epoch": 10.473223440334412,
+      "grad_norm": 1.21875,
+      "learning_rate": 9.652684210526316e-05,
+      "loss": 1.1287,
+      "step": 16600
+    },
+    {
+      "epoch": 10.504771669690038,
+      "grad_norm": 1.21875,
+      "learning_rate": 9.650052631578947e-05,
+      "loss": 1.2609,
+      "step": 16650
+    },
+    {
+      "epoch": 10.536319899045665,
+      "grad_norm": 1.296875,
+      "learning_rate": 9.647421052631579e-05,
+      "loss": 1.2599,
+      "step": 16700
+    },
+    {
+      "epoch": 10.567868128401294,
+      "grad_norm": 1.3671875,
+      "learning_rate": 9.644789473684212e-05,
+      "loss": 1.184,
+      "step": 16750
+    },
+    {
+      "epoch": 10.59941635775692,
+      "grad_norm": 1.3359375,
+      "learning_rate": 9.642157894736843e-05,
+      "loss": 1.207,
+      "step": 16800
+    },
+    {
+      "epoch": 10.630964587112548,
+      "grad_norm": 1.3984375,
+      "learning_rate": 9.639526315789474e-05,
+      "loss": 1.2456,
+      "step": 16850
+    },
+    {
+      "epoch": 10.662512816468176,
+      "grad_norm": 1.0703125,
+      "learning_rate": 9.636894736842105e-05,
+      "loss": 1.1748,
+      "step": 16900
+    },
+    {
+      "epoch": 10.694061045823803,
+      "grad_norm": 1.1875,
+      "learning_rate": 9.634263157894738e-05,
+      "loss": 1.2294,
+      "step": 16950
+    },
+    {
+      "epoch": 10.72560927517943,
+      "grad_norm": 1.0859375,
+      "learning_rate": 9.631631578947369e-05,
+      "loss": 1.2622,
+      "step": 17000
+    },
+    {
+      "epoch": 10.72560927517943,
+      "eval_loss": 1.6381371021270752,
+      "eval_runtime": 266.9095,
+      "eval_samples_per_second": 95.002,
+      "eval_steps_per_second": 5.938,
+      "step": 17000
+    },
+    {
+      "epoch": 10.757157504535058,
+      "grad_norm": 1.5234375,
+      "learning_rate": 9.629e-05,
+      "loss": 1.2294,
+      "step": 17050
+    },
+    {
+      "epoch": 10.788705733890685,
+      "grad_norm": 1.2734375,
+      "learning_rate": 9.626368421052631e-05,
+      "loss": 1.1825,
+      "step": 17100
+    },
+    {
+      "epoch": 10.820253963246312,
+      "grad_norm": 1.1328125,
+      "learning_rate": 9.623736842105264e-05,
+      "loss": 1.2099,
+      "step": 17150
+    },
+    {
+      "epoch": 10.85180219260194,
+      "grad_norm": 1.234375,
+      "learning_rate": 9.621105263157896e-05,
+      "loss": 1.2358,
+      "step": 17200
+    },
+    {
+      "epoch": 10.883350421957568,
+      "grad_norm": 1.0546875,
+      "learning_rate": 9.618473684210527e-05,
+      "loss": 1.1026,
+      "step": 17250
+    },
+    {
+      "epoch": 10.914898651313194,
+      "grad_norm": 1.1640625,
+      "learning_rate": 9.615842105263159e-05,
+      "loss": 1.1988,
+      "step": 17300
+    },
+    {
+      "epoch": 10.946446880668823,
+      "grad_norm": 1.234375,
+      "learning_rate": 9.61321052631579e-05,
+      "loss": 1.2846,
+      "step": 17350
+    },
+    {
+      "epoch": 10.97799511002445,
+      "grad_norm": 0.91015625,
+      "learning_rate": 9.610578947368421e-05,
+      "loss": 1.1801,
+      "step": 17400
+    },
+    {
+      "epoch": 11.009464468806689,
+      "grad_norm": 1.3984375,
+      "learning_rate": 9.607947368421053e-05,
+      "loss": 1.2068,
+      "step": 17450
+    },
+    {
+      "epoch": 11.041012698162316,
+      "grad_norm": 1.2265625,
+      "learning_rate": 9.605315789473685e-05,
+      "loss": 1.2886,
+      "step": 17500
+    },
+    {
+      "epoch": 11.072560927517943,
+      "grad_norm": 1.0546875,
+      "learning_rate": 9.602684210526316e-05,
+      "loss": 1.2498,
+      "step": 17550
+    },
+    {
+      "epoch": 11.104109156873571,
+      "grad_norm": 1.015625,
+      "learning_rate": 9.600052631578947e-05,
+      "loss": 1.2796,
+      "step": 17600
+    },
+    {
+      "epoch": 11.135657386229198,
+      "grad_norm": 1.71875,
+      "learning_rate": 9.59742105263158e-05,
+      "loss": 1.2457,
+      "step": 17650
+    },
+    {
+      "epoch": 11.167205615584825,
+      "grad_norm": 1.453125,
+      "learning_rate": 9.594789473684212e-05,
+      "loss": 1.1583,
+      "step": 17700
+    },
+    {
+      "epoch": 11.198753844940454,
+      "grad_norm": 1.2109375,
+      "learning_rate": 9.592157894736843e-05,
+      "loss": 1.1494,
+      "step": 17750
+    },
+    {
+      "epoch": 11.23030207429608,
+      "grad_norm": 1.3125,
+      "learning_rate": 9.589526315789474e-05,
+      "loss": 1.1675,
+      "step": 17800
+    },
+    {
+      "epoch": 11.261850303651707,
+      "grad_norm": 1.0,
+      "learning_rate": 9.586894736842105e-05,
+      "loss": 1.1923,
+      "step": 17850
+    },
+    {
+      "epoch": 11.293398533007334,
+      "grad_norm": 1.2421875,
+      "learning_rate": 9.584263157894737e-05,
+      "loss": 1.1461,
+      "step": 17900
+    },
+    {
+      "epoch": 11.324946762362963,
+      "grad_norm": 1.15625,
+      "learning_rate": 9.581631578947369e-05,
+      "loss": 1.1753,
+      "step": 17950
+    },
+    {
+      "epoch": 11.35649499171859,
+      "grad_norm": 1.1015625,
+      "learning_rate": 9.579e-05,
+      "loss": 1.2912,
+      "step": 18000
+    },
+    {
+      "epoch": 11.35649499171859,
+      "eval_loss": 1.6246815919876099,
+      "eval_runtime": 266.8814,
+      "eval_samples_per_second": 95.012,
+      "eval_steps_per_second": 5.939,
+      "step": 18000
+    },
+    {
+      "epoch": 11.388043221074216,
+      "grad_norm": 1.2265625,
+      "learning_rate": 9.576368421052631e-05,
+      "loss": 1.1746,
+      "step": 18050
+    },
+    {
+      "epoch": 11.419591450429845,
+      "grad_norm": 1.4140625,
+      "learning_rate": 9.573736842105263e-05,
+      "loss": 1.2125,
+      "step": 18100
+    },
+    {
+      "epoch": 11.451139679785472,
+      "grad_norm": 1.09375,
+      "learning_rate": 9.571105263157895e-05,
+      "loss": 1.2383,
+      "step": 18150
+    },
+    {
+      "epoch": 11.482687909141099,
+      "grad_norm": 1.7890625,
+      "learning_rate": 9.568473684210528e-05,
+      "loss": 1.1907,
+      "step": 18200
+    },
+    {
+      "epoch": 11.514236138496727,
+      "grad_norm": 1.0078125,
+      "learning_rate": 9.565842105263159e-05,
+      "loss": 1.2293,
+      "step": 18250
+    },
+    {
+      "epoch": 11.545784367852354,
+      "grad_norm": 1.1328125,
+      "learning_rate": 9.56321052631579e-05,
+      "loss": 1.2854,
+      "step": 18300
+    },
+    {
+      "epoch": 11.577332597207981,
+      "grad_norm": 1.1015625,
+      "learning_rate": 9.560578947368421e-05,
+      "loss": 1.2259,
+      "step": 18350
+    },
+    {
+      "epoch": 11.60888082656361,
+      "grad_norm": 1.1953125,
+      "learning_rate": 9.557947368421054e-05,
+      "loss": 1.1974,
+      "step": 18400
+    },
+    {
+      "epoch": 11.640429055919236,
+      "grad_norm": 1.3828125,
+      "learning_rate": 9.555315789473685e-05,
+      "loss": 1.1581,
+      "step": 18450
+    },
+    {
+      "epoch": 11.671977285274863,
+      "grad_norm": 1.3515625,
+      "learning_rate": 9.552684210526316e-05,
+      "loss": 1.3139,
+      "step": 18500
+    },
+    {
+      "epoch": 11.703525514630492,
+      "grad_norm": 1.015625,
+      "learning_rate": 9.550052631578947e-05,
+      "loss": 1.2093,
+      "step": 18550
+    },
+    {
+      "epoch": 11.735073743986119,
+      "grad_norm": 1.203125,
+      "learning_rate": 9.54742105263158e-05,
+      "loss": 1.1618,
+      "step": 18600
+    },
+    {
+      "epoch": 11.766621973341746,
+      "grad_norm": 1.15625,
+      "learning_rate": 9.544789473684211e-05,
+      "loss": 1.2333,
+      "step": 18650
+    },
+    {
+      "epoch": 11.798170202697374,
+      "grad_norm": 1.34375,
+      "learning_rate": 9.542157894736843e-05,
+      "loss": 1.2568,
+      "step": 18700
+    },
+    {
+      "epoch": 11.829718432053001,
+      "grad_norm": 1.1640625,
+      "learning_rate": 9.539526315789474e-05,
+      "loss": 1.2077,
+      "step": 18750
+    },
+    {
+      "epoch": 11.861266661408628,
+      "grad_norm": 1.1015625,
+      "learning_rate": 9.536894736842106e-05,
+      "loss": 1.1648,
+      "step": 18800
+    },
+    {
+      "epoch": 11.892814890764257,
+      "grad_norm": 1.28125,
+      "learning_rate": 9.534263157894737e-05,
+      "loss": 1.201,
+      "step": 18850
+    },
+    {
+      "epoch": 11.924363120119883,
+      "grad_norm": 0.97265625,
+      "learning_rate": 9.531631578947369e-05,
+      "loss": 1.1852,
+      "step": 18900
+    },
+    {
+      "epoch": 11.95591134947551,
+      "grad_norm": 1.2421875,
+      "learning_rate": 9.529e-05,
+      "loss": 1.2017,
+      "step": 18950
+    },
+    {
+      "epoch": 11.987459578831139,
+      "grad_norm": 1.21875,
+      "learning_rate": 9.526368421052632e-05,
+      "loss": 1.24,
+      "step": 19000
+    },
+    {
+      "epoch": 11.987459578831139,
+      "eval_loss": 1.6129719018936157,
+      "eval_runtime": 267.2903,
+      "eval_samples_per_second": 94.867,
+      "eval_steps_per_second": 5.93,
+      "step": 19000
+    },
+    {
+      "epoch": 12.018928937613376,
+      "grad_norm": 1.296875,
+      "learning_rate": 9.523736842105263e-05,
+      "loss": 1.1555,
+      "step": 19050
+    },
+    {
+      "epoch": 12.050477166969005,
+      "grad_norm": 1.2578125,
+      "learning_rate": 9.521105263157895e-05,
+      "loss": 1.2965,
+      "step": 19100
+    },
+    {
+      "epoch": 12.082025396324632,
+      "grad_norm": 1.2734375,
+      "learning_rate": 9.518473684210528e-05,
+      "loss": 1.2043,
+      "step": 19150
+    },
+    {
+      "epoch": 12.113573625680258,
+      "grad_norm": 1.3203125,
+      "learning_rate": 9.515842105263159e-05,
+      "loss": 1.1309,
+      "step": 19200
+    },
+    {
+      "epoch": 12.145121855035885,
+      "grad_norm": 1.265625,
+      "learning_rate": 9.51321052631579e-05,
+      "loss": 1.1215,
+      "step": 19250
+    },
+    {
+      "epoch": 12.176670084391514,
+      "grad_norm": 1.2109375,
+      "learning_rate": 9.510578947368421e-05,
+      "loss": 1.1511,
+      "step": 19300
+    },
+    {
+      "epoch": 12.20821831374714,
+      "grad_norm": 1.3125,
+      "learning_rate": 9.507947368421052e-05,
+      "loss": 1.1652,
+      "step": 19350
+    },
+    {
+      "epoch": 12.239766543102768,
+      "grad_norm": 0.73046875,
+      "learning_rate": 9.505315789473685e-05,
+      "loss": 1.2012,
+      "step": 19400
+    },
+    {
+      "epoch": 12.271314772458396,
+      "grad_norm": 1.25,
+      "learning_rate": 9.502684210526316e-05,
+      "loss": 1.0918,
+      "step": 19450
+    },
+    {
+      "epoch": 12.302863001814023,
+      "grad_norm": 1.09375,
+      "learning_rate": 9.500052631578947e-05,
+      "loss": 1.249,
+      "step": 19500
+    },
+    {
+      "epoch": 12.33441123116965,
+      "grad_norm": 1.0546875,
+      "learning_rate": 9.49742105263158e-05,
+      "loss": 1.1607,
+      "step": 19550
+    },
+    {
+      "epoch": 12.365959460525279,
+      "grad_norm": 1.3984375,
+      "learning_rate": 9.494789473684211e-05,
+      "loss": 1.21,
+      "step": 19600
+    },
+    {
+      "epoch": 12.397507689880905,
+      "grad_norm": 0.94140625,
+      "learning_rate": 9.492157894736843e-05,
+      "loss": 1.23,
+      "step": 19650
+    },
+    {
+      "epoch": 12.429055919236532,
+      "grad_norm": 1.3671875,
+      "learning_rate": 9.489526315789475e-05,
+      "loss": 1.1942,
+      "step": 19700
+    },
+    {
+      "epoch": 12.46060414859216,
+      "grad_norm": 1.328125,
+      "learning_rate": 9.486894736842106e-05,
+      "loss": 1.1034,
+      "step": 19750
+    },
+    {
+      "epoch": 12.492152377947788,
+      "grad_norm": 1.5546875,
+      "learning_rate": 9.484263157894737e-05,
+      "loss": 1.1593,
+      "step": 19800
+    },
+    {
+      "epoch": 12.523700607303415,
+      "grad_norm": 1.0234375,
+      "learning_rate": 9.481631578947368e-05,
+      "loss": 1.2432,
+      "step": 19850
+    },
+    {
+      "epoch": 12.555248836659043,
+      "grad_norm": 1.203125,
+      "learning_rate": 9.479e-05,
+      "loss": 1.2159,
+      "step": 19900
+    },
+    {
+      "epoch": 12.58679706601467,
+      "grad_norm": 1.359375,
+      "learning_rate": 9.476368421052632e-05,
+      "loss": 1.2214,
+      "step": 19950
+    },
+    {
+      "epoch": 12.618345295370297,
+      "grad_norm": 1.3125,
+      "learning_rate": 9.473736842105264e-05,
+      "loss": 1.2311,
+      "step": 20000
+    },
+    {
+      "epoch": 12.618345295370297,
+      "eval_loss": 1.6025601625442505,
+      "eval_runtime": 267.1111,
+      "eval_samples_per_second": 94.931,
+      "eval_steps_per_second": 5.934,
+      "step": 20000
+    },
+    {
+      "epoch": 12.649893524725925,
+      "grad_norm": 1.0703125,
+      "learning_rate": 9.471105263157895e-05,
+      "loss": 1.178,
+      "step": 20050
+    },
+    {
+      "epoch": 12.681441754081552,
+      "grad_norm": 1.15625,
+      "learning_rate": 9.468473684210527e-05,
+      "loss": 1.145,
+      "step": 20100
+    },
+    {
+      "epoch": 12.71298998343718,
+      "grad_norm": 1.3671875,
+      "learning_rate": 9.465842105263159e-05,
+      "loss": 1.1935,
+      "step": 20150
+    },
+    {
+      "epoch": 12.744538212792808,
+      "grad_norm": 1.109375,
+      "learning_rate": 9.46321052631579e-05,
+      "loss": 1.2261,
+      "step": 20200
+    },
+    {
+      "epoch": 12.776086442148435,
+      "grad_norm": 1.03125,
+      "learning_rate": 9.460578947368421e-05,
+      "loss": 1.1438,
+      "step": 20250
+    },
+    {
+      "epoch": 12.807634671504061,
+      "grad_norm": 1.0625,
+      "learning_rate": 9.457947368421053e-05,
+      "loss": 1.1981,
+      "step": 20300
+    },
+    {
+      "epoch": 12.83918290085969,
+      "grad_norm": 1.421875,
+      "learning_rate": 9.455315789473684e-05,
+      "loss": 1.2022,
+      "step": 20350
+    },
+    {
+      "epoch": 12.870731130215317,
+      "grad_norm": 1.2734375,
+      "learning_rate": 9.452684210526316e-05,
+      "loss": 1.1456,
+      "step": 20400
+    },
+    {
+      "epoch": 12.902279359570944,
+      "grad_norm": 1.4140625,
+      "learning_rate": 9.450052631578947e-05,
+      "loss": 1.3042,
+      "step": 20450
+    },
+    {
+      "epoch": 12.933827588926572,
+      "grad_norm": 1.1015625,
+      "learning_rate": 9.44742105263158e-05,
+      "loss": 1.1636,
+      "step": 20500
+    },
+    {
+      "epoch": 12.9653758182822,
+      "grad_norm": 1.140625,
+      "learning_rate": 9.444789473684211e-05,
+      "loss": 1.1927,
+      "step": 20550
+    },
+    {
+      "epoch": 12.996924047637826,
+      "grad_norm": 1.21875,
+      "learning_rate": 9.442157894736842e-05,
+      "loss": 1.19,
+      "step": 20600
+    },
+    {
+      "epoch": 13.028393406420065,
+      "grad_norm": 1.265625,
+      "learning_rate": 9.439526315789475e-05,
+      "loss": 1.1314,
+      "step": 20650
+    },
+    {
+      "epoch": 13.059941635775692,
+      "grad_norm": 1.359375,
+      "learning_rate": 9.436894736842106e-05,
+      "loss": 1.1814,
+      "step": 20700
+    },
+    {
+      "epoch": 13.091489865131319,
+      "grad_norm": 2.484375,
+      "learning_rate": 9.434263157894737e-05,
+      "loss": 1.2089,
+      "step": 20750
+    },
+    {
+      "epoch": 13.123038094486947,
+      "grad_norm": 10.9375,
+      "learning_rate": 9.431631578947368e-05,
+      "loss": 1.1614,
+      "step": 20800
+    },
+    {
+      "epoch": 13.154586323842574,
+      "grad_norm": 1.484375,
+      "learning_rate": 9.429000000000001e-05,
+      "loss": 1.2293,
+      "step": 20850
+    },
+    {
+      "epoch": 13.186134553198201,
+      "grad_norm": 1.2109375,
+      "learning_rate": 9.426368421052632e-05,
+      "loss": 1.1544,
+      "step": 20900
+    },
+    {
+      "epoch": 13.21768278255383,
+      "grad_norm": 0.95703125,
+      "learning_rate": 9.423736842105264e-05,
+      "loss": 1.2051,
+      "step": 20950
+    },
+    {
+      "epoch": 13.249231011909457,
+      "grad_norm": 1.15625,
+      "learning_rate": 9.421105263157896e-05,
+      "loss": 1.1744,
+      "step": 21000
+    },
+    {
+      "epoch": 13.249231011909457,
+      "eval_loss": 1.5931901931762695,
+      "eval_runtime": 266.8272,
+      "eval_samples_per_second": 95.032,
+      "eval_steps_per_second": 5.94,
+      "step": 21000
+    },
+    {
+      "epoch": 13.280779241265083,
+      "grad_norm": 1.2890625,
+      "learning_rate": 9.418473684210527e-05,
+      "loss": 1.2007,
+      "step": 21050
+    },
+    {
+      "epoch": 13.312327470620712,
+      "grad_norm": 1.0703125,
+      "learning_rate": 9.415842105263158e-05,
+      "loss": 1.1093,
+      "step": 21100
+    },
+    {
+      "epoch": 13.343875699976339,
+      "grad_norm": 1.5546875,
+      "learning_rate": 9.41321052631579e-05,
+      "loss": 1.2242,
+      "step": 21150
+    },
+    {
+      "epoch": 13.375423929331966,
+      "grad_norm": 1.234375,
+      "learning_rate": 9.410578947368422e-05,
+      "loss": 1.2096,
+      "step": 21200
+    },
+    {
+      "epoch": 13.406972158687594,
+      "grad_norm": 3.234375,
+      "learning_rate": 9.407947368421053e-05,
+      "loss": 1.1943,
+      "step": 21250
+    },
+    {
+      "epoch": 13.438520388043221,
+      "grad_norm": 1.09375,
+      "learning_rate": 9.405315789473684e-05,
+      "loss": 1.1824,
+      "step": 21300
+    },
+    {
+      "epoch": 13.470068617398848,
+      "grad_norm": 0.95703125,
+      "learning_rate": 9.402684210526316e-05,
+      "loss": 1.2037,
+      "step": 21350
+    },
+    {
+      "epoch": 13.501616846754477,
+      "grad_norm": 1.3125,
+      "learning_rate": 9.400052631578949e-05,
+      "loss": 1.1881,
+      "step": 21400
+    },
+    {
+      "epoch": 13.533165076110103,
+      "grad_norm": 1.625,
+      "learning_rate": 9.39742105263158e-05,
+      "loss": 1.0935,
+      "step": 21450
+    },
+    {
+      "epoch": 13.56471330546573,
+      "grad_norm": 1.1015625,
+      "learning_rate": 9.394789473684211e-05,
+      "loss": 1.1977,
+      "step": 21500
+    },
+    {
+      "epoch": 13.596261534821359,
+      "grad_norm": 1.2265625,
+      "learning_rate": 9.392157894736842e-05,
+      "loss": 1.2146,
+      "step": 21550
+    },
+    {
+      "epoch": 13.627809764176986,
+      "grad_norm": 1.625,
+      "learning_rate": 9.389526315789475e-05,
+      "loss": 1.1826,
+      "step": 21600
+    },
+    {
+      "epoch": 13.659357993532613,
+      "grad_norm": 1.0546875,
+      "learning_rate": 9.386894736842106e-05,
+      "loss": 1.1387,
+      "step": 21650
+    },
+    {
+      "epoch": 13.690906222888241,
+      "grad_norm": 1.53125,
+      "learning_rate": 9.384263157894737e-05,
+      "loss": 1.1169,
+      "step": 21700
+    },
+    {
+      "epoch": 13.722454452243868,
+      "grad_norm": 1.375,
+      "learning_rate": 9.381631578947368e-05,
+      "loss": 1.2465,
+      "step": 21750
+    },
+    {
+      "epoch": 13.754002681599495,
+      "grad_norm": 1.3203125,
+      "learning_rate": 9.379e-05,
+      "loss": 1.2198,
+      "step": 21800
+    },
+    {
+      "epoch": 13.785550910955124,
+      "grad_norm": 1.40625,
+      "learning_rate": 9.376368421052632e-05,
+      "loss": 1.1828,
+      "step": 21850
+    },
+    {
+      "epoch": 13.81709914031075,
+      "grad_norm": 1.2578125,
+      "learning_rate": 9.373736842105265e-05,
+      "loss": 1.1286,
+      "step": 21900
+    },
+    {
+      "epoch": 13.848647369666377,
+      "grad_norm": 1.2421875,
+      "learning_rate": 9.371105263157896e-05,
+      "loss": 1.1864,
+      "step": 21950
+    },
+    {
+      "epoch": 13.880195599022004,
+      "grad_norm": 1.765625,
+      "learning_rate": 9.368473684210527e-05,
+      "loss": 1.1334,
+      "step": 22000
+    },
+    {
+      "epoch": 13.880195599022004,
+      "eval_loss": 1.5843102931976318,
+      "eval_runtime": 267.102,
+      "eval_samples_per_second": 94.934,
+      "eval_steps_per_second": 5.934,
+      "step": 22000
+    },
+    {
+      "epoch": 13.911743828377633,
+      "grad_norm": 1.3828125,
+      "learning_rate": 9.365842105263158e-05,
+      "loss": 1.1686,
+      "step": 22050
+    },
+    {
+      "epoch": 13.94329205773326,
+      "grad_norm": 1.1328125,
+      "learning_rate": 9.36321052631579e-05,
+      "loss": 1.2305,
+      "step": 22100
+    },
+    {
+      "epoch": 13.974840287088886,
+      "grad_norm": 1.171875,
+      "learning_rate": 9.360578947368422e-05,
+      "loss": 1.2467,
+      "step": 22150
+    },
+    {
+      "epoch": 14.006309645871125,
+      "grad_norm": 0.8671875,
+      "learning_rate": 9.357947368421053e-05,
+      "loss": 1.1825,
+      "step": 22200
+    },
+    {
+      "epoch": 14.037857875226752,
+      "grad_norm": 0.9453125,
+      "learning_rate": 9.355315789473684e-05,
+      "loss": 1.2244,
+      "step": 22250
+    },
+    {
+      "epoch": 14.06940610458238,
+      "grad_norm": 1.109375,
+      "learning_rate": 9.352684210526315e-05,
+      "loss": 1.2818,
+      "step": 22300
+    },
+    {
+      "epoch": 14.100954333938008,
+      "grad_norm": 1.4296875,
+      "learning_rate": 9.350052631578948e-05,
+      "loss": 1.2653,
+      "step": 22350
+    },
+    {
+      "epoch": 14.132502563293635,
+      "grad_norm": 1.734375,
+      "learning_rate": 9.34742105263158e-05,
+      "loss": 1.1813,
+      "step": 22400
+    },
+    {
+      "epoch": 14.164050792649263,
+      "grad_norm": 1.0859375,
+      "learning_rate": 9.344789473684211e-05,
+      "loss": 1.1284,
+      "step": 22450
+    },
+    {
+      "epoch": 14.19559902200489,
+      "grad_norm": 1.2578125,
+      "learning_rate": 9.342157894736842e-05,
+      "loss": 1.1927,
+      "step": 22500
+    },
+    {
+      "epoch": 14.227147251360517,
+      "grad_norm": 0.98046875,
+      "learning_rate": 9.339526315789474e-05,
+      "loss": 1.2206,
+      "step": 22550
+    },
+    {
+      "epoch": 14.258695480716145,
+      "grad_norm": 3.734375,
+      "learning_rate": 9.336894736842106e-05,
+      "loss": 1.2683,
+      "step": 22600
+    },
+    {
+      "epoch": 14.290243710071772,
+      "grad_norm": 1.0703125,
+      "learning_rate": 9.334263157894737e-05,
+      "loss": 1.1436,
+      "step": 22650
+    },
+    {
+      "epoch": 14.3217919394274,
+      "grad_norm": 1.3828125,
+      "learning_rate": 9.331631578947368e-05,
+      "loss": 1.1272,
+      "step": 22700
+    },
+    {
+      "epoch": 14.353340168783028,
+      "grad_norm": 1.1875,
+      "learning_rate": 9.329e-05,
+      "loss": 1.1771,
+      "step": 22750
+    },
+    {
+      "epoch": 14.384888398138655,
+      "grad_norm": 1.421875,
+      "learning_rate": 9.326368421052632e-05,
+      "loss": 1.2135,
+      "step": 22800
+    },
+    {
+      "epoch": 14.416436627494281,
+      "grad_norm": 1.296875,
+      "learning_rate": 9.323736842105265e-05,
+      "loss": 1.1247,
+      "step": 22850
+    },
+    {
+      "epoch": 14.44798485684991,
+      "grad_norm": 1.09375,
+      "learning_rate": 9.321105263157896e-05,
+      "loss": 1.0685,
+      "step": 22900
+    },
+    {
+      "epoch": 14.479533086205537,
+      "grad_norm": 1.1328125,
+      "learning_rate": 9.318473684210527e-05,
+      "loss": 1.1276,
+      "step": 22950
+    },
+    {
+      "epoch": 14.511081315561164,
+      "grad_norm": 13.0625,
+      "learning_rate": 9.315842105263158e-05,
+      "loss": 1.1941,
+      "step": 23000
+    },
+    {
+      "epoch": 14.511081315561164,
+      "eval_loss": 1.5769537687301636,
+      "eval_runtime": 266.8284,
+      "eval_samples_per_second": 95.031,
+      "eval_steps_per_second": 5.94,
+      "step": 23000
+    },
+    {
+      "epoch": 14.542629544916792,
+      "grad_norm": 1.34375,
+      "learning_rate": 9.313210526315789e-05,
+      "loss": 1.1838,
+      "step": 23050
+    },
+    {
+      "epoch": 14.57417777427242,
+      "grad_norm": 1.6171875,
+      "learning_rate": 9.310578947368422e-05,
+      "loss": 1.0749,
+      "step": 23100
+    },
+    {
+      "epoch": 14.605726003628046,
+      "grad_norm": 1.21875,
+      "learning_rate": 9.307947368421053e-05,
+      "loss": 1.1389,
+      "step": 23150
+    },
+    {
+      "epoch": 14.637274232983675,
+      "grad_norm": 1.3203125,
+      "learning_rate": 9.305315789473684e-05,
+      "loss": 1.1979,
+      "step": 23200
+    },
+    {
+      "epoch": 14.668822462339302,
+      "grad_norm": 1.2421875,
+      "learning_rate": 9.302684210526317e-05,
+      "loss": 1.1649,
+      "step": 23250
+    },
+    {
+      "epoch": 14.700370691694928,
+      "grad_norm": 0.8671875,
+      "learning_rate": 9.300052631578948e-05,
+      "loss": 1.1631,
+      "step": 23300
+    },
+    {
+      "epoch": 14.731918921050555,
+      "grad_norm": 1.6875,
+      "learning_rate": 9.29742105263158e-05,
+      "loss": 1.2639,
+      "step": 23350
+    },
+    {
+      "epoch": 14.763467150406184,
+      "grad_norm": 1.265625,
+      "learning_rate": 9.294789473684211e-05,
+      "loss": 1.0786,
+      "step": 23400
+    },
+    {
+      "epoch": 14.79501537976181,
+      "grad_norm": 1.28125,
+      "learning_rate": 9.292157894736843e-05,
+      "loss": 1.1023,
+      "step": 23450
+    },
+    {
+      "epoch": 14.826563609117438,
+      "grad_norm": 1.0234375,
+      "learning_rate": 9.289526315789474e-05,
+      "loss": 1.2058,
+      "step": 23500
+    },
+    {
+      "epoch": 14.858111838473066,
+      "grad_norm": 1.1328125,
+      "learning_rate": 9.286894736842105e-05,
+      "loss": 1.2545,
+      "step": 23550
+    },
+    {
+      "epoch": 14.889660067828693,
+      "grad_norm": 1.5,
+      "learning_rate": 9.284263157894737e-05,
+      "loss": 1.1352,
+      "step": 23600
+    },
+    {
+      "epoch": 14.92120829718432,
+      "grad_norm": 1.4453125,
+      "learning_rate": 9.281631578947369e-05,
+      "loss": 1.1925,
+      "step": 23650
+    },
+    {
+      "epoch": 14.952756526539948,
+      "grad_norm": 0.8359375,
+      "learning_rate": 9.279e-05,
+      "loss": 1.1737,
+      "step": 23700
+    },
+    {
+      "epoch": 14.984304755895575,
+      "grad_norm": 1.2265625,
+      "learning_rate": 9.276368421052632e-05,
+      "loss": 1.2321,
+      "step": 23750
+    },
+    {
+      "epoch": 15.015774114677814,
+      "grad_norm": 0.80078125,
+      "learning_rate": 9.273736842105263e-05,
+      "loss": 1.104,
+      "step": 23800
+    },
+    {
+      "epoch": 15.047322344033441,
+      "grad_norm": 1.3046875,
+      "learning_rate": 9.271105263157896e-05,
+      "loss": 1.2197,
+      "step": 23850
+    },
+    {
+      "epoch": 15.078870573389068,
+      "grad_norm": 1.078125,
+      "learning_rate": 9.268473684210527e-05,
+      "loss": 1.1403,
+      "step": 23900
+    },
+    {
+      "epoch": 15.110418802744697,
+      "grad_norm": 1.2578125,
+      "learning_rate": 9.265842105263158e-05,
+      "loss": 1.1551,
+      "step": 23950
+    },
+    {
+      "epoch": 15.141967032100323,
+      "grad_norm": 1.1328125,
+      "learning_rate": 9.26321052631579e-05,
+      "loss": 1.2111,
+      "step": 24000
+    },
+    {
+      "epoch": 15.141967032100323,
+      "eval_loss": 1.570060133934021,
+      "eval_runtime": 266.8513,
+      "eval_samples_per_second": 95.023,
+      "eval_steps_per_second": 5.94,
+      "step": 24000
+    },
+    {
+      "epoch": 15.17351526145595,
+      "grad_norm": 0.98828125,
+      "learning_rate": 9.26057894736842e-05,
+      "loss": 1.193,
+      "step": 24050
+    },
+    {
+      "epoch": 15.205063490811579,
+      "grad_norm": 0.8515625,
+      "learning_rate": 9.257947368421053e-05,
+      "loss": 1.1775,
+      "step": 24100
+    },
+    {
+      "epoch": 15.236611720167206,
+      "grad_norm": 1.1640625,
+      "learning_rate": 9.255315789473684e-05,
+      "loss": 1.0497,
+      "step": 24150
+    },
+    {
+      "epoch": 15.268159949522833,
+      "grad_norm": 1.375,
+      "learning_rate": 9.252684210526317e-05,
+      "loss": 1.2102,
+      "step": 24200
+    },
+    {
+      "epoch": 15.299708178878461,
+      "grad_norm": 0.87890625,
+      "learning_rate": 9.250052631578948e-05,
+      "loss": 1.2412,
+      "step": 24250
+    },
+    {
+      "epoch": 15.331256408234088,
+      "grad_norm": 1.265625,
+      "learning_rate": 9.247421052631579e-05,
+      "loss": 1.1529,
+      "step": 24300
+    },
+    {
+      "epoch": 15.362804637589715,
+      "grad_norm": 1.2421875,
+      "learning_rate": 9.244789473684212e-05,
+      "loss": 1.1953,
+      "step": 24350
+    },
+    {
+      "epoch": 15.394352866945344,
+      "grad_norm": 0.94140625,
+      "learning_rate": 9.242157894736843e-05,
+      "loss": 1.1939,
+      "step": 24400
+    },
+    {
+      "epoch": 15.42590109630097,
+      "grad_norm": 1.3125,
+      "learning_rate": 9.239526315789474e-05,
+      "loss": 1.0984,
+      "step": 24450
+    },
+    {
+      "epoch": 15.457449325656597,
+      "grad_norm": 1.6953125,
+      "learning_rate": 9.236894736842105e-05,
+      "loss": 1.1652,
+      "step": 24500
+    },
+    {
+      "epoch": 15.488997555012224,
+      "grad_norm": 1.34375,
+      "learning_rate": 9.234263157894738e-05,
+      "loss": 1.0468,
+      "step": 24550
+    },
+    {
+      "epoch": 15.520545784367853,
+      "grad_norm": 1.359375,
+      "learning_rate": 9.231631578947369e-05,
+      "loss": 1.2469,
+      "step": 24600
+    },
+    {
+      "epoch": 15.55209401372348,
+      "grad_norm": 1.2734375,
+      "learning_rate": 9.229000000000001e-05,
+      "loss": 1.2097,
+      "step": 24650
+    },
+    {
+      "epoch": 15.583642243079106,
+      "grad_norm": 1.5859375,
+      "learning_rate": 9.226368421052632e-05,
+      "loss": 1.1814,
+      "step": 24700
+    },
+    {
+      "epoch": 15.615190472434735,
+      "grad_norm": 1.6796875,
+      "learning_rate": 9.223736842105264e-05,
+      "loss": 1.1004,
+      "step": 24750
+    },
+    {
+      "epoch": 15.646738701790362,
+      "grad_norm": 1.3125,
+      "learning_rate": 9.221105263157895e-05,
+      "loss": 1.0581,
+      "step": 24800
+    },
+    {
+      "epoch": 15.678286931145989,
+      "grad_norm": 1.4375,
+      "learning_rate": 9.218473684210527e-05,
+      "loss": 1.0528,
+      "step": 24850
+    },
+    {
+      "epoch": 15.709835160501617,
+      "grad_norm": 1.453125,
+      "learning_rate": 9.215842105263158e-05,
+      "loss": 1.1313,
+      "step": 24900
+    },
+    {
+      "epoch": 15.741383389857244,
+      "grad_norm": 1.4765625,
+      "learning_rate": 9.21321052631579e-05,
+      "loss": 1.132,
+      "step": 24950
+    },
+    {
+      "epoch": 15.772931619212871,
+      "grad_norm": 82.0,
+      "learning_rate": 9.210578947368421e-05,
+      "loss": 1.2929,
+      "step": 25000
+    },
+    {
+      "epoch": 15.772931619212871,
+      "eval_loss": 1.5642035007476807,
+      "eval_runtime": 267.4044,
+      "eval_samples_per_second": 94.826,
+      "eval_steps_per_second": 5.927,
+      "step": 25000
+    },
+    {
+      "epoch": 15.8044798485685,
+      "grad_norm": 1.40625,
+      "learning_rate": 9.207947368421053e-05,
+      "loss": 1.1608,
+      "step": 25050
+    },
+    {
+      "epoch": 15.836028077924126,
+      "grad_norm": 1.125,
+      "learning_rate": 9.205315789473684e-05,
+      "loss": 1.2047,
+      "step": 25100
+    },
+    {
+      "epoch": 15.867576307279753,
+      "grad_norm": 1.2265625,
+      "learning_rate": 9.202684210526317e-05,
+      "loss": 1.1987,
+      "step": 25150
+    },
+    {
+      "epoch": 15.899124536635382,
+      "grad_norm": 0.98046875,
+      "learning_rate": 9.200052631578948e-05,
+      "loss": 1.136,
+      "step": 25200
+    },
+    {
+      "epoch": 15.930672765991009,
+      "grad_norm": 1.3046875,
+      "learning_rate": 9.197421052631579e-05,
+      "loss": 1.1869,
+      "step": 25250
+    },
+    {
+      "epoch": 15.962220995346636,
+      "grad_norm": 1.09375,
+      "learning_rate": 9.194789473684212e-05,
+      "loss": 1.1783,
+      "step": 25300
+    },
+    {
+      "epoch": 15.993769224702264,
+      "grad_norm": 1.53125,
+      "learning_rate": 9.192157894736843e-05,
+      "loss": 1.1213,
+      "step": 25350
+    },
+    {
+      "epoch": 16.0252385834845,
+      "grad_norm": 1.375,
+      "learning_rate": 9.189526315789474e-05,
+      "loss": 1.1586,
+      "step": 25400
+    },
+    {
+      "epoch": 16.05678681284013,
+      "grad_norm": 1.6484375,
+      "learning_rate": 9.186894736842105e-05,
+      "loss": 1.1799,
+      "step": 25450
+    },
+    {
+      "epoch": 16.088335042195755,
+      "grad_norm": 0.953125,
+      "learning_rate": 9.184263157894736e-05,
+      "loss": 1.1178,
+      "step": 25500
+    },
+    {
+      "epoch": 16.119883271551384,
+      "grad_norm": 1.1015625,
+      "learning_rate": 9.181631578947369e-05,
+      "loss": 1.1504,
+      "step": 25550
+    },
+    {
+      "epoch": 16.151431500907012,
+      "grad_norm": 1.5078125,
+      "learning_rate": 9.179000000000001e-05,
+      "loss": 1.2006,
+      "step": 25600
+    },
+    {
+      "epoch": 16.182979730262637,
+      "grad_norm": 1.2109375,
+      "learning_rate": 9.176368421052633e-05,
+      "loss": 1.1127,
+      "step": 25650
+    },
+    {
+      "epoch": 16.214527959618266,
+      "grad_norm": 1.3515625,
+      "learning_rate": 9.173736842105264e-05,
+      "loss": 1.1468,
+      "step": 25700
+    },
+    {
+      "epoch": 16.246076188973895,
+      "grad_norm": 1.171875,
+      "learning_rate": 9.171105263157895e-05,
+      "loss": 1.1406,
+      "step": 25750
+    },
+    {
+      "epoch": 16.27762441832952,
+      "grad_norm": 1.4765625,
+      "learning_rate": 9.168473684210527e-05,
+      "loss": 1.0935,
+      "step": 25800
+    },
+    {
+      "epoch": 16.30917264768515,
+      "grad_norm": 1.2109375,
+      "learning_rate": 9.165842105263159e-05,
+      "loss": 1.1796,
+      "step": 25850
+    },
+    {
+      "epoch": 16.340720877040777,
+      "grad_norm": 1.5546875,
+      "learning_rate": 9.16321052631579e-05,
+      "loss": 1.1953,
+      "step": 25900
+    },
+    {
+      "epoch": 16.372269106396402,
+      "grad_norm": 1.3203125,
+      "learning_rate": 9.160578947368421e-05,
+      "loss": 1.137,
+      "step": 25950
+    },
+    {
+      "epoch": 16.40381733575203,
+      "grad_norm": 2.6875,
+      "learning_rate": 9.157947368421052e-05,
+      "loss": 1.1988,
+      "step": 26000
+    },
+    {
+      "epoch": 16.40381733575203,
+      "eval_loss": 1.5581732988357544,
+      "eval_runtime": 267.3647,
+      "eval_samples_per_second": 94.841,
+      "eval_steps_per_second": 5.928,
+      "step": 26000
+    },
+    {
+      "epoch": 16.43536556510766,
+      "grad_norm": 1.1796875,
+      "learning_rate": 9.155315789473685e-05,
+      "loss": 1.0596,
+      "step": 26050
+    },
+    {
+      "epoch": 16.466913794463284,
+      "grad_norm": 1.6171875,
+      "learning_rate": 9.152684210526317e-05,
+      "loss": 1.0931,
+      "step": 26100
+    },
+    {
+      "epoch": 16.498462023818913,
+      "grad_norm": 1.3125,
+      "learning_rate": 9.150052631578948e-05,
+      "loss": 1.1419,
+      "step": 26150
+    },
+    {
+      "epoch": 16.53001025317454,
+      "grad_norm": 1.28125,
+      "learning_rate": 9.14742105263158e-05,
+      "loss": 1.1514,
+      "step": 26200
+    },
+    {
+      "epoch": 16.561558482530167,
+      "grad_norm": 1.0625,
+      "learning_rate": 9.14478947368421e-05,
+      "loss": 1.1669,
+      "step": 26250
+    },
+    {
+      "epoch": 16.593106711885795,
+      "grad_norm": 1.546875,
+      "learning_rate": 9.142157894736843e-05,
+      "loss": 1.1621,
+      "step": 26300
+    },
+    {
+      "epoch": 16.624654941241424,
+      "grad_norm": 1.28125,
+      "learning_rate": 9.139526315789474e-05,
+      "loss": 1.263,
+      "step": 26350
+    },
+    {
+      "epoch": 16.65620317059705,
+      "grad_norm": 1.1953125,
+      "learning_rate": 9.136894736842105e-05,
+      "loss": 1.1713,
+      "step": 26400
+    },
+    {
+      "epoch": 16.687751399952678,
+      "grad_norm": 1.5859375,
+      "learning_rate": 9.134263157894737e-05,
+      "loss": 1.2331,
+      "step": 26450
+    },
+    {
+      "epoch": 16.719299629308306,
+      "grad_norm": 1.3125,
+      "learning_rate": 9.131631578947368e-05,
+      "loss": 1.1634,
+      "step": 26500
+    },
+    {
+      "epoch": 16.75084785866393,
+      "grad_norm": 1.71875,
+      "learning_rate": 9.129000000000002e-05,
+      "loss": 1.0728,
+      "step": 26550
+    },
+    {
+      "epoch": 16.78239608801956,
+      "grad_norm": 1.4453125,
+      "learning_rate": 9.126368421052633e-05,
+      "loss": 1.279,
+      "step": 26600
+    },
+    {
+      "epoch": 16.81394431737519,
+      "grad_norm": 1.234375,
+      "learning_rate": 9.123736842105264e-05,
+      "loss": 1.1402,
+      "step": 26650
+    },
+    {
+      "epoch": 16.845492546730814,
+      "grad_norm": 1.6328125,
+      "learning_rate": 9.121105263157895e-05,
+      "loss": 1.1017,
+      "step": 26700
+    },
+    {
+      "epoch": 16.877040776086442,
+      "grad_norm": 1.4296875,
+      "learning_rate": 9.118473684210526e-05,
+      "loss": 1.0707,
+      "step": 26750
+    },
+    {
+      "epoch": 16.90858900544207,
+      "grad_norm": 1.859375,
+      "learning_rate": 9.115842105263159e-05,
+      "loss": 1.2186,
+      "step": 26800
+    },
+    {
+      "epoch": 16.940137234797696,
+      "grad_norm": 1.2578125,
+      "learning_rate": 9.11321052631579e-05,
+      "loss": 1.2004,
+      "step": 26850
+    },
+    {
+      "epoch": 16.971685464153325,
+      "grad_norm": 1.5078125,
+      "learning_rate": 9.110578947368421e-05,
+      "loss": 1.2102,
+      "step": 26900
+    },
+    {
+      "epoch": 17.003154822935564,
+      "grad_norm": 1.3203125,
+      "learning_rate": 9.107947368421052e-05,
+      "loss": 1.1377,
+      "step": 26950
+    },
+    {
+      "epoch": 17.03470305229119,
+      "grad_norm": 1.28125,
+      "learning_rate": 9.105315789473685e-05,
+      "loss": 1.1954,
+      "step": 27000
+    },
+    {
+      "epoch": 17.03470305229119,
+      "eval_loss": 1.5533775091171265,
+      "eval_runtime": 267.3193,
+      "eval_samples_per_second": 94.857,
+      "eval_steps_per_second": 5.929,
+      "step": 27000
+    },
+    {
+      "epoch": 17.066251281646817,
+      "grad_norm": 1.3984375,
+      "learning_rate": 9.102684210526317e-05,
+      "loss": 1.1347,
+      "step": 27050
+    },
+    {
+      "epoch": 17.097799511002446,
+      "grad_norm": 0.72265625,
+      "learning_rate": 9.100052631578948e-05,
+      "loss": 1.2264,
+      "step": 27100
+    },
+    {
+      "epoch": 17.12934774035807,
+      "grad_norm": 1.078125,
+      "learning_rate": 9.09742105263158e-05,
+      "loss": 1.1454,
+      "step": 27150
+    },
+    {
+      "epoch": 17.1608959697137,
+      "grad_norm": 1.3359375,
+      "learning_rate": 9.09478947368421e-05,
+      "loss": 1.136,
+      "step": 27200
+    },
+    {
+      "epoch": 17.192444199069328,
+      "grad_norm": 1.3125,
+      "learning_rate": 9.092157894736842e-05,
+      "loss": 1.1995,
+      "step": 27250
+    },
+    {
+      "epoch": 17.223992428424953,
+      "grad_norm": 1.1875,
+      "learning_rate": 9.089526315789474e-05,
+      "loss": 1.1956,
+      "step": 27300
+    },
+    {
+      "epoch": 17.255540657780582,
+      "grad_norm": 0.91015625,
+      "learning_rate": 9.086894736842105e-05,
+      "loss": 1.1551,
+      "step": 27350
+    },
+    {
+      "epoch": 17.28708888713621,
+      "grad_norm": 2.765625,
+      "learning_rate": 9.084263157894737e-05,
+      "loss": 1.192,
+      "step": 27400
+    },
+    {
+      "epoch": 17.318637116491836,
+      "grad_norm": 1.2578125,
+      "learning_rate": 9.081631578947369e-05,
+      "loss": 1.1508,
+      "step": 27450
+    },
+    {
+      "epoch": 17.350185345847464,
+      "grad_norm": 1.3515625,
+      "learning_rate": 9.079e-05,
+      "loss": 1.105,
+      "step": 27500
+    },
+    {
+      "epoch": 17.381733575203093,
+      "grad_norm": 1.25,
+      "learning_rate": 9.076368421052633e-05,
+      "loss": 1.2126,
+      "step": 27550
+    },
+    {
+      "epoch": 17.413281804558718,
+      "grad_norm": 1.1328125,
+      "learning_rate": 9.073736842105264e-05,
+      "loss": 1.2555,
+      "step": 27600
+    },
+    {
+      "epoch": 17.444830033914347,
+      "grad_norm": 1.0078125,
+      "learning_rate": 9.071105263157895e-05,
+      "loss": 1.1146,
+      "step": 27650
+    },
+    {
+      "epoch": 17.476378263269975,
+      "grad_norm": 1.25,
+      "learning_rate": 9.068473684210526e-05,
+      "loss": 1.2614,
+      "step": 27700
+    },
+    {
+      "epoch": 17.5079264926256,
+      "grad_norm": 1.421875,
+      "learning_rate": 9.065842105263157e-05,
+      "loss": 1.0694,
+      "step": 27750
+    },
+    {
+      "epoch": 17.53947472198123,
+      "grad_norm": 0.859375,
+      "learning_rate": 9.06321052631579e-05,
+      "loss": 1.0467,
+      "step": 27800
+    },
+    {
+      "epoch": 17.571022951336857,
+      "grad_norm": 1.2578125,
+      "learning_rate": 9.060578947368421e-05,
+      "loss": 1.1542,
+      "step": 27850
+    },
+    {
+      "epoch": 17.602571180692482,
+      "grad_norm": 1.0625,
+      "learning_rate": 9.057947368421052e-05,
+      "loss": 1.1161,
+      "step": 27900
+    },
+    {
+      "epoch": 17.63411941004811,
+      "grad_norm": 1.0546875,
+      "learning_rate": 9.055315789473685e-05,
+      "loss": 1.1102,
+      "step": 27950
+    },
+    {
+      "epoch": 17.66566763940374,
+      "grad_norm": 1.125,
+      "learning_rate": 9.052684210526316e-05,
+      "loss": 1.1637,
+      "step": 28000
+    },
+    {
+      "epoch": 17.66566763940374,
+      "eval_loss": 1.5486234426498413,
+      "eval_runtime": 267.4326,
+      "eval_samples_per_second": 94.816,
+      "eval_steps_per_second": 5.927,
+      "step": 28000
+    },
+    {
+      "epoch": 17.697215868759365,
+      "grad_norm": 1.34375,
+      "learning_rate": 9.050052631578948e-05,
+      "loss": 1.0892,
+      "step": 28050
+    },
+    {
+      "epoch": 17.728764098114993,
+      "grad_norm": 1.1015625,
+      "learning_rate": 9.04742105263158e-05,
+      "loss": 1.1039,
+      "step": 28100
+    },
+    {
+      "epoch": 17.760312327470622,
+      "grad_norm": 1.2890625,
+      "learning_rate": 9.044789473684211e-05,
+      "loss": 1.0568,
+      "step": 28150
+    },
+    {
+      "epoch": 17.791860556826247,
+      "grad_norm": 1.328125,
+      "learning_rate": 9.042157894736842e-05,
+      "loss": 1.2409,
+      "step": 28200
+    },
+    {
+      "epoch": 17.823408786181876,
+      "grad_norm": 1.140625,
+      "learning_rate": 9.039526315789474e-05,
+      "loss": 1.1879,
+      "step": 28250
+    },
+    {
+      "epoch": 17.854957015537504,
+      "grad_norm": 1.109375,
+      "learning_rate": 9.036894736842106e-05,
+      "loss": 1.2804,
+      "step": 28300
+    },
+    {
+      "epoch": 17.88650524489313,
+      "grad_norm": 1.1796875,
+      "learning_rate": 9.034263157894737e-05,
+      "loss": 1.1923,
+      "step": 28350
+    },
+    {
+      "epoch": 17.918053474248758,
+      "grad_norm": 1.4296875,
+      "learning_rate": 9.031631578947369e-05,
+      "loss": 1.0509,
+      "step": 28400
+    },
+    {
+      "epoch": 17.949601703604387,
+      "grad_norm": 1.3125,
+      "learning_rate": 9.029e-05,
+      "loss": 1.2324,
+      "step": 28450
+    },
+    {
+      "epoch": 17.98114993296001,
+      "grad_norm": 1.3203125,
+      "learning_rate": 9.026368421052632e-05,
+      "loss": 1.1335,
+      "step": 28500
+    },
+    {
+      "epoch": 18.01261929174225,
+      "grad_norm": 1.5390625,
+      "learning_rate": 9.023736842105264e-05,
+      "loss": 1.1168,
+      "step": 28550
+    },
+    {
+      "epoch": 18.04416752109788,
+      "grad_norm": 1.1015625,
+      "learning_rate": 9.021105263157895e-05,
+      "loss": 1.1836,
+      "step": 28600
+    },
+    {
+      "epoch": 18.075715750453504,
+      "grad_norm": 1.359375,
+      "learning_rate": 9.018473684210526e-05,
+      "loss": 1.2124,
+      "step": 28650
+    },
+    {
+      "epoch": 18.107263979809133,
+      "grad_norm": 1.328125,
+      "learning_rate": 9.015842105263158e-05,
+      "loss": 1.041,
+      "step": 28700
+    },
+    {
+      "epoch": 18.13881220916476,
+      "grad_norm": 1.0625,
+      "learning_rate": 9.01321052631579e-05,
+      "loss": 1.1059,
+      "step": 28750
+    },
+    {
+      "epoch": 18.170360438520387,
+      "grad_norm": 1.328125,
+      "learning_rate": 9.010578947368421e-05,
+      "loss": 1.1117,
+      "step": 28800
+    },
+    {
+      "epoch": 18.201908667876015,
+      "grad_norm": 1.1640625,
+      "learning_rate": 9.007947368421054e-05,
+      "loss": 1.1449,
+      "step": 28850
+    },
+    {
+      "epoch": 18.233456897231644,
+      "grad_norm": 1.078125,
+      "learning_rate": 9.005315789473685e-05,
+      "loss": 1.0887,
+      "step": 28900
+    },
+    {
+      "epoch": 18.26500512658727,
+      "grad_norm": 1.203125,
+      "learning_rate": 9.002684210526316e-05,
+      "loss": 1.1533,
+      "step": 28950
+    },
+    {
+      "epoch": 18.296553355942898,
+      "grad_norm": 1.2109375,
+      "learning_rate": 9.000052631578949e-05,
+      "loss": 1.114,
+      "step": 29000
+    },
+    {
+      "epoch": 18.296553355942898,
+      "eval_loss": 1.5437445640563965,
+      "eval_runtime": 266.5209,
+      "eval_samples_per_second": 95.141,
+      "eval_steps_per_second": 5.947,
+      "step": 29000
+    },
+    {
+      "epoch": 18.328101585298526,
+      "grad_norm": 1.3671875,
+      "learning_rate": 8.99742105263158e-05,
+      "loss": 1.1433,
+      "step": 29050
+    },
+    {
+      "epoch": 18.35964981465415,
+      "grad_norm": 1.1640625,
+      "learning_rate": 8.994789473684211e-05,
+      "loss": 1.2408,
+      "step": 29100
+    },
+    {
+      "epoch": 18.39119804400978,
+      "grad_norm": 1.1796875,
+      "learning_rate": 8.992157894736842e-05,
+      "loss": 1.2331,
+      "step": 29150
+    },
+    {
+      "epoch": 18.42274627336541,
+      "grad_norm": 1.1796875,
+      "learning_rate": 8.989526315789473e-05,
+      "loss": 1.1272,
+      "step": 29200
+    },
+    {
+      "epoch": 18.454294502721034,
+      "grad_norm": 1.21875,
+      "learning_rate": 8.986894736842106e-05,
+      "loss": 1.1593,
+      "step": 29250
+    },
+    {
+      "epoch": 18.485842732076662,
+      "grad_norm": 1.09375,
+      "learning_rate": 8.984263157894738e-05,
+      "loss": 1.1064,
+      "step": 29300
+    },
+    {
+      "epoch": 18.51739096143229,
+      "grad_norm": 1.484375,
+      "learning_rate": 8.98163157894737e-05,
+      "loss": 1.1021,
+      "step": 29350
+    },
+    {
+      "epoch": 18.548939190787916,
+      "grad_norm": 1.3828125,
+      "learning_rate": 8.979e-05,
+      "loss": 1.1155,
+      "step": 29400
+    },
+    {
+      "epoch": 18.580487420143545,
+      "grad_norm": 1.359375,
+      "learning_rate": 8.976368421052632e-05,
+      "loss": 1.0826,
+      "step": 29450
+    },
+    {
+      "epoch": 18.612035649499173,
+      "grad_norm": 1.0625,
+      "learning_rate": 8.973736842105264e-05,
+      "loss": 1.1426,
+      "step": 29500
+    },
+    {
+      "epoch": 18.6435838788548,
+      "grad_norm": 1.1875,
+      "learning_rate": 8.971105263157895e-05,
+      "loss": 1.2192,
+      "step": 29550
+    },
+    {
+      "epoch": 18.675132108210427,
+      "grad_norm": 3.21875,
+      "learning_rate": 8.968473684210527e-05,
+      "loss": 1.2204,
+      "step": 29600
+    },
+    {
+      "epoch": 18.706680337566056,
+      "grad_norm": 1.515625,
+      "learning_rate": 8.965842105263158e-05,
+      "loss": 1.2113,
+      "step": 29650
+    },
+    {
+      "epoch": 18.73822856692168,
+      "grad_norm": 1.0703125,
+      "learning_rate": 8.963210526315789e-05,
+      "loss": 1.0822,
+      "step": 29700
+    },
+    {
+      "epoch": 18.76977679627731,
+      "grad_norm": 1.203125,
+      "learning_rate": 8.960578947368421e-05,
+      "loss": 1.1272,
+      "step": 29750
+    },
+    {
+      "epoch": 18.801325025632938,
+      "grad_norm": 1.3203125,
+      "learning_rate": 8.957947368421054e-05,
+      "loss": 1.0672,
+      "step": 29800
+    },
+    {
+      "epoch": 18.832873254988563,
+      "grad_norm": 1.6015625,
+      "learning_rate": 8.955315789473685e-05,
+      "loss": 1.1568,
+      "step": 29850
+    },
+    {
+      "epoch": 18.86442148434419,
+      "grad_norm": 0.8828125,
+      "learning_rate": 8.952684210526316e-05,
+      "loss": 1.1823,
+      "step": 29900
+    },
+    {
+      "epoch": 18.89596971369982,
+      "grad_norm": 1.453125,
+      "learning_rate": 8.950052631578947e-05,
+      "loss": 1.2598,
+      "step": 29950
+    },
+    {
+      "epoch": 18.927517943055445,
+      "grad_norm": 1.9375,
+      "learning_rate": 8.94742105263158e-05,
+      "loss": 1.2443,
+      "step": 30000
+    },
+    {
+      "epoch": 18.927517943055445,
+      "eval_loss": 1.5400972366333008,
+      "eval_runtime": 266.768,
+      "eval_samples_per_second": 95.053,
+      "eval_steps_per_second": 5.941,
+      "step": 30000
+    },
+    {
+      "epoch": 18.959066172411074,
+      "grad_norm": 1.0625,
+      "learning_rate": 8.944789473684211e-05,
+      "loss": 1.2057,
+      "step": 30050
+    },
+    {
+      "epoch": 18.990614401766702,
+      "grad_norm": 1.3359375,
+      "learning_rate": 8.942157894736842e-05,
+      "loss": 1.2,
+      "step": 30100
+    },
+    {
+      "epoch": 19.022083760548938,
+      "grad_norm": 1.125,
+      "learning_rate": 8.939526315789473e-05,
+      "loss": 1.231,
+      "step": 30150
+    },
+    {
+      "epoch": 19.053631989904567,
+      "grad_norm": 1.2421875,
+      "learning_rate": 8.936894736842105e-05,
+      "loss": 1.133,
+      "step": 30200
+    },
+    {
+      "epoch": 19.085180219260195,
+      "grad_norm": 1.359375,
+      "learning_rate": 8.934263157894738e-05,
+      "loss": 1.2142,
+      "step": 30250
+    },
+    {
+      "epoch": 19.11672844861582,
+      "grad_norm": 3.65625,
+      "learning_rate": 8.93163157894737e-05,
+      "loss": 1.1454,
+      "step": 30300
+    },
+    {
+      "epoch": 19.14827667797145,
+      "grad_norm": 2.109375,
+      "learning_rate": 8.929000000000001e-05,
+      "loss": 1.1329,
+      "step": 30350
+    },
+    {
+      "epoch": 19.179824907327077,
+      "grad_norm": 1.3046875,
+      "learning_rate": 8.926368421052632e-05,
+      "loss": 1.1663,
+      "step": 30400
+    },
+    {
+      "epoch": 19.211373136682703,
+      "grad_norm": 1.0703125,
+      "learning_rate": 8.923736842105263e-05,
+      "loss": 1.1911,
+      "step": 30450
+    },
+    {
+      "epoch": 19.24292136603833,
+      "grad_norm": 0.75,
+      "learning_rate": 8.921105263157896e-05,
+      "loss": 1.1312,
+      "step": 30500
+    },
+    {
+      "epoch": 19.27446959539396,
+      "grad_norm": 1.4375,
+      "learning_rate": 8.918473684210527e-05,
+      "loss": 1.0483,
+      "step": 30550
+    },
+    {
+      "epoch": 19.306017824749585,
+      "grad_norm": 1.296875,
+      "learning_rate": 8.915842105263158e-05,
+      "loss": 1.2329,
+      "step": 30600
+    },
+    {
+      "epoch": 19.337566054105213,
+      "grad_norm": 1.1328125,
+      "learning_rate": 8.913210526315789e-05,
+      "loss": 1.096,
+      "step": 30650
+    },
+    {
+      "epoch": 19.369114283460842,
+      "grad_norm": 2.203125,
+      "learning_rate": 8.910578947368422e-05,
+      "loss": 1.1441,
+      "step": 30700
+    },
+    {
+      "epoch": 19.400662512816467,
+      "grad_norm": 1.375,
+      "learning_rate": 8.907947368421054e-05,
+      "loss": 1.1642,
+      "step": 30750
+    },
+    {
+      "epoch": 19.432210742172096,
+      "grad_norm": 0.984375,
+      "learning_rate": 8.905315789473685e-05,
+      "loss": 1.1549,
+      "step": 30800
+    },
+    {
+      "epoch": 19.463758971527724,
+      "grad_norm": 3.25,
+      "learning_rate": 8.902684210526316e-05,
+      "loss": 1.1997,
+      "step": 30850
+    },
+    {
+      "epoch": 19.49530720088335,
+      "grad_norm": 1.2578125,
+      "learning_rate": 8.900052631578948e-05,
+      "loss": 1.1528,
+      "step": 30900
+    },
+    {
+      "epoch": 19.526855430238978,
+      "grad_norm": 1.1953125,
+      "learning_rate": 8.897421052631579e-05,
+      "loss": 1.093,
+      "step": 30950
+    },
+    {
+      "epoch": 19.558403659594607,
+      "grad_norm": 1.3359375,
+      "learning_rate": 8.894789473684211e-05,
+      "loss": 1.2028,
+      "step": 31000
+    },
+    {
+      "epoch": 19.558403659594607,
+      "eval_loss": 1.5360363721847534,
+      "eval_runtime": 267.2349,
+      "eval_samples_per_second": 94.887,
+      "eval_steps_per_second": 5.931,
+      "step": 31000
+    },
+    {
+      "epoch": 19.58995188895023,
+      "grad_norm": 1.2421875,
+      "learning_rate": 8.892157894736842e-05,
+      "loss": 1.1147,
+      "step": 31050
+    },
+    {
+      "epoch": 19.62150011830586,
+      "grad_norm": 1.0859375,
+      "learning_rate": 8.889526315789474e-05,
+      "loss": 1.1735,
+      "step": 31100
+    },
+    {
+      "epoch": 19.65304834766149,
+      "grad_norm": 0.9921875,
+      "learning_rate": 8.886894736842105e-05,
+      "loss": 1.1778,
+      "step": 31150
+    },
+    {
+      "epoch": 19.684596577017114,
+      "grad_norm": 1.328125,
+      "learning_rate": 8.884263157894737e-05,
+      "loss": 1.1468,
+      "step": 31200
+    },
+    {
+      "epoch": 19.716144806372743,
+      "grad_norm": 1.0703125,
+      "learning_rate": 8.88163157894737e-05,
+      "loss": 1.1041,
+      "step": 31250
+    },
+    {
+      "epoch": 19.74769303572837,
+      "grad_norm": 0.9921875,
+      "learning_rate": 8.879000000000001e-05,
+      "loss": 1.144,
+      "step": 31300
+    },
+    {
+      "epoch": 19.779241265083996,
+      "grad_norm": 0.9296875,
+      "learning_rate": 8.876368421052632e-05,
+      "loss": 1.1797,
+      "step": 31350
+    },
+    {
+      "epoch": 19.810789494439625,
+      "grad_norm": 1.2265625,
+      "learning_rate": 8.873736842105263e-05,
+      "loss": 1.1098,
+      "step": 31400
+    },
+    {
+      "epoch": 19.842337723795254,
+      "grad_norm": 9.75,
+      "learning_rate": 8.871105263157894e-05,
+      "loss": 1.1241,
+      "step": 31450
+    },
+    {
+      "epoch": 19.87388595315088,
+      "grad_norm": 1.234375,
+      "learning_rate": 8.868473684210527e-05,
+      "loss": 1.0782,
+      "step": 31500
+    },
+    {
+      "epoch": 19.905434182506507,
+      "grad_norm": 1.828125,
+      "learning_rate": 8.865842105263158e-05,
+      "loss": 1.1342,
+      "step": 31550
+    },
+    {
+      "epoch": 19.936982411862136,
+      "grad_norm": 1.1171875,
+      "learning_rate": 8.863210526315789e-05,
+      "loss": 1.0846,
+      "step": 31600
+    },
+    {
+      "epoch": 19.96853064121776,
+      "grad_norm": 1.2890625,
+      "learning_rate": 8.860578947368422e-05,
+      "loss": 1.1876,
+      "step": 31650
+    },
+    {
+      "epoch": 20.0,
+      "grad_norm": 1.328125,
+      "learning_rate": 8.857947368421053e-05,
+      "loss": 1.2064,
+      "step": 31700
+    },
+    {
+      "epoch": 20.03154822935563,
+      "grad_norm": 1.5234375,
+      "learning_rate": 8.855315789473685e-05,
+      "loss": 1.2347,
+      "step": 31750
+    },
+    {
+      "epoch": 20.063096458711254,
+      "grad_norm": 1.375,
+      "learning_rate": 8.852684210526317e-05,
+      "loss": 1.1243,
+      "step": 31800
+    },
+    {
+      "epoch": 20.094644688066882,
+      "grad_norm": 1.2421875,
+      "learning_rate": 8.850052631578948e-05,
+      "loss": 1.1414,
+      "step": 31850
+    },
+    {
+      "epoch": 20.12619291742251,
+      "grad_norm": 1.3203125,
+      "learning_rate": 8.847421052631579e-05,
+      "loss": 1.1736,
+      "step": 31900
+    },
+    {
+      "epoch": 20.157741146778136,
+      "grad_norm": 1.125,
+      "learning_rate": 8.844789473684211e-05,
+      "loss": 1.1636,
+      "step": 31950
+    },
+    {
+      "epoch": 20.189289376133765,
+      "grad_norm": 1.484375,
+      "learning_rate": 8.842157894736843e-05,
+      "loss": 1.1606,
+      "step": 32000
+    },
+    {
+      "epoch": 20.189289376133765,
+      "eval_loss": 1.5331028699874878,
+      "eval_runtime": 266.9276,
+      "eval_samples_per_second": 94.996,
+      "eval_steps_per_second": 5.938,
+      "step": 32000
+    },
+    {
+      "epoch": 20.220837605489393,
+      "grad_norm": 1.015625,
+      "learning_rate": 8.839526315789474e-05,
+      "loss": 1.0825,
+      "step": 32050
+    },
+    {
+      "epoch": 20.25238583484502,
+      "grad_norm": 1.28125,
+      "learning_rate": 8.836894736842106e-05,
+      "loss": 1.1555,
+      "step": 32100
+    },
+    {
+      "epoch": 20.283934064200647,
+      "grad_norm": 1.4765625,
+      "learning_rate": 8.834263157894737e-05,
+      "loss": 1.1658,
+      "step": 32150
+    },
+    {
+      "epoch": 20.315482293556276,
+      "grad_norm": 1.3046875,
+      "learning_rate": 8.831631578947369e-05,
+      "loss": 1.1245,
+      "step": 32200
+    },
+    {
+      "epoch": 20.3470305229119,
+      "grad_norm": 1.4375,
+      "learning_rate": 8.829000000000001e-05,
+      "loss": 1.2383,
+      "step": 32250
+    },
+    {
+      "epoch": 20.37857875226753,
+      "grad_norm": 0.90625,
+      "learning_rate": 8.826368421052632e-05,
+      "loss": 1.1005,
+      "step": 32300
+    },
+    {
+      "epoch": 20.410126981623158,
+      "grad_norm": 0.984375,
+      "learning_rate": 8.823736842105263e-05,
+      "loss": 1.1315,
+      "step": 32350
+    },
+    {
+      "epoch": 20.441675210978783,
+      "grad_norm": 1.0703125,
+      "learning_rate": 8.821105263157894e-05,
+      "loss": 1.1512,
+      "step": 32400
+    },
+    {
+      "epoch": 20.47322344033441,
+      "grad_norm": 1.25,
+      "learning_rate": 8.818473684210527e-05,
+      "loss": 1.1167,
+      "step": 32450
+    },
+    {
+      "epoch": 20.50477166969004,
+      "grad_norm": 1.1171875,
+      "learning_rate": 8.815842105263158e-05,
+      "loss": 1.1824,
+      "step": 32500
+    },
+    {
+      "epoch": 20.536319899045665,
+      "grad_norm": 1.3671875,
+      "learning_rate": 8.81321052631579e-05,
+      "loss": 1.1381,
+      "step": 32550
+    },
+    {
+      "epoch": 20.567868128401294,
+      "grad_norm": 1.03125,
+      "learning_rate": 8.810578947368422e-05,
+      "loss": 1.1448,
+      "step": 32600
+    },
+    {
+      "epoch": 20.599416357756922,
+      "grad_norm": 0.72265625,
+      "learning_rate": 8.807947368421053e-05,
+      "loss": 1.1,
+      "step": 32650
+    },
+    {
+      "epoch": 20.630964587112548,
+      "grad_norm": 1.5859375,
+      "learning_rate": 8.805315789473686e-05,
+      "loss": 1.1527,
+      "step": 32700
+    },
+    {
+      "epoch": 20.662512816468176,
+      "grad_norm": 1.3984375,
+      "learning_rate": 8.802684210526317e-05,
+      "loss": 1.1162,
+      "step": 32750
+    },
+    {
+      "epoch": 20.694061045823805,
+      "grad_norm": 0.984375,
+      "learning_rate": 8.800052631578948e-05,
+      "loss": 1.127,
+      "step": 32800
+    },
+    {
+      "epoch": 20.72560927517943,
+      "grad_norm": 1.28125,
+      "learning_rate": 8.797421052631579e-05,
+      "loss": 1.1265,
+      "step": 32850
+    },
+    {
+      "epoch": 20.75715750453506,
+      "grad_norm": 1.078125,
+      "learning_rate": 8.79478947368421e-05,
+      "loss": 1.1595,
+      "step": 32900
+    },
+    {
+      "epoch": 20.788705733890687,
+      "grad_norm": 1.46875,
+      "learning_rate": 8.792157894736843e-05,
+      "loss": 1.1433,
+      "step": 32950
+    },
+    {
+      "epoch": 20.820253963246312,
+      "grad_norm": 1.5234375,
+      "learning_rate": 8.789526315789474e-05,
+      "loss": 1.1601,
+      "step": 33000
+    },
+    {
+      "epoch": 20.820253963246312,
+      "eval_loss": 1.5301166772842407,
+      "eval_runtime": 267.0084,
+      "eval_samples_per_second": 94.967,
+      "eval_steps_per_second": 5.936,
+      "step": 33000
+    },
+    {
+      "epoch": 20.85180219260194,
+      "grad_norm": 1.0859375,
+      "learning_rate": 8.786894736842106e-05,
+      "loss": 1.154,
+      "step": 33050
+    },
+    {
+      "epoch": 20.883350421957566,
+      "grad_norm": 1.1875,
+      "learning_rate": 8.784263157894737e-05,
+      "loss": 1.276,
+      "step": 33100
+    },
+    {
+      "epoch": 20.914898651313194,
+      "grad_norm": 1.75,
+      "learning_rate": 8.781631578947369e-05,
+      "loss": 1.1272,
+      "step": 33150
+    },
+    {
+      "epoch": 20.946446880668823,
+      "grad_norm": 1.6015625,
+      "learning_rate": 8.779000000000001e-05,
+      "loss": 1.1104,
+      "step": 33200
+    },
+    {
+      "epoch": 20.977995110024448,
+      "grad_norm": 1.046875,
+      "learning_rate": 8.776368421052632e-05,
+      "loss": 1.0886,
+      "step": 33250
+    },
+    {
+      "epoch": 21.009464468806687,
+      "grad_norm": 1.390625,
+      "learning_rate": 8.773736842105263e-05,
+      "loss": 1.0329,
+      "step": 33300
+    },
+    {
+      "epoch": 21.041012698162316,
+      "grad_norm": 8.75,
+      "learning_rate": 8.771105263157895e-05,
+      "loss": 1.1335,
+      "step": 33350
+    },
+    {
+      "epoch": 21.072560927517944,
+      "grad_norm": 1.609375,
+      "learning_rate": 8.768473684210526e-05,
+      "loss": 1.2062,
+      "step": 33400
+    },
+    {
+      "epoch": 21.10410915687357,
+      "grad_norm": 1.3125,
+      "learning_rate": 8.765842105263158e-05,
+      "loss": 1.1686,
+      "step": 33450
+    },
+    {
+      "epoch": 21.135657386229198,
+      "grad_norm": 1.625,
+      "learning_rate": 8.763210526315791e-05,
+      "loss": 1.1086,
+      "step": 33500
+    },
+    {
+      "epoch": 21.167205615584827,
+      "grad_norm": 1.3671875,
+      "learning_rate": 8.760578947368422e-05,
+      "loss": 1.1313,
+      "step": 33550
+    },
+    {
+      "epoch": 21.198753844940452,
+      "grad_norm": 1.25,
+      "learning_rate": 8.757947368421053e-05,
+      "loss": 1.1474,
+      "step": 33600
+    },
+    {
+      "epoch": 21.23030207429608,
+      "grad_norm": 0.9921875,
+      "learning_rate": 8.755315789473684e-05,
+      "loss": 1.2645,
+      "step": 33650
+    },
+    {
+      "epoch": 21.26185030365171,
+      "grad_norm": 1.1171875,
+      "learning_rate": 8.752684210526317e-05,
+      "loss": 1.107,
+      "step": 33700
+    },
+    {
+      "epoch": 21.293398533007334,
+      "grad_norm": 1.8046875,
+      "learning_rate": 8.750052631578948e-05,
+      "loss": 1.083,
+      "step": 33750
+    },
+    {
+      "epoch": 21.324946762362963,
+      "grad_norm": 0.97265625,
+      "learning_rate": 8.747421052631579e-05,
+      "loss": 1.1479,
+      "step": 33800
+    },
+    {
+      "epoch": 21.35649499171859,
+      "grad_norm": 1.453125,
+      "learning_rate": 8.74478947368421e-05,
+      "loss": 1.1635,
+      "step": 33850
+    },
+    {
+      "epoch": 21.388043221074216,
+      "grad_norm": 1.71875,
+      "learning_rate": 8.742157894736841e-05,
+      "loss": 1.1679,
+      "step": 33900
+    },
+    {
+      "epoch": 21.419591450429845,
+      "grad_norm": 1.15625,
+      "learning_rate": 8.739526315789474e-05,
+      "loss": 1.093,
+      "step": 33950
+    },
+    {
+      "epoch": 21.451139679785474,
+      "grad_norm": 1.03125,
+      "learning_rate": 8.736894736842106e-05,
+      "loss": 1.0564,
+      "step": 34000
+    },
+    {
+      "epoch": 21.451139679785474,
+      "eval_loss": 1.5272431373596191,
+      "eval_runtime": 266.9593,
+      "eval_samples_per_second": 94.985,
+      "eval_steps_per_second": 5.937,
+      "step": 34000
+    },
+    {
+      "epoch": 21.4826879091411,
+      "grad_norm": 1.1796875,
+      "learning_rate": 8.734263157894738e-05,
+      "loss": 1.1702,
+      "step": 34050
+    },
+    {
+      "epoch": 21.514236138496727,
+      "grad_norm": 1.4453125,
+      "learning_rate": 8.731631578947369e-05,
+      "loss": 1.0621,
+      "step": 34100
+    },
+    {
+      "epoch": 21.545784367852356,
+      "grad_norm": 1.5390625,
+      "learning_rate": 8.729e-05,
+      "loss": 1.2028,
+      "step": 34150
+    },
+    {
+      "epoch": 21.57733259720798,
+      "grad_norm": 1.3125,
+      "learning_rate": 8.726368421052632e-05,
+      "loss": 1.1803,
+      "step": 34200
+    },
+    {
+      "epoch": 21.60888082656361,
+      "grad_norm": 1.5625,
+      "learning_rate": 8.723736842105264e-05,
+      "loss": 1.1601,
+      "step": 34250
+    },
+    {
+      "epoch": 21.640429055919235,
+      "grad_norm": 1.140625,
+      "learning_rate": 8.721105263157895e-05,
+      "loss": 1.1953,
+      "step": 34300
+    },
+    {
+      "epoch": 21.671977285274863,
+      "grad_norm": 1.296875,
+      "learning_rate": 8.718473684210526e-05,
+      "loss": 1.1671,
+      "step": 34350
+    },
+    {
+      "epoch": 21.703525514630492,
+      "grad_norm": 1.0078125,
+      "learning_rate": 8.715842105263158e-05,
+      "loss": 1.0822,
+      "step": 34400
+    },
+    {
+      "epoch": 21.73507374398612,
+      "grad_norm": 1.4375,
+      "learning_rate": 8.713210526315791e-05,
+      "loss": 1.1072,
+      "step": 34450
+    },
+    {
+      "epoch": 21.766621973341746,
+      "grad_norm": 1.7265625,
+      "learning_rate": 8.710578947368422e-05,
+      "loss": 1.1747,
+      "step": 34500
+    },
+    {
+      "epoch": 21.798170202697374,
+      "grad_norm": 1.421875,
+      "learning_rate": 8.707947368421053e-05,
+      "loss": 1.1389,
+      "step": 34550
+    },
+    {
+      "epoch": 21.829718432053,
+      "grad_norm": 1.3046875,
+      "learning_rate": 8.705315789473684e-05,
+      "loss": 1.1176,
+      "step": 34600
+    },
+    {
+      "epoch": 21.861266661408628,
+      "grad_norm": 1.1328125,
+      "learning_rate": 8.702684210526316e-05,
+      "loss": 1.2013,
+      "step": 34650
+    },
+    {
+      "epoch": 21.892814890764257,
+      "grad_norm": 1.59375,
+      "learning_rate": 8.700052631578948e-05,
+      "loss": 1.1344,
+      "step": 34700
+    },
+    {
+      "epoch": 21.92436312011988,
+      "grad_norm": 1.5390625,
+      "learning_rate": 8.697421052631579e-05,
+      "loss": 1.1601,
+      "step": 34750
+    },
+    {
+      "epoch": 21.95591134947551,
+      "grad_norm": 1.96875,
+      "learning_rate": 8.69478947368421e-05,
+      "loss": 1.11,
+      "step": 34800
+    },
+    {
+      "epoch": 21.98745957883114,
+      "grad_norm": 1.0546875,
+      "learning_rate": 8.692157894736842e-05,
+      "loss": 1.1303,
+      "step": 34850
+    },
+    {
+      "epoch": 22.018928937613378,
+      "grad_norm": 1.171875,
+      "learning_rate": 8.689526315789474e-05,
+      "loss": 1.1803,
+      "step": 34900
+    },
+    {
+      "epoch": 22.050477166969003,
+      "grad_norm": 1.4609375,
+      "learning_rate": 8.686894736842107e-05,
+      "loss": 1.0775,
+      "step": 34950
+    },
+    {
+      "epoch": 22.08202539632463,
+      "grad_norm": 1.4375,
+      "learning_rate": 8.684263157894738e-05,
+      "loss": 1.1902,
+      "step": 35000
+    },
+    {
+      "epoch": 22.08202539632463,
+      "eval_loss": 1.524708867073059,
+      "eval_runtime": 267.2268,
+      "eval_samples_per_second": 94.889,
+      "eval_steps_per_second": 5.931,
+      "step": 35000
+    },
+    {
+      "epoch": 22.11357362568026,
+      "grad_norm": 1.0859375,
+      "learning_rate": 8.681631578947369e-05,
+      "loss": 1.1983,
+      "step": 35050
+    },
+    {
+      "epoch": 22.145121855035885,
+      "grad_norm": 1.1875,
+      "learning_rate": 8.679e-05,
+      "loss": 1.1345,
+      "step": 35100
+    },
+    {
+      "epoch": 22.176670084391514,
+      "grad_norm": 1.7265625,
+      "learning_rate": 8.676368421052633e-05,
+      "loss": 1.1564,
+      "step": 35150
+    },
+    {
+      "epoch": 22.208218313747143,
+      "grad_norm": 1.0859375,
+      "learning_rate": 8.673736842105264e-05,
+      "loss": 1.1506,
+      "step": 35200
+    },
+    {
+      "epoch": 22.239766543102768,
+      "grad_norm": 1.1796875,
+      "learning_rate": 8.671105263157895e-05,
+      "loss": 1.2614,
+      "step": 35250
+    },
+    {
+      "epoch": 22.271314772458396,
+      "grad_norm": 1.234375,
+      "learning_rate": 8.668473684210526e-05,
+      "loss": 1.1369,
+      "step": 35300
+    },
+    {
+      "epoch": 22.302863001814025,
+      "grad_norm": 1.1015625,
+      "learning_rate": 8.665842105263157e-05,
+      "loss": 1.1186,
+      "step": 35350
+    },
+    {
+      "epoch": 22.33441123116965,
+      "grad_norm": 1.0546875,
+      "learning_rate": 8.66321052631579e-05,
+      "loss": 1.1106,
+      "step": 35400
+    },
+    {
+      "epoch": 22.36595946052528,
+      "grad_norm": 1.75,
+      "learning_rate": 8.660578947368422e-05,
+      "loss": 1.1323,
+      "step": 35450
+    },
+    {
+      "epoch": 22.397507689880907,
+      "grad_norm": 1.84375,
+      "learning_rate": 8.657947368421053e-05,
+      "loss": 1.1145,
+      "step": 35500
+    },
+    {
+      "epoch": 22.429055919236532,
+      "grad_norm": 1.328125,
+      "learning_rate": 8.655315789473685e-05,
+      "loss": 1.2457,
+      "step": 35550
+    },
+    {
+      "epoch": 22.46060414859216,
+      "grad_norm": 2.0625,
+      "learning_rate": 8.652684210526316e-05,
+      "loss": 1.1645,
+      "step": 35600
+    },
+    {
+      "epoch": 22.49215237794779,
+      "grad_norm": 1.578125,
+      "learning_rate": 8.650052631578948e-05,
+      "loss": 1.0613,
+      "step": 35650
+    },
+    {
+      "epoch": 22.523700607303415,
+      "grad_norm": 1.234375,
+      "learning_rate": 8.64742105263158e-05,
+      "loss": 1.1264,
+      "step": 35700
+    },
+    {
+      "epoch": 22.555248836659043,
+      "grad_norm": 1.3828125,
+      "learning_rate": 8.64478947368421e-05,
+      "loss": 1.0699,
+      "step": 35750
+    },
+    {
+      "epoch": 22.586797066014668,
+      "grad_norm": 1.265625,
+      "learning_rate": 8.642157894736842e-05,
+      "loss": 1.1516,
+      "step": 35800
+    },
+    {
+      "epoch": 22.618345295370297,
+      "grad_norm": 0.94921875,
+      "learning_rate": 8.639526315789474e-05,
+      "loss": 1.2029,
+      "step": 35850
+    },
+    {
+      "epoch": 22.649893524725925,
+      "grad_norm": 1.546875,
+      "learning_rate": 8.636894736842105e-05,
+      "loss": 1.1092,
+      "step": 35900
+    },
+    {
+      "epoch": 22.68144175408155,
+      "grad_norm": 1.1328125,
+      "learning_rate": 8.634263157894738e-05,
+      "loss": 1.1379,
+      "step": 35950
+    },
+    {
+      "epoch": 22.71298998343718,
+      "grad_norm": 1.671875,
+      "learning_rate": 8.631631578947369e-05,
+      "loss": 1.1758,
+      "step": 36000
+    },
+    {
+      "epoch": 22.71298998343718,
+      "eval_loss": 1.5223606824874878,
+      "eval_runtime": 267.1866,
+      "eval_samples_per_second": 94.904,
+      "eval_steps_per_second": 5.932,
+      "step": 36000
+    },
+    {
+      "epoch": 22.744538212792808,
+      "grad_norm": 1.3671875,
+      "learning_rate": 8.629e-05,
+      "loss": 1.1285,
+      "step": 36050
+    },
+    {
+      "epoch": 22.776086442148433,
+      "grad_norm": 1.3359375,
+      "learning_rate": 8.626368421052631e-05,
+      "loss": 1.094,
+      "step": 36100
+    },
+    {
+      "epoch": 22.80763467150406,
+      "grad_norm": 1.15625,
+      "learning_rate": 8.623736842105264e-05,
+      "loss": 1.1294,
+      "step": 36150
+    },
+    {
+      "epoch": 22.83918290085969,
+      "grad_norm": 1.796875,
+      "learning_rate": 8.621105263157895e-05,
+      "loss": 1.0608,
+      "step": 36200
+    },
+    {
+      "epoch": 22.870731130215315,
+      "grad_norm": 1.40625,
+      "learning_rate": 8.618473684210526e-05,
+      "loss": 1.1035,
+      "step": 36250
+    },
+    {
+      "epoch": 22.902279359570944,
+      "grad_norm": 1.46875,
+      "learning_rate": 8.615842105263159e-05,
+      "loss": 1.1301,
+      "step": 36300
+    },
+    {
+      "epoch": 22.933827588926572,
+      "grad_norm": 1.0625,
+      "learning_rate": 8.61321052631579e-05,
+      "loss": 1.1169,
+      "step": 36350
+    },
+    {
+      "epoch": 22.965375818282197,
+      "grad_norm": 1.234375,
+      "learning_rate": 8.610578947368422e-05,
+      "loss": 1.1624,
+      "step": 36400
+    },
+    {
+      "epoch": 22.996924047637826,
+      "grad_norm": 0.8984375,
+      "learning_rate": 8.607947368421054e-05,
+      "loss": 1.0599,
+      "step": 36450
+    },
+    {
+      "epoch": 23.028393406420065,
+      "grad_norm": 1.078125,
+      "learning_rate": 8.605315789473685e-05,
+      "loss": 1.0562,
+      "step": 36500
+    },
+    {
+      "epoch": 23.059941635775694,
+      "grad_norm": 2.015625,
+      "learning_rate": 8.602684210526316e-05,
+      "loss": 1.1849,
+      "step": 36550
+    },
+    {
+      "epoch": 23.09148986513132,
+      "grad_norm": 1.6015625,
+      "learning_rate": 8.600052631578947e-05,
+      "loss": 1.1411,
+      "step": 36600
+    },
+    {
+      "epoch": 23.123038094486947,
+      "grad_norm": 1.203125,
+      "learning_rate": 8.59742105263158e-05,
+      "loss": 1.1676,
+      "step": 36650
+    },
+    {
+      "epoch": 23.154586323842576,
+      "grad_norm": 1.1484375,
+      "learning_rate": 8.594789473684211e-05,
+      "loss": 1.1663,
+      "step": 36700
+    },
+    {
+      "epoch": 23.1861345531982,
+      "grad_norm": 1.1953125,
+      "learning_rate": 8.592157894736842e-05,
+      "loss": 1.1226,
+      "step": 36750
+    },
+    {
+      "epoch": 23.21768278255383,
+      "grad_norm": 1.1171875,
+      "learning_rate": 8.589526315789474e-05,
+      "loss": 1.1417,
+      "step": 36800
+    },
+    {
+      "epoch": 23.24923101190946,
+      "grad_norm": 1.3359375,
+      "learning_rate": 8.586894736842106e-05,
+      "loss": 1.1659,
+      "step": 36850
+    },
+    {
+      "epoch": 23.280779241265083,
+      "grad_norm": 1.0,
+      "learning_rate": 8.584263157894738e-05,
+      "loss": 1.137,
+      "step": 36900
+    },
+    {
+      "epoch": 23.312327470620712,
+      "grad_norm": 1.5390625,
+      "learning_rate": 8.581631578947369e-05,
+      "loss": 1.2469,
+      "step": 36950
+    },
+    {
+      "epoch": 23.343875699976337,
+      "grad_norm": 1.3984375,
+      "learning_rate": 8.579e-05,
+      "loss": 1.2458,
+      "step": 37000
+    },
+    {
+      "epoch": 23.343875699976337,
+      "eval_loss": 1.5205219984054565,
+      "eval_runtime": 267.0201,
+      "eval_samples_per_second": 94.963,
+      "eval_steps_per_second": 5.936,
+      "step": 37000
+    },
+    {
+      "epoch": 23.375423929331966,
+      "grad_norm": 1.015625,
+      "learning_rate": 8.576368421052632e-05,
+      "loss": 1.1369,
+      "step": 37050
+    },
+    {
+      "epoch": 23.406972158687594,
+      "grad_norm": 1.390625,
+      "learning_rate": 8.573736842105263e-05,
+      "loss": 1.1991,
+      "step": 37100
+    },
+    {
+      "epoch": 23.43852038804322,
+      "grad_norm": 1.4765625,
+      "learning_rate": 8.571105263157895e-05,
+      "loss": 1.185,
+      "step": 37150
+    },
+    {
+      "epoch": 23.470068617398848,
+      "grad_norm": 0.87109375,
+      "learning_rate": 8.568473684210526e-05,
+      "loss": 1.1244,
+      "step": 37200
+    },
+    {
+      "epoch": 23.501616846754477,
+      "grad_norm": 1.5078125,
+      "learning_rate": 8.565842105263159e-05,
+      "loss": 1.1446,
+      "step": 37250
+    },
+    {
+      "epoch": 23.5331650761101,
+      "grad_norm": 1.046875,
+      "learning_rate": 8.56321052631579e-05,
+      "loss": 1.0467,
+      "step": 37300
+    },
+    {
+      "epoch": 23.56471330546573,
+      "grad_norm": 0.91015625,
+      "learning_rate": 8.560578947368421e-05,
+      "loss": 1.0834,
+      "step": 37350
+    },
+    {
+      "epoch": 23.59626153482136,
+      "grad_norm": 1.1875,
+      "learning_rate": 8.557947368421054e-05,
+      "loss": 1.0935,
+      "step": 37400
+    },
+    {
+      "epoch": 23.627809764176984,
+      "grad_norm": 1.046875,
+      "learning_rate": 8.555315789473685e-05,
+      "loss": 1.1686,
+      "step": 37450
+    },
+    {
+      "epoch": 23.659357993532613,
+      "grad_norm": 1.1875,
+      "learning_rate": 8.552684210526316e-05,
+      "loss": 1.0235,
+      "step": 37500
+    },
+    {
+      "epoch": 23.69090622288824,
+      "grad_norm": 1.2109375,
+      "learning_rate": 8.550052631578947e-05,
+      "loss": 1.0734,
+      "step": 37550
+    },
+    {
+      "epoch": 23.722454452243866,
+      "grad_norm": 1.1171875,
+      "learning_rate": 8.547421052631578e-05,
+      "loss": 1.1121,
+      "step": 37600
+    },
+    {
+      "epoch": 23.754002681599495,
+      "grad_norm": 1.1484375,
+      "learning_rate": 8.544789473684211e-05,
+      "loss": 1.1587,
+      "step": 37650
+    },
+    {
+      "epoch": 23.785550910955124,
+      "grad_norm": 1.765625,
+      "learning_rate": 8.542157894736843e-05,
+      "loss": 1.0876,
+      "step": 37700
+    },
+    {
+      "epoch": 23.81709914031075,
+      "grad_norm": 1.2734375,
+      "learning_rate": 8.539526315789475e-05,
+      "loss": 1.1147,
+      "step": 37750
+    },
+    {
+      "epoch": 23.848647369666377,
+      "grad_norm": 1.21875,
+      "learning_rate": 8.536894736842106e-05,
+      "loss": 1.1459,
+      "step": 37800
+    },
+    {
+      "epoch": 23.880195599022006,
+      "grad_norm": 1.40625,
+      "learning_rate": 8.534263157894737e-05,
+      "loss": 1.0868,
+      "step": 37850
+    },
+    {
+      "epoch": 23.91174382837763,
+      "grad_norm": 1.3203125,
+      "learning_rate": 8.53163157894737e-05,
+      "loss": 1.1512,
+      "step": 37900
+    },
+    {
+      "epoch": 23.94329205773326,
+      "grad_norm": 1.2890625,
+      "learning_rate": 8.529e-05,
+      "loss": 1.0586,
+      "step": 37950
+    },
+    {
+      "epoch": 23.974840287088888,
+      "grad_norm": 0.953125,
+      "learning_rate": 8.526368421052632e-05,
+      "loss": 1.1604,
+      "step": 38000
+    },
+    {
+      "epoch": 23.974840287088888,
+      "eval_loss": 1.5178626775741577,
+      "eval_runtime": 267.4107,
+      "eval_samples_per_second": 94.824,
+      "eval_steps_per_second": 5.927,
+      "step": 38000
+    },
+    {
+      "epoch": 24.006309645871127,
+      "grad_norm": 1.3125,
+      "learning_rate": 8.523736842105263e-05,
+      "loss": 1.1604,
+      "step": 38050
+    },
+    {
+      "epoch": 24.037857875226752,
+      "grad_norm": 1.4609375,
+      "learning_rate": 8.521105263157895e-05,
+      "loss": 1.1294,
+      "step": 38100
+    },
+    {
+      "epoch": 24.06940610458238,
+      "grad_norm": 1.1484375,
+      "learning_rate": 8.518473684210528e-05,
+      "loss": 1.2385,
+      "step": 38150
+    },
+    {
+      "epoch": 24.10095433393801,
+      "grad_norm": 1.3203125,
+      "learning_rate": 8.515842105263159e-05,
+      "loss": 1.2034,
+      "step": 38200
+    },
+    {
+      "epoch": 24.132502563293635,
+      "grad_norm": 1.5234375,
+      "learning_rate": 8.51321052631579e-05,
+      "loss": 1.1935,
+      "step": 38250
+    },
+    {
+      "epoch": 24.164050792649263,
+      "grad_norm": 1.234375,
+      "learning_rate": 8.510578947368421e-05,
+      "loss": 1.1127,
+      "step": 38300
+    },
+    {
+      "epoch": 24.19559902200489,
+      "grad_norm": 1.2890625,
+      "learning_rate": 8.507947368421052e-05,
+      "loss": 1.0757,
+      "step": 38350
+    },
+    {
+      "epoch": 24.227147251360517,
+      "grad_norm": 1.3359375,
+      "learning_rate": 8.505315789473685e-05,
+      "loss": 1.1407,
+      "step": 38400
+    },
+    {
+      "epoch": 24.258695480716145,
+      "grad_norm": 2.359375,
+      "learning_rate": 8.502684210526316e-05,
+      "loss": 1.1599,
+      "step": 38450
+    },
+    {
+      "epoch": 24.29024371007177,
+      "grad_norm": 1.265625,
+      "learning_rate": 8.500052631578947e-05,
+      "loss": 1.0929,
+      "step": 38500
+    },
+    {
+      "epoch": 24.3217919394274,
+      "grad_norm": 1.25,
+      "learning_rate": 8.497421052631578e-05,
+      "loss": 1.0718,
+      "step": 38550
+    },
+    {
+      "epoch": 24.353340168783028,
+      "grad_norm": 1.078125,
+      "learning_rate": 8.494789473684211e-05,
+      "loss": 1.0182,
+      "step": 38600
+    },
+    {
+      "epoch": 24.384888398138653,
+      "grad_norm": 1.234375,
+      "learning_rate": 8.492157894736843e-05,
+      "loss": 1.1052,
+      "step": 38650
+    },
+    {
+      "epoch": 24.41643662749428,
+      "grad_norm": 3.0,
+      "learning_rate": 8.489526315789475e-05,
+      "loss": 1.059,
+      "step": 38700
+    },
+    {
+      "epoch": 24.44798485684991,
+      "grad_norm": 1.2734375,
+      "learning_rate": 8.486894736842106e-05,
+      "loss": 1.1979,
+      "step": 38750
+    },
+    {
+      "epoch": 24.479533086205535,
+      "grad_norm": 0.953125,
+      "learning_rate": 8.484263157894737e-05,
+      "loss": 1.0467,
+      "step": 38800
+    },
+    {
+      "epoch": 24.511081315561164,
+      "grad_norm": 1.6328125,
+      "learning_rate": 8.48163157894737e-05,
+      "loss": 1.2318,
+      "step": 38850
+    },
+    {
+      "epoch": 24.542629544916792,
+      "grad_norm": 1.0859375,
+      "learning_rate": 8.479e-05,
+      "loss": 1.1296,
+      "step": 38900
+    },
+    {
+      "epoch": 24.574177774272417,
+      "grad_norm": 0.828125,
+      "learning_rate": 8.476368421052632e-05,
+      "loss": 1.1352,
+      "step": 38950
+    },
+    {
+      "epoch": 24.605726003628046,
+      "grad_norm": 1.203125,
+      "learning_rate": 8.473736842105263e-05,
+      "loss": 1.1311,
+      "step": 39000
+    },
+    {
+      "epoch": 24.605726003628046,
+      "eval_loss": 1.5162802934646606,
+      "eval_runtime": 266.8244,
+      "eval_samples_per_second": 95.033,
+      "eval_steps_per_second": 5.94,
+      "step": 39000
+    },
+    {
+      "epoch": 24.637274232983675,
+      "grad_norm": 1.203125,
+      "learning_rate": 8.471105263157894e-05,
+      "loss": 1.1194,
+      "step": 39050
+    },
+    {
+      "epoch": 24.6688224623393,
+      "grad_norm": 1.6015625,
+      "learning_rate": 8.468473684210527e-05,
+      "loss": 1.1584,
+      "step": 39100
+    },
+    {
+      "epoch": 24.70037069169493,
+      "grad_norm": 1.546875,
+      "learning_rate": 8.465842105263159e-05,
+      "loss": 1.1765,
+      "step": 39150
+    },
+    {
+      "epoch": 24.731918921050557,
+      "grad_norm": 1.3125,
+      "learning_rate": 8.46321052631579e-05,
+      "loss": 1.1289,
+      "step": 39200
+    },
+    {
+      "epoch": 24.763467150406182,
+      "grad_norm": 1.703125,
+      "learning_rate": 8.460578947368421e-05,
+      "loss": 1.129,
+      "step": 39250
+    },
+    {
+      "epoch": 24.79501537976181,
+      "grad_norm": 1.4609375,
+      "learning_rate": 8.457947368421053e-05,
+      "loss": 1.1591,
+      "step": 39300
+    },
+    {
+      "epoch": 24.82656360911744,
+      "grad_norm": 1.3671875,
+      "learning_rate": 8.455315789473685e-05,
+      "loss": 1.2381,
+      "step": 39350
+    },
+    {
+      "epoch": 24.858111838473064,
+      "grad_norm": 0.94140625,
+      "learning_rate": 8.452684210526316e-05,
+      "loss": 1.1744,
+      "step": 39400
+    },
+    {
+      "epoch": 24.889660067828693,
+      "grad_norm": 2.046875,
+      "learning_rate": 8.450052631578947e-05,
+      "loss": 1.1799,
+      "step": 39450
+    },
+    {
+      "epoch": 24.92120829718432,
+      "grad_norm": 1.4140625,
+      "learning_rate": 8.447421052631579e-05,
+      "loss": 1.0675,
+      "step": 39500
+    },
+    {
+      "epoch": 24.952756526539947,
+      "grad_norm": 1.484375,
+      "learning_rate": 8.444789473684211e-05,
+      "loss": 1.1453,
+      "step": 39550
+    },
+    {
+      "epoch": 24.984304755895575,
+      "grad_norm": 1.765625,
+      "learning_rate": 8.442157894736842e-05,
+      "loss": 1.086,
+      "step": 39600
+    },
+    {
+      "epoch": 25.015774114677814,
+      "grad_norm": 2.421875,
+      "learning_rate": 8.439526315789475e-05,
+      "loss": 1.1468,
+      "step": 39650
+    },
+    {
+      "epoch": 25.04732234403344,
+      "grad_norm": 1.1640625,
+      "learning_rate": 8.436894736842106e-05,
+      "loss": 1.0776,
+      "step": 39700
+    },
+    {
+      "epoch": 25.078870573389068,
+      "grad_norm": 1.4140625,
+      "learning_rate": 8.434263157894737e-05,
+      "loss": 0.9669,
+      "step": 39750
+    },
+    {
+      "epoch": 25.110418802744697,
+      "grad_norm": 1.21875,
+      "learning_rate": 8.431631578947368e-05,
+      "loss": 1.1222,
+      "step": 39800
+    },
+    {
+      "epoch": 25.14196703210032,
+      "grad_norm": 1.65625,
+      "learning_rate": 8.429000000000001e-05,
+      "loss": 1.1037,
+      "step": 39850
+    },
+    {
+      "epoch": 25.17351526145595,
+      "grad_norm": 1.5625,
+      "learning_rate": 8.426368421052632e-05,
+      "loss": 1.1482,
+      "step": 39900
+    },
+    {
+      "epoch": 25.20506349081158,
+      "grad_norm": 1.328125,
+      "learning_rate": 8.423736842105263e-05,
+      "loss": 1.0434,
+      "step": 39950
+    },
+    {
+      "epoch": 25.236611720167204,
+      "grad_norm": 1.390625,
+      "learning_rate": 8.421105263157894e-05,
+      "loss": 1.112,
+      "step": 40000
+    },
+    {
+      "epoch": 25.236611720167204,
+      "eval_loss": 1.5145944356918335,
+      "eval_runtime": 267.137,
+      "eval_samples_per_second": 94.921,
+      "eval_steps_per_second": 5.933,
+      "step": 40000
+    },
+    {
+      "epoch": 25.268159949522833,
+      "grad_norm": 1.109375,
+      "learning_rate": 8.418473684210527e-05,
+      "loss": 1.0981,
+      "step": 40050
+    },
+    {
+      "epoch": 25.29970817887846,
+      "grad_norm": 1.7890625,
+      "learning_rate": 8.415842105263159e-05,
+      "loss": 1.156,
+      "step": 40100
+    },
+    {
+      "epoch": 25.331256408234086,
+      "grad_norm": 0.8828125,
+      "learning_rate": 8.41321052631579e-05,
+      "loss": 1.2156,
+      "step": 40150
+    },
+    {
+      "epoch": 25.362804637589715,
+      "grad_norm": 1.328125,
+      "learning_rate": 8.410578947368422e-05,
+      "loss": 1.1541,
+      "step": 40200
+    },
+    {
+      "epoch": 25.394352866945344,
+      "grad_norm": 1.1484375,
+      "learning_rate": 8.407947368421053e-05,
+      "loss": 1.0822,
+      "step": 40250
+    },
+    {
+      "epoch": 25.42590109630097,
+      "grad_norm": 1.171875,
+      "learning_rate": 8.405315789473684e-05,
+      "loss": 1.1099,
+      "step": 40300
+    },
+    {
+      "epoch": 25.457449325656597,
+      "grad_norm": 1.328125,
+      "learning_rate": 8.402684210526316e-05,
+      "loss": 1.1289,
+      "step": 40350
+    },
+    {
+      "epoch": 25.488997555012226,
+      "grad_norm": 1.703125,
+      "learning_rate": 8.400052631578948e-05,
+      "loss": 1.1307,
+      "step": 40400
+    },
+    {
+      "epoch": 25.52054578436785,
+      "grad_norm": 1.78125,
+      "learning_rate": 8.397421052631579e-05,
+      "loss": 1.1502,
+      "step": 40450
+    },
+    {
+      "epoch": 25.55209401372348,
+      "grad_norm": 1.3984375,
+      "learning_rate": 8.394789473684211e-05,
+      "loss": 1.1708,
+      "step": 40500
+    },
+    {
+      "epoch": 25.583642243079108,
+      "grad_norm": 1.140625,
+      "learning_rate": 8.392157894736842e-05,
+      "loss": 1.1826,
+      "step": 40550
+    },
+    {
+      "epoch": 25.615190472434733,
+      "grad_norm": 1.5546875,
+      "learning_rate": 8.389526315789475e-05,
+      "loss": 1.1273,
+      "step": 40600
+    },
+    {
+      "epoch": 25.646738701790362,
+      "grad_norm": 1.296875,
+      "learning_rate": 8.386894736842106e-05,
+      "loss": 1.1207,
+      "step": 40650
+    },
+    {
+      "epoch": 25.67828693114599,
+      "grad_norm": 0.9453125,
+      "learning_rate": 8.384263157894737e-05,
+      "loss": 1.0559,
+      "step": 40700
+    },
+    {
+      "epoch": 25.709835160501616,
+      "grad_norm": 1.359375,
+      "learning_rate": 8.381631578947368e-05,
+      "loss": 1.097,
+      "step": 40750
+    },
+    {
+      "epoch": 25.741383389857244,
+      "grad_norm": 1.2265625,
+      "learning_rate": 8.379e-05,
+      "loss": 1.1353,
+      "step": 40800
+    },
+    {
+      "epoch": 25.772931619212873,
+      "grad_norm": 1.3359375,
+      "learning_rate": 8.376368421052632e-05,
+      "loss": 1.2467,
+      "step": 40850
+    },
+    {
+      "epoch": 25.804479848568498,
+      "grad_norm": 2.296875,
+      "learning_rate": 8.373736842105263e-05,
+      "loss": 1.2264,
+      "step": 40900
+    },
+    {
+      "epoch": 25.836028077924126,
+      "grad_norm": 1.5859375,
+      "learning_rate": 8.371105263157896e-05,
+      "loss": 1.1003,
+      "step": 40950
+    },
+    {
+      "epoch": 25.867576307279755,
+      "grad_norm": 1.1015625,
+      "learning_rate": 8.368473684210527e-05,
+      "loss": 1.0826,
+      "step": 41000
+    },
+    {
+      "epoch": 25.867576307279755,
+      "eval_loss": 1.512583613395691,
+      "eval_runtime": 267.7473,
+      "eval_samples_per_second": 94.705,
+      "eval_steps_per_second": 5.92,
+      "step": 41000
+    },
+    {
+      "epoch": 25.89912453663538,
+      "grad_norm": 1.3828125,
+      "learning_rate": 8.365842105263158e-05,
+      "loss": 1.2477,
+      "step": 41050
+    },
+    {
+      "epoch": 25.93067276599101,
+      "grad_norm": 0.93359375,
+      "learning_rate": 8.36321052631579e-05,
+      "loss": 1.16,
+      "step": 41100
+    },
+    {
+      "epoch": 25.962220995346637,
+      "grad_norm": 6.53125,
+      "learning_rate": 8.360578947368422e-05,
+      "loss": 1.1243,
+      "step": 41150
+    },
+    {
+      "epoch": 25.993769224702262,
+      "grad_norm": 1.203125,
+      "learning_rate": 8.357947368421053e-05,
+      "loss": 1.0781,
+      "step": 41200
+    },
+    {
+      "epoch": 26.0252385834845,
+      "grad_norm": 1.1953125,
+      "learning_rate": 8.355315789473684e-05,
+      "loss": 1.1761,
+      "step": 41250
+    },
+    {
+      "epoch": 26.05678681284013,
+      "grad_norm": 1.015625,
+      "learning_rate": 8.352684210526315e-05,
+      "loss": 1.1105,
+      "step": 41300
+    },
+    {
+      "epoch": 26.088335042195755,
+      "grad_norm": 2.296875,
+      "learning_rate": 8.350052631578948e-05,
+      "loss": 1.1172,
+      "step": 41350
+    },
+    {
+      "epoch": 26.119883271551384,
+      "grad_norm": 1.1015625,
+      "learning_rate": 8.347421052631579e-05,
+      "loss": 1.0579,
+      "step": 41400
+    },
+    {
+      "epoch": 26.151431500907012,
+      "grad_norm": 1.7578125,
+      "learning_rate": 8.344789473684211e-05,
+      "loss": 1.1108,
+      "step": 41450
+    },
+    {
+      "epoch": 26.182979730262637,
+      "grad_norm": 1.1640625,
+      "learning_rate": 8.342157894736843e-05,
+      "loss": 1.1698,
+      "step": 41500
+    },
+    {
+      "epoch": 26.214527959618266,
+      "grad_norm": 1.015625,
+      "learning_rate": 8.339526315789474e-05,
+      "loss": 1.1217,
+      "step": 41550
+    },
+    {
+      "epoch": 26.246076188973895,
+      "grad_norm": 2.109375,
+      "learning_rate": 8.336894736842106e-05,
+      "loss": 1.1791,
+      "step": 41600
+    },
+    {
+      "epoch": 26.27762441832952,
+      "grad_norm": 1.046875,
+      "learning_rate": 8.334263157894737e-05,
+      "loss": 1.1931,
+      "step": 41650
+    },
+    {
+      "epoch": 26.30917264768515,
+      "grad_norm": 1.34375,
+      "learning_rate": 8.331631578947369e-05,
+      "loss": 1.1063,
+      "step": 41700
+    },
+    {
+      "epoch": 26.340720877040777,
+      "grad_norm": 1.3046875,
+      "learning_rate": 8.329e-05,
+      "loss": 1.1386,
+      "step": 41750
+    },
+    {
+      "epoch": 26.372269106396402,
+      "grad_norm": 1.40625,
+      "learning_rate": 8.326368421052632e-05,
+      "loss": 1.1654,
+      "step": 41800
+    },
+    {
+      "epoch": 26.40381733575203,
+      "grad_norm": 1.3828125,
+      "learning_rate": 8.323736842105263e-05,
+      "loss": 1.1421,
+      "step": 41850
+    },
+    {
+      "epoch": 26.43536556510766,
+      "grad_norm": 1.1484375,
+      "learning_rate": 8.321105263157896e-05,
+      "loss": 1.1093,
+      "step": 41900
+    },
+    {
+      "epoch": 26.466913794463284,
+      "grad_norm": 1.0390625,
+      "learning_rate": 8.318473684210527e-05,
+      "loss": 1.1201,
+      "step": 41950
+    },
+    {
+      "epoch": 26.498462023818913,
+      "grad_norm": 1.5,
+      "learning_rate": 8.315842105263158e-05,
+      "loss": 1.24,
+      "step": 42000
+    },
+    {
+      "epoch": 26.498462023818913,
+      "eval_loss": 1.5111093521118164,
+      "eval_runtime": 267.2484,
+      "eval_samples_per_second": 94.882,
+      "eval_steps_per_second": 5.931,
+      "step": 42000
+    },
+    {
+      "epoch": 26.53001025317454,
+      "grad_norm": 1.3125,
+      "learning_rate": 8.31321052631579e-05,
+      "loss": 1.1231,
+      "step": 42050
+    },
+    {
+      "epoch": 26.561558482530167,
+      "grad_norm": 1.2890625,
+      "learning_rate": 8.310578947368422e-05,
+      "loss": 1.1488,
+      "step": 42100
+    },
+    {
+      "epoch": 26.593106711885795,
+      "grad_norm": 1.71875,
+      "learning_rate": 8.307947368421053e-05,
+      "loss": 1.1427,
+      "step": 42150
+    },
+    {
+      "epoch": 26.624654941241424,
+      "grad_norm": 1.46875,
+      "learning_rate": 8.305315789473684e-05,
+      "loss": 1.1537,
+      "step": 42200
+    },
+    {
+      "epoch": 26.65620317059705,
+      "grad_norm": 0.953125,
+      "learning_rate": 8.302684210526315e-05,
+      "loss": 1.116,
+      "step": 42250
+    },
+    {
+      "epoch": 26.687751399952678,
+      "grad_norm": 0.99609375,
+      "learning_rate": 8.300052631578948e-05,
+      "loss": 1.0836,
+      "step": 42300
+    },
+    {
+      "epoch": 26.719299629308306,
+      "grad_norm": 1.2109375,
+      "learning_rate": 8.29742105263158e-05,
+      "loss": 1.1648,
+      "step": 42350
+    },
+    {
+      "epoch": 26.75084785866393,
+      "grad_norm": 1.40625,
+      "learning_rate": 8.294789473684212e-05,
+      "loss": 1.0502,
+      "step": 42400
+    },
+    {
+      "epoch": 26.78239608801956,
+      "grad_norm": 2.34375,
+      "learning_rate": 8.292157894736843e-05,
+      "loss": 1.2082,
+      "step": 42450
+    },
+    {
+      "epoch": 26.81394431737519,
+      "grad_norm": 1.3671875,
+      "learning_rate": 8.289526315789474e-05,
+      "loss": 1.1388,
+      "step": 42500
+    },
+    {
+      "epoch": 26.845492546730814,
+      "grad_norm": 1.3125,
+      "learning_rate": 8.286894736842106e-05,
+      "loss": 1.1777,
+      "step": 42550
+    },
+    {
+      "epoch": 26.877040776086442,
+      "grad_norm": 1.3046875,
+      "learning_rate": 8.284263157894738e-05,
+      "loss": 1.1671,
+      "step": 42600
+    },
+    {
+      "epoch": 26.90858900544207,
+      "grad_norm": 1.109375,
+      "learning_rate": 8.281631578947369e-05,
+      "loss": 1.1348,
+      "step": 42650
+    },
+    {
+      "epoch": 26.940137234797696,
+      "grad_norm": 1.2265625,
+      "learning_rate": 8.279e-05,
+      "loss": 1.0797,
+      "step": 42700
+    },
+    {
+      "epoch": 26.971685464153325,
+      "grad_norm": 1.5,
+      "learning_rate": 8.276368421052631e-05,
+      "loss": 1.0572,
+      "step": 42750
+    },
+    {
+      "epoch": 27.003154822935564,
+      "grad_norm": 1.3359375,
+      "learning_rate": 8.273736842105264e-05,
+      "loss": 1.1384,
+      "step": 42800
+    },
+    {
+      "epoch": 27.03470305229119,
+      "grad_norm": 1.265625,
+      "learning_rate": 8.271105263157896e-05,
+      "loss": 1.0789,
+      "step": 42850
+    },
+    {
+      "epoch": 27.066251281646817,
+      "grad_norm": 1.2421875,
+      "learning_rate": 8.268473684210527e-05,
+      "loss": 1.0927,
+      "step": 42900
+    },
+    {
+      "epoch": 27.097799511002446,
+      "grad_norm": 1.2109375,
+      "learning_rate": 8.265842105263158e-05,
+      "loss": 1.0928,
+      "step": 42950
+    },
+    {
+      "epoch": 27.12934774035807,
+      "grad_norm": 1.0859375,
+      "learning_rate": 8.26321052631579e-05,
+      "loss": 1.1963,
+      "step": 43000
+    },
+    {
+      "epoch": 27.12934774035807,
+      "eval_loss": 1.5096408128738403,
+      "eval_runtime": 268.4062,
+      "eval_samples_per_second": 94.472,
+      "eval_steps_per_second": 5.905,
+      "step": 43000
+    },
+    {
+      "epoch": 27.1608959697137,
+      "grad_norm": 1.2734375,
+      "learning_rate": 8.260578947368422e-05,
+      "loss": 1.0704,
+      "step": 43050
+    },
+    {
+      "epoch": 27.192444199069328,
+      "grad_norm": 1.2734375,
+      "learning_rate": 8.257947368421053e-05,
+      "loss": 1.208,
+      "step": 43100
+    },
+    {
+      "epoch": 27.223992428424953,
+      "grad_norm": 1.4921875,
+      "learning_rate": 8.255315789473684e-05,
+      "loss": 1.0895,
+      "step": 43150
+    },
+    {
+      "epoch": 27.255540657780582,
+      "grad_norm": 1.0859375,
+      "learning_rate": 8.252684210526315e-05,
+      "loss": 1.1455,
+      "step": 43200
+    },
+    {
+      "epoch": 27.28708888713621,
+      "grad_norm": 1.53125,
+      "learning_rate": 8.250052631578947e-05,
+      "loss": 1.2282,
+      "step": 43250
+    },
+    {
+      "epoch": 27.318637116491836,
+      "grad_norm": 1.6953125,
+      "learning_rate": 8.247421052631579e-05,
+      "loss": 1.0985,
+      "step": 43300
+    },
+    {
+      "epoch": 27.350185345847464,
+      "grad_norm": 1.0859375,
+      "learning_rate": 8.244789473684212e-05,
+      "loss": 1.1631,
+      "step": 43350
+    },
+    {
+      "epoch": 27.381733575203093,
+      "grad_norm": 1.4453125,
+      "learning_rate": 8.242157894736843e-05,
+      "loss": 1.105,
+      "step": 43400
+    },
+    {
+      "epoch": 27.413281804558718,
+      "grad_norm": 0.78515625,
+      "learning_rate": 8.239526315789474e-05,
+      "loss": 1.1341,
+      "step": 43450
+    },
+    {
+      "epoch": 27.444830033914347,
+      "grad_norm": 0.9921875,
+      "learning_rate": 8.236894736842105e-05,
+      "loss": 1.0494,
+      "step": 43500
+    },
+    {
+      "epoch": 27.476378263269975,
+      "grad_norm": 0.7890625,
+      "learning_rate": 8.234263157894738e-05,
+      "loss": 1.1368,
+      "step": 43550
+    },
+    {
+      "epoch": 27.5079264926256,
+      "grad_norm": 0.91796875,
+      "learning_rate": 8.231631578947369e-05,
+      "loss": 1.1349,
+      "step": 43600
+    },
+    {
+      "epoch": 27.53947472198123,
+      "grad_norm": 1.109375,
+      "learning_rate": 8.229e-05,
+      "loss": 1.1768,
+      "step": 43650
+    },
+    {
+      "epoch": 27.571022951336857,
+      "grad_norm": 1.1015625,
+      "learning_rate": 8.226368421052631e-05,
+      "loss": 1.0864,
+      "step": 43700
+    },
+    {
+      "epoch": 27.602571180692482,
+      "grad_norm": 1.0,
+      "learning_rate": 8.223736842105264e-05,
+      "loss": 1.0461,
+      "step": 43750
+    },
+    {
+      "epoch": 27.63411941004811,
+      "grad_norm": 2.640625,
+      "learning_rate": 8.221105263157896e-05,
+      "loss": 1.203,
+      "step": 43800
+    },
+    {
+      "epoch": 27.66566763940374,
+      "grad_norm": 1.265625,
+      "learning_rate": 8.218473684210527e-05,
+      "loss": 1.0647,
+      "step": 43850
+    },
+    {
+      "epoch": 27.697215868759365,
+      "grad_norm": 1.5390625,
+      "learning_rate": 8.215842105263158e-05,
+      "loss": 1.0393,
+      "step": 43900
+    },
+    {
+      "epoch": 27.728764098114993,
+      "grad_norm": 2.453125,
+      "learning_rate": 8.21321052631579e-05,
+      "loss": 1.0748,
+      "step": 43950
+    },
+    {
+      "epoch": 27.760312327470622,
+      "grad_norm": 2.71875,
+      "learning_rate": 8.210578947368421e-05,
+      "loss": 1.1257,
+      "step": 44000
+    },
+    {
+      "epoch": 27.760312327470622,
+      "eval_loss": 1.508254885673523,
+      "eval_runtime": 267.4817,
+      "eval_samples_per_second": 94.799,
+      "eval_steps_per_second": 5.926,
+      "step": 44000
+    },
+    {
+      "epoch": 27.791860556826247,
+      "grad_norm": 1.3671875,
+      "learning_rate": 8.207947368421053e-05,
+      "loss": 1.1718,
+      "step": 44050
+    },
+    {
+      "epoch": 27.823408786181876,
+      "grad_norm": 0.9921875,
+      "learning_rate": 8.205315789473684e-05,
+      "loss": 1.1457,
+      "step": 44100
+    },
+    {
+      "epoch": 27.854957015537504,
+      "grad_norm": 1.421875,
+      "learning_rate": 8.202684210526316e-05,
+      "loss": 1.0853,
+      "step": 44150
+    },
+    {
+      "epoch": 27.88650524489313,
+      "grad_norm": 1.140625,
+      "learning_rate": 8.200052631578947e-05,
+      "loss": 1.1679,
+      "step": 44200
+    },
+    {
+      "epoch": 27.918053474248758,
+      "grad_norm": 1.2734375,
+      "learning_rate": 8.197421052631579e-05,
+      "loss": 1.1635,
+      "step": 44250
+    },
+    {
+      "epoch": 27.949601703604387,
+      "grad_norm": 0.84765625,
+      "learning_rate": 8.194789473684212e-05,
+      "loss": 1.1341,
+      "step": 44300
+    },
+    {
+      "epoch": 27.98114993296001,
+      "grad_norm": 1.0703125,
+      "learning_rate": 8.192157894736843e-05,
+      "loss": 1.0777,
+      "step": 44350
+    },
+    {
+      "epoch": 28.01261929174225,
+      "grad_norm": 1.2421875,
+      "learning_rate": 8.189526315789474e-05,
+      "loss": 1.1823,
+      "step": 44400
+    },
+    {
+      "epoch": 28.04416752109788,
+      "grad_norm": 1.1171875,
+      "learning_rate": 8.186894736842105e-05,
+      "loss": 1.0569,
+      "step": 44450
+    },
+    {
+      "epoch": 28.075715750453504,
+      "grad_norm": 1.40625,
+      "learning_rate": 8.184263157894736e-05,
+      "loss": 1.0733,
+      "step": 44500
+    },
+    {
+      "epoch": 28.107263979809133,
+      "grad_norm": 1.03125,
+      "learning_rate": 8.181631578947369e-05,
+      "loss": 1.1116,
+      "step": 44550
+    },
+    {
+      "epoch": 28.13881220916476,
+      "grad_norm": 2.1875,
+      "learning_rate": 8.179e-05,
+      "loss": 1.1406,
+      "step": 44600
+    },
+    {
+      "epoch": 28.170360438520387,
+      "grad_norm": 1.28125,
+      "learning_rate": 8.176368421052631e-05,
+      "loss": 1.07,
+      "step": 44650
+    },
+    {
+      "epoch": 28.201908667876015,
+      "grad_norm": 1.375,
+      "learning_rate": 8.173736842105264e-05,
+      "loss": 1.0451,
+      "step": 44700
+    },
+    {
+      "epoch": 28.233456897231644,
+      "grad_norm": 1.4140625,
+      "learning_rate": 8.171105263157895e-05,
+      "loss": 1.0857,
+      "step": 44750
+    },
+    {
+      "epoch": 28.26500512658727,
+      "grad_norm": 1.2734375,
+      "learning_rate": 8.168473684210527e-05,
+      "loss": 1.1488,
+      "step": 44800
+    },
+    {
+      "epoch": 28.296553355942898,
+      "grad_norm": 1.09375,
+      "learning_rate": 8.165842105263159e-05,
+      "loss": 1.0355,
+      "step": 44850
+    },
+    {
+      "epoch": 28.328101585298526,
+      "grad_norm": 1.1484375,
+      "learning_rate": 8.16321052631579e-05,
+      "loss": 1.1336,
+      "step": 44900
+    },
+    {
+      "epoch": 28.35964981465415,
+      "grad_norm": 1.09375,
+      "learning_rate": 8.160578947368421e-05,
+      "loss": 1.1152,
+      "step": 44950
+    },
+    {
+      "epoch": 28.39119804400978,
+      "grad_norm": 1.2265625,
+      "learning_rate": 8.157947368421053e-05,
+      "loss": 1.1292,
+      "step": 45000
+    },
+    {
+      "epoch": 28.39119804400978,
+      "eval_loss": 1.5078208446502686,
+      "eval_runtime": 267.4752,
+      "eval_samples_per_second": 94.801,
+      "eval_steps_per_second": 5.926,
+      "step": 45000
+    },
+    {
+      "epoch": 28.42274627336541,
+      "grad_norm": 1.2109375,
+      "learning_rate": 8.155315789473685e-05,
+      "loss": 1.2702,
+      "step": 45050
+    },
+    {
+      "epoch": 28.454294502721034,
+      "grad_norm": 1.2265625,
+      "learning_rate": 8.152684210526316e-05,
+      "loss": 1.2059,
+      "step": 45100
+    },
+    {
+      "epoch": 28.485842732076662,
+      "grad_norm": 1.3125,
+      "learning_rate": 8.150052631578948e-05,
+      "loss": 1.2433,
+      "step": 45150
+    },
+    {
+      "epoch": 28.51739096143229,
+      "grad_norm": 1.21875,
+      "learning_rate": 8.14742105263158e-05,
+      "loss": 1.1294,
+      "step": 45200
+    },
+    {
+      "epoch": 28.548939190787916,
+      "grad_norm": 1.296875,
+      "learning_rate": 8.14478947368421e-05,
+      "loss": 1.1268,
+      "step": 45250
+    },
+    {
+      "epoch": 28.580487420143545,
+      "grad_norm": 1.1640625,
+      "learning_rate": 8.142157894736843e-05,
+      "loss": 1.134,
+      "step": 45300
+    },
+    {
+      "epoch": 28.612035649499173,
+      "grad_norm": 1.28125,
+      "learning_rate": 8.139526315789474e-05,
+      "loss": 1.1095,
+      "step": 45350
+    },
+    {
+      "epoch": 28.6435838788548,
+      "grad_norm": 0.9375,
+      "learning_rate": 8.136894736842105e-05,
+      "loss": 1.1825,
+      "step": 45400
+    },
+    {
+      "epoch": 28.675132108210427,
+      "grad_norm": 1.15625,
+      "learning_rate": 8.134263157894737e-05,
+      "loss": 1.1514,
+      "step": 45450
+    },
+    {
+      "epoch": 28.706680337566056,
+      "grad_norm": 1.46875,
+      "learning_rate": 8.131631578947369e-05,
+      "loss": 1.0768,
+      "step": 45500
+    },
+    {
+      "epoch": 28.73822856692168,
+      "grad_norm": 0.91796875,
+      "learning_rate": 8.129e-05,
+      "loss": 1.1478,
+      "step": 45550
+    },
+    {
+      "epoch": 28.76977679627731,
+      "grad_norm": 1.484375,
+      "learning_rate": 8.126368421052633e-05,
+      "loss": 1.1336,
+      "step": 45600
+    },
+    {
+      "epoch": 28.801325025632938,
+      "grad_norm": 1.015625,
+      "learning_rate": 8.123736842105264e-05,
+      "loss": 1.0757,
+      "step": 45650
+    },
+    {
+      "epoch": 28.832873254988563,
+      "grad_norm": 1.6015625,
+      "learning_rate": 8.121105263157895e-05,
+      "loss": 1.1605,
+      "step": 45700
+    },
+    {
+      "epoch": 28.86442148434419,
+      "grad_norm": 1.4921875,
+      "learning_rate": 8.118473684210526e-05,
+      "loss": 1.1447,
+      "step": 45750
+    },
+    {
+      "epoch": 28.89596971369982,
+      "grad_norm": 1.28125,
+      "learning_rate": 8.115842105263159e-05,
+      "loss": 1.0994,
+      "step": 45800
+    },
+    {
+      "epoch": 28.927517943055445,
+      "grad_norm": 1.53125,
+      "learning_rate": 8.11321052631579e-05,
+      "loss": 1.146,
+      "step": 45850
+    },
+    {
+      "epoch": 28.959066172411074,
+      "grad_norm": 1.1171875,
+      "learning_rate": 8.110578947368421e-05,
+      "loss": 1.1763,
+      "step": 45900
+    },
+    {
+      "epoch": 28.990614401766702,
+      "grad_norm": 1.25,
+      "learning_rate": 8.107947368421052e-05,
+      "loss": 1.1102,
+      "step": 45950
+    },
+    {
+      "epoch": 29.022083760548938,
+      "grad_norm": 1.5,
+      "learning_rate": 8.105315789473685e-05,
+      "loss": 1.0882,
+      "step": 46000
+    },
+    {
+      "epoch": 29.022083760548938,
+      "eval_loss": 1.5057809352874756,
+      "eval_runtime": 267.5234,
+      "eval_samples_per_second": 94.784,
+      "eval_steps_per_second": 5.925,
+      "step": 46000
+    },
+    {
+      "epoch": 29.053631989904567,
+      "grad_norm": 1.1015625,
+      "learning_rate": 8.102684210526316e-05,
+      "loss": 1.0811,
+      "step": 46050
+    },
+    {
+      "epoch": 29.085180219260195,
+      "grad_norm": 1.2734375,
+      "learning_rate": 8.100052631578948e-05,
+      "loss": 1.1022,
+      "step": 46100
+    },
+    {
+      "epoch": 29.11672844861582,
+      "grad_norm": 1.34375,
+      "learning_rate": 8.09742105263158e-05,
+      "loss": 1.1502,
+      "step": 46150
+    },
+    {
+      "epoch": 29.14827667797145,
+      "grad_norm": 1.609375,
+      "learning_rate": 8.094789473684211e-05,
+      "loss": 1.0814,
+      "step": 46200
+    },
+    {
+      "epoch": 29.179824907327077,
+      "grad_norm": 1.3203125,
+      "learning_rate": 8.092157894736843e-05,
+      "loss": 1.1044,
+      "step": 46250
+    },
+    {
+      "epoch": 29.211373136682703,
+      "grad_norm": 1.5546875,
+      "learning_rate": 8.089526315789474e-05,
+      "loss": 1.0989,
+      "step": 46300
+    },
+    {
+      "epoch": 29.24292136603833,
+      "grad_norm": 1.4453125,
+      "learning_rate": 8.086894736842106e-05,
+      "loss": 1.1061,
+      "step": 46350
+    },
+    {
+      "epoch": 29.27446959539396,
+      "grad_norm": 1.265625,
+      "learning_rate": 8.084263157894737e-05,
+      "loss": 1.0923,
+      "step": 46400
+    },
+    {
+      "epoch": 29.306017824749585,
+      "grad_norm": 1.2109375,
+      "learning_rate": 8.081631578947368e-05,
+      "loss": 1.1236,
+      "step": 46450
+    },
+    {
+      "epoch": 29.337566054105213,
+      "grad_norm": 0.98828125,
+      "learning_rate": 8.079e-05,
+      "loss": 1.1432,
+      "step": 46500
+    },
+    {
+      "epoch": 29.369114283460842,
+      "grad_norm": 1.2265625,
+      "learning_rate": 8.076368421052633e-05,
+      "loss": 1.1098,
+      "step": 46550
+    },
+    {
+      "epoch": 29.400662512816467,
+      "grad_norm": 1.28125,
+      "learning_rate": 8.073736842105264e-05,
+      "loss": 1.1515,
+      "step": 46600
+    },
+    {
+      "epoch": 29.432210742172096,
+      "grad_norm": 1.0234375,
+      "learning_rate": 8.071105263157895e-05,
+      "loss": 1.1382,
+      "step": 46650
+    },
+    {
+      "epoch": 29.463758971527724,
+      "grad_norm": 1.2265625,
+      "learning_rate": 8.068473684210526e-05,
+      "loss": 1.1341,
+      "step": 46700
+    },
+    {
+      "epoch": 29.49530720088335,
+      "grad_norm": 1.2109375,
+      "learning_rate": 8.065842105263159e-05,
+      "loss": 1.0388,
+      "step": 46750
+    },
+    {
+      "epoch": 29.526855430238978,
+      "grad_norm": 1.296875,
+      "learning_rate": 8.06321052631579e-05,
+      "loss": 1.1351,
+      "step": 46800
+    },
+    {
+      "epoch": 29.558403659594607,
+      "grad_norm": 1.265625,
+      "learning_rate": 8.060578947368421e-05,
+      "loss": 1.1116,
+      "step": 46850
+    },
+    {
+      "epoch": 29.58995188895023,
+      "grad_norm": 1.3828125,
+      "learning_rate": 8.057947368421052e-05,
+      "loss": 1.1026,
+      "step": 46900
+    },
+    {
+      "epoch": 29.62150011830586,
+      "grad_norm": 1.5625,
+      "learning_rate": 8.055315789473684e-05,
+      "loss": 1.1015,
+      "step": 46950
+    },
+    {
+      "epoch": 29.65304834766149,
+      "grad_norm": 1.453125,
+      "learning_rate": 8.052684210526316e-05,
+      "loss": 1.2103,
+      "step": 47000
+    },
+    {
+      "epoch": 29.65304834766149,
+      "eval_loss": 1.504706859588623,
+      "eval_runtime": 267.8989,
+      "eval_samples_per_second": 94.651,
+      "eval_steps_per_second": 5.916,
+      "step": 47000
+    },
+    {
+      "epoch": 29.684596577017114,
+      "grad_norm": 1.2890625,
+      "learning_rate": 8.050052631578949e-05,
+      "loss": 1.0734,
+      "step": 47050
+    },
+    {
+      "epoch": 29.716144806372743,
+      "grad_norm": 1.25,
+      "learning_rate": 8.04742105263158e-05,
+      "loss": 1.1404,
+      "step": 47100
+    },
+    {
+      "epoch": 29.74769303572837,
+      "grad_norm": 0.8125,
+      "learning_rate": 8.044789473684211e-05,
+      "loss": 1.1454,
+      "step": 47150
+    },
+    {
+      "epoch": 29.779241265083996,
+      "grad_norm": 1.1015625,
+      "learning_rate": 8.042157894736842e-05,
+      "loss": 1.1346,
+      "step": 47200
+    },
+    {
+      "epoch": 29.810789494439625,
+      "grad_norm": 1.2265625,
+      "learning_rate": 8.039526315789475e-05,
+      "loss": 1.202,
+      "step": 47250
+    },
+    {
+      "epoch": 29.842337723795254,
+      "grad_norm": 1.234375,
+      "learning_rate": 8.036894736842106e-05,
+      "loss": 1.1173,
+      "step": 47300
+    },
+    {
+      "epoch": 29.87388595315088,
+      "grad_norm": 1.2109375,
+      "learning_rate": 8.034263157894737e-05,
+      "loss": 1.1577,
+      "step": 47350
+    },
+    {
+      "epoch": 29.905434182506507,
+      "grad_norm": 2.828125,
+      "learning_rate": 8.031631578947368e-05,
+      "loss": 1.088,
+      "step": 47400
+    },
+    {
+      "epoch": 29.936982411862136,
+      "grad_norm": 1.6640625,
+      "learning_rate": 8.028999999999999e-05,
+      "loss": 1.1395,
+      "step": 47450
+    },
+    {
+      "epoch": 29.96853064121776,
+      "grad_norm": 1.0390625,
+      "learning_rate": 8.026368421052633e-05,
+      "loss": 1.1112,
+      "step": 47500
+    },
+    {
+      "epoch": 30.0,
+      "grad_norm": 1.8984375,
+      "learning_rate": 8.023736842105264e-05,
+      "loss": 1.0993,
+      "step": 47550
+    },
+    {
+      "epoch": 30.03154822935563,
+      "grad_norm": 0.87890625,
+      "learning_rate": 8.021105263157895e-05,
+      "loss": 1.1962,
+      "step": 47600
+    },
+    {
+      "epoch": 30.063096458711254,
+      "grad_norm": 1.0625,
+      "learning_rate": 8.018473684210527e-05,
+      "loss": 1.0634,
+      "step": 47650
+    },
+    {
+      "epoch": 30.094644688066882,
+      "grad_norm": 1.2890625,
+      "learning_rate": 8.015842105263158e-05,
+      "loss": 1.1137,
+      "step": 47700
+    },
+    {
+      "epoch": 30.12619291742251,
+      "grad_norm": 1.109375,
+      "learning_rate": 8.01321052631579e-05,
+      "loss": 1.1681,
+      "step": 47750
+    },
+    {
+      "epoch": 30.157741146778136,
+      "grad_norm": 1.3984375,
+      "learning_rate": 8.010578947368421e-05,
+      "loss": 1.0424,
+      "step": 47800
+    },
+    {
+      "epoch": 30.189289376133765,
+      "grad_norm": 1.3046875,
+      "learning_rate": 8.007947368421053e-05,
+      "loss": 1.1927,
+      "step": 47850
+    },
+    {
+      "epoch": 30.220837605489393,
+      "grad_norm": 1.03125,
+      "learning_rate": 8.005315789473684e-05,
+      "loss": 1.0464,
+      "step": 47900
+    },
+    {
+      "epoch": 30.25238583484502,
+      "grad_norm": 1.2109375,
+      "learning_rate": 8.002684210526316e-05,
+      "loss": 1.0849,
+      "step": 47950
+    },
+    {
+      "epoch": 30.283934064200647,
+      "grad_norm": 1.484375,
+      "learning_rate": 8.000052631578949e-05,
+      "loss": 1.1061,
+      "step": 48000
+    },
+    {
+      "epoch": 30.283934064200647,
+      "eval_loss": 1.5036768913269043,
+      "eval_runtime": 267.7531,
+      "eval_samples_per_second": 94.703,
+      "eval_steps_per_second": 5.92,
+      "step": 48000
+    },
+    {
+      "epoch": 30.315482293556276,
+      "grad_norm": 1.4765625,
+      "learning_rate": 7.99742105263158e-05,
+      "loss": 1.1553,
+      "step": 48050
+    },
+    {
+      "epoch": 30.3470305229119,
+      "grad_norm": 1.375,
+      "learning_rate": 7.994789473684211e-05,
+      "loss": 1.0855,
+      "step": 48100
+    },
+    {
+      "epoch": 30.37857875226753,
+      "grad_norm": 1.6015625,
+      "learning_rate": 7.992157894736842e-05,
+      "loss": 1.1316,
+      "step": 48150
+    },
+    {
+      "epoch": 30.410126981623158,
+      "grad_norm": 1.6328125,
+      "learning_rate": 7.989526315789473e-05,
+      "loss": 1.1295,
+      "step": 48200
+    },
+    {
+      "epoch": 30.441675210978783,
+      "grad_norm": 1.7109375,
+      "learning_rate": 7.986894736842106e-05,
+      "loss": 1.1605,
+      "step": 48250
+    },
+    {
+      "epoch": 30.47322344033441,
+      "grad_norm": 1.421875,
+      "learning_rate": 7.984263157894737e-05,
+      "loss": 1.1485,
+      "step": 48300
+    },
+    {
+      "epoch": 30.50477166969004,
+      "grad_norm": 1.1328125,
+      "learning_rate": 7.981631578947368e-05,
+      "loss": 1.1767,
+      "step": 48350
+    },
+    {
+      "epoch": 30.536319899045665,
+      "grad_norm": 1.046875,
+      "learning_rate": 7.979000000000001e-05,
+      "loss": 1.1085,
+      "step": 48400
+    },
+    {
+      "epoch": 30.567868128401294,
+      "grad_norm": 1.0859375,
+      "learning_rate": 7.976368421052632e-05,
+      "loss": 1.081,
+      "step": 48450
+    },
+    {
+      "epoch": 30.599416357756922,
+      "grad_norm": 2.25,
+      "learning_rate": 7.973736842105264e-05,
+      "loss": 1.1187,
+      "step": 48500
+    },
+    {
+      "epoch": 30.630964587112548,
+      "grad_norm": 1.328125,
+      "learning_rate": 7.971105263157896e-05,
+      "loss": 1.1455,
+      "step": 48550
+    },
+    {
+      "epoch": 30.662512816468176,
+      "grad_norm": 1.1796875,
+      "learning_rate": 7.968473684210527e-05,
+      "loss": 1.0459,
+      "step": 48600
+    },
+    {
+      "epoch": 30.694061045823805,
+      "grad_norm": 1.078125,
+      "learning_rate": 7.965842105263158e-05,
+      "loss": 1.1228,
+      "step": 48650
+    },
+    {
+      "epoch": 30.72560927517943,
+      "grad_norm": 1.6328125,
+      "learning_rate": 7.96321052631579e-05,
+      "loss": 1.1881,
+      "step": 48700
+    },
+    {
+      "epoch": 30.75715750453506,
+      "grad_norm": 1.0625,
+      "learning_rate": 7.960578947368422e-05,
+      "loss": 1.1357,
+      "step": 48750
+    },
+    {
+      "epoch": 30.788705733890687,
+      "grad_norm": 1.328125,
+      "learning_rate": 7.957947368421053e-05,
+      "loss": 1.1449,
+      "step": 48800
+    },
+    {
+      "epoch": 30.820253963246312,
+      "grad_norm": 1.2265625,
+      "learning_rate": 7.955315789473684e-05,
+      "loss": 1.0769,
+      "step": 48850
+    },
+    {
+      "epoch": 30.85180219260194,
+      "grad_norm": 1.3203125,
+      "learning_rate": 7.952684210526316e-05,
+      "loss": 1.1382,
+      "step": 48900
+    },
+    {
+      "epoch": 30.883350421957566,
+      "grad_norm": 1.1328125,
+      "learning_rate": 7.950052631578947e-05,
+      "loss": 1.1921,
+      "step": 48950
+    },
+    {
+      "epoch": 30.914898651313194,
+      "grad_norm": 1.3671875,
+      "learning_rate": 7.94742105263158e-05,
+      "loss": 1.1503,
+      "step": 49000
+    },
+    {
+      "epoch": 30.914898651313194,
+      "eval_loss": 1.5025678873062134,
+      "eval_runtime": 268.5667,
+      "eval_samples_per_second": 94.416,
+      "eval_steps_per_second": 5.902,
+      "step": 49000
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 200000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 127,
+  "save_steps": 1000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.634955028608e+17,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}