diff --git "a/checkpoint-1695/trainer_state.json" "b/checkpoint-1695/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-1695/trainer_state.json"
@@ -0,0 +1,11982 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.9986731534719153,
+  "eval_steps": 142,
+  "global_step": 1695,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 5.333217144012451,
+      "learning_rate": 3e-06,
+      "loss": 3.4697,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "eval_loss": 3.547457695007324,
+      "eval_runtime": 14.4761,
+      "eval_samples_per_second": 32.882,
+      "eval_steps_per_second": 8.22,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 5.45876932144165,
+      "learning_rate": 6e-06,
+      "loss": 3.4361,
+      "step": 2
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.8387346267700195,
+      "learning_rate": 9e-06,
+      "loss": 3.6111,
+      "step": 3
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.629228591918945,
+      "learning_rate": 1.2e-05,
+      "loss": 3.468,
+      "step": 4
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.132396697998047,
+      "learning_rate": 1.5e-05,
+      "loss": 3.4794,
+      "step": 5
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.605806350708008,
+      "learning_rate": 1.8e-05,
+      "loss": 3.4557,
+      "step": 6
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.7556891441345215,
+      "learning_rate": 2.1e-05,
+      "loss": 3.4051,
+      "step": 7
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.8776655197143555,
+      "learning_rate": 2.4e-05,
+      "loss": 3.1824,
+      "step": 8
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.992830753326416,
+      "learning_rate": 2.7000000000000002e-05,
+      "loss": 3.0846,
+      "step": 9
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.38725471496582,
+      "learning_rate": 3e-05,
+      "loss": 2.8616,
+      "step": 10
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.4103684425354,
+      "learning_rate": 2.9999973928796923e-05,
+      "loss": 2.5679,
+      "step": 11
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.2000298500061035,
+      "learning_rate": 2.999989571527831e-05,
+      "loss": 2.2378,
+      "step": 12
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.7659687995910645,
+      "learning_rate": 2.9999765359716046e-05,
+      "loss": 1.8411,
+      "step": 13
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.1735832691192627,
+      "learning_rate": 2.999958286256327e-05,
+      "loss": 1.519,
+      "step": 14
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.7938477993011475,
+      "learning_rate": 2.9999348224454367e-05,
+      "loss": 1.2343,
+      "step": 15
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.5651497840881348,
+      "learning_rate": 2.9999061446204985e-05,
+      "loss": 1.0183,
+      "step": 16
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.3572280406951904,
+      "learning_rate": 2.9998722528812e-05,
+      "loss": 0.7906,
+      "step": 17
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.2292940616607666,
+      "learning_rate": 2.9998331473453557e-05,
+      "loss": 0.6625,
+      "step": 18
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.9572314023971558,
+      "learning_rate": 2.9997888281489015e-05,
+      "loss": 0.5271,
+      "step": 19
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.7180088758468628,
+      "learning_rate": 2.9997392954458985e-05,
+      "loss": 0.4319,
+      "step": 20
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.5089068412780762,
+      "learning_rate": 2.9996845494085306e-05,
+      "loss": 0.351,
+      "step": 21
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.2355144023895264,
+      "learning_rate": 2.999624590227103e-05,
+      "loss": 0.2994,
+      "step": 22
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.7736936211585999,
+      "learning_rate": 2.9995594181100443e-05,
+      "loss": 0.1704,
+      "step": 23
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.5642102956771851,
+      "learning_rate": 2.9994890332839027e-05,
+      "loss": 0.1217,
+      "step": 24
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.34946027398109436,
+      "learning_rate": 2.9994134359933476e-05,
+      "loss": 0.0992,
+      "step": 25
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.24627894163131714,
+      "learning_rate": 2.999332626501167e-05,
+      "loss": 0.0828,
+      "step": 26
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.28634899854660034,
+      "learning_rate": 2.9992466050882673e-05,
+      "loss": 0.0945,
+      "step": 27
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.1457609236240387,
+      "learning_rate": 2.9991553720536733e-05,
+      "loss": 0.0924,
+      "step": 28
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.3852764964103699,
+      "learning_rate": 2.9990589277145254e-05,
+      "loss": 0.0868,
+      "step": 29
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.28345659375190735,
+      "learning_rate": 2.9989572724060797e-05,
+      "loss": 0.0845,
+      "step": 30
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.24844343960285187,
+      "learning_rate": 2.998850406481707e-05,
+      "loss": 0.0756,
+      "step": 31
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.2215878665447235,
+      "learning_rate": 2.9987383303128887e-05,
+      "loss": 0.0799,
+      "step": 32
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.20273327827453613,
+      "learning_rate": 2.9986210442892215e-05,
+      "loss": 0.099,
+      "step": 33
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.23849628865718842,
+      "learning_rate": 2.9984985488184086e-05,
+      "loss": 0.0709,
+      "step": 34
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.28950855135917664,
+      "learning_rate": 2.9983708443262656e-05,
+      "loss": 0.0736,
+      "step": 35
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.4729451537132263,
+      "learning_rate": 2.9982379312567126e-05,
+      "loss": 0.0898,
+      "step": 36
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.26402196288108826,
+      "learning_rate": 2.998099810071777e-05,
+      "loss": 0.0989,
+      "step": 37
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.21811941266059875,
+      "learning_rate": 2.9979564812515906e-05,
+      "loss": 0.0816,
+      "step": 38
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.4151253402233124,
+      "learning_rate": 2.9978079452943875e-05,
+      "loss": 0.0646,
+      "step": 39
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.5114066004753113,
+      "learning_rate": 2.9976542027165016e-05,
+      "loss": 0.0986,
+      "step": 40
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.23638691008090973,
+      "learning_rate": 2.9974952540523676e-05,
+      "loss": 0.0929,
+      "step": 41
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.18284691870212555,
+      "learning_rate": 2.997331099854516e-05,
+      "loss": 0.0707,
+      "step": 42
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.415445476770401,
+      "learning_rate": 2.9971617406935735e-05,
+      "loss": 0.0875,
+      "step": 43
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.35435256361961365,
+      "learning_rate": 2.9969871771582596e-05,
+      "loss": 0.0917,
+      "step": 44
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.5223746299743652,
+      "learning_rate": 2.996807409855385e-05,
+      "loss": 0.0839,
+      "step": 45
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.2852335274219513,
+      "learning_rate": 2.99662243940985e-05,
+      "loss": 0.0581,
+      "step": 46
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.19847406446933746,
+      "learning_rate": 2.9964322664646412e-05,
+      "loss": 0.0581,
+      "step": 47
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.42156773805618286,
+      "learning_rate": 2.9962368916808308e-05,
+      "loss": 0.0949,
+      "step": 48
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.1932952105998993,
+      "learning_rate": 2.9960363157375724e-05,
+      "loss": 0.0936,
+      "step": 49
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.28940048813819885,
+      "learning_rate": 2.9958305393321e-05,
+      "loss": 0.1074,
+      "step": 50
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.4381609857082367,
+      "learning_rate": 2.995619563179726e-05,
+      "loss": 0.1053,
+      "step": 51
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.16727806627750397,
+      "learning_rate": 2.9954033880138368e-05,
+      "loss": 0.0825,
+      "step": 52
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.3134240210056305,
+      "learning_rate": 2.995182014585892e-05,
+      "loss": 0.0565,
+      "step": 53
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.16883538663387299,
+      "learning_rate": 2.9949554436654215e-05,
+      "loss": 0.0805,
+      "step": 54
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.4369119107723236,
+      "learning_rate": 2.994723676040022e-05,
+      "loss": 0.079,
+      "step": 55
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.22722230851650238,
+      "learning_rate": 2.9944867125153548e-05,
+      "loss": 0.0707,
+      "step": 56
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.2767599821090698,
+      "learning_rate": 2.9942445539151432e-05,
+      "loss": 0.0877,
+      "step": 57
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.2087458074092865,
+      "learning_rate": 2.9939972010811693e-05,
+      "loss": 0.0746,
+      "step": 58
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.2058565467596054,
+      "learning_rate": 2.993744654873272e-05,
+      "loss": 0.0783,
+      "step": 59
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.2181982547044754,
+      "learning_rate": 2.993486916169341e-05,
+      "loss": 0.0764,
+      "step": 60
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.18020497262477875,
+      "learning_rate": 2.9932239858653183e-05,
+      "loss": 0.0675,
+      "step": 61
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.2599628269672394,
+      "learning_rate": 2.992955864875192e-05,
+      "loss": 0.0957,
+      "step": 62
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.2814478576183319,
+      "learning_rate": 2.9926825541309928e-05,
+      "loss": 0.0926,
+      "step": 63
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.21382613480091095,
+      "learning_rate": 2.9924040545827936e-05,
+      "loss": 0.0944,
+      "step": 64
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.16025784611701965,
+      "learning_rate": 2.9921203671987025e-05,
+      "loss": 0.0727,
+      "step": 65
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.22130753099918365,
+      "learning_rate": 2.9918314929648637e-05,
+      "loss": 0.0906,
+      "step": 66
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.16255636513233185,
+      "learning_rate": 2.991537432885449e-05,
+      "loss": 0.057,
+      "step": 67
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.3057488203048706,
+      "learning_rate": 2.991238187982659e-05,
+      "loss": 0.0866,
+      "step": 68
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.24336600303649902,
+      "learning_rate": 2.9909337592967176e-05,
+      "loss": 0.0509,
+      "step": 69
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.25192490220069885,
+      "learning_rate": 2.9906241478858666e-05,
+      "loss": 0.0677,
+      "step": 70
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.6552413105964661,
+      "learning_rate": 2.990309354826366e-05,
+      "loss": 0.1014,
+      "step": 71
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.22106248140335083,
+      "learning_rate": 2.9899893812124862e-05,
+      "loss": 0.0779,
+      "step": 72
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.2502892017364502,
+      "learning_rate": 2.989664228156507e-05,
+      "loss": 0.0759,
+      "step": 73
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.23127524554729462,
+      "learning_rate": 2.9893338967887128e-05,
+      "loss": 0.0735,
+      "step": 74
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.22794046998023987,
+      "learning_rate": 2.988998388257388e-05,
+      "loss": 0.048,
+      "step": 75
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.2867235541343689,
+      "learning_rate": 2.988657703728815e-05,
+      "loss": 0.0449,
+      "step": 76
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.28723156452178955,
+      "learning_rate": 2.9883118443872662e-05,
+      "loss": 0.0525,
+      "step": 77
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.21423418819904327,
+      "learning_rate": 2.9879608114350064e-05,
+      "loss": 0.0684,
+      "step": 78
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.22717736661434174,
+      "learning_rate": 2.9876046060922803e-05,
+      "loss": 0.0496,
+      "step": 79
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.52623051404953,
+      "learning_rate": 2.987243229597316e-05,
+      "loss": 0.1125,
+      "step": 80
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.41985878348350525,
+      "learning_rate": 2.9868766832063156e-05,
+      "loss": 0.1115,
+      "step": 81
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.5244089365005493,
+      "learning_rate": 2.986504968193454e-05,
+      "loss": 0.1185,
+      "step": 82
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.4969002604484558,
+      "learning_rate": 2.9861280858508712e-05,
+      "loss": 0.1123,
+      "step": 83
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.31229168176651,
+      "learning_rate": 2.9857460374886717e-05,
+      "loss": 0.082,
+      "step": 84
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.34874776005744934,
+      "learning_rate": 2.985358824434916e-05,
+      "loss": 0.0732,
+      "step": 85
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.33042728900909424,
+      "learning_rate": 2.984966448035619e-05,
+      "loss": 0.0719,
+      "step": 86
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.9238921403884888,
+      "learning_rate": 2.9845689096547442e-05,
+      "loss": 0.1171,
+      "step": 87
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.24976597726345062,
+      "learning_rate": 2.9841662106741986e-05,
+      "loss": 0.0875,
+      "step": 88
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.23768329620361328,
+      "learning_rate": 2.983758352493829e-05,
+      "loss": 0.0821,
+      "step": 89
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.1419605314731598,
+      "learning_rate": 2.983345336531415e-05,
+      "loss": 0.0696,
+      "step": 90
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.15112988650798798,
+      "learning_rate": 2.9829271642226665e-05,
+      "loss": 0.0958,
+      "step": 91
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.11599217355251312,
+      "learning_rate": 2.9825038370212183e-05,
+      "loss": 0.0838,
+      "step": 92
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.34909361600875854,
+      "learning_rate": 2.982075356398623e-05,
+      "loss": 0.0922,
+      "step": 93
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.3156541585922241,
+      "learning_rate": 2.9816417238443482e-05,
+      "loss": 0.106,
+      "step": 94
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.21198609471321106,
+      "learning_rate": 2.9812029408657698e-05,
+      "loss": 0.0832,
+      "step": 95
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.17507284879684448,
+      "learning_rate": 2.9807590089881687e-05,
+      "loss": 0.0736,
+      "step": 96
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.09285402297973633,
+      "learning_rate": 2.980309929754722e-05,
+      "loss": 0.0816,
+      "step": 97
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.21642349660396576,
+      "learning_rate": 2.9798557047265023e-05,
+      "loss": 0.0806,
+      "step": 98
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.20919114351272583,
+      "learning_rate": 2.979396335482469e-05,
+      "loss": 0.0852,
+      "step": 99
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.1738620400428772,
+      "learning_rate": 2.9789318236194618e-05,
+      "loss": 0.079,
+      "step": 100
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.19599463045597076,
+      "learning_rate": 2.9784621707521993e-05,
+      "loss": 0.0635,
+      "step": 101
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.18762755393981934,
+      "learning_rate": 2.97798737851327e-05,
+      "loss": 0.0863,
+      "step": 102
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.15935586392879486,
+      "learning_rate": 2.977507448553128e-05,
+      "loss": 0.0692,
+      "step": 103
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.25695595145225525,
+      "learning_rate": 2.9770223825400872e-05,
+      "loss": 0.0914,
+      "step": 104
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.17666970193386078,
+      "learning_rate": 2.9765321821603143e-05,
+      "loss": 0.0617,
+      "step": 105
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.1571740359067917,
+      "learning_rate": 2.9760368491178244e-05,
+      "loss": 0.0797,
+      "step": 106
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.13504642248153687,
+      "learning_rate": 2.9755363851344753e-05,
+      "loss": 0.0647,
+      "step": 107
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.12887655198574066,
+      "learning_rate": 2.9750307919499595e-05,
+      "loss": 0.0662,
+      "step": 108
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.22609621286392212,
+      "learning_rate": 2.9745200713218002e-05,
+      "loss": 0.0809,
+      "step": 109
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.2685830295085907,
+      "learning_rate": 2.9740042250253443e-05,
+      "loss": 0.084,
+      "step": 110
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.18346565961837769,
+      "learning_rate": 2.973483254853756e-05,
+      "loss": 0.0568,
+      "step": 111
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.4906081259250641,
+      "learning_rate": 2.9729571626180116e-05,
+      "loss": 0.0911,
+      "step": 112
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.28486496210098267,
+      "learning_rate": 2.972425950146891e-05,
+      "loss": 0.0671,
+      "step": 113
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.35737311840057373,
+      "learning_rate": 2.9718896192869758e-05,
+      "loss": 0.1005,
+      "step": 114
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.27113574743270874,
+      "learning_rate": 2.9713481719026368e-05,
+      "loss": 0.0655,
+      "step": 115
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.17216627299785614,
+      "learning_rate": 2.970801609876032e-05,
+      "loss": 0.0463,
+      "step": 116
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.1476421356201172,
+      "learning_rate": 2.9702499351070992e-05,
+      "loss": 0.0534,
+      "step": 117
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.3753887116909027,
+      "learning_rate": 2.969693149513548e-05,
+      "loss": 0.0763,
+      "step": 118
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.22127705812454224,
+      "learning_rate": 2.969131255030855e-05,
+      "loss": 0.0548,
+      "step": 119
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.18454331159591675,
+      "learning_rate": 2.9685642536122545e-05,
+      "loss": 0.0494,
+      "step": 120
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.24991334974765778,
+      "learning_rate": 2.9679921472287358e-05,
+      "loss": 0.089,
+      "step": 121
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.17822043597698212,
+      "learning_rate": 2.967414937869031e-05,
+      "loss": 0.0337,
+      "step": 122
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.19183677434921265,
+      "learning_rate": 2.9668326275396133e-05,
+      "loss": 0.0627,
+      "step": 123
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.1919647753238678,
+      "learning_rate": 2.966245218264687e-05,
+      "loss": 0.0549,
+      "step": 124
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.3507068455219269,
+      "learning_rate": 2.96565271208618e-05,
+      "loss": 0.0802,
+      "step": 125
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.3020438253879547,
+      "learning_rate": 2.9650551110637397e-05,
+      "loss": 0.0803,
+      "step": 126
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.951285183429718,
+      "learning_rate": 2.964452417274723e-05,
+      "loss": 0.1049,
+      "step": 127
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.6102455258369446,
+      "learning_rate": 2.96384463281419e-05,
+      "loss": 0.0711,
+      "step": 128
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.3394455313682556,
+      "learning_rate": 2.9632317597948968e-05,
+      "loss": 0.0704,
+      "step": 129
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.296131432056427,
+      "learning_rate": 2.9626138003472884e-05,
+      "loss": 0.0726,
+      "step": 130
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.37837737798690796,
+      "learning_rate": 2.9619907566194915e-05,
+      "loss": 0.092,
+      "step": 131
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.7398345470428467,
+      "learning_rate": 2.9613626307773055e-05,
+      "loss": 0.1055,
+      "step": 132
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.18485528230667114,
+      "learning_rate": 2.9607294250041965e-05,
+      "loss": 0.0615,
+      "step": 133
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.4245338439941406,
+      "learning_rate": 2.96009114150129e-05,
+      "loss": 0.0938,
+      "step": 134
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.25269758701324463,
+      "learning_rate": 2.959447782487361e-05,
+      "loss": 0.0686,
+      "step": 135
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.1531674861907959,
+      "learning_rate": 2.9587993501988292e-05,
+      "loss": 0.0515,
+      "step": 136
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.1748104691505432,
+      "learning_rate": 2.958145846889749e-05,
+      "loss": 0.0841,
+      "step": 137
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.16989848017692566,
+      "learning_rate": 2.957487274831803e-05,
+      "loss": 0.0822,
+      "step": 138
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.16533434391021729,
+      "learning_rate": 2.9568236363142927e-05,
+      "loss": 0.0676,
+      "step": 139
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.12993450462818146,
+      "learning_rate": 2.9561549336441333e-05,
+      "loss": 0.0837,
+      "step": 140
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.23538929224014282,
+      "learning_rate": 2.955481169145841e-05,
+      "loss": 0.0784,
+      "step": 141
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.1557569056749344,
+      "learning_rate": 2.95480234516153e-05,
+      "loss": 0.0788,
+      "step": 142
+    },
+    {
+      "epoch": 0.25,
+      "eval_loss": 0.08073563128709793,
+      "eval_runtime": 14.7663,
+      "eval_samples_per_second": 32.236,
+      "eval_steps_per_second": 8.059,
+      "step": 142
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.217232346534729,
+      "learning_rate": 2.9541184640509015e-05,
+      "loss": 0.0598,
+      "step": 143
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.2527540326118469,
+      "learning_rate": 2.953429528191236e-05,
+      "loss": 0.0604,
+      "step": 144
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.1690036803483963,
+      "learning_rate": 2.9527355399773847e-05,
+      "loss": 0.0815,
+      "step": 145
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.15807238221168518,
+      "learning_rate": 2.9520365018217622e-05,
+      "loss": 0.0695,
+      "step": 146
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.12181195616722107,
+      "learning_rate": 2.951332416154337e-05,
+      "loss": 0.0869,
+      "step": 147
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.27408546209335327,
+      "learning_rate": 2.9506232854226242e-05,
+      "loss": 0.0725,
+      "step": 148
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.1542072892189026,
+      "learning_rate": 2.9499091120916757e-05,
+      "loss": 0.0616,
+      "step": 149
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.31103819608688354,
+      "learning_rate": 2.9491898986440726e-05,
+      "loss": 0.0791,
+      "step": 150
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.15201374888420105,
+      "learning_rate": 2.9484656475799164e-05,
+      "loss": 0.0681,
+      "step": 151
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.33369091153144836,
+      "learning_rate": 2.9477363614168197e-05,
+      "loss": 0.0673,
+      "step": 152
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.1663801521062851,
+      "learning_rate": 2.9470020426898983e-05,
+      "loss": 0.0514,
+      "step": 153
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.3342381715774536,
+      "learning_rate": 2.946262693951762e-05,
+      "loss": 0.0824,
+      "step": 154
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.15362265706062317,
+      "learning_rate": 2.9455183177725055e-05,
+      "loss": 0.0801,
+      "step": 155
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.1435208022594452,
+      "learning_rate": 2.9447689167397e-05,
+      "loss": 0.0625,
+      "step": 156
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.23910465836524963,
+      "learning_rate": 2.944014493458383e-05,
+      "loss": 0.0977,
+      "step": 157
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.15610359609127045,
+      "learning_rate": 2.9432550505510516e-05,
+      "loss": 0.0329,
+      "step": 158
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.23283672332763672,
+      "learning_rate": 2.942490590657651e-05,
+      "loss": 0.1056,
+      "step": 159
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.21941441297531128,
+      "learning_rate": 2.9417211164355668e-05,
+      "loss": 0.0765,
+      "step": 160
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.16233329474925995,
+      "learning_rate": 2.9409466305596135e-05,
+      "loss": 0.056,
+      "step": 161
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.22437363862991333,
+      "learning_rate": 2.9401671357220297e-05,
+      "loss": 0.0845,
+      "step": 162
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.24630329012870789,
+      "learning_rate": 2.9393826346324634e-05,
+      "loss": 0.0588,
+      "step": 163
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.17304874956607819,
+      "learning_rate": 2.9385931300179675e-05,
+      "loss": 0.0582,
+      "step": 164
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.35216909646987915,
+      "learning_rate": 2.9377986246229853e-05,
+      "loss": 0.1085,
+      "step": 165
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.13712701201438904,
+      "learning_rate": 2.9369991212093462e-05,
+      "loss": 0.0451,
+      "step": 166
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.15788517892360687,
+      "learning_rate": 2.9361946225562516e-05,
+      "loss": 0.0572,
+      "step": 167
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.15666936337947845,
+      "learning_rate": 2.9353851314602676e-05,
+      "loss": 0.0528,
+      "step": 168
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.17414148151874542,
+      "learning_rate": 2.9345706507353158e-05,
+      "loss": 0.0548,
+      "step": 169
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.2488938271999359,
+      "learning_rate": 2.9337511832126616e-05,
+      "loss": 0.0904,
+      "step": 170
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.2884049415588379,
+      "learning_rate": 2.9329267317409053e-05,
+      "loss": 0.0721,
+      "step": 171
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.3515964150428772,
+      "learning_rate": 2.9320972991859728e-05,
+      "loss": 0.1108,
+      "step": 172
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.49220752716064453,
+      "learning_rate": 2.9312628884311048e-05,
+      "loss": 0.1056,
+      "step": 173
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.1283823549747467,
+      "learning_rate": 2.9304235023768465e-05,
+      "loss": 0.062,
+      "step": 174
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.20535992085933685,
+      "learning_rate": 2.9295791439410387e-05,
+      "loss": 0.0591,
+      "step": 175
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.548753559589386,
+      "learning_rate": 2.9287298160588073e-05,
+      "loss": 0.1252,
+      "step": 176
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.17065873742103577,
+      "learning_rate": 2.927875521682551e-05,
+      "loss": 0.0717,
+      "step": 177
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.15028174221515656,
+      "learning_rate": 2.9270162637819352e-05,
+      "loss": 0.059,
+      "step": 178
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.3822297751903534,
+      "learning_rate": 2.9261520453438775e-05,
+      "loss": 0.0752,
+      "step": 179
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.17701248824596405,
+      "learning_rate": 2.9252828693725404e-05,
+      "loss": 0.0872,
+      "step": 180
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.1596059501171112,
+      "learning_rate": 2.9244087388893187e-05,
+      "loss": 0.051,
+      "step": 181
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.13807597756385803,
+      "learning_rate": 2.9235296569328303e-05,
+      "loss": 0.0518,
+      "step": 182
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.14610590040683746,
+      "learning_rate": 2.922645626558905e-05,
+      "loss": 0.08,
+      "step": 183
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.13961593806743622,
+      "learning_rate": 2.921756650840574e-05,
+      "loss": 0.0496,
+      "step": 184
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.12945222854614258,
+      "learning_rate": 2.92086273286806e-05,
+      "loss": 0.0725,
+      "step": 185
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.16739709675312042,
+      "learning_rate": 2.9199638757487648e-05,
+      "loss": 0.0514,
+      "step": 186
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.1626630425453186,
+      "learning_rate": 2.9190600826072603e-05,
+      "loss": 0.0506,
+      "step": 187
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.17024777829647064,
+      "learning_rate": 2.918151356585276e-05,
+      "loss": 0.0646,
+      "step": 188
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.12996181845664978,
+      "learning_rate": 2.9172377008416898e-05,
+      "loss": 0.0421,
+      "step": 189
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.1544611155986786,
+      "learning_rate": 2.916319118552515e-05,
+      "loss": 0.0665,
+      "step": 190
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.37734368443489075,
+      "learning_rate": 2.9153956129108918e-05,
+      "loss": 0.0851,
+      "step": 191
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.28739050030708313,
+      "learning_rate": 2.9144671871270734e-05,
+      "loss": 0.0786,
+      "step": 192
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.2730344533920288,
+      "learning_rate": 2.913533844428417e-05,
+      "loss": 0.0923,
+      "step": 193
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.26277613639831543,
+      "learning_rate": 2.912595588059371e-05,
+      "loss": 0.0737,
+      "step": 194
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.15028347074985504,
+      "learning_rate": 2.9116524212814653e-05,
+      "loss": 0.0528,
+      "step": 195
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.5112478733062744,
+      "learning_rate": 2.9107043473733e-05,
+      "loss": 0.0744,
+      "step": 196
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.1651376336812973,
+      "learning_rate": 2.9097513696305304e-05,
+      "loss": 0.0421,
+      "step": 197
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.18412701785564423,
+      "learning_rate": 2.908793491365861e-05,
+      "loss": 0.0602,
+      "step": 198
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.26233047246932983,
+      "learning_rate": 2.90783071590903e-05,
+      "loss": 0.086,
+      "step": 199
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.17222779989242554,
+      "learning_rate": 2.9068630466067997e-05,
+      "loss": 0.0634,
+      "step": 200
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.4317842125892639,
+      "learning_rate": 2.905890486822943e-05,
+      "loss": 0.0904,
+      "step": 201
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.1886034905910492,
+      "learning_rate": 2.9049130399382345e-05,
+      "loss": 0.0699,
+      "step": 202
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.21989533305168152,
+      "learning_rate": 2.903930709350436e-05,
+      "loss": 0.0623,
+      "step": 203
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.19172674417495728,
+      "learning_rate": 2.9029434984742866e-05,
+      "loss": 0.0956,
+      "step": 204
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.20230571925640106,
+      "learning_rate": 2.9019514107414888e-05,
+      "loss": 0.0865,
+      "step": 205
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.258619487285614,
+      "learning_rate": 2.9009544496006998e-05,
+      "loss": 0.077,
+      "step": 206
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.37927359342575073,
+      "learning_rate": 2.8999526185175155e-05,
+      "loss": 0.0955,
+      "step": 207
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.1734340786933899,
+      "learning_rate": 2.898945920974462e-05,
+      "loss": 0.0698,
+      "step": 208
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.2214702069759369,
+      "learning_rate": 2.8979343604709818e-05,
+      "loss": 0.0656,
+      "step": 209
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.13421665132045746,
+      "learning_rate": 2.8969179405234202e-05,
+      "loss": 0.0801,
+      "step": 210
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.1412036120891571,
+      "learning_rate": 2.8958966646650172e-05,
+      "loss": 0.0413,
+      "step": 211
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.14668437838554382,
+      "learning_rate": 2.894870536445891e-05,
+      "loss": 0.067,
+      "step": 212
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.17726927995681763,
+      "learning_rate": 2.893839559433028e-05,
+      "loss": 0.0572,
+      "step": 213
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.1916431039571762,
+      "learning_rate": 2.8928037372102698e-05,
+      "loss": 0.0761,
+      "step": 214
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.16705051064491272,
+      "learning_rate": 2.8917630733783004e-05,
+      "loss": 0.0604,
+      "step": 215
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.17496813833713531,
+      "learning_rate": 2.8907175715546337e-05,
+      "loss": 0.0705,
+      "step": 216
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.15736040472984314,
+      "learning_rate": 2.889667235373603e-05,
+      "loss": 0.0305,
+      "step": 217
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.274248868227005,
+      "learning_rate": 2.888612068486344e-05,
+      "loss": 0.0783,
+      "step": 218
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.24461400508880615,
+      "learning_rate": 2.887552074560787e-05,
+      "loss": 0.0643,
+      "step": 219
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.20980973541736603,
+      "learning_rate": 2.8864872572816407e-05,
+      "loss": 0.0625,
+      "step": 220
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.20946985483169556,
+      "learning_rate": 2.885417620350381e-05,
+      "loss": 0.0834,
+      "step": 221
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.16951480507850647,
+      "learning_rate": 2.8843431674852366e-05,
+      "loss": 0.0528,
+      "step": 222
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.1708109974861145,
+      "learning_rate": 2.883263902421179e-05,
+      "loss": 0.0728,
+      "step": 223
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.14461833238601685,
+      "learning_rate": 2.8821798289099054e-05,
+      "loss": 0.0421,
+      "step": 224
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.24623633921146393,
+      "learning_rate": 2.881090950719831e-05,
+      "loss": 0.0854,
+      "step": 225
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.21715879440307617,
+      "learning_rate": 2.87999727163607e-05,
+      "loss": 0.0384,
+      "step": 226
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.4943262040615082,
+      "learning_rate": 2.878898795460426e-05,
+      "loss": 0.0727,
+      "step": 227
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.21625903248786926,
+      "learning_rate": 2.8777955260113794e-05,
+      "loss": 0.0764,
+      "step": 228
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.49219760298728943,
+      "learning_rate": 2.8766874671240708e-05,
+      "loss": 0.0645,
+      "step": 229
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.48366281390190125,
+      "learning_rate": 2.8755746226502914e-05,
+      "loss": 0.0812,
+      "step": 230
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.23804642260074615,
+      "learning_rate": 2.874456996458467e-05,
+      "loss": 0.0412,
+      "step": 231
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.15954728424549103,
+      "learning_rate": 2.8733345924336448e-05,
+      "loss": 0.0319,
+      "step": 232
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.1840825229883194,
+      "learning_rate": 2.872207414477482e-05,
+      "loss": 0.0391,
+      "step": 233
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.34154170751571655,
+      "learning_rate": 2.8710754665082295e-05,
+      "loss": 0.0871,
+      "step": 234
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.28852567076683044,
+      "learning_rate": 2.8699387524607206e-05,
+      "loss": 0.0743,
+      "step": 235
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.3130442798137665,
+      "learning_rate": 2.868797276286355e-05,
+      "loss": 0.0596,
+      "step": 236
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.15511904656887054,
+      "learning_rate": 2.8676510419530875e-05,
+      "loss": 0.0345,
+      "step": 237
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.3511506915092468,
+      "learning_rate": 2.866500053445412e-05,
+      "loss": 0.0795,
+      "step": 238
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.21553099155426025,
+      "learning_rate": 2.86534431476435e-05,
+      "loss": 0.0685,
+      "step": 239
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.2557514011859894,
+      "learning_rate": 2.8641838299274336e-05,
+      "loss": 0.0386,
+      "step": 240
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.22085203230381012,
+      "learning_rate": 2.863018602968695e-05,
+      "loss": 0.0741,
+      "step": 241
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.3036896884441376,
+      "learning_rate": 2.8618486379386496e-05,
+      "loss": 0.0553,
+      "step": 242
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.24089917540550232,
+      "learning_rate": 2.8606739389042838e-05,
+      "loss": 0.0699,
+      "step": 243
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.3548223376274109,
+      "learning_rate": 2.8594945099490395e-05,
+      "loss": 0.0573,
+      "step": 244
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.288339763879776,
+      "learning_rate": 2.8583103551728008e-05,
+      "loss": 0.0592,
+      "step": 245
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.3822185695171356,
+      "learning_rate": 2.857121478691881e-05,
+      "loss": 0.0788,
+      "step": 246
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.3023355305194855,
+      "learning_rate": 2.855927884639004e-05,
+      "loss": 0.08,
+      "step": 247
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.15000642836093903,
+      "learning_rate": 2.854729577163294e-05,
+      "loss": 0.053,
+      "step": 248
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.38528746366500854,
+      "learning_rate": 2.8535265604302614e-05,
+      "loss": 0.132,
+      "step": 249
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.3799440562725067,
+      "learning_rate": 2.852318838621784e-05,
+      "loss": 0.0824,
+      "step": 250
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.1958758383989334,
+      "learning_rate": 2.851106415936098e-05,
+      "loss": 0.0501,
+      "step": 251
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.47345441579818726,
+      "learning_rate": 2.8498892965877776e-05,
+      "loss": 0.076,
+      "step": 252
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.4306192398071289,
+      "learning_rate": 2.848667484807726e-05,
+      "loss": 0.0907,
+      "step": 253
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.15393877029418945,
+      "learning_rate": 2.8474409848431562e-05,
+      "loss": 0.0471,
+      "step": 254
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.1548716276884079,
+      "learning_rate": 2.8462098009575793e-05,
+      "loss": 0.0545,
+      "step": 255
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.2719079256057739,
+      "learning_rate": 2.8449739374307877e-05,
+      "loss": 0.0712,
+      "step": 256
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.14566250145435333,
+      "learning_rate": 2.8437333985588418e-05,
+      "loss": 0.0491,
+      "step": 257
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.17003220319747925,
+      "learning_rate": 2.8424881886540527e-05,
+      "loss": 0.078,
+      "step": 258
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.16778719425201416,
+      "learning_rate": 2.8412383120449707e-05,
+      "loss": 0.0688,
+      "step": 259
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.2298566699028015,
+      "learning_rate": 2.839983773076367e-05,
+      "loss": 0.0654,
+      "step": 260
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.14038778841495514,
+      "learning_rate": 2.8387245761092203e-05,
+      "loss": 0.0656,
+      "step": 261
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.2299400120973587,
+      "learning_rate": 2.8374607255207012e-05,
+      "loss": 0.0563,
+      "step": 262
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.12228001654148102,
+      "learning_rate": 2.8361922257041577e-05,
+      "loss": 0.0426,
+      "step": 263
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.2170531004667282,
+      "learning_rate": 2.8349190810690977e-05,
+      "loss": 0.0739,
+      "step": 264
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.25972139835357666,
+      "learning_rate": 2.8336412960411765e-05,
+      "loss": 0.0449,
+      "step": 265
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.16110770404338837,
+      "learning_rate": 2.8323588750621802e-05,
+      "loss": 0.0429,
+      "step": 266
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.36092159152030945,
+      "learning_rate": 2.8310718225900095e-05,
+      "loss": 0.0532,
+      "step": 267
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.13183802366256714,
+      "learning_rate": 2.8297801430986652e-05,
+      "loss": 0.0306,
+      "step": 268
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.1969520002603531,
+      "learning_rate": 2.8284838410782327e-05,
+      "loss": 0.0729,
+      "step": 269
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.3288170397281647,
+      "learning_rate": 2.8271829210348657e-05,
+      "loss": 0.0657,
+      "step": 270
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.17383714020252228,
+      "learning_rate": 2.82587738749077e-05,
+      "loss": 0.0296,
+      "step": 271
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.3440845310688019,
+      "learning_rate": 2.824567244984192e-05,
+      "loss": 0.055,
+      "step": 272
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.5535409450531006,
+      "learning_rate": 2.8232524980693947e-05,
+      "loss": 0.0845,
+      "step": 273
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.3418692946434021,
+      "learning_rate": 2.8219331513166503e-05,
+      "loss": 0.0501,
+      "step": 274
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.24258548021316528,
+      "learning_rate": 2.8206092093122195e-05,
+      "loss": 0.0572,
+      "step": 275
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.2323545515537262,
+      "learning_rate": 2.8192806766583373e-05,
+      "loss": 0.036,
+      "step": 276
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.18079668283462524,
+      "learning_rate": 2.8179475579731967e-05,
+      "loss": 0.039,
+      "step": 277
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.6126664280891418,
+      "learning_rate": 2.8166098578909315e-05,
+      "loss": 0.0964,
+      "step": 278
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.29580938816070557,
+      "learning_rate": 2.815267581061602e-05,
+      "loss": 0.0682,
+      "step": 279
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.25424325466156006,
+      "learning_rate": 2.8139207321511778e-05,
+      "loss": 0.0682,
+      "step": 280
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.20733873546123505,
+      "learning_rate": 2.8125693158415217e-05,
+      "loss": 0.0369,
+      "step": 281
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.28711849451065063,
+      "learning_rate": 2.8112133368303737e-05,
+      "loss": 0.0734,
+      "step": 282
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.18627013266086578,
+      "learning_rate": 2.809852799831334e-05,
+      "loss": 0.052,
+      "step": 283
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.3224278688430786,
+      "learning_rate": 2.8084877095738477e-05,
+      "loss": 0.0973,
+      "step": 284
+    },
+    {
+      "epoch": 0.5,
+      "eval_loss": 0.06570792198181152,
+      "eval_runtime": 14.7272,
+      "eval_samples_per_second": 32.321,
+      "eval_steps_per_second": 8.08,
+      "step": 284
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.38838616013526917,
+      "learning_rate": 2.8071180708031874e-05,
+      "loss": 0.0916,
+      "step": 285
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.24502405524253845,
+      "learning_rate": 2.8057438882804372e-05,
+      "loss": 0.0816,
+      "step": 286
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.4185783863067627,
+      "learning_rate": 2.8043651667824767e-05,
+      "loss": 0.0776,
+      "step": 287
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.2056495100259781,
+      "learning_rate": 2.8029819111019618e-05,
+      "loss": 0.0266,
+      "step": 288
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.19617842137813568,
+      "learning_rate": 2.8015941260473117e-05,
+      "loss": 0.0456,
+      "step": 289
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.3548401892185211,
+      "learning_rate": 2.8002018164426896e-05,
+      "loss": 0.0507,
+      "step": 290
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.2668496370315552,
+      "learning_rate": 2.798804987127988e-05,
+      "loss": 0.0539,
+      "step": 291
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.2864347994327545,
+      "learning_rate": 2.7974036429588082e-05,
+      "loss": 0.076,
+      "step": 292
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.26988208293914795,
+      "learning_rate": 2.7959977888064484e-05,
+      "loss": 0.0599,
+      "step": 293
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.451043039560318,
+      "learning_rate": 2.7945874295578827e-05,
+      "loss": 0.0676,
+      "step": 294
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.2092967927455902,
+      "learning_rate": 2.7931725701157462e-05,
+      "loss": 0.0653,
+      "step": 295
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.25097522139549255,
+      "learning_rate": 2.7917532153983176e-05,
+      "loss": 0.086,
+      "step": 296
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.1981886476278305,
+      "learning_rate": 2.790329370339501e-05,
+      "loss": 0.0616,
+      "step": 297
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.20687969028949738,
+      "learning_rate": 2.7889010398888104e-05,
+      "loss": 0.0479,
+      "step": 298
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.16657814383506775,
+      "learning_rate": 2.7874682290113514e-05,
+      "loss": 0.0432,
+      "step": 299
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.1697748750448227,
+      "learning_rate": 2.786030942687805e-05,
+      "loss": 0.0624,
+      "step": 300
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.17153337597846985,
+      "learning_rate": 2.7845891859144088e-05,
+      "loss": 0.035,
+      "step": 301
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.20956017076969147,
+      "learning_rate": 2.7831429637029402e-05,
+      "loss": 0.0468,
+      "step": 302
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.1893073171377182,
+      "learning_rate": 2.7816922810807e-05,
+      "loss": 0.0783,
+      "step": 303
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.24053029716014862,
+      "learning_rate": 2.7802371430904936e-05,
+      "loss": 0.0534,
+      "step": 304
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.15913282334804535,
+      "learning_rate": 2.7787775547906142e-05,
+      "loss": 0.0478,
+      "step": 305
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.3303963840007782,
+      "learning_rate": 2.7773135212548247e-05,
+      "loss": 0.0698,
+      "step": 306
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.2255176603794098,
+      "learning_rate": 2.7758450475723405e-05,
+      "loss": 0.0345,
+      "step": 307
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.20534120500087738,
+      "learning_rate": 2.774372138847812e-05,
+      "loss": 0.0668,
+      "step": 308
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.25963813066482544,
+      "learning_rate": 2.7728948002013054e-05,
+      "loss": 0.0741,
+      "step": 309
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2108840048313141,
+      "learning_rate": 2.771413036768288e-05,
+      "loss": 0.0691,
+      "step": 310
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.18770939111709595,
+      "learning_rate": 2.769926853699606e-05,
+      "loss": 0.0701,
+      "step": 311
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.1566968858242035,
+      "learning_rate": 2.7684362561614714e-05,
+      "loss": 0.0196,
+      "step": 312
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.3223302364349365,
+      "learning_rate": 2.766941249335439e-05,
+      "loss": 0.0752,
+      "step": 313
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.21899370849132538,
+      "learning_rate": 2.765441838418393e-05,
+      "loss": 0.0703,
+      "step": 314
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.21244989335536957,
+      "learning_rate": 2.7639380286225264e-05,
+      "loss": 0.0672,
+      "step": 315
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.19558708369731903,
+      "learning_rate": 2.7624298251753232e-05,
+      "loss": 0.042,
+      "step": 316
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.2814396917819977,
+      "learning_rate": 2.7609172333195406e-05,
+      "loss": 0.0434,
+      "step": 317
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.2828751504421234,
+      "learning_rate": 2.75940025831319e-05,
+      "loss": 0.047,
+      "step": 318
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.18190258741378784,
+      "learning_rate": 2.7578789054295202e-05,
+      "loss": 0.0619,
+      "step": 319
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.16573399305343628,
+      "learning_rate": 2.7563531799569984e-05,
+      "loss": 0.0681,
+      "step": 320
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.23267892003059387,
+      "learning_rate": 2.75482308719929e-05,
+      "loss": 0.0518,
+      "step": 321
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.21713323891162872,
+      "learning_rate": 2.753288632475244e-05,
+      "loss": 0.0437,
+      "step": 322
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.26459431648254395,
+      "learning_rate": 2.75174982111887e-05,
+      "loss": 0.0616,
+      "step": 323
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.5927810668945312,
+      "learning_rate": 2.7502066584793243e-05,
+      "loss": 0.0931,
+      "step": 324
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.32426661252975464,
+      "learning_rate": 2.7486591499208867e-05,
+      "loss": 0.08,
+      "step": 325
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.18653956055641174,
+      "learning_rate": 2.7471073008229462e-05,
+      "loss": 0.0679,
+      "step": 326
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.3730272948741913,
+      "learning_rate": 2.7455511165799783e-05,
+      "loss": 0.0399,
+      "step": 327
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.2434634417295456,
+      "learning_rate": 2.7439906026015297e-05,
+      "loss": 0.0608,
+      "step": 328
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.27506130933761597,
+      "learning_rate": 2.742425764312197e-05,
+      "loss": 0.0586,
+      "step": 329
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.2162889540195465,
+      "learning_rate": 2.7408566071516087e-05,
+      "loss": 0.0587,
+      "step": 330
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.2727173864841461,
+      "learning_rate": 2.7392831365744074e-05,
+      "loss": 0.0774,
+      "step": 331
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.2565726339817047,
+      "learning_rate": 2.7377053580502298e-05,
+      "loss": 0.0961,
+      "step": 332
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.2756386995315552,
+      "learning_rate": 2.7361232770636856e-05,
+      "loss": 0.0465,
+      "step": 333
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.30874213576316833,
+      "learning_rate": 2.7345368991143433e-05,
+      "loss": 0.0908,
+      "step": 334
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.21297310292720795,
+      "learning_rate": 2.732946229716707e-05,
+      "loss": 0.0822,
+      "step": 335
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.21286709606647491,
+      "learning_rate": 2.7313512744001982e-05,
+      "loss": 0.0916,
+      "step": 336
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.18058332800865173,
+      "learning_rate": 2.7297520387091376e-05,
+      "loss": 0.0459,
+      "step": 337
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.1574922353029251,
+      "learning_rate": 2.7281485282027252e-05,
+      "loss": 0.048,
+      "step": 338
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.1783793717622757,
+      "learning_rate": 2.7265407484550206e-05,
+      "loss": 0.0733,
+      "step": 339
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.17018386721611023,
+      "learning_rate": 2.724928705054924e-05,
+      "loss": 0.064,
+      "step": 340
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.12130982428789139,
+      "learning_rate": 2.7233124036061575e-05,
+      "loss": 0.0603,
+      "step": 341
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.19916939735412598,
+      "learning_rate": 2.7216918497272426e-05,
+      "loss": 0.0885,
+      "step": 342
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.23629699647426605,
+      "learning_rate": 2.7200670490514865e-05,
+      "loss": 0.0696,
+      "step": 343
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.1870480626821518,
+      "learning_rate": 2.7184380072269558e-05,
+      "loss": 0.0797,
+      "step": 344
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.22087708115577698,
+      "learning_rate": 2.7168047299164614e-05,
+      "loss": 0.0755,
+      "step": 345
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.17880262434482574,
+      "learning_rate": 2.7151672227975377e-05,
+      "loss": 0.0582,
+      "step": 346
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.15626394748687744,
+      "learning_rate": 2.7135254915624213e-05,
+      "loss": 0.072,
+      "step": 347
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.11785972118377686,
+      "learning_rate": 2.711879541918034e-05,
+      "loss": 0.0515,
+      "step": 348
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.13664838671684265,
+      "learning_rate": 2.71022937958596e-05,
+      "loss": 0.0542,
+      "step": 349
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.22781230509281158,
+      "learning_rate": 2.7085750103024296e-05,
+      "loss": 0.1037,
+      "step": 350
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.1862461417913437,
+      "learning_rate": 2.7069164398182948e-05,
+      "loss": 0.0617,
+      "step": 351
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.315283864736557,
+      "learning_rate": 2.7052536738990125e-05,
+      "loss": 0.0756,
+      "step": 352
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.22472089529037476,
+      "learning_rate": 2.7035867183246247e-05,
+      "loss": 0.0792,
+      "step": 353
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.1294025033712387,
+      "learning_rate": 2.7019155788897357e-05,
+      "loss": 0.0738,
+      "step": 354
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.2340669184923172,
+      "learning_rate": 2.700240261403494e-05,
+      "loss": 0.0925,
+      "step": 355
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.27781906723976135,
+      "learning_rate": 2.6985607716895727e-05,
+      "loss": 0.0793,
+      "step": 356
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.18572118878364563,
+      "learning_rate": 2.6968771155861464e-05,
+      "loss": 0.065,
+      "step": 357
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.2246127426624298,
+      "learning_rate": 2.695189298945875e-05,
+      "loss": 0.0825,
+      "step": 358
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.23202070593833923,
+      "learning_rate": 2.6934973276358792e-05,
+      "loss": 0.0598,
+      "step": 359
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.16124321520328522,
+      "learning_rate": 2.6918012075377226e-05,
+      "loss": 0.0825,
+      "step": 360
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.15348608791828156,
+      "learning_rate": 2.6901009445473912e-05,
+      "loss": 0.0442,
+      "step": 361
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.18639832735061646,
+      "learning_rate": 2.6883965445752714e-05,
+      "loss": 0.0699,
+      "step": 362
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.2920565605163574,
+      "learning_rate": 2.6866880135461314e-05,
+      "loss": 0.0798,
+      "step": 363
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.16075679659843445,
+      "learning_rate": 2.684975357399099e-05,
+      "loss": 0.0691,
+      "step": 364
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.17678283154964447,
+      "learning_rate": 2.683258582087641e-05,
+      "loss": 0.0386,
+      "step": 365
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.4795132875442505,
+      "learning_rate": 2.681537693579545e-05,
+      "loss": 0.0988,
+      "step": 366
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.18269479274749756,
+      "learning_rate": 2.6798126978568942e-05,
+      "loss": 0.074,
+      "step": 367
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.2764766812324524,
+      "learning_rate": 2.6780836009160514e-05,
+      "loss": 0.0565,
+      "step": 368
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.19089283049106598,
+      "learning_rate": 2.6763504087676346e-05,
+      "loss": 0.0389,
+      "step": 369
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.15806300938129425,
+      "learning_rate": 2.674613127436498e-05,
+      "loss": 0.0482,
+      "step": 370
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.2809430658817291,
+      "learning_rate": 2.6728717629617093e-05,
+      "loss": 0.0429,
+      "step": 371
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.12930545210838318,
+      "learning_rate": 2.671126321396532e-05,
+      "loss": 0.0255,
+      "step": 372
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.17408722639083862,
+      "learning_rate": 2.6693768088083994e-05,
+      "loss": 0.0703,
+      "step": 373
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.3387722671031952,
+      "learning_rate": 2.6676232312788998e-05,
+      "loss": 0.0569,
+      "step": 374
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.25352975726127625,
+      "learning_rate": 2.6658655949037482e-05,
+      "loss": 0.0914,
+      "step": 375
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.3007981777191162,
+      "learning_rate": 2.6641039057927724e-05,
+      "loss": 0.0502,
+      "step": 376
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.20817789435386658,
+      "learning_rate": 2.662338170069885e-05,
+      "loss": 0.0408,
+      "step": 377
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.15569907426834106,
+      "learning_rate": 2.6605683938730666e-05,
+      "loss": 0.029,
+      "step": 378
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.25259971618652344,
+      "learning_rate": 2.6587945833543432e-05,
+      "loss": 0.0336,
+      "step": 379
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.13589264452457428,
+      "learning_rate": 2.6570167446797657e-05,
+      "loss": 0.0218,
+      "step": 380
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.34168577194213867,
+      "learning_rate": 2.6552348840293856e-05,
+      "loss": 0.0882,
+      "step": 381
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.26590269804000854,
+      "learning_rate": 2.6534490075972368e-05,
+      "loss": 0.0576,
+      "step": 382
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.2801959812641144,
+      "learning_rate": 2.6516591215913118e-05,
+      "loss": 0.0586,
+      "step": 383
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.3427751660346985,
+      "learning_rate": 2.6498652322335416e-05,
+      "loss": 0.0575,
+      "step": 384
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.5352123975753784,
+      "learning_rate": 2.6480673457597737e-05,
+      "loss": 0.0915,
+      "step": 385
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.45933809876441956,
+      "learning_rate": 2.646265468419749e-05,
+      "loss": 0.0865,
+      "step": 386
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.14740170538425446,
+      "learning_rate": 2.6444596064770837e-05,
+      "loss": 0.0447,
+      "step": 387
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.18046963214874268,
+      "learning_rate": 2.6426497662092424e-05,
+      "loss": 0.0302,
+      "step": 388
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.2271748036146164,
+      "learning_rate": 2.6408359539075204e-05,
+      "loss": 0.067,
+      "step": 389
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.345716267824173,
+      "learning_rate": 2.6390181758770208e-05,
+      "loss": 0.0461,
+      "step": 390
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.2197301983833313,
+      "learning_rate": 2.6371964384366305e-05,
+      "loss": 0.0252,
+      "step": 391
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.2628111243247986,
+      "learning_rate": 2.6353707479190022e-05,
+      "loss": 0.0616,
+      "step": 392
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.41908586025238037,
+      "learning_rate": 2.6335411106705283e-05,
+      "loss": 0.0599,
+      "step": 393
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.13220524787902832,
+      "learning_rate": 2.6317075330513212e-05,
+      "loss": 0.0116,
+      "step": 394
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.21840068697929382,
+      "learning_rate": 2.6298700214351922e-05,
+      "loss": 0.0413,
+      "step": 395
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.3715525269508362,
+      "learning_rate": 2.628028582209625e-05,
+      "loss": 0.0643,
+      "step": 396
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.21229352056980133,
+      "learning_rate": 2.626183221775758e-05,
+      "loss": 0.0369,
+      "step": 397
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.3171239495277405,
+      "learning_rate": 2.6243339465483605e-05,
+      "loss": 0.0712,
+      "step": 398
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.33259809017181396,
+      "learning_rate": 2.6224807629558094e-05,
+      "loss": 0.1066,
+      "step": 399
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.14465388655662537,
+      "learning_rate": 2.6206236774400684e-05,
+      "loss": 0.034,
+      "step": 400
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.5533297061920166,
+      "learning_rate": 2.6187626964566644e-05,
+      "loss": 0.1046,
+      "step": 401
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.17661869525909424,
+      "learning_rate": 2.6168978264746663e-05,
+      "loss": 0.0322,
+      "step": 402
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.19023488461971283,
+      "learning_rate": 2.615029073976661e-05,
+      "loss": 0.0493,
+      "step": 403
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.1495324969291687,
+      "learning_rate": 2.6131564454587316e-05,
+      "loss": 0.0404,
+      "step": 404
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.20712687075138092,
+      "learning_rate": 2.611279947430436e-05,
+      "loss": 0.0472,
+      "step": 405
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.233810156583786,
+      "learning_rate": 2.609399586414782e-05,
+      "loss": 0.0424,
+      "step": 406
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.14824284613132477,
+      "learning_rate": 2.607515368948206e-05,
+      "loss": 0.0318,
+      "step": 407
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.3361953794956207,
+      "learning_rate": 2.60562730158055e-05,
+      "loss": 0.0763,
+      "step": 408
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.2170048952102661,
+      "learning_rate": 2.6037353908750394e-05,
+      "loss": 0.0643,
+      "step": 409
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.18919886648654938,
+      "learning_rate": 2.601839643408259e-05,
+      "loss": 0.0583,
+      "step": 410
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.42277276515960693,
+      "learning_rate": 2.5999400657701314e-05,
+      "loss": 0.0698,
+      "step": 411
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.25986483693122864,
+      "learning_rate": 2.598036664563893e-05,
+      "loss": 0.041,
+      "step": 412
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2869338095188141,
+      "learning_rate": 2.596129446406072e-05,
+      "loss": 0.0503,
+      "step": 413
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.41842973232269287,
+      "learning_rate": 2.594218417926464e-05,
+      "loss": 0.0786,
+      "step": 414
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.4030296504497528,
+      "learning_rate": 2.592303585768111e-05,
+      "loss": 0.067,
+      "step": 415
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.31492769718170166,
+      "learning_rate": 2.590384956587277e-05,
+      "loss": 0.0479,
+      "step": 416
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.5827561020851135,
+      "learning_rate": 2.5884625370534242e-05,
+      "loss": 0.1315,
+      "step": 417
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.30939584970474243,
+      "learning_rate": 2.5865363338491916e-05,
+      "loss": 0.0759,
+      "step": 418
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.20597557723522186,
+      "learning_rate": 2.5846063536703706e-05,
+      "loss": 0.0739,
+      "step": 419
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.31797170639038086,
+      "learning_rate": 2.582672603225882e-05,
+      "loss": 0.0892,
+      "step": 420
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.4155197739601135,
+      "learning_rate": 2.5807350892377517e-05,
+      "loss": 0.13,
+      "step": 421
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.14263832569122314,
+      "learning_rate": 2.5787938184410902e-05,
+      "loss": 0.0489,
+      "step": 422
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.1489766389131546,
+      "learning_rate": 2.5768487975840655e-05,
+      "loss": 0.0638,
+      "step": 423
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.30038881301879883,
+      "learning_rate": 2.5749000334278825e-05,
+      "loss": 0.0455,
+      "step": 424
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.2359144389629364,
+      "learning_rate": 2.572947532746758e-05,
+      "loss": 0.0822,
+      "step": 425
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.16958381235599518,
+      "learning_rate": 2.570991302327897e-05,
+      "loss": 0.0454,
+      "step": 426
+    },
+    {
+      "epoch": 0.75,
+      "eval_loss": 0.0626850351691246,
+      "eval_runtime": 14.7469,
+      "eval_samples_per_second": 32.278,
+      "eval_steps_per_second": 8.07,
+      "step": 426
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.40241333842277527,
+      "learning_rate": 2.569031348971471e-05,
+      "loss": 0.0926,
+      "step": 427
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.2126995176076889,
+      "learning_rate": 2.5670676794905915e-05,
+      "loss": 0.085,
+      "step": 428
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.1960797905921936,
+      "learning_rate": 2.5651003007112892e-05,
+      "loss": 0.0853,
+      "step": 429
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.13784633576869965,
+      "learning_rate": 2.5631292194724883e-05,
+      "loss": 0.0453,
+      "step": 430
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.14578142762184143,
+      "learning_rate": 2.561154442625983e-05,
+      "loss": 0.0765,
+      "step": 431
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.14827878773212433,
+      "learning_rate": 2.559175977036415e-05,
+      "loss": 0.0594,
+      "step": 432
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.19840463995933533,
+      "learning_rate": 2.5571938295812476e-05,
+      "loss": 0.0692,
+      "step": 433
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.11538412421941757,
+      "learning_rate": 2.555208007150743e-05,
+      "loss": 0.067,
+      "step": 434
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.3484029173851013,
+      "learning_rate": 2.553218516647939e-05,
+      "loss": 0.0799,
+      "step": 435
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.20653250813484192,
+      "learning_rate": 2.5512253649886237e-05,
+      "loss": 0.0741,
+      "step": 436
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.18202582001686096,
+      "learning_rate": 2.5492285591013118e-05,
+      "loss": 0.0636,
+      "step": 437
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.17838889360427856,
+      "learning_rate": 2.5472281059272213e-05,
+      "loss": 0.067,
+      "step": 438
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.2804567217826843,
+      "learning_rate": 2.5452240124202477e-05,
+      "loss": 0.0795,
+      "step": 439
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.1692652553319931,
+      "learning_rate": 2.5432162855469422e-05,
+      "loss": 0.0722,
+      "step": 440
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.1247120052576065,
+      "learning_rate": 2.5412049322864847e-05,
+      "loss": 0.0332,
+      "step": 441
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.1898394376039505,
+      "learning_rate": 2.539189959630662e-05,
+      "loss": 0.055,
+      "step": 442
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.20923753082752228,
+      "learning_rate": 2.537171374583843e-05,
+      "loss": 0.0728,
+      "step": 443
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.1551828235387802,
+      "learning_rate": 2.535149184162952e-05,
+      "loss": 0.0629,
+      "step": 444
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.1762448400259018,
+      "learning_rate": 2.533123395397448e-05,
+      "loss": 0.0673,
+      "step": 445
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.12544862926006317,
+      "learning_rate": 2.5310940153292978e-05,
+      "loss": 0.0311,
+      "step": 446
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.15880614519119263,
+      "learning_rate": 2.5290610510129518e-05,
+      "loss": 0.0402,
+      "step": 447
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.17256905138492584,
+      "learning_rate": 2.5270245095153198e-05,
+      "loss": 0.0633,
+      "step": 448
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.1660386472940445,
+      "learning_rate": 2.524984397915747e-05,
+      "loss": 0.077,
+      "step": 449
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.13268114626407623,
+      "learning_rate": 2.5229407233059887e-05,
+      "loss": 0.0426,
+      "step": 450
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.2337772697210312,
+      "learning_rate": 2.5208934927901857e-05,
+      "loss": 0.0836,
+      "step": 451
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.13446028530597687,
+      "learning_rate": 2.5188427134848395e-05,
+      "loss": 0.0382,
+      "step": 452
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.18553531169891357,
+      "learning_rate": 2.5167883925187878e-05,
+      "loss": 0.0548,
+      "step": 453
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.25128498673439026,
+      "learning_rate": 2.51473053703318e-05,
+      "loss": 0.0766,
+      "step": 454
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.1686338186264038,
+      "learning_rate": 2.5126691541814518e-05,
+      "loss": 0.0434,
+      "step": 455
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.18022535741329193,
+      "learning_rate": 2.510604251129301e-05,
+      "loss": 0.0688,
+      "step": 456
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.17681004106998444,
+      "learning_rate": 2.5085358350546612e-05,
+      "loss": 0.0533,
+      "step": 457
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.2677585482597351,
+      "learning_rate": 2.506463913147679e-05,
+      "loss": 0.0818,
+      "step": 458
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.5509142875671387,
+      "learning_rate": 2.5043884926106873e-05,
+      "loss": 0.1005,
+      "step": 459
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.14625947177410126,
+      "learning_rate": 2.5023095806581802e-05,
+      "loss": 0.0426,
+      "step": 460
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.3046765923500061,
+      "learning_rate": 2.5002271845167896e-05,
+      "loss": 0.0515,
+      "step": 461
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.16383983194828033,
+      "learning_rate": 2.4981413114252588e-05,
+      "loss": 0.0455,
+      "step": 462
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.3099363148212433,
+      "learning_rate": 2.4960519686344168e-05,
+      "loss": 0.0924,
+      "step": 463
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.25622445344924927,
+      "learning_rate": 2.4939591634071544e-05,
+      "loss": 0.0737,
+      "step": 464
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.21525193750858307,
+      "learning_rate": 2.491862903018398e-05,
+      "loss": 0.0816,
+      "step": 465
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.18544378876686096,
+      "learning_rate": 2.4897631947550857e-05,
+      "loss": 0.0453,
+      "step": 466
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.15406207740306854,
+      "learning_rate": 2.4876600459161397e-05,
+      "loss": 0.0477,
+      "step": 467
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.21517229080200195,
+      "learning_rate": 2.4855534638124427e-05,
+      "loss": 0.0467,
+      "step": 468
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.27017998695373535,
+      "learning_rate": 2.4834434557668126e-05,
+      "loss": 0.0593,
+      "step": 469
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.24762241542339325,
+      "learning_rate": 2.4813300291139754e-05,
+      "loss": 0.0587,
+      "step": 470
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.29878756403923035,
+      "learning_rate": 2.4792131912005407e-05,
+      "loss": 0.0635,
+      "step": 471
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.18735036253929138,
+      "learning_rate": 2.4770929493849773e-05,
+      "loss": 0.0312,
+      "step": 472
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.27927571535110474,
+      "learning_rate": 2.4749693110375856e-05,
+      "loss": 0.0668,
+      "step": 473
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.19749711453914642,
+      "learning_rate": 2.4728422835404735e-05,
+      "loss": 0.0321,
+      "step": 474
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.12730471789836884,
+      "learning_rate": 2.4707118742875293e-05,
+      "loss": 0.0183,
+      "step": 475
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.2711328864097595,
+      "learning_rate": 2.4685780906843975e-05,
+      "loss": 0.0677,
+      "step": 476
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.26742562651634216,
+      "learning_rate": 2.4664409401484522e-05,
+      "loss": 0.0617,
+      "step": 477
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.5431042313575745,
+      "learning_rate": 2.4643004301087716e-05,
+      "loss": 0.1233,
+      "step": 478
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.13498014211654663,
+      "learning_rate": 2.462156568006112e-05,
+      "loss": 0.0268,
+      "step": 479
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.22139033675193787,
+      "learning_rate": 2.4600093612928812e-05,
+      "loss": 0.0493,
+      "step": 480
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.15270984172821045,
+      "learning_rate": 2.457858817433115e-05,
+      "loss": 0.0255,
+      "step": 481
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.158059224486351,
+      "learning_rate": 2.4557049439024488e-05,
+      "loss": 0.0574,
+      "step": 482
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.17642827332019806,
+      "learning_rate": 2.4535477481880923e-05,
+      "loss": 0.0351,
+      "step": 483
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.1525094211101532,
+      "learning_rate": 2.451387237788804e-05,
+      "loss": 0.0242,
+      "step": 484
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.21704228222370148,
+      "learning_rate": 2.4492234202148643e-05,
+      "loss": 0.0582,
+      "step": 485
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.2657334804534912,
+      "learning_rate": 2.44705630298805e-05,
+      "loss": 0.0458,
+      "step": 486
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.23600801825523376,
+      "learning_rate": 2.4448858936416093e-05,
+      "loss": 0.0447,
+      "step": 487
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.20092691481113434,
+      "learning_rate": 2.442712199720232e-05,
+      "loss": 0.0462,
+      "step": 488
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.23618610203266144,
+      "learning_rate": 2.4405352287800268e-05,
+      "loss": 0.0383,
+      "step": 489
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.18307435512542725,
+      "learning_rate": 2.4383549883884954e-05,
+      "loss": 0.0312,
+      "step": 490
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.20053623616695404,
+      "learning_rate": 2.4361714861245017e-05,
+      "loss": 0.0352,
+      "step": 491
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.23183472454547882,
+      "learning_rate": 2.4339847295782508e-05,
+      "loss": 0.0445,
+      "step": 492
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.554999589920044,
+      "learning_rate": 2.431794726351258e-05,
+      "loss": 0.1143,
+      "step": 493
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.3061026334762573,
+      "learning_rate": 2.4296014840563266e-05,
+      "loss": 0.0488,
+      "step": 494
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.23472747206687927,
+      "learning_rate": 2.4274050103175195e-05,
+      "loss": 0.0549,
+      "step": 495
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.36875736713409424,
+      "learning_rate": 2.42520531277013e-05,
+      "loss": 0.0472,
+      "step": 496
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.3055475950241089,
+      "learning_rate": 2.423002399060661e-05,
+      "loss": 0.0485,
+      "step": 497
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.33722659945487976,
+      "learning_rate": 2.420796276846793e-05,
+      "loss": 0.1004,
+      "step": 498
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.3110818862915039,
+      "learning_rate": 2.4185869537973613e-05,
+      "loss": 0.0893,
+      "step": 499
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.5243550539016724,
+      "learning_rate": 2.4163744375923272e-05,
+      "loss": 0.0742,
+      "step": 500
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.18637774884700775,
+      "learning_rate": 2.4141587359227514e-05,
+      "loss": 0.0453,
+      "step": 501
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.11755359172821045,
+      "learning_rate": 2.4119398564907685e-05,
+      "loss": 0.0236,
+      "step": 502
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.145331472158432,
+      "learning_rate": 2.4097178070095602e-05,
+      "loss": 0.0432,
+      "step": 503
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.1340416669845581,
+      "learning_rate": 2.4074925952033263e-05,
+      "loss": 0.0339,
+      "step": 504
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.2009749859571457,
+      "learning_rate": 2.4052642288072594e-05,
+      "loss": 0.033,
+      "step": 505
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.25946885347366333,
+      "learning_rate": 2.4030327155675192e-05,
+      "loss": 0.0618,
+      "step": 506
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.338084876537323,
+      "learning_rate": 2.4007980632412034e-05,
+      "loss": 0.076,
+      "step": 507
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.16105656325817108,
+      "learning_rate": 2.398560279596323e-05,
+      "loss": 0.055,
+      "step": 508
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.2381068766117096,
+      "learning_rate": 2.3963193724117715e-05,
+      "loss": 0.0847,
+      "step": 509
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.148463174700737,
+      "learning_rate": 2.394075349477302e-05,
+      "loss": 0.0404,
+      "step": 510
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.359811007976532,
+      "learning_rate": 2.3918282185934986e-05,
+      "loss": 0.0508,
+      "step": 511
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.2325233817100525,
+      "learning_rate": 2.3895779875717486e-05,
+      "loss": 0.0594,
+      "step": 512
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.18048043549060822,
+      "learning_rate": 2.3873246642342163e-05,
+      "loss": 0.0267,
+      "step": 513
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.14463412761688232,
+      "learning_rate": 2.3850682564138145e-05,
+      "loss": 0.0519,
+      "step": 514
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.17916260659694672,
+      "learning_rate": 2.3828087719541787e-05,
+      "loss": 0.0535,
+      "step": 515
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.2539409399032593,
+      "learning_rate": 2.3805462187096402e-05,
+      "loss": 0.0579,
+      "step": 516
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.20096299052238464,
+      "learning_rate": 2.3782806045451963e-05,
+      "loss": 0.0581,
+      "step": 517
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.15366317331790924,
+      "learning_rate": 2.376011937336485e-05,
+      "loss": 0.0392,
+      "step": 518
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.28773725032806396,
+      "learning_rate": 2.373740224969758e-05,
+      "loss": 0.1054,
+      "step": 519
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.18706980347633362,
+      "learning_rate": 2.371465475341852e-05,
+      "loss": 0.0422,
+      "step": 520
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.263085275888443,
+      "learning_rate": 2.369187696360161e-05,
+      "loss": 0.0533,
+      "step": 521
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.35978153347969055,
+      "learning_rate": 2.3669068959426107e-05,
+      "loss": 0.0871,
+      "step": 522
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.3314432203769684,
+      "learning_rate": 2.364623082017629e-05,
+      "loss": 0.0481,
+      "step": 523
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.19833628833293915,
+      "learning_rate": 2.3623362625241193e-05,
+      "loss": 0.0563,
+      "step": 524
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.17452682554721832,
+      "learning_rate": 2.3600464454114326e-05,
+      "loss": 0.0464,
+      "step": 525
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.24458369612693787,
+      "learning_rate": 2.3577536386393416e-05,
+      "loss": 0.0722,
+      "step": 526
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.309350848197937,
+      "learning_rate": 2.35545785017801e-05,
+      "loss": 0.0794,
+      "step": 527
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.42290210723876953,
+      "learning_rate": 2.3531590880079663e-05,
+      "loss": 0.0753,
+      "step": 528
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.1993589699268341,
+      "learning_rate": 2.3508573601200767e-05,
+      "loss": 0.0589,
+      "step": 529
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.1793728619813919,
+      "learning_rate": 2.348552674515517e-05,
+      "loss": 0.0587,
+      "step": 530
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.1571635901927948,
+      "learning_rate": 2.3462450392057437e-05,
+      "loss": 0.0427,
+      "step": 531
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.2197573184967041,
+      "learning_rate": 2.343934462212467e-05,
+      "loss": 0.0569,
+      "step": 532
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.1468045711517334,
+      "learning_rate": 2.341620951567624e-05,
+      "loss": 0.0501,
+      "step": 533
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.1460326462984085,
+      "learning_rate": 2.339304515313348e-05,
+      "loss": 0.0446,
+      "step": 534
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.20360475778579712,
+      "learning_rate": 2.3369851615019436e-05,
+      "loss": 0.056,
+      "step": 535
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.17234598100185394,
+      "learning_rate": 2.3346628981958565e-05,
+      "loss": 0.0493,
+      "step": 536
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.21084345877170563,
+      "learning_rate": 2.332337733467646e-05,
+      "loss": 0.0781,
+      "step": 537
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.15120622515678406,
+      "learning_rate": 2.3300096753999585e-05,
+      "loss": 0.0508,
+      "step": 538
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.19097232818603516,
+      "learning_rate": 2.3276787320854967e-05,
+      "loss": 0.0558,
+      "step": 539
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.14518237113952637,
+      "learning_rate": 2.3253449116269937e-05,
+      "loss": 0.0363,
+      "step": 540
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.38005223870277405,
+      "learning_rate": 2.3230082221371834e-05,
+      "loss": 0.1022,
+      "step": 541
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.15401460230350494,
+      "learning_rate": 2.3206686717387742e-05,
+      "loss": 0.0745,
+      "step": 542
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.15603908896446228,
+      "learning_rate": 2.3183262685644177e-05,
+      "loss": 0.035,
+      "step": 543
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.17525097727775574,
+      "learning_rate": 2.3159810207566832e-05,
+      "loss": 0.0584,
+      "step": 544
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.17069534957408905,
+      "learning_rate": 2.3136329364680288e-05,
+      "loss": 0.0569,
+      "step": 545
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.3221694231033325,
+      "learning_rate": 2.3112820238607716e-05,
+      "loss": 0.0878,
+      "step": 546
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.2566680908203125,
+      "learning_rate": 2.3089282911070613e-05,
+      "loss": 0.0394,
+      "step": 547
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.17850065231323242,
+      "learning_rate": 2.3065717463888505e-05,
+      "loss": 0.0351,
+      "step": 548
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.25362351536750793,
+      "learning_rate": 2.3042123978978665e-05,
+      "loss": 0.0596,
+      "step": 549
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.20955722033977509,
+      "learning_rate": 2.3018502538355827e-05,
+      "loss": 0.0534,
+      "step": 550
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.19723302125930786,
+      "learning_rate": 2.2994853224131915e-05,
+      "loss": 0.0674,
+      "step": 551
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2044544219970703,
+      "learning_rate": 2.2971176118515734e-05,
+      "loss": 0.0505,
+      "step": 552
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2544882893562317,
+      "learning_rate": 2.2947471303812708e-05,
+      "loss": 0.0967,
+      "step": 553
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.16680030524730682,
+      "learning_rate": 2.2923738862424565e-05,
+      "loss": 0.0395,
+      "step": 554
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.27806153893470764,
+      "learning_rate": 2.2899978876849084e-05,
+      "loss": 0.0788,
+      "step": 555
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.1830155849456787,
+      "learning_rate": 2.287619142967979e-05,
+      "loss": 0.0538,
+      "step": 556
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.33155858516693115,
+      "learning_rate": 2.285237660360566e-05,
+      "loss": 0.0417,
+      "step": 557
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.2150736302137375,
+      "learning_rate": 2.2828534481410847e-05,
+      "loss": 0.0396,
+      "step": 558
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.2404957413673401,
+      "learning_rate": 2.28046651459744e-05,
+      "loss": 0.0554,
+      "step": 559
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.22239123284816742,
+      "learning_rate": 2.278076868026995e-05,
+      "loss": 0.0592,
+      "step": 560
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.24049235880374908,
+      "learning_rate": 2.2756845167365452e-05,
+      "loss": 0.0462,
+      "step": 561
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.24881111085414886,
+      "learning_rate": 2.273289469042287e-05,
+      "loss": 0.0483,
+      "step": 562
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.5261463522911072,
+      "learning_rate": 2.2708917332697908e-05,
+      "loss": 0.0748,
+      "step": 563
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.3064204752445221,
+      "learning_rate": 2.26849131775397e-05,
+      "loss": 0.0532,
+      "step": 564
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.6424417495727539,
+      "learning_rate": 2.2660882308390547e-05,
+      "loss": 0.1566,
+      "step": 565
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.19905412197113037,
+      "learning_rate": 2.2636824808785604e-05,
+      "loss": 0.0456,
+      "step": 566
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.25375813245773315,
+      "learning_rate": 2.2612740762352594e-05,
+      "loss": 0.0548,
+      "step": 567
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.17399847507476807,
+      "learning_rate": 2.258863025281153e-05,
+      "loss": 0.0263,
+      "step": 568
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 0.05472303926944733,
+      "eval_runtime": 14.6946,
+      "eval_samples_per_second": 32.393,
+      "eval_steps_per_second": 8.098,
+      "step": 568
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.27021336555480957,
+      "learning_rate": 2.2564493363974416e-05,
+      "loss": 0.1,
+      "step": 569
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.20826458930969238,
+      "learning_rate": 2.2540330179744935e-05,
+      "loss": 0.0422,
+      "step": 570
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.20264066755771637,
+      "learning_rate": 2.2516140784118202e-05,
+      "loss": 0.0462,
+      "step": 571
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.16754667460918427,
+      "learning_rate": 2.2491925261180433e-05,
+      "loss": 0.0745,
+      "step": 572
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.2399885356426239,
+      "learning_rate": 2.2467683695108676e-05,
+      "loss": 0.071,
+      "step": 573
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.18665122985839844,
+      "learning_rate": 2.244341617017049e-05,
+      "loss": 0.0673,
+      "step": 574
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.2963515520095825,
+      "learning_rate": 2.24191227707237e-05,
+      "loss": 0.0486,
+      "step": 575
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.24179871380329132,
+      "learning_rate": 2.2394803581216047e-05,
+      "loss": 0.0484,
+      "step": 576
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.22649294137954712,
+      "learning_rate": 2.2370458686184945e-05,
+      "loss": 0.0433,
+      "step": 577
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.1680031716823578,
+      "learning_rate": 2.234608817025715e-05,
+      "loss": 0.0573,
+      "step": 578
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.12036772817373276,
+      "learning_rate": 2.232169211814848e-05,
+      "loss": 0.0359,
+      "step": 579
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.11121578514575958,
+      "learning_rate": 2.2297270614663535e-05,
+      "loss": 0.0275,
+      "step": 580
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.2029242366552353,
+      "learning_rate": 2.2272823744695366e-05,
+      "loss": 0.0379,
+      "step": 581
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.15203304588794708,
+      "learning_rate": 2.2248351593225225e-05,
+      "loss": 0.0416,
+      "step": 582
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.1620362251996994,
+      "learning_rate": 2.2223854245322236e-05,
+      "loss": 0.0491,
+      "step": 583
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.22791963815689087,
+      "learning_rate": 2.21993317861431e-05,
+      "loss": 0.066,
+      "step": 584
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.13969655334949493,
+      "learning_rate": 2.217478430093183e-05,
+      "loss": 0.0316,
+      "step": 585
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.1673637479543686,
+      "learning_rate": 2.2150211875019417e-05,
+      "loss": 0.0444,
+      "step": 586
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.26466965675354004,
+      "learning_rate": 2.2125614593823556e-05,
+      "loss": 0.0603,
+      "step": 587
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.25127851963043213,
+      "learning_rate": 2.2100992542848344e-05,
+      "loss": 0.0653,
+      "step": 588
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.14236998558044434,
+      "learning_rate": 2.2076345807683975e-05,
+      "loss": 0.0347,
+      "step": 589
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.2006857842206955,
+      "learning_rate": 2.2051674474006457e-05,
+      "loss": 0.0261,
+      "step": 590
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.3477920591831207,
+      "learning_rate": 2.2026978627577303e-05,
+      "loss": 0.0552,
+      "step": 591
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.23367953300476074,
+      "learning_rate": 2.2002258354243242e-05,
+      "loss": 0.0583,
+      "step": 592
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.2674083411693573,
+      "learning_rate": 2.1977513739935897e-05,
+      "loss": 0.0552,
+      "step": 593
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.16168631613254547,
+      "learning_rate": 2.195274487067153e-05,
+      "loss": 0.0292,
+      "step": 594
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.24982483685016632,
+      "learning_rate": 2.1927951832550696e-05,
+      "loss": 0.0487,
+      "step": 595
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.3051944673061371,
+      "learning_rate": 2.1903134711757974e-05,
+      "loss": 0.051,
+      "step": 596
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.40649136900901794,
+      "learning_rate": 2.187829359456166e-05,
+      "loss": 0.0709,
+      "step": 597
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.203045055270195,
+      "learning_rate": 2.1853428567313455e-05,
+      "loss": 0.0426,
+      "step": 598
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.2079191356897354,
+      "learning_rate": 2.1828539716448188e-05,
+      "loss": 0.0362,
+      "step": 599
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.3522377610206604,
+      "learning_rate": 2.1803627128483497e-05,
+      "loss": 0.0828,
+      "step": 600
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.2527177631855011,
+      "learning_rate": 2.1778690890019535e-05,
+      "loss": 0.0512,
+      "step": 601
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.239670068025589,
+      "learning_rate": 2.1753731087738657e-05,
+      "loss": 0.058,
+      "step": 602
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.28101757168769836,
+      "learning_rate": 2.1728747808405154e-05,
+      "loss": 0.0474,
+      "step": 603
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.3522912561893463,
+      "learning_rate": 2.1703741138864908e-05,
+      "loss": 0.0867,
+      "step": 604
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.20509746670722961,
+      "learning_rate": 2.167871116604511e-05,
+      "loss": 0.0452,
+      "step": 605
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.25338292121887207,
+      "learning_rate": 2.1653657976953956e-05,
+      "loss": 0.0271,
+      "step": 606
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.30135923624038696,
+      "learning_rate": 2.1628581658680358e-05,
+      "loss": 0.0834,
+      "step": 607
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.4553530216217041,
+      "learning_rate": 2.1603482298393622e-05,
+      "loss": 0.0674,
+      "step": 608
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.5035529136657715,
+      "learning_rate": 2.157835998334315e-05,
+      "loss": 0.0693,
+      "step": 609
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.24384114146232605,
+      "learning_rate": 2.155321480085813e-05,
+      "loss": 0.0588,
+      "step": 610
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.19528117775917053,
+      "learning_rate": 2.1528046838347262e-05,
+      "loss": 0.0257,
+      "step": 611
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.36607757210731506,
+      "learning_rate": 2.1502856183298414e-05,
+      "loss": 0.0793,
+      "step": 612
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.17997542023658752,
+      "learning_rate": 2.147764292327834e-05,
+      "loss": 0.0364,
+      "step": 613
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.2820572257041931,
+      "learning_rate": 2.145240714593239e-05,
+      "loss": 0.047,
+      "step": 614
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.2257036566734314,
+      "learning_rate": 2.142714893898416e-05,
+      "loss": 0.0257,
+      "step": 615
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.1840316653251648,
+      "learning_rate": 2.1401868390235236e-05,
+      "loss": 0.0387,
+      "step": 616
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.1750848889350891,
+      "learning_rate": 2.137656558756486e-05,
+      "loss": 0.0542,
+      "step": 617
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.2306337058544159,
+      "learning_rate": 2.1351240618929633e-05,
+      "loss": 0.0536,
+      "step": 618
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.5436125993728638,
+      "learning_rate": 2.1325893572363205e-05,
+      "loss": 0.1015,
+      "step": 619
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.22225777804851532,
+      "learning_rate": 2.130052453597598e-05,
+      "loss": 0.058,
+      "step": 620
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.19612561166286469,
+      "learning_rate": 2.1275133597954794e-05,
+      "loss": 0.0601,
+      "step": 621
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.19745659828186035,
+      "learning_rate": 2.1249720846562626e-05,
+      "loss": 0.0339,
+      "step": 622
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.22013776004314423,
+      "learning_rate": 2.122428637013827e-05,
+      "loss": 0.0381,
+      "step": 623
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.22055767476558685,
+      "learning_rate": 2.1198830257096055e-05,
+      "loss": 0.0329,
+      "step": 624
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.19469846785068512,
+      "learning_rate": 2.1173352595925505e-05,
+      "loss": 0.0514,
+      "step": 625
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.29255107045173645,
+      "learning_rate": 2.114785347519106e-05,
+      "loss": 0.0756,
+      "step": 626
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.19103175401687622,
+      "learning_rate": 2.112233298353175e-05,
+      "loss": 0.034,
+      "step": 627
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.16312550008296967,
+      "learning_rate": 2.1096791209660905e-05,
+      "loss": 0.0389,
+      "step": 628
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.19477924704551697,
+      "learning_rate": 2.1071228242365826e-05,
+      "loss": 0.0457,
+      "step": 629
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.21488840878009796,
+      "learning_rate": 2.1045644170507486e-05,
+      "loss": 0.0692,
+      "step": 630
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.24809981882572174,
+      "learning_rate": 2.1020039083020222e-05,
+      "loss": 0.0667,
+      "step": 631
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.13645565509796143,
+      "learning_rate": 2.0994413068911425e-05,
+      "loss": 0.0182,
+      "step": 632
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.2085627168416977,
+      "learning_rate": 2.0968766217261236e-05,
+      "loss": 0.0431,
+      "step": 633
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.49600788950920105,
+      "learning_rate": 2.0943098617222216e-05,
+      "loss": 0.0863,
+      "step": 634
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.16732609272003174,
+      "learning_rate": 2.0917410358019073e-05,
+      "loss": 0.0399,
+      "step": 635
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.4573953449726105,
+      "learning_rate": 2.0891701528948302e-05,
+      "loss": 0.0757,
+      "step": 636
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.21385256946086884,
+      "learning_rate": 2.0865972219377935e-05,
+      "loss": 0.0514,
+      "step": 637
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.3950164318084717,
+      "learning_rate": 2.0840222518747162e-05,
+      "loss": 0.0611,
+      "step": 638
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.1962326318025589,
+      "learning_rate": 2.081445251656608e-05,
+      "loss": 0.0315,
+      "step": 639
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.2456423044204712,
+      "learning_rate": 2.0788662302415352e-05,
+      "loss": 0.0497,
+      "step": 640
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.13714934885501862,
+      "learning_rate": 2.0762851965945894e-05,
+      "loss": 0.0224,
+      "step": 641
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.2559102475643158,
+      "learning_rate": 2.073702159687858e-05,
+      "loss": 0.074,
+      "step": 642
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.3649904429912567,
+      "learning_rate": 2.0711171285003917e-05,
+      "loss": 0.0573,
+      "step": 643
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.4410479962825775,
+      "learning_rate": 2.0685301120181737e-05,
+      "loss": 0.0913,
+      "step": 644
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.15331228077411652,
+      "learning_rate": 2.0659411192340875e-05,
+      "loss": 0.0458,
+      "step": 645
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.19452711939811707,
+      "learning_rate": 2.0633501591478886e-05,
+      "loss": 0.0412,
+      "step": 646
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.18934063613414764,
+      "learning_rate": 2.0607572407661683e-05,
+      "loss": 0.058,
+      "step": 647
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.09978535771369934,
+      "learning_rate": 2.058162373102328e-05,
+      "loss": 0.0166,
+      "step": 648
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.2150053083896637,
+      "learning_rate": 2.0555655651765436e-05,
+      "loss": 0.0641,
+      "step": 649
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.30375051498413086,
+      "learning_rate": 2.0529668260157354e-05,
+      "loss": 0.0382,
+      "step": 650
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.18037180602550507,
+      "learning_rate": 2.050366164653539e-05,
+      "loss": 0.0408,
+      "step": 651
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.14954544603824615,
+      "learning_rate": 2.047763590130269e-05,
+      "loss": 0.0374,
+      "step": 652
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.1657114028930664,
+      "learning_rate": 2.045159111492893e-05,
+      "loss": 0.0266,
+      "step": 653
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.27676835656166077,
+      "learning_rate": 2.042552737794996e-05,
+      "loss": 0.0688,
+      "step": 654
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.19456683099269867,
+      "learning_rate": 2.039944478096751e-05,
+      "loss": 0.0576,
+      "step": 655
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.15780119597911835,
+      "learning_rate": 2.037334341464888e-05,
+      "loss": 0.0254,
+      "step": 656
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.18241529166698456,
+      "learning_rate": 2.034722336972659e-05,
+      "loss": 0.0444,
+      "step": 657
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.21890577673912048,
+      "learning_rate": 2.032108473699812e-05,
+      "loss": 0.0496,
+      "step": 658
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.17364642024040222,
+      "learning_rate": 2.0294927607325542e-05,
+      "loss": 0.0426,
+      "step": 659
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.1697612851858139,
+      "learning_rate": 2.0268752071635233e-05,
+      "loss": 0.0367,
+      "step": 660
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.2621804475784302,
+      "learning_rate": 2.0242558220917573e-05,
+      "loss": 0.0395,
+      "step": 661
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.3021070063114166,
+      "learning_rate": 2.021634614622656e-05,
+      "loss": 0.0768,
+      "step": 662
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.35511207580566406,
+      "learning_rate": 2.0190115938679594e-05,
+      "loss": 0.0305,
+      "step": 663
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.49793681502342224,
+      "learning_rate": 2.0163867689457072e-05,
+      "loss": 0.066,
+      "step": 664
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.38811054825782776,
+      "learning_rate": 2.0137601489802126e-05,
+      "loss": 0.0389,
+      "step": 665
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.1782839596271515,
+      "learning_rate": 2.0111317431020272e-05,
+      "loss": 0.0185,
+      "step": 666
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.22854508459568024,
+      "learning_rate": 2.0085015604479113e-05,
+      "loss": 0.0741,
+      "step": 667
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.24257494509220123,
+      "learning_rate": 2.005869610160803e-05,
+      "loss": 0.0546,
+      "step": 668
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.1760607361793518,
+      "learning_rate": 2.0032359013897825e-05,
+      "loss": 0.0374,
+      "step": 669
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.38422301411628723,
+      "learning_rate": 2.0006004432900446e-05,
+      "loss": 0.0623,
+      "step": 670
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.39798131585121155,
+      "learning_rate": 1.9979632450228632e-05,
+      "loss": 0.0547,
+      "step": 671
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.2755303680896759,
+      "learning_rate": 1.9953243157555648e-05,
+      "loss": 0.0911,
+      "step": 672
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.26932981610298157,
+      "learning_rate": 1.9926836646614888e-05,
+      "loss": 0.0614,
+      "step": 673
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.2523995041847229,
+      "learning_rate": 1.9900413009199627e-05,
+      "loss": 0.0423,
+      "step": 674
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.21099168062210083,
+      "learning_rate": 1.987397233716267e-05,
+      "loss": 0.0652,
+      "step": 675
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.6092944741249084,
+      "learning_rate": 1.9847514722416028e-05,
+      "loss": 0.091,
+      "step": 676
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.2873200476169586,
+      "learning_rate": 1.982104025693062e-05,
+      "loss": 0.0265,
+      "step": 677
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.2782742381095886,
+      "learning_rate": 1.9794549032735935e-05,
+      "loss": 0.0632,
+      "step": 678
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.33089664578437805,
+      "learning_rate": 1.9768041141919713e-05,
+      "loss": 0.0548,
+      "step": 679
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.21904997527599335,
+      "learning_rate": 1.9741516676627632e-05,
+      "loss": 0.0384,
+      "step": 680
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.16128236055374146,
+      "learning_rate": 1.9714975729063e-05,
+      "loss": 0.0634,
+      "step": 681
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.18856774270534515,
+      "learning_rate": 1.96884183914864e-05,
+      "loss": 0.049,
+      "step": 682
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.16085284948349,
+      "learning_rate": 1.96618447562154e-05,
+      "loss": 0.0454,
+      "step": 683
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.2683512568473816,
+      "learning_rate": 1.963525491562421e-05,
+      "loss": 0.049,
+      "step": 684
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.1554921716451645,
+      "learning_rate": 1.9608648962143395e-05,
+      "loss": 0.0348,
+      "step": 685
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.48349252343177795,
+      "learning_rate": 1.958202698825951e-05,
+      "loss": 0.0956,
+      "step": 686
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.23943151533603668,
+      "learning_rate": 1.9555389086514813e-05,
+      "loss": 0.0308,
+      "step": 687
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.15932539105415344,
+      "learning_rate": 1.9528735349506906e-05,
+      "loss": 0.0229,
+      "step": 688
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.2831600308418274,
+      "learning_rate": 1.9502065869888473e-05,
+      "loss": 0.0599,
+      "step": 689
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.3630116879940033,
+      "learning_rate": 1.94753807403669e-05,
+      "loss": 0.0561,
+      "step": 690
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.19034230709075928,
+      "learning_rate": 1.9448680053703974e-05,
+      "loss": 0.0389,
+      "step": 691
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.2033299058675766,
+      "learning_rate": 1.9421963902715565e-05,
+      "loss": 0.0363,
+      "step": 692
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.1664944291114807,
+      "learning_rate": 1.9395232380271292e-05,
+      "loss": 0.0306,
+      "step": 693
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.23537740111351013,
+      "learning_rate": 1.9368485579294233e-05,
+      "loss": 0.0332,
+      "step": 694
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.3242640495300293,
+      "learning_rate": 1.9341723592760542e-05,
+      "loss": 0.0796,
+      "step": 695
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.39729979634284973,
+      "learning_rate": 1.931494651369919e-05,
+      "loss": 0.1146,
+      "step": 696
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.22055041790008545,
+      "learning_rate": 1.9288154435191583e-05,
+      "loss": 0.0392,
+      "step": 697
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.26389139890670776,
+      "learning_rate": 1.92613474503713e-05,
+      "loss": 0.0564,
+      "step": 698
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.22827556729316711,
+      "learning_rate": 1.9234525652423714e-05,
+      "loss": 0.0504,
+      "step": 699
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.3187848627567291,
+      "learning_rate": 1.9207689134585698e-05,
+      "loss": 0.0361,
+      "step": 700
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.22649644315242767,
+      "learning_rate": 1.918083799014529e-05,
+      "loss": 0.0637,
+      "step": 701
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.17243173718452454,
+      "learning_rate": 1.9153972312441374e-05,
+      "loss": 0.0378,
+      "step": 702
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.25477367639541626,
+      "learning_rate": 1.912709219486336e-05,
+      "loss": 0.0365,
+      "step": 703
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.15195806324481964,
+      "learning_rate": 1.910019773085085e-05,
+      "loss": 0.0584,
+      "step": 704
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.18720103800296783,
+      "learning_rate": 1.907328901389331e-05,
+      "loss": 0.043,
+      "step": 705
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.26587921380996704,
+      "learning_rate": 1.904636613752976e-05,
+      "loss": 0.0456,
+      "step": 706
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.17741888761520386,
+      "learning_rate": 1.9019429195348433e-05,
+      "loss": 0.0399,
+      "step": 707
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.24706660211086273,
+      "learning_rate": 1.8992478280986468e-05,
+      "loss": 0.049,
+      "step": 708
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.7546937465667725,
+      "learning_rate": 1.896551348812956e-05,
+      "loss": 0.1339,
+      "step": 709
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.3364761769771576,
+      "learning_rate": 1.893853491051165e-05,
+      "loss": 0.0819,
+      "step": 710
+    },
+    {
+      "epoch": 1.26,
+      "eval_loss": 0.05212438479065895,
+      "eval_runtime": 14.7307,
+      "eval_samples_per_second": 32.314,
+      "eval_steps_per_second": 8.078,
+      "step": 710
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.24806012213230133,
+      "learning_rate": 1.8911542641914617e-05,
+      "loss": 0.062,
+      "step": 711
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.2110578715801239,
+      "learning_rate": 1.88845367761679e-05,
+      "loss": 0.06,
+      "step": 712
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.11818195879459381,
+      "learning_rate": 1.8857517407148236e-05,
+      "loss": 0.0415,
+      "step": 713
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.2232706993818283,
+      "learning_rate": 1.8830484628779267e-05,
+      "loss": 0.0271,
+      "step": 714
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.2241712361574173,
+      "learning_rate": 1.880343853503129e-05,
+      "loss": 0.0458,
+      "step": 715
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.23428824543952942,
+      "learning_rate": 1.8776379219920847e-05,
+      "loss": 0.0553,
+      "step": 716
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.2631600499153137,
+      "learning_rate": 1.8749306777510465e-05,
+      "loss": 0.0653,
+      "step": 717
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.2828924059867859,
+      "learning_rate": 1.8722221301908295e-05,
+      "loss": 0.0703,
+      "step": 718
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.14960721135139465,
+      "learning_rate": 1.8695122887267788e-05,
+      "loss": 0.0538,
+      "step": 719
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.11142805963754654,
+      "learning_rate": 1.8668011627787398e-05,
+      "loss": 0.023,
+      "step": 720
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.25347310304641724,
+      "learning_rate": 1.86408876177102e-05,
+      "loss": 0.0387,
+      "step": 721
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.1751774698495865,
+      "learning_rate": 1.8613750951323602e-05,
+      "loss": 0.0522,
+      "step": 722
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.34507665038108826,
+      "learning_rate": 1.858660172295901e-05,
+      "loss": 0.0655,
+      "step": 723
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.31746798753738403,
+      "learning_rate": 1.8559440026991508e-05,
+      "loss": 0.0441,
+      "step": 724
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.1273573487997055,
+      "learning_rate": 1.8532265957839497e-05,
+      "loss": 0.0191,
+      "step": 725
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.19146591424942017,
+      "learning_rate": 1.8505079609964403e-05,
+      "loss": 0.0652,
+      "step": 726
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.3864126205444336,
+      "learning_rate": 1.847788107787033e-05,
+      "loss": 0.0682,
+      "step": 727
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.17144039273262024,
+      "learning_rate": 1.8450670456103743e-05,
+      "loss": 0.0565,
+      "step": 728
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.2258003205060959,
+      "learning_rate": 1.8423447839253128e-05,
+      "loss": 0.0356,
+      "step": 729
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.22176998853683472,
+      "learning_rate": 1.8396213321948662e-05,
+      "loss": 0.06,
+      "step": 730
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.14529289305210114,
+      "learning_rate": 1.83689669988619e-05,
+      "loss": 0.0173,
+      "step": 731
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.3319777846336365,
+      "learning_rate": 1.8341708964705434e-05,
+      "loss": 0.1009,
+      "step": 732
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.22678159177303314,
+      "learning_rate": 1.831443931423256e-05,
+      "loss": 0.0417,
+      "step": 733
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.18460430204868317,
+      "learning_rate": 1.828715814223696e-05,
+      "loss": 0.0415,
+      "step": 734
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.31541454792022705,
+      "learning_rate": 1.825986554355236e-05,
+      "loss": 0.0743,
+      "step": 735
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.4476594626903534,
+      "learning_rate": 1.8232561613052214e-05,
+      "loss": 0.0598,
+      "step": 736
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.18413041532039642,
+      "learning_rate": 1.8205246445649366e-05,
+      "loss": 0.0388,
+      "step": 737
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.2517610192298889,
+      "learning_rate": 1.8177920136295715e-05,
+      "loss": 0.0705,
+      "step": 738
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.14027471840381622,
+      "learning_rate": 1.81505827799819e-05,
+      "loss": 0.0303,
+      "step": 739
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.2901992201805115,
+      "learning_rate": 1.8123234471736947e-05,
+      "loss": 0.0824,
+      "step": 740
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.31951528787612915,
+      "learning_rate": 1.809587530662798e-05,
+      "loss": 0.1095,
+      "step": 741
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.414044052362442,
+      "learning_rate": 1.8068505379759828e-05,
+      "loss": 0.0438,
+      "step": 742
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.1758764535188675,
+      "learning_rate": 1.8041124786274757e-05,
+      "loss": 0.0452,
+      "step": 743
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.16384004056453705,
+      "learning_rate": 1.80137336213521e-05,
+      "loss": 0.0417,
+      "step": 744
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.2656683027744293,
+      "learning_rate": 1.7986331980207942e-05,
+      "loss": 0.0526,
+      "step": 745
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.22542397677898407,
+      "learning_rate": 1.795891995809478e-05,
+      "loss": 0.0637,
+      "step": 746
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.16901203989982605,
+      "learning_rate": 1.79314976503012e-05,
+      "loss": 0.0343,
+      "step": 747
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.31080687046051025,
+      "learning_rate": 1.7904065152151547e-05,
+      "loss": 0.0765,
+      "step": 748
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.15974196791648865,
+      "learning_rate": 1.7876622559005578e-05,
+      "loss": 0.0342,
+      "step": 749
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.21243633329868317,
+      "learning_rate": 1.7849169966258158e-05,
+      "loss": 0.0428,
+      "step": 750
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.34151706099510193,
+      "learning_rate": 1.7821707469338894e-05,
+      "loss": 0.0721,
+      "step": 751
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.14494989812374115,
+      "learning_rate": 1.7794235163711837e-05,
+      "loss": 0.0299,
+      "step": 752
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.1844569444656372,
+      "learning_rate": 1.776675314487512e-05,
+      "loss": 0.0362,
+      "step": 753
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.3417447805404663,
+      "learning_rate": 1.7739261508360662e-05,
+      "loss": 0.0594,
+      "step": 754
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.255882203578949,
+      "learning_rate": 1.7711760349733794e-05,
+      "loss": 0.0322,
+      "step": 755
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.20268352329730988,
+      "learning_rate": 1.7684249764592948e-05,
+      "loss": 0.0457,
+      "step": 756
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.16957353055477142,
+      "learning_rate": 1.7656729848569342e-05,
+      "loss": 0.0288,
+      "step": 757
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.5046777725219727,
+      "learning_rate": 1.7629200697326617e-05,
+      "loss": 0.0738,
+      "step": 758
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.12323735654354095,
+      "learning_rate": 1.760166240656051e-05,
+      "loss": 0.0228,
+      "step": 759
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.2688758373260498,
+      "learning_rate": 1.757411507199855e-05,
+      "loss": 0.0517,
+      "step": 760
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.45129191875457764,
+      "learning_rate": 1.754655878939968e-05,
+      "loss": 0.0684,
+      "step": 761
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.1658601313829422,
+      "learning_rate": 1.7518993654553963e-05,
+      "loss": 0.0293,
+      "step": 762
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.293125718832016,
+      "learning_rate": 1.7491419763282228e-05,
+      "loss": 0.041,
+      "step": 763
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.2707131505012512,
+      "learning_rate": 1.7463837211435746e-05,
+      "loss": 0.0327,
+      "step": 764
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.36076560616493225,
+      "learning_rate": 1.7436246094895896e-05,
+      "loss": 0.0768,
+      "step": 765
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.2822013795375824,
+      "learning_rate": 1.7408646509573815e-05,
+      "loss": 0.0492,
+      "step": 766
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.21052542328834534,
+      "learning_rate": 1.73810385514101e-05,
+      "loss": 0.0444,
+      "step": 767
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.24180252850055695,
+      "learning_rate": 1.7353422316374428e-05,
+      "loss": 0.0494,
+      "step": 768
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.26525041460990906,
+      "learning_rate": 1.732579790046528e-05,
+      "loss": 0.0458,
+      "step": 769
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.18544049561023712,
+      "learning_rate": 1.729816539970954e-05,
+      "loss": 0.0466,
+      "step": 770
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.22581623494625092,
+      "learning_rate": 1.727052491016222e-05,
+      "loss": 0.0433,
+      "step": 771
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.31744149327278137,
+      "learning_rate": 1.7242876527906096e-05,
+      "loss": 0.0537,
+      "step": 772
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.3232523202896118,
+      "learning_rate": 1.721522034905138e-05,
+      "loss": 0.0691,
+      "step": 773
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.2335691899061203,
+      "learning_rate": 1.718755646973539e-05,
+      "loss": 0.045,
+      "step": 774
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.2773955464363098,
+      "learning_rate": 1.7159884986122196e-05,
+      "loss": 0.0603,
+      "step": 775
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.44610023498535156,
+      "learning_rate": 1.713220599440233e-05,
+      "loss": 0.0982,
+      "step": 776
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.122254878282547,
+      "learning_rate": 1.7104519590792396e-05,
+      "loss": 0.016,
+      "step": 777
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.3145179748535156,
+      "learning_rate": 1.7076825871534775e-05,
+      "loss": 0.0639,
+      "step": 778
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.22903162240982056,
+      "learning_rate": 1.7049124932897278e-05,
+      "loss": 0.0388,
+      "step": 779
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.453363299369812,
+      "learning_rate": 1.7021416871172817e-05,
+      "loss": 0.0677,
+      "step": 780
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.30594778060913086,
+      "learning_rate": 1.6993701782679056e-05,
+      "loss": 0.0497,
+      "step": 781
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.20155024528503418,
+      "learning_rate": 1.696597976375808e-05,
+      "loss": 0.0539,
+      "step": 782
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.19732704758644104,
+      "learning_rate": 1.6938250910776086e-05,
+      "loss": 0.035,
+      "step": 783
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.3609841465950012,
+      "learning_rate": 1.691051532012301e-05,
+      "loss": 0.0478,
+      "step": 784
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.19694820046424866,
+      "learning_rate": 1.6882773088212213e-05,
+      "loss": 0.0498,
+      "step": 785
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.19805662333965302,
+      "learning_rate": 1.6855024311480147e-05,
+      "loss": 0.0327,
+      "step": 786
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.18435846269130707,
+      "learning_rate": 1.6827269086386006e-05,
+      "loss": 0.0335,
+      "step": 787
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.16236752271652222,
+      "learning_rate": 1.6799507509411408e-05,
+      "loss": 0.0337,
+      "step": 788
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.16657193005084991,
+      "learning_rate": 1.6771739677060042e-05,
+      "loss": 0.0336,
+      "step": 789
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.36152663826942444,
+      "learning_rate": 1.6743965685857362e-05,
+      "loss": 0.0798,
+      "step": 790
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.1292068213224411,
+      "learning_rate": 1.6716185632350204e-05,
+      "loss": 0.032,
+      "step": 791
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.18294057250022888,
+      "learning_rate": 1.6688399613106488e-05,
+      "loss": 0.0358,
+      "step": 792
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.30948176980018616,
+      "learning_rate": 1.666060772471488e-05,
+      "loss": 0.0695,
+      "step": 793
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.27416589856147766,
+      "learning_rate": 1.663281006378445e-05,
+      "loss": 0.076,
+      "step": 794
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.650515615940094,
+      "learning_rate": 1.6605006726944317e-05,
+      "loss": 0.1093,
+      "step": 795
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.32595914602279663,
+      "learning_rate": 1.6577197810843332e-05,
+      "loss": 0.0535,
+      "step": 796
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.15904508531093597,
+      "learning_rate": 1.6549383412149764e-05,
+      "loss": 0.0275,
+      "step": 797
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.19686171412467957,
+      "learning_rate": 1.6521563627550915e-05,
+      "loss": 0.0288,
+      "step": 798
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.17189526557922363,
+      "learning_rate": 1.6493738553752822e-05,
+      "loss": 0.044,
+      "step": 799
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.19990110397338867,
+      "learning_rate": 1.646590828747991e-05,
+      "loss": 0.0419,
+      "step": 800
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.27069151401519775,
+      "learning_rate": 1.643807292547463e-05,
+      "loss": 0.0713,
+      "step": 801
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.25677499175071716,
+      "learning_rate": 1.6410232564497188e-05,
+      "loss": 0.0482,
+      "step": 802
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.390544056892395,
+      "learning_rate": 1.6382387301325136e-05,
+      "loss": 0.0549,
+      "step": 803
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.1617233157157898,
+      "learning_rate": 1.6354537232753072e-05,
+      "loss": 0.0257,
+      "step": 804
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.2596289813518524,
+      "learning_rate": 1.6326682455592305e-05,
+      "loss": 0.0776,
+      "step": 805
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.21102601289749146,
+      "learning_rate": 1.629882306667051e-05,
+      "loss": 0.0208,
+      "step": 806
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.1607803851366043,
+      "learning_rate": 1.627095916283139e-05,
+      "loss": 0.0484,
+      "step": 807
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.27364104986190796,
+      "learning_rate": 1.6243090840934344e-05,
+      "loss": 0.0274,
+      "step": 808
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.19951047003269196,
+      "learning_rate": 1.621521819785413e-05,
+      "loss": 0.0323,
+      "step": 809
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.28387781977653503,
+      "learning_rate": 1.618734133048052e-05,
+      "loss": 0.0601,
+      "step": 810
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.4012148678302765,
+      "learning_rate": 1.6159460335717992e-05,
+      "loss": 0.0429,
+      "step": 811
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.21728171408176422,
+      "learning_rate": 1.6131575310485346e-05,
+      "loss": 0.0364,
+      "step": 812
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.2689477503299713,
+      "learning_rate": 1.6103686351715403e-05,
+      "loss": 0.0466,
+      "step": 813
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.38029181957244873,
+      "learning_rate": 1.6075793556354655e-05,
+      "loss": 0.1038,
+      "step": 814
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.1651802659034729,
+      "learning_rate": 1.604789702136294e-05,
+      "loss": 0.0411,
+      "step": 815
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.3085605502128601,
+      "learning_rate": 1.6019996843713088e-05,
+      "loss": 0.0523,
+      "step": 816
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.37859705090522766,
+      "learning_rate": 1.599209312039058e-05,
+      "loss": 0.0514,
+      "step": 817
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.23541994392871857,
+      "learning_rate": 1.5964185948393248e-05,
+      "loss": 0.0551,
+      "step": 818
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.23828637599945068,
+      "learning_rate": 1.5936275424730892e-05,
+      "loss": 0.0382,
+      "step": 819
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.3132995069026947,
+      "learning_rate": 1.5908361646424973e-05,
+      "loss": 0.0604,
+      "step": 820
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.26171785593032837,
+      "learning_rate": 1.5880444710508263e-05,
+      "loss": 0.0498,
+      "step": 821
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.36306342482566833,
+      "learning_rate": 1.5852524714024502e-05,
+      "loss": 0.0631,
+      "step": 822
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.26050522923469543,
+      "learning_rate": 1.5824601754028083e-05,
+      "loss": 0.0447,
+      "step": 823
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.26902008056640625,
+      "learning_rate": 1.579667592758369e-05,
+      "loss": 0.0729,
+      "step": 824
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.332304447889328,
+      "learning_rate": 1.5768747331765975e-05,
+      "loss": 0.0691,
+      "step": 825
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.35629650950431824,
+      "learning_rate": 1.5740816063659226e-05,
+      "loss": 0.0893,
+      "step": 826
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.3809410035610199,
+      "learning_rate": 1.5712882220357e-05,
+      "loss": 0.0696,
+      "step": 827
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.5881456136703491,
+      "learning_rate": 1.5684945898961826e-05,
+      "loss": 0.1203,
+      "step": 828
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.24559205770492554,
+      "learning_rate": 1.5657007196584833e-05,
+      "loss": 0.0668,
+      "step": 829
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.243602454662323,
+      "learning_rate": 1.562906621034543e-05,
+      "loss": 0.0734,
+      "step": 830
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.23130877315998077,
+      "learning_rate": 1.5601123037370977e-05,
+      "loss": 0.0576,
+      "step": 831
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.24903638660907745,
+      "learning_rate": 1.5573177774796415e-05,
+      "loss": 0.0697,
+      "step": 832
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.1651858389377594,
+      "learning_rate": 1.5545230519763967e-05,
+      "loss": 0.0451,
+      "step": 833
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.33191120624542236,
+      "learning_rate": 1.5517281369422767e-05,
+      "loss": 0.0981,
+      "step": 834
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.21882258355617523,
+      "learning_rate": 1.5489330420928554e-05,
+      "loss": 0.0526,
+      "step": 835
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.16961008310317993,
+      "learning_rate": 1.54613777714433e-05,
+      "loss": 0.064,
+      "step": 836
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.1839754283428192,
+      "learning_rate": 1.5433423518134906e-05,
+      "loss": 0.0505,
+      "step": 837
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.14958679676055908,
+      "learning_rate": 1.5405467758176836e-05,
+      "loss": 0.0411,
+      "step": 838
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.19069762527942657,
+      "learning_rate": 1.53775105887478e-05,
+      "loss": 0.0444,
+      "step": 839
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.18522578477859497,
+      "learning_rate": 1.5349552107031397e-05,
+      "loss": 0.0434,
+      "step": 840
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.20363502204418182,
+      "learning_rate": 1.5321592410215797e-05,
+      "loss": 0.0505,
+      "step": 841
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.19764205813407898,
+      "learning_rate": 1.5293631595493402e-05,
+      "loss": 0.0656,
+      "step": 842
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.13997752964496613,
+      "learning_rate": 1.5265669760060465e-05,
+      "loss": 0.0289,
+      "step": 843
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.1866457611322403,
+      "learning_rate": 1.5237707001116832e-05,
+      "loss": 0.0729,
+      "step": 844
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.2979126274585724,
+      "learning_rate": 1.5209743415865535e-05,
+      "loss": 0.0857,
+      "step": 845
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.13852760195732117,
+      "learning_rate": 1.5181779101512479e-05,
+      "loss": 0.0243,
+      "step": 846
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.26966992020606995,
+      "learning_rate": 1.5153814155266105e-05,
+      "loss": 0.0602,
+      "step": 847
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.15731748938560486,
+      "learning_rate": 1.5125848674337052e-05,
+      "loss": 0.041,
+      "step": 848
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.12978889048099518,
+      "learning_rate": 1.5097882755937824e-05,
+      "loss": 0.0401,
+      "step": 849
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.24159477651119232,
+      "learning_rate": 1.5069916497282432e-05,
+      "loss": 0.0558,
+      "step": 850
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.24909768998622894,
+      "learning_rate": 1.504194999558608e-05,
+      "loss": 0.0725,
+      "step": 851
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.2621814012527466,
+      "learning_rate": 1.5013983348064814e-05,
+      "loss": 0.0552,
+      "step": 852
+    },
+    {
+      "epoch": 1.51,
+      "eval_loss": 0.05297964811325073,
+      "eval_runtime": 14.7241,
+      "eval_samples_per_second": 32.328,
+      "eval_steps_per_second": 8.082,
+      "step": 852
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.1669163703918457,
+      "learning_rate": 1.4986016651935187e-05,
+      "loss": 0.0544,
+      "step": 853
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.16181348264217377,
+      "learning_rate": 1.4958050004413922e-05,
+      "loss": 0.0475,
+      "step": 854
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.25394901633262634,
+      "learning_rate": 1.4930083502717573e-05,
+      "loss": 0.0471,
+      "step": 855
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.23477764427661896,
+      "learning_rate": 1.490211724406218e-05,
+      "loss": 0.0615,
+      "step": 856
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.22598886489868164,
+      "learning_rate": 1.4874151325662952e-05,
+      "loss": 0.05,
+      "step": 857
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.21471647918224335,
+      "learning_rate": 1.4846185844733896e-05,
+      "loss": 0.0421,
+      "step": 858
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.21471573412418365,
+      "learning_rate": 1.4818220898487522e-05,
+      "loss": 0.062,
+      "step": 859
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.32464271783828735,
+      "learning_rate": 1.4790256584134467e-05,
+      "loss": 0.0951,
+      "step": 860
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.2311284989118576,
+      "learning_rate": 1.4762292998883167e-05,
+      "loss": 0.0541,
+      "step": 861
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.1898965984582901,
+      "learning_rate": 1.4734330239939537e-05,
+      "loss": 0.0357,
+      "step": 862
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.2285154163837433,
+      "learning_rate": 1.4706368404506604e-05,
+      "loss": 0.03,
+      "step": 863
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.1730206459760666,
+      "learning_rate": 1.4678407589784202e-05,
+      "loss": 0.0304,
+      "step": 864
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.1433994174003601,
+      "learning_rate": 1.4650447892968606e-05,
+      "loss": 0.0201,
+      "step": 865
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.2737201750278473,
+      "learning_rate": 1.4622489411252203e-05,
+      "loss": 0.0513,
+      "step": 866
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.3572165369987488,
+      "learning_rate": 1.4594532241823168e-05,
+      "loss": 0.0426,
+      "step": 867
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.20711761713027954,
+      "learning_rate": 1.4566576481865097e-05,
+      "loss": 0.0435,
+      "step": 868
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.14759641885757446,
+      "learning_rate": 1.4538622228556698e-05,
+      "loss": 0.0178,
+      "step": 869
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.25281572341918945,
+      "learning_rate": 1.4510669579071449e-05,
+      "loss": 0.0483,
+      "step": 870
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.1265438199043274,
+      "learning_rate": 1.4482718630577234e-05,
+      "loss": 0.0122,
+      "step": 871
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.2462211698293686,
+      "learning_rate": 1.4454769480236034e-05,
+      "loss": 0.0259,
+      "step": 872
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.4634632468223572,
+      "learning_rate": 1.442682222520359e-05,
+      "loss": 0.0612,
+      "step": 873
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.08901804685592651,
+      "learning_rate": 1.4398876962629026e-05,
+      "loss": 0.0076,
+      "step": 874
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.25518471002578735,
+      "learning_rate": 1.4370933789654571e-05,
+      "loss": 0.0715,
+      "step": 875
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.3362948000431061,
+      "learning_rate": 1.4342992803415171e-05,
+      "loss": 0.0417,
+      "step": 876
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.2440776228904724,
+      "learning_rate": 1.4315054101038176e-05,
+      "loss": 0.0473,
+      "step": 877
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.22846253216266632,
+      "learning_rate": 1.4287117779643005e-05,
+      "loss": 0.0372,
+      "step": 878
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.16435202956199646,
+      "learning_rate": 1.4259183936340775e-05,
+      "loss": 0.0236,
+      "step": 879
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.16216912865638733,
+      "learning_rate": 1.4231252668234025e-05,
+      "loss": 0.0194,
+      "step": 880
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.590180516242981,
+      "learning_rate": 1.4203324072416313e-05,
+      "loss": 0.0959,
+      "step": 881
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.2067226618528366,
+      "learning_rate": 1.4175398245971921e-05,
+      "loss": 0.0162,
+      "step": 882
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.23973137140274048,
+      "learning_rate": 1.4147475285975504e-05,
+      "loss": 0.0553,
+      "step": 883
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.23057444393634796,
+      "learning_rate": 1.411955528949174e-05,
+      "loss": 0.0339,
+      "step": 884
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.44691550731658936,
+      "learning_rate": 1.4091638353575025e-05,
+      "loss": 0.0607,
+      "step": 885
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.3693409264087677,
+      "learning_rate": 1.4063724575269109e-05,
+      "loss": 0.0841,
+      "step": 886
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.36656439304351807,
+      "learning_rate": 1.4035814051606753e-05,
+      "loss": 0.0592,
+      "step": 887
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.37300795316696167,
+      "learning_rate": 1.4007906879609424e-05,
+      "loss": 0.0431,
+      "step": 888
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.1838502585887909,
+      "learning_rate": 1.3980003156286916e-05,
+      "loss": 0.035,
+      "step": 889
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.33711016178131104,
+      "learning_rate": 1.3952102978637058e-05,
+      "loss": 0.0515,
+      "step": 890
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.3825240433216095,
+      "learning_rate": 1.3924206443645345e-05,
+      "loss": 0.059,
+      "step": 891
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.22543422877788544,
+      "learning_rate": 1.38963136482846e-05,
+      "loss": 0.0289,
+      "step": 892
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.37494131922721863,
+      "learning_rate": 1.3868424689514658e-05,
+      "loss": 0.0762,
+      "step": 893
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.6698411703109741,
+      "learning_rate": 1.384053966428201e-05,
+      "loss": 0.0848,
+      "step": 894
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.2594004273414612,
+      "learning_rate": 1.3812658669519474e-05,
+      "loss": 0.0281,
+      "step": 895
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.39433106780052185,
+      "learning_rate": 1.3784781802145875e-05,
+      "loss": 0.0985,
+      "step": 896
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.20175516605377197,
+      "learning_rate": 1.3756909159065659e-05,
+      "loss": 0.0408,
+      "step": 897
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.18394209444522858,
+      "learning_rate": 1.372904083716861e-05,
+      "loss": 0.0263,
+      "step": 898
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.14119882881641388,
+      "learning_rate": 1.3701176933329495e-05,
+      "loss": 0.0264,
+      "step": 899
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.14260679483413696,
+      "learning_rate": 1.3673317544407694e-05,
+      "loss": 0.0312,
+      "step": 900
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.2470966875553131,
+      "learning_rate": 1.364546276724693e-05,
+      "loss": 0.0403,
+      "step": 901
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.25093796849250793,
+      "learning_rate": 1.3617612698674867e-05,
+      "loss": 0.0511,
+      "step": 902
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.25882598757743835,
+      "learning_rate": 1.3589767435502813e-05,
+      "loss": 0.0303,
+      "step": 903
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.13052983582019806,
+      "learning_rate": 1.3561927074525372e-05,
+      "loss": 0.0176,
+      "step": 904
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.1630779206752777,
+      "learning_rate": 1.3534091712520097e-05,
+      "loss": 0.0377,
+      "step": 905
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.3587014973163605,
+      "learning_rate": 1.3506261446247182e-05,
+      "loss": 0.0856,
+      "step": 906
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.25035837292671204,
+      "learning_rate": 1.3478436372449089e-05,
+      "loss": 0.0612,
+      "step": 907
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.26720166206359863,
+      "learning_rate": 1.345061658785024e-05,
+      "loss": 0.0853,
+      "step": 908
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.20204173028469086,
+      "learning_rate": 1.3422802189156672e-05,
+      "loss": 0.046,
+      "step": 909
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.3688868284225464,
+      "learning_rate": 1.339499327305569e-05,
+      "loss": 0.0629,
+      "step": 910
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.13798101246356964,
+      "learning_rate": 1.3367189936215552e-05,
+      "loss": 0.0289,
+      "step": 911
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.2122255116701126,
+      "learning_rate": 1.3339392275285118e-05,
+      "loss": 0.041,
+      "step": 912
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.34295904636383057,
+      "learning_rate": 1.3311600386893513e-05,
+      "loss": 0.0599,
+      "step": 913
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.2708895802497864,
+      "learning_rate": 1.3283814367649804e-05,
+      "loss": 0.0672,
+      "step": 914
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.24829035997390747,
+      "learning_rate": 1.3256034314142639e-05,
+      "loss": 0.05,
+      "step": 915
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.2881320118904114,
+      "learning_rate": 1.3228260322939955e-05,
+      "loss": 0.0319,
+      "step": 916
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.1695733219385147,
+      "learning_rate": 1.3200492490588596e-05,
+      "loss": 0.0482,
+      "step": 917
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.2604435980319977,
+      "learning_rate": 1.3172730913613996e-05,
+      "loss": 0.0547,
+      "step": 918
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.3750756084918976,
+      "learning_rate": 1.3144975688519859e-05,
+      "loss": 0.0839,
+      "step": 919
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.1456877589225769,
+      "learning_rate": 1.3117226911787791e-05,
+      "loss": 0.0277,
+      "step": 920
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.2657271921634674,
+      "learning_rate": 1.3089484679876989e-05,
+      "loss": 0.033,
+      "step": 921
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.26634252071380615,
+      "learning_rate": 1.3061749089223914e-05,
+      "loss": 0.0389,
+      "step": 922
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.2602170705795288,
+      "learning_rate": 1.303402023624192e-05,
+      "loss": 0.0503,
+      "step": 923
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.19216115772724152,
+      "learning_rate": 1.3006298217320948e-05,
+      "loss": 0.0298,
+      "step": 924
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.3049064576625824,
+      "learning_rate": 1.2978583128827187e-05,
+      "loss": 0.0755,
+      "step": 925
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.3116673529148102,
+      "learning_rate": 1.2950875067102719e-05,
+      "loss": 0.0728,
+      "step": 926
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.17638739943504333,
+      "learning_rate": 1.2923174128465227e-05,
+      "loss": 0.0252,
+      "step": 927
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.1943345069885254,
+      "learning_rate": 1.2895480409207608e-05,
+      "loss": 0.061,
+      "step": 928
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.2744896709918976,
+      "learning_rate": 1.286779400559767e-05,
+      "loss": 0.0633,
+      "step": 929
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.18297532200813293,
+      "learning_rate": 1.2840115013877803e-05,
+      "loss": 0.0272,
+      "step": 930
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.29986709356307983,
+      "learning_rate": 1.2812443530264611e-05,
+      "loss": 0.0517,
+      "step": 931
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.21048623323440552,
+      "learning_rate": 1.2784779650948622e-05,
+      "loss": 0.0514,
+      "step": 932
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.21968035399913788,
+      "learning_rate": 1.2757123472093904e-05,
+      "loss": 0.0527,
+      "step": 933
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.3859691023826599,
+      "learning_rate": 1.272947508983778e-05,
+      "loss": 0.0593,
+      "step": 934
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.29459449648857117,
+      "learning_rate": 1.2701834600290465e-05,
+      "loss": 0.0517,
+      "step": 935
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.23370222747325897,
+      "learning_rate": 1.2674202099534725e-05,
+      "loss": 0.0473,
+      "step": 936
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.3279827833175659,
+      "learning_rate": 1.264657768362557e-05,
+      "loss": 0.0432,
+      "step": 937
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.3182874917984009,
+      "learning_rate": 1.2618961448589903e-05,
+      "loss": 0.0642,
+      "step": 938
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.10656490176916122,
+      "learning_rate": 1.2591353490426185e-05,
+      "loss": 0.0138,
+      "step": 939
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.1540316343307495,
+      "learning_rate": 1.2563753905104108e-05,
+      "loss": 0.0318,
+      "step": 940
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.19106361269950867,
+      "learning_rate": 1.2536162788564253e-05,
+      "loss": 0.0439,
+      "step": 941
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.3189610242843628,
+      "learning_rate": 1.250858023671777e-05,
+      "loss": 0.0542,
+      "step": 942
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.25459712743759155,
+      "learning_rate": 1.2481006345446039e-05,
+      "loss": 0.0446,
+      "step": 943
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.26181209087371826,
+      "learning_rate": 1.2453441210600323e-05,
+      "loss": 0.0806,
+      "step": 944
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.32628104090690613,
+      "learning_rate": 1.2425884928001455e-05,
+      "loss": 0.056,
+      "step": 945
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.2178797721862793,
+      "learning_rate": 1.2398337593439491e-05,
+      "loss": 0.0323,
+      "step": 946
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.518593430519104,
+      "learning_rate": 1.2370799302673386e-05,
+      "loss": 0.0711,
+      "step": 947
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.19468623399734497,
+      "learning_rate": 1.2343270151430659e-05,
+      "loss": 0.0228,
+      "step": 948
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.3345097303390503,
+      "learning_rate": 1.2315750235407053e-05,
+      "loss": 0.0657,
+      "step": 949
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.3418566584587097,
+      "learning_rate": 1.2288239650266211e-05,
+      "loss": 0.0374,
+      "step": 950
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.22158052027225494,
+      "learning_rate": 1.2260738491639342e-05,
+      "loss": 0.0451,
+      "step": 951
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.3689166009426117,
+      "learning_rate": 1.2233246855124876e-05,
+      "loss": 0.0742,
+      "step": 952
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.21123674511909485,
+      "learning_rate": 1.2205764836288167e-05,
+      "loss": 0.0398,
+      "step": 953
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.21988846361637115,
+      "learning_rate": 1.2178292530661108e-05,
+      "loss": 0.0285,
+      "step": 954
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.33735549449920654,
+      "learning_rate": 1.2150830033741845e-05,
+      "loss": 0.0585,
+      "step": 955
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.1364648938179016,
+      "learning_rate": 1.2123377440994424e-05,
+      "loss": 0.0177,
+      "step": 956
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.23270867764949799,
+      "learning_rate": 1.2095934847848454e-05,
+      "loss": 0.0533,
+      "step": 957
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.19464728236198425,
+      "learning_rate": 1.2068502349698806e-05,
+      "loss": 0.0383,
+      "step": 958
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.23923218250274658,
+      "learning_rate": 1.2041080041905221e-05,
+      "loss": 0.0406,
+      "step": 959
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.2545720338821411,
+      "learning_rate": 1.2013668019792059e-05,
+      "loss": 0.04,
+      "step": 960
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.2737041115760803,
+      "learning_rate": 1.1986266378647903e-05,
+      "loss": 0.0688,
+      "step": 961
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.32465556263923645,
+      "learning_rate": 1.1958875213725242e-05,
+      "loss": 0.053,
+      "step": 962
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.5943883657455444,
+      "learning_rate": 1.1931494620240173e-05,
+      "loss": 0.0711,
+      "step": 963
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.36020776629447937,
+      "learning_rate": 1.1904124693372027e-05,
+      "loss": 0.0431,
+      "step": 964
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.16541947424411774,
+      "learning_rate": 1.1876765528263054e-05,
+      "loss": 0.0306,
+      "step": 965
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.21830913424491882,
+      "learning_rate": 1.1849417220018108e-05,
+      "loss": 0.0556,
+      "step": 966
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.4576943814754486,
+      "learning_rate": 1.1822079863704291e-05,
+      "loss": 0.1155,
+      "step": 967
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.34350088238716125,
+      "learning_rate": 1.1794753554350635e-05,
+      "loss": 0.0616,
+      "step": 968
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.24154837429523468,
+      "learning_rate": 1.1767438386947789e-05,
+      "loss": 0.041,
+      "step": 969
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.3688625395298004,
+      "learning_rate": 1.1740134456447642e-05,
+      "loss": 0.0789,
+      "step": 970
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.15884725749492645,
+      "learning_rate": 1.1712841857763043e-05,
+      "loss": 0.037,
+      "step": 971
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.2018052637577057,
+      "learning_rate": 1.1685560685767441e-05,
+      "loss": 0.0581,
+      "step": 972
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.2525852620601654,
+      "learning_rate": 1.1658291035294565e-05,
+      "loss": 0.0388,
+      "step": 973
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.13577726483345032,
+      "learning_rate": 1.16310330011381e-05,
+      "loss": 0.0207,
+      "step": 974
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.1704084426164627,
+      "learning_rate": 1.160378667805134e-05,
+      "loss": 0.0336,
+      "step": 975
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.2135757952928543,
+      "learning_rate": 1.1576552160746876e-05,
+      "loss": 0.0428,
+      "step": 976
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.21314018964767456,
+      "learning_rate": 1.1549329543896261e-05,
+      "loss": 0.0469,
+      "step": 977
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.29755663871765137,
+      "learning_rate": 1.152211892212967e-05,
+      "loss": 0.0567,
+      "step": 978
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.18695975840091705,
+      "learning_rate": 1.1494920390035603e-05,
+      "loss": 0.0343,
+      "step": 979
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.15595993399620056,
+      "learning_rate": 1.1467734042160506e-05,
+      "loss": 0.027,
+      "step": 980
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.16335420310497284,
+      "learning_rate": 1.1440559973008493e-05,
+      "loss": 0.029,
+      "step": 981
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.22777877748012543,
+      "learning_rate": 1.141339827704099e-05,
+      "loss": 0.0259,
+      "step": 982
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.31491386890411377,
+      "learning_rate": 1.1386249048676399e-05,
+      "loss": 0.0737,
+      "step": 983
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.19159311056137085,
+      "learning_rate": 1.1359112382289807e-05,
+      "loss": 0.0235,
+      "step": 984
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.21155066788196564,
+      "learning_rate": 1.1331988372212604e-05,
+      "loss": 0.055,
+      "step": 985
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.40876173973083496,
+      "learning_rate": 1.1304877112732209e-05,
+      "loss": 0.0654,
+      "step": 986
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.2856805622577667,
+      "learning_rate": 1.1277778698091711e-05,
+      "loss": 0.0503,
+      "step": 987
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.25982338190078735,
+      "learning_rate": 1.1250693222489538e-05,
+      "loss": 0.0348,
+      "step": 988
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.15217123925685883,
+      "learning_rate": 1.1223620780079155e-05,
+      "loss": 0.0273,
+      "step": 989
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.19661454856395721,
+      "learning_rate": 1.1196561464968715e-05,
+      "loss": 0.038,
+      "step": 990
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.27082914113998413,
+      "learning_rate": 1.1169515371220729e-05,
+      "loss": 0.0547,
+      "step": 991
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.2935861349105835,
+      "learning_rate": 1.114248259285177e-05,
+      "loss": 0.052,
+      "step": 992
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.3520117998123169,
+      "learning_rate": 1.1115463223832101e-05,
+      "loss": 0.0447,
+      "step": 993
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.34975117444992065,
+      "learning_rate": 1.1088457358085384e-05,
+      "loss": 0.0764,
+      "step": 994
+    },
+    {
+      "epoch": 1.76,
+      "eval_loss": 0.052343208342790604,
+      "eval_runtime": 14.7406,
+      "eval_samples_per_second": 32.292,
+      "eval_steps_per_second": 8.073,
+      "step": 994
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.22226814925670624,
+      "learning_rate": 1.106146508948835e-05,
+      "loss": 0.0274,
+      "step": 995
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.27738481760025024,
+      "learning_rate": 1.1034486511870444e-05,
+      "loss": 0.0394,
+      "step": 996
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.5362345576286316,
+      "learning_rate": 1.1007521719013538e-05,
+      "loss": 0.0669,
+      "step": 997
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.2275220900774002,
+      "learning_rate": 1.0980570804651568e-05,
+      "loss": 0.0266,
+      "step": 998
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.16260147094726562,
+      "learning_rate": 1.095363386247024e-05,
+      "loss": 0.0127,
+      "step": 999
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.281746506690979,
+      "learning_rate": 1.0926710986106691e-05,
+      "loss": 0.0313,
+      "step": 1000
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.26501229405403137,
+      "learning_rate": 1.0899802269149151e-05,
+      "loss": 0.0452,
+      "step": 1001
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.23682145774364471,
+      "learning_rate": 1.0872907805136637e-05,
+      "loss": 0.0332,
+      "step": 1002
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.29634806513786316,
+      "learning_rate": 1.0846027687558627e-05,
+      "loss": 0.0475,
+      "step": 1003
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.34405043721199036,
+      "learning_rate": 1.081916200985471e-05,
+      "loss": 0.1156,
+      "step": 1004
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.21926525235176086,
+      "learning_rate": 1.0792310865414307e-05,
+      "loss": 0.0345,
+      "step": 1005
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.24628080427646637,
+      "learning_rate": 1.0765474347576288e-05,
+      "loss": 0.0361,
+      "step": 1006
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.3084715008735657,
+      "learning_rate": 1.07386525496287e-05,
+      "loss": 0.0627,
+      "step": 1007
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.21348412334918976,
+      "learning_rate": 1.0711845564808418e-05,
+      "loss": 0.0208,
+      "step": 1008
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.24961091578006744,
+      "learning_rate": 1.0685053486300814e-05,
+      "loss": 0.0347,
+      "step": 1009
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.3914209306240082,
+      "learning_rate": 1.0658276407239462e-05,
+      "loss": 0.0583,
+      "step": 1010
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.23963287472724915,
+      "learning_rate": 1.063151442070577e-05,
+      "loss": 0.0574,
+      "step": 1011
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.18053047358989716,
+      "learning_rate": 1.0604767619728707e-05,
+      "loss": 0.0233,
+      "step": 1012
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.34937530755996704,
+      "learning_rate": 1.0578036097284442e-05,
+      "loss": 0.0422,
+      "step": 1013
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.36412081122398376,
+      "learning_rate": 1.0551319946296027e-05,
+      "loss": 0.0832,
+      "step": 1014
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.5039358735084534,
+      "learning_rate": 1.05246192596331e-05,
+      "loss": 0.0665,
+      "step": 1015
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.381173700094223,
+      "learning_rate": 1.0497934130111526e-05,
+      "loss": 0.0455,
+      "step": 1016
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.30506399273872375,
+      "learning_rate": 1.0471264650493093e-05,
+      "loss": 0.0539,
+      "step": 1017
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.38967427611351013,
+      "learning_rate": 1.0444610913485195e-05,
+      "loss": 0.0767,
+      "step": 1018
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.24847134947776794,
+      "learning_rate": 1.0417973011740492e-05,
+      "loss": 0.028,
+      "step": 1019
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.19682088494300842,
+      "learning_rate": 1.0391351037856604e-05,
+      "loss": 0.0277,
+      "step": 1020
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.22912351787090302,
+      "learning_rate": 1.036474508437579e-05,
+      "loss": 0.0456,
+      "step": 1021
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.22427204251289368,
+      "learning_rate": 1.0338155243784606e-05,
+      "loss": 0.0298,
+      "step": 1022
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.38963979482650757,
+      "learning_rate": 1.0311581608513603e-05,
+      "loss": 0.0831,
+      "step": 1023
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.28928878903388977,
+      "learning_rate": 1.0285024270937003e-05,
+      "loss": 0.0482,
+      "step": 1024
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.24627822637557983,
+      "learning_rate": 1.0258483323372365e-05,
+      "loss": 0.0234,
+      "step": 1025
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.37990662455558777,
+      "learning_rate": 1.0231958858080291e-05,
+      "loss": 0.0683,
+      "step": 1026
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.13172967731952667,
+      "learning_rate": 1.0205450967264067e-05,
+      "loss": 0.017,
+      "step": 1027
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.5053113102912903,
+      "learning_rate": 1.017895974306938e-05,
+      "loss": 0.0538,
+      "step": 1028
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.2872104346752167,
+      "learning_rate": 1.0152485277583975e-05,
+      "loss": 0.0334,
+      "step": 1029
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.2830204367637634,
+      "learning_rate": 1.0126027662837332e-05,
+      "loss": 0.0541,
+      "step": 1030
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.16259455680847168,
+      "learning_rate": 1.0099586990800375e-05,
+      "loss": 0.0265,
+      "step": 1031
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.18389791250228882,
+      "learning_rate": 1.0073163353385114e-05,
+      "loss": 0.0382,
+      "step": 1032
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.21226352453231812,
+      "learning_rate": 1.0046756842444356e-05,
+      "loss": 0.0495,
+      "step": 1033
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.34458455443382263,
+      "learning_rate": 1.0020367549771367e-05,
+      "loss": 0.0524,
+      "step": 1034
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.2613154649734497,
+      "learning_rate": 9.993995567099557e-06,
+      "loss": 0.0383,
+      "step": 1035
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.3383423984050751,
+      "learning_rate": 9.967640986102181e-06,
+      "loss": 0.0539,
+      "step": 1036
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.22641730308532715,
+      "learning_rate": 9.941303898391973e-06,
+      "loss": 0.0676,
+      "step": 1037
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.38731008768081665,
+      "learning_rate": 9.914984395520885e-06,
+      "loss": 0.0518,
+      "step": 1038
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.21816083788871765,
+      "learning_rate": 9.888682568979735e-06,
+      "loss": 0.0499,
+      "step": 1039
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.23305702209472656,
+      "learning_rate": 9.862398510197875e-06,
+      "loss": 0.0465,
+      "step": 1040
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.19520357251167297,
+      "learning_rate": 9.836132310542926e-06,
+      "loss": 0.0353,
+      "step": 1041
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.25492751598358154,
+      "learning_rate": 9.809884061320407e-06,
+      "loss": 0.0396,
+      "step": 1042
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.15815380215644836,
+      "learning_rate": 9.783653853773438e-06,
+      "loss": 0.0221,
+      "step": 1043
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.514136791229248,
+      "learning_rate": 9.757441779082434e-06,
+      "loss": 0.0624,
+      "step": 1044
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.4792744517326355,
+      "learning_rate": 9.731247928364766e-06,
+      "loss": 0.0666,
+      "step": 1045
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.165685772895813,
+      "learning_rate": 9.705072392674459e-06,
+      "loss": 0.0314,
+      "step": 1046
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.31327199935913086,
+      "learning_rate": 9.678915263001884e-06,
+      "loss": 0.0269,
+      "step": 1047
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.20864950120449066,
+      "learning_rate": 9.65277663027341e-06,
+      "loss": 0.0389,
+      "step": 1048
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.6159412860870361,
+      "learning_rate": 9.626656585351127e-06,
+      "loss": 0.0844,
+      "step": 1049
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.17599329352378845,
+      "learning_rate": 9.600555219032493e-06,
+      "loss": 0.0198,
+      "step": 1050
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.46148017048835754,
+      "learning_rate": 9.574472622050037e-06,
+      "loss": 0.0821,
+      "step": 1051
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.33153191208839417,
+      "learning_rate": 9.548408885071071e-06,
+      "loss": 0.0707,
+      "step": 1052
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.43923747539520264,
+      "learning_rate": 9.52236409869731e-06,
+      "loss": 0.0726,
+      "step": 1053
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.29100126028060913,
+      "learning_rate": 9.496338353464613e-06,
+      "loss": 0.0581,
+      "step": 1054
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.08649720996618271,
+      "learning_rate": 9.470331739842645e-06,
+      "loss": 0.0089,
+      "step": 1055
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.28912803530693054,
+      "learning_rate": 9.444344348234566e-06,
+      "loss": 0.0345,
+      "step": 1056
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.1984950602054596,
+      "learning_rate": 9.418376268976727e-06,
+      "loss": 0.0242,
+      "step": 1057
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.13096588850021362,
+      "learning_rate": 9.392427592338318e-06,
+      "loss": 0.0195,
+      "step": 1058
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.414447546005249,
+      "learning_rate": 9.36649840852112e-06,
+      "loss": 0.0605,
+      "step": 1059
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.7186275124549866,
+      "learning_rate": 9.340588807659128e-06,
+      "loss": 0.0952,
+      "step": 1060
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.43872931599617004,
+      "learning_rate": 9.314698879818264e-06,
+      "loss": 0.0652,
+      "step": 1061
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.12227629125118256,
+      "learning_rate": 9.288828714996086e-06,
+      "loss": 0.0223,
+      "step": 1062
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.31879130005836487,
+      "learning_rate": 9.26297840312142e-06,
+      "loss": 0.0681,
+      "step": 1063
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.29195696115493774,
+      "learning_rate": 9.237148034054107e-06,
+      "loss": 0.0666,
+      "step": 1064
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.27950647473335266,
+      "learning_rate": 9.211337697584654e-06,
+      "loss": 0.0278,
+      "step": 1065
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.22695614397525787,
+      "learning_rate": 9.185547483433923e-06,
+      "loss": 0.0521,
+      "step": 1066
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.2461085170507431,
+      "learning_rate": 9.159777481252838e-06,
+      "loss": 0.0648,
+      "step": 1067
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.33142539858818054,
+      "learning_rate": 9.134027780622069e-06,
+      "loss": 0.0372,
+      "step": 1068
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.23242291808128357,
+      "learning_rate": 9.108298471051695e-06,
+      "loss": 0.0472,
+      "step": 1069
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.22239136695861816,
+      "learning_rate": 9.082589641980931e-06,
+      "loss": 0.0467,
+      "step": 1070
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.14557255804538727,
+      "learning_rate": 9.056901382777784e-06,
+      "loss": 0.033,
+      "step": 1071
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.29809367656707764,
+      "learning_rate": 9.031233782738765e-06,
+      "loss": 0.0662,
+      "step": 1072
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.12355267256498337,
+      "learning_rate": 9.00558693108858e-06,
+      "loss": 0.0245,
+      "step": 1073
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.3000732958316803,
+      "learning_rate": 8.979960916979782e-06,
+      "loss": 0.0894,
+      "step": 1074
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.405777245759964,
+      "learning_rate": 8.954355829492522e-06,
+      "loss": 0.0683,
+      "step": 1075
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.2747291028499603,
+      "learning_rate": 8.928771757634179e-06,
+      "loss": 0.067,
+      "step": 1076
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.20261916518211365,
+      "learning_rate": 8.903208790339094e-06,
+      "loss": 0.0419,
+      "step": 1077
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.18399080634117126,
+      "learning_rate": 8.877667016468251e-06,
+      "loss": 0.0458,
+      "step": 1078
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.22144752740859985,
+      "learning_rate": 8.852146524808945e-06,
+      "loss": 0.0605,
+      "step": 1079
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.24125979840755463,
+      "learning_rate": 8.826647404074497e-06,
+      "loss": 0.0494,
+      "step": 1080
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.19941553473472595,
+      "learning_rate": 8.80116974290395e-06,
+      "loss": 0.0382,
+      "step": 1081
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.15817837417125702,
+      "learning_rate": 8.775713629861726e-06,
+      "loss": 0.0459,
+      "step": 1082
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.17759451270103455,
+      "learning_rate": 8.750279153437375e-06,
+      "loss": 0.0378,
+      "step": 1083
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.1938961297273636,
+      "learning_rate": 8.72486640204521e-06,
+      "loss": 0.0451,
+      "step": 1084
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.3393564224243164,
+      "learning_rate": 8.699475464024022e-06,
+      "loss": 0.0581,
+      "step": 1085
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.22199125587940216,
+      "learning_rate": 8.6741064276368e-06,
+      "loss": 0.0492,
+      "step": 1086
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.14225240051746368,
+      "learning_rate": 8.648759381070369e-06,
+      "loss": 0.0251,
+      "step": 1087
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.23262937366962433,
+      "learning_rate": 8.623434412435144e-06,
+      "loss": 0.0339,
+      "step": 1088
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.32259851694107056,
+      "learning_rate": 8.59813160976477e-06,
+      "loss": 0.0544,
+      "step": 1089
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.25321877002716064,
+      "learning_rate": 8.572851061015842e-06,
+      "loss": 0.0443,
+      "step": 1090
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.19444559514522552,
+      "learning_rate": 8.547592854067616e-06,
+      "loss": 0.0466,
+      "step": 1091
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.20306558907032013,
+      "learning_rate": 8.522357076721658e-06,
+      "loss": 0.0359,
+      "step": 1092
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.3852079212665558,
+      "learning_rate": 8.497143816701592e-06,
+      "loss": 0.0784,
+      "step": 1093
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.2700721323490143,
+      "learning_rate": 8.471953161652744e-06,
+      "loss": 0.0453,
+      "step": 1094
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.39073994755744934,
+      "learning_rate": 8.44678519914187e-06,
+      "loss": 0.0683,
+      "step": 1095
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.24957872927188873,
+      "learning_rate": 8.421640016656857e-06,
+      "loss": 0.0422,
+      "step": 1096
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.17364278435707092,
+      "learning_rate": 8.396517701606377e-06,
+      "loss": 0.0441,
+      "step": 1097
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.3036033511161804,
+      "learning_rate": 8.371418341319643e-06,
+      "loss": 0.0496,
+      "step": 1098
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.3998473584651947,
+      "learning_rate": 8.346342023046049e-06,
+      "loss": 0.0849,
+      "step": 1099
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.19726194441318512,
+      "learning_rate": 8.321288833954896e-06,
+      "loss": 0.0435,
+      "step": 1100
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.24617618322372437,
+      "learning_rate": 8.296258861135098e-06,
+      "loss": 0.0544,
+      "step": 1101
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.169749453663826,
+      "learning_rate": 8.271252191594846e-06,
+      "loss": 0.0424,
+      "step": 1102
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.17742374539375305,
+      "learning_rate": 8.246268912261344e-06,
+      "loss": 0.0325,
+      "step": 1103
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.2964729070663452,
+      "learning_rate": 8.221309109980472e-06,
+      "loss": 0.0812,
+      "step": 1104
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.16450470685958862,
+      "learning_rate": 8.196372871516502e-06,
+      "loss": 0.0259,
+      "step": 1105
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.2630455195903778,
+      "learning_rate": 8.17146028355181e-06,
+      "loss": 0.0326,
+      "step": 1106
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.6756049394607544,
+      "learning_rate": 8.146571432686545e-06,
+      "loss": 0.1267,
+      "step": 1107
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.3916427791118622,
+      "learning_rate": 8.121706405438346e-06,
+      "loss": 0.0579,
+      "step": 1108
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.22415930032730103,
+      "learning_rate": 8.09686528824203e-06,
+      "loss": 0.0364,
+      "step": 1109
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.2058791220188141,
+      "learning_rate": 8.072048167449306e-06,
+      "loss": 0.0437,
+      "step": 1110
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.16186773777008057,
+      "learning_rate": 8.047255129328467e-06,
+      "loss": 0.0383,
+      "step": 1111
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.3812834620475769,
+      "learning_rate": 8.0224862600641e-06,
+      "loss": 0.1065,
+      "step": 1112
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.23627111315727234,
+      "learning_rate": 7.997741645756764e-06,
+      "loss": 0.0501,
+      "step": 1113
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.574120283126831,
+      "learning_rate": 7.9730213724227e-06,
+      "loss": 0.0565,
+      "step": 1114
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.17482930421829224,
+      "learning_rate": 7.948325525993545e-06,
+      "loss": 0.0475,
+      "step": 1115
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.27232393622398376,
+      "learning_rate": 7.923654192316023e-06,
+      "loss": 0.0471,
+      "step": 1116
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.2575230300426483,
+      "learning_rate": 7.899007457151658e-06,
+      "loss": 0.0437,
+      "step": 1117
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.2097906619310379,
+      "learning_rate": 7.874385406176445e-06,
+      "loss": 0.0268,
+      "step": 1118
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.30610811710357666,
+      "learning_rate": 7.849788124980583e-06,
+      "loss": 0.0594,
+      "step": 1119
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.177670419216156,
+      "learning_rate": 7.82521569906817e-06,
+      "loss": 0.0505,
+      "step": 1120
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.26744842529296875,
+      "learning_rate": 7.800668213856897e-06,
+      "loss": 0.0438,
+      "step": 1121
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.155733123421669,
+      "learning_rate": 7.776145754677773e-06,
+      "loss": 0.0428,
+      "step": 1122
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.1795520782470703,
+      "learning_rate": 7.751648406774778e-06,
+      "loss": 0.0312,
+      "step": 1123
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.23328830301761627,
+      "learning_rate": 7.727176255304633e-06,
+      "loss": 0.0263,
+      "step": 1124
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.23346711695194244,
+      "learning_rate": 7.70272938533647e-06,
+      "loss": 0.0369,
+      "step": 1125
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.23599651455879211,
+      "learning_rate": 7.678307881851517e-06,
+      "loss": 0.0421,
+      "step": 1126
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.2010120004415512,
+      "learning_rate": 7.653911829742858e-06,
+      "loss": 0.0543,
+      "step": 1127
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.1991390883922577,
+      "learning_rate": 7.629541313815057e-06,
+      "loss": 0.0343,
+      "step": 1128
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.23358415067195892,
+      "learning_rate": 7.605196418783949e-06,
+      "loss": 0.0302,
+      "step": 1129
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.31206679344177246,
+      "learning_rate": 7.580877229276302e-06,
+      "loss": 0.0794,
+      "step": 1130
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.15948724746704102,
+      "learning_rate": 7.556583829829512e-06,
+      "loss": 0.026,
+      "step": 1131
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.22460494935512543,
+      "learning_rate": 7.532316304891327e-06,
+      "loss": 0.0351,
+      "step": 1132
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.15963564813137054,
+      "learning_rate": 7.508074738819569e-06,
+      "loss": 0.0169,
+      "step": 1133
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.13531796634197235,
+      "learning_rate": 7.483859215881799e-06,
+      "loss": 0.018,
+      "step": 1134
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.2731406092643738,
+      "learning_rate": 7.459669820255068e-06,
+      "loss": 0.0611,
+      "step": 1135
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.35660219192504883,
+      "learning_rate": 7.4355066360255916e-06,
+      "loss": 0.0503,
+      "step": 1136
+    },
+    {
+      "epoch": 2.01,
+      "eval_loss": 0.051356323063373566,
+      "eval_runtime": 14.7225,
+      "eval_samples_per_second": 32.332,
+      "eval_steps_per_second": 8.083,
+      "step": 1136
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.24530379474163055,
+      "learning_rate": 7.411369747188469e-06,
+      "loss": 0.0442,
+      "step": 1137
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.17080162465572357,
+      "learning_rate": 7.387259237647408e-06,
+      "loss": 0.0262,
+      "step": 1138
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.15803413093090057,
+      "learning_rate": 7.3631751912143954e-06,
+      "loss": 0.0164,
+      "step": 1139
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.24637369811534882,
+      "learning_rate": 7.339117691609454e-06,
+      "loss": 0.0325,
+      "step": 1140
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.35546156764030457,
+      "learning_rate": 7.3150868224603034e-06,
+      "loss": 0.0824,
+      "step": 1141
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.2265998274087906,
+      "learning_rate": 7.2910826673020954e-06,
+      "loss": 0.0265,
+      "step": 1142
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.24802450835704803,
+      "learning_rate": 7.267105309577133e-06,
+      "loss": 0.0552,
+      "step": 1143
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.372810035943985,
+      "learning_rate": 7.243154832634547e-06,
+      "loss": 0.0353,
+      "step": 1144
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.13953986763954163,
+      "learning_rate": 7.21923131973005e-06,
+      "loss": 0.017,
+      "step": 1145
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.16603930294513702,
+      "learning_rate": 7.195334854025607e-06,
+      "loss": 0.0252,
+      "step": 1146
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.2538008689880371,
+      "learning_rate": 7.171465518589153e-06,
+      "loss": 0.0337,
+      "step": 1147
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.32962384819984436,
+      "learning_rate": 7.147623396394347e-06,
+      "loss": 0.0369,
+      "step": 1148
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.27777454257011414,
+      "learning_rate": 7.123808570320212e-06,
+      "loss": 0.0389,
+      "step": 1149
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.36999934911727905,
+      "learning_rate": 7.100021123150918e-06,
+      "loss": 0.0484,
+      "step": 1150
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.23858538269996643,
+      "learning_rate": 7.076261137575441e-06,
+      "loss": 0.0303,
+      "step": 1151
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.33956798911094666,
+      "learning_rate": 7.052528696187296e-06,
+      "loss": 0.0283,
+      "step": 1152
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.18158185482025146,
+      "learning_rate": 7.028823881484269e-06,
+      "loss": 0.021,
+      "step": 1153
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.2082119882106781,
+      "learning_rate": 7.005146775868086e-06,
+      "loss": 0.0246,
+      "step": 1154
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.1464511901140213,
+      "learning_rate": 6.981497461644177e-06,
+      "loss": 0.0139,
+      "step": 1155
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.21138913929462433,
+      "learning_rate": 6.957876021021344e-06,
+      "loss": 0.048,
+      "step": 1156
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.15928184986114502,
+      "learning_rate": 6.9342825361114975e-06,
+      "loss": 0.0168,
+      "step": 1157
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.19654129445552826,
+      "learning_rate": 6.910717088929385e-06,
+      "loss": 0.0206,
+      "step": 1158
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.5633344054222107,
+      "learning_rate": 6.887179761392283e-06,
+      "loss": 0.0832,
+      "step": 1159
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.2124648541212082,
+      "learning_rate": 6.863670635319715e-06,
+      "loss": 0.0221,
+      "step": 1160
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.24031034111976624,
+      "learning_rate": 6.8401897924331705e-06,
+      "loss": 0.0263,
+      "step": 1161
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.2149808555841446,
+      "learning_rate": 6.816737314355826e-06,
+      "loss": 0.0274,
+      "step": 1162
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.28008171916007996,
+      "learning_rate": 6.793313282612257e-06,
+      "loss": 0.0323,
+      "step": 1163
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.26243600249290466,
+      "learning_rate": 6.769917778628165e-06,
+      "loss": 0.0381,
+      "step": 1164
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.2561109960079193,
+      "learning_rate": 6.746550883730068e-06,
+      "loss": 0.0246,
+      "step": 1165
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.3455231189727783,
+      "learning_rate": 6.723212679145039e-06,
+      "loss": 0.0359,
+      "step": 1166
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.260616660118103,
+      "learning_rate": 6.699903246000418e-06,
+      "loss": 0.0258,
+      "step": 1167
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.26885679364204407,
+      "learning_rate": 6.676622665323538e-06,
+      "loss": 0.0371,
+      "step": 1168
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.2691458463668823,
+      "learning_rate": 6.653371018041438e-06,
+      "loss": 0.0443,
+      "step": 1169
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.3132081627845764,
+      "learning_rate": 6.630148384980567e-06,
+      "loss": 0.0302,
+      "step": 1170
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.338286429643631,
+      "learning_rate": 6.60695484686652e-06,
+      "loss": 0.0639,
+      "step": 1171
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.25535184144973755,
+      "learning_rate": 6.5837904843237624e-06,
+      "loss": 0.0173,
+      "step": 1172
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.3836885988712311,
+      "learning_rate": 6.560655377875328e-06,
+      "loss": 0.0365,
+      "step": 1173
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.6522020101547241,
+      "learning_rate": 6.537549607942572e-06,
+      "loss": 0.0564,
+      "step": 1174
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.2585732638835907,
+      "learning_rate": 6.514473254844833e-06,
+      "loss": 0.023,
+      "step": 1175
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.4598050117492676,
+      "learning_rate": 6.491426398799232e-06,
+      "loss": 0.0789,
+      "step": 1176
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.2599335014820099,
+      "learning_rate": 6.468409119920339e-06,
+      "loss": 0.0259,
+      "step": 1177
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.3798498511314392,
+      "learning_rate": 6.445421498219905e-06,
+      "loss": 0.056,
+      "step": 1178
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.21937690675258636,
+      "learning_rate": 6.422463613606587e-06,
+      "loss": 0.0233,
+      "step": 1179
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.5562123656272888,
+      "learning_rate": 6.399535545885673e-06,
+      "loss": 0.0778,
+      "step": 1180
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.27208957076072693,
+      "learning_rate": 6.376637374758807e-06,
+      "loss": 0.0273,
+      "step": 1181
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.21259506046772003,
+      "learning_rate": 6.353769179823713e-06,
+      "loss": 0.0203,
+      "step": 1182
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.45079004764556885,
+      "learning_rate": 6.330931040573896e-06,
+      "loss": 0.0587,
+      "step": 1183
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.30730047821998596,
+      "learning_rate": 6.308123036398389e-06,
+      "loss": 0.0345,
+      "step": 1184
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.18655456602573395,
+      "learning_rate": 6.285345246581483e-06,
+      "loss": 0.0252,
+      "step": 1185
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.288705050945282,
+      "learning_rate": 6.2625977503024175e-06,
+      "loss": 0.035,
+      "step": 1186
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.3843291997909546,
+      "learning_rate": 6.239880626635151e-06,
+      "loss": 0.0594,
+      "step": 1187
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.22076481580734253,
+      "learning_rate": 6.2171939545480424e-06,
+      "loss": 0.014,
+      "step": 1188
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.22325800359249115,
+      "learning_rate": 6.194537812903599e-06,
+      "loss": 0.0367,
+      "step": 1189
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.17269223928451538,
+      "learning_rate": 6.171912280458214e-06,
+      "loss": 0.0244,
+      "step": 1190
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.2758122384548187,
+      "learning_rate": 6.1493174358618556e-06,
+      "loss": 0.0263,
+      "step": 1191
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.34955108165740967,
+      "learning_rate": 6.12675335765784e-06,
+      "loss": 0.0538,
+      "step": 1192
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.18005065619945526,
+      "learning_rate": 6.104220124282517e-06,
+      "loss": 0.0151,
+      "step": 1193
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.24025125801563263,
+      "learning_rate": 6.081717814065014e-06,
+      "loss": 0.036,
+      "step": 1194
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.13868770003318787,
+      "learning_rate": 6.0592465052269845e-06,
+      "loss": 0.0129,
+      "step": 1195
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.19439572095870972,
+      "learning_rate": 6.0368062758822885e-06,
+      "loss": 0.0168,
+      "step": 1196
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.3090403378009796,
+      "learning_rate": 6.014397204036776e-06,
+      "loss": 0.0316,
+      "step": 1197
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.30013999342918396,
+      "learning_rate": 5.992019367587968e-06,
+      "loss": 0.0257,
+      "step": 1198
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.42233145236968994,
+      "learning_rate": 5.96967284432481e-06,
+      "loss": 0.0705,
+      "step": 1199
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.28180763125419617,
+      "learning_rate": 5.94735771192741e-06,
+      "loss": 0.0454,
+      "step": 1200
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.18875925242900848,
+      "learning_rate": 5.9250740479667405e-06,
+      "loss": 0.0253,
+      "step": 1201
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.25488123297691345,
+      "learning_rate": 5.902821929904401e-06,
+      "loss": 0.0291,
+      "step": 1202
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.46456852555274963,
+      "learning_rate": 5.880601435092316e-06,
+      "loss": 0.0398,
+      "step": 1203
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.353292852640152,
+      "learning_rate": 5.858412640772488e-06,
+      "loss": 0.0254,
+      "step": 1204
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.2687720060348511,
+      "learning_rate": 5.836255624076733e-06,
+      "loss": 0.0255,
+      "step": 1205
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.3558889329433441,
+      "learning_rate": 5.814130462026388e-06,
+      "loss": 0.0502,
+      "step": 1206
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.33219629526138306,
+      "learning_rate": 5.792037231532072e-06,
+      "loss": 0.0214,
+      "step": 1207
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.21416199207305908,
+      "learning_rate": 5.769976009393396e-06,
+      "loss": 0.0189,
+      "step": 1208
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.18990960717201233,
+      "learning_rate": 5.7479468722987e-06,
+      "loss": 0.0104,
+      "step": 1209
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.5913635492324829,
+      "learning_rate": 5.725949896824805e-06,
+      "loss": 0.0624,
+      "step": 1210
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.32329440116882324,
+      "learning_rate": 5.703985159436732e-06,
+      "loss": 0.0456,
+      "step": 1211
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.318105012178421,
+      "learning_rate": 5.682052736487422e-06,
+      "loss": 0.0314,
+      "step": 1212
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.3486747443675995,
+      "learning_rate": 5.6601527042175e-06,
+      "loss": 0.0393,
+      "step": 1213
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.3139636516571045,
+      "learning_rate": 5.638285138754984e-06,
+      "loss": 0.0238,
+      "step": 1214
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.18708229064941406,
+      "learning_rate": 5.616450116115045e-06,
+      "loss": 0.0216,
+      "step": 1215
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.637169599533081,
+      "learning_rate": 5.59464771219973e-06,
+      "loss": 0.1016,
+      "step": 1216
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.21339960396289825,
+      "learning_rate": 5.572878002797686e-06,
+      "loss": 0.0255,
+      "step": 1217
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.4058300852775574,
+      "learning_rate": 5.551141063583915e-06,
+      "loss": 0.0733,
+      "step": 1218
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.5156813859939575,
+      "learning_rate": 5.529436970119501e-06,
+      "loss": 0.0519,
+      "step": 1219
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.2935110330581665,
+      "learning_rate": 5.5077657978513566e-06,
+      "loss": 0.031,
+      "step": 1220
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.49626970291137695,
+      "learning_rate": 5.486127622111967e-06,
+      "loss": 0.0498,
+      "step": 1221
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.3934653699398041,
+      "learning_rate": 5.464522518119079e-06,
+      "loss": 0.0651,
+      "step": 1222
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.2802502512931824,
+      "learning_rate": 5.44295056097551e-06,
+      "loss": 0.0284,
+      "step": 1223
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.40204301476478577,
+      "learning_rate": 5.42141182566885e-06,
+      "loss": 0.0423,
+      "step": 1224
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.5832412838935852,
+      "learning_rate": 5.3999063870711865e-06,
+      "loss": 0.0436,
+      "step": 1225
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.3613007068634033,
+      "learning_rate": 5.378434319938889e-06,
+      "loss": 0.054,
+      "step": 1226
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.13747632503509521,
+      "learning_rate": 5.356995698912287e-06,
+      "loss": 0.017,
+      "step": 1227
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.24265992641448975,
+      "learning_rate": 5.3355905985154765e-06,
+      "loss": 0.0318,
+      "step": 1228
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.36003297567367554,
+      "learning_rate": 5.314219093156027e-06,
+      "loss": 0.034,
+      "step": 1229
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.36421242356300354,
+      "learning_rate": 5.292881257124711e-06,
+      "loss": 0.0572,
+      "step": 1230
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.540324330329895,
+      "learning_rate": 5.271577164595271e-06,
+      "loss": 0.0808,
+      "step": 1231
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.5269570350646973,
+      "learning_rate": 5.250306889624147e-06,
+      "loss": 0.105,
+      "step": 1232
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.23268546164035797,
+      "learning_rate": 5.229070506150227e-06,
+      "loss": 0.0271,
+      "step": 1233
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.3838222324848175,
+      "learning_rate": 5.207868087994596e-06,
+      "loss": 0.0935,
+      "step": 1234
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.3672729432582855,
+      "learning_rate": 5.186699708860253e-06,
+      "loss": 0.0435,
+      "step": 1235
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.3133617341518402,
+      "learning_rate": 5.165565442331876e-06,
+      "loss": 0.031,
+      "step": 1236
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.42213746905326843,
+      "learning_rate": 5.144465361875574e-06,
+      "loss": 0.0705,
+      "step": 1237
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.4637577533721924,
+      "learning_rate": 5.123399540838603e-06,
+      "loss": 0.0665,
+      "step": 1238
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.11997988075017929,
+      "learning_rate": 5.102368052449145e-06,
+      "loss": 0.0175,
+      "step": 1239
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.28916677832603455,
+      "learning_rate": 5.0813709698160224e-06,
+      "loss": 0.0457,
+      "step": 1240
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.418958842754364,
+      "learning_rate": 5.060408365928459e-06,
+      "loss": 0.0518,
+      "step": 1241
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.18787431716918945,
+      "learning_rate": 5.039480313655836e-06,
+      "loss": 0.0344,
+      "step": 1242
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.280153751373291,
+      "learning_rate": 5.018586885747414e-06,
+      "loss": 0.0483,
+      "step": 1243
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.22574745118618011,
+      "learning_rate": 4.997728154832106e-06,
+      "loss": 0.0308,
+      "step": 1244
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.2624027132987976,
+      "learning_rate": 4.976904193418204e-06,
+      "loss": 0.0375,
+      "step": 1245
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.22533202171325684,
+      "learning_rate": 4.95611507389313e-06,
+      "loss": 0.0355,
+      "step": 1246
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.2835668921470642,
+      "learning_rate": 4.935360868523213e-06,
+      "loss": 0.0296,
+      "step": 1247
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.13780538737773895,
+      "learning_rate": 4.914641649453387e-06,
+      "loss": 0.0207,
+      "step": 1248
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.35207000374794006,
+      "learning_rate": 4.893957488706993e-06,
+      "loss": 0.0673,
+      "step": 1249
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.5085780620574951,
+      "learning_rate": 4.873308458185485e-06,
+      "loss": 0.0509,
+      "step": 1250
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.17727850377559662,
+      "learning_rate": 4.852694629668201e-06,
+      "loss": 0.0249,
+      "step": 1251
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.16062287986278534,
+      "learning_rate": 4.8321160748121245e-06,
+      "loss": 0.0239,
+      "step": 1252
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.2764899730682373,
+      "learning_rate": 4.811572865151605e-06,
+      "loss": 0.0481,
+      "step": 1253
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.1882406324148178,
+      "learning_rate": 4.791065072098145e-06,
+      "loss": 0.0274,
+      "step": 1254
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.14640513062477112,
+      "learning_rate": 4.770592766940116e-06,
+      "loss": 0.0168,
+      "step": 1255
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.24956367909908295,
+      "learning_rate": 4.75015602084253e-06,
+      "loss": 0.0417,
+      "step": 1256
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.1868850588798523,
+      "learning_rate": 4.729754904846806e-06,
+      "loss": 0.0199,
+      "step": 1257
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.21410945057868958,
+      "learning_rate": 4.7093894898704835e-06,
+      "loss": 0.0216,
+      "step": 1258
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.2581914961338043,
+      "learning_rate": 4.689059846707025e-06,
+      "loss": 0.0454,
+      "step": 1259
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.20490798354148865,
+      "learning_rate": 4.668766046025522e-06,
+      "loss": 0.0269,
+      "step": 1260
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.21419018507003784,
+      "learning_rate": 4.64850815837048e-06,
+      "loss": 0.0299,
+      "step": 1261
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.2651609778404236,
+      "learning_rate": 4.628286254161571e-06,
+      "loss": 0.0481,
+      "step": 1262
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.2107803374528885,
+      "learning_rate": 4.60810040369338e-06,
+      "loss": 0.0327,
+      "step": 1263
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.31488195061683655,
+      "learning_rate": 4.587950677135157e-06,
+      "loss": 0.0466,
+      "step": 1264
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.43546855449676514,
+      "learning_rate": 4.567837144530585e-06,
+      "loss": 0.0677,
+      "step": 1265
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.42556390166282654,
+      "learning_rate": 4.547759875797524e-06,
+      "loss": 0.0716,
+      "step": 1266
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.29258009791374207,
+      "learning_rate": 4.5277189407277855e-06,
+      "loss": 0.0616,
+      "step": 1267
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.44819411635398865,
+      "learning_rate": 4.50771440898688e-06,
+      "loss": 0.0375,
+      "step": 1268
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.18072977662086487,
+      "learning_rate": 4.487746350113764e-06,
+      "loss": 0.0165,
+      "step": 1269
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.2837827801704407,
+      "learning_rate": 4.467814833520613e-06,
+      "loss": 0.0471,
+      "step": 1270
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.27946364879608154,
+      "learning_rate": 4.447919928492572e-06,
+      "loss": 0.0343,
+      "step": 1271
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.12713082134723663,
+      "learning_rate": 4.428061704187525e-06,
+      "loss": 0.012,
+      "step": 1272
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.1154218316078186,
+      "learning_rate": 4.408240229635855e-06,
+      "loss": 0.0165,
+      "step": 1273
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.2695019543170929,
+      "learning_rate": 4.3884555737401705e-06,
+      "loss": 0.0488,
+      "step": 1274
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.26274654269218445,
+      "learning_rate": 4.368707805275116e-06,
+      "loss": 0.0452,
+      "step": 1275
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.1521860957145691,
+      "learning_rate": 4.348996992887107e-06,
+      "loss": 0.0085,
+      "step": 1276
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.30315443873405457,
+      "learning_rate": 4.329323205094083e-06,
+      "loss": 0.0533,
+      "step": 1277
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.2060246765613556,
+      "learning_rate": 4.309686510285296e-06,
+      "loss": 0.0221,
+      "step": 1278
+    },
+    {
+      "epoch": 2.26,
+      "eval_loss": 0.051851145923137665,
+      "eval_runtime": 14.738,
+      "eval_samples_per_second": 32.298,
+      "eval_steps_per_second": 8.074,
+      "step": 1278
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.2752631902694702,
+      "learning_rate": 4.2900869767210326e-06,
+      "loss": 0.0543,
+      "step": 1279
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.26388072967529297,
+      "learning_rate": 4.270524672532422e-06,
+      "loss": 0.0291,
+      "step": 1280
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.4224337935447693,
+      "learning_rate": 4.250999665721177e-06,
+      "loss": 0.0293,
+      "step": 1281
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.3607558310031891,
+      "learning_rate": 4.2315120241593486e-06,
+      "loss": 0.0448,
+      "step": 1282
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.5090684294700623,
+      "learning_rate": 4.212061815589103e-06,
+      "loss": 0.0541,
+      "step": 1283
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.23229840397834778,
+      "learning_rate": 4.192649107622485e-06,
+      "loss": 0.0366,
+      "step": 1284
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.3348425328731537,
+      "learning_rate": 4.173273967741184e-06,
+      "loss": 0.0556,
+      "step": 1285
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.20915712416172028,
+      "learning_rate": 4.1539364632962955e-06,
+      "loss": 0.0376,
+      "step": 1286
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.18740570545196533,
+      "learning_rate": 4.134636661508088e-06,
+      "loss": 0.0218,
+      "step": 1287
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.42705151438713074,
+      "learning_rate": 4.115374629465759e-06,
+      "loss": 0.0424,
+      "step": 1288
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.19738562405109406,
+      "learning_rate": 4.096150434127234e-06,
+      "loss": 0.0265,
+      "step": 1289
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.373444139957428,
+      "learning_rate": 4.07696414231889e-06,
+      "loss": 0.0586,
+      "step": 1290
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.26447492837905884,
+      "learning_rate": 4.0578158207353615e-06,
+      "loss": 0.0345,
+      "step": 1291
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.35752755403518677,
+      "learning_rate": 4.038705535939285e-06,
+      "loss": 0.0434,
+      "step": 1292
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.22995081543922424,
+      "learning_rate": 4.019633354361071e-06,
+      "loss": 0.0238,
+      "step": 1293
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.23020121455192566,
+      "learning_rate": 4.000599342298688e-06,
+      "loss": 0.0268,
+      "step": 1294
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.2527749240398407,
+      "learning_rate": 3.98160356591741e-06,
+      "loss": 0.0345,
+      "step": 1295
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.42879918217658997,
+      "learning_rate": 3.962646091249609e-06,
+      "loss": 0.0682,
+      "step": 1296
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.1271456778049469,
+      "learning_rate": 3.943726984194504e-06,
+      "loss": 0.0091,
+      "step": 1297
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.3033767640590668,
+      "learning_rate": 3.9248463105179434e-06,
+      "loss": 0.0546,
+      "step": 1298
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.18214237689971924,
+      "learning_rate": 3.9060041358521836e-06,
+      "loss": 0.0151,
+      "step": 1299
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.16898174583911896,
+      "learning_rate": 3.8872005256956384e-06,
+      "loss": 0.0214,
+      "step": 1300
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.28705576062202454,
+      "learning_rate": 3.868435545412683e-06,
+      "loss": 0.0347,
+      "step": 1301
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.3911001682281494,
+      "learning_rate": 3.849709260233394e-06,
+      "loss": 0.0344,
+      "step": 1302
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.417367160320282,
+      "learning_rate": 3.831021735253338e-06,
+      "loss": 0.0519,
+      "step": 1303
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.2624224126338959,
+      "learning_rate": 3.812373035433358e-06,
+      "loss": 0.0272,
+      "step": 1304
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.27091771364212036,
+      "learning_rate": 3.7937632255993176e-06,
+      "loss": 0.0319,
+      "step": 1305
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.2036266028881073,
+      "learning_rate": 3.77519237044191e-06,
+      "loss": 0.0279,
+      "step": 1306
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.3872365355491638,
+      "learning_rate": 3.7566605345164014e-06,
+      "loss": 0.022,
+      "step": 1307
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.3360632061958313,
+      "learning_rate": 3.738167782242423e-06,
+      "loss": 0.0516,
+      "step": 1308
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.3032302260398865,
+      "learning_rate": 3.7197141779037543e-06,
+      "loss": 0.0278,
+      "step": 1309
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.3070181608200073,
+      "learning_rate": 3.7012997856480797e-06,
+      "loss": 0.0408,
+      "step": 1310
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.34049341082572937,
+      "learning_rate": 3.682924669486785e-06,
+      "loss": 0.0446,
+      "step": 1311
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.27505141496658325,
+      "learning_rate": 3.664588893294721e-06,
+      "loss": 0.0331,
+      "step": 1312
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.2924882769584656,
+      "learning_rate": 3.6462925208099796e-06,
+      "loss": 0.0385,
+      "step": 1313
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.17688198387622833,
+      "learning_rate": 3.6280356156336924e-06,
+      "loss": 0.0171,
+      "step": 1314
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.24592575430870056,
+      "learning_rate": 3.6098182412297943e-06,
+      "loss": 0.0319,
+      "step": 1315
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.2453385591506958,
+      "learning_rate": 3.591640460924797e-06,
+      "loss": 0.0188,
+      "step": 1316
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.16710466146469116,
+      "learning_rate": 3.573502337907579e-06,
+      "loss": 0.0214,
+      "step": 1317
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.22846581041812897,
+      "learning_rate": 3.5554039352291633e-06,
+      "loss": 0.0213,
+      "step": 1318
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.21333551406860352,
+      "learning_rate": 3.537345315802506e-06,
+      "loss": 0.0225,
+      "step": 1319
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.1734769195318222,
+      "learning_rate": 3.51932654240227e-06,
+      "loss": 0.0212,
+      "step": 1320
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.3820168673992157,
+      "learning_rate": 3.5013476776645874e-06,
+      "loss": 0.0417,
+      "step": 1321
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.5832388401031494,
+      "learning_rate": 3.4834087840868883e-06,
+      "loss": 0.0636,
+      "step": 1322
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.34285444021224976,
+      "learning_rate": 3.4655099240276363e-06,
+      "loss": 0.0251,
+      "step": 1323
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.277201384305954,
+      "learning_rate": 3.447651159706144e-06,
+      "loss": 0.0104,
+      "step": 1324
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.36982035636901855,
+      "learning_rate": 3.4298325532023494e-06,
+      "loss": 0.0609,
+      "step": 1325
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.21037155389785767,
+      "learning_rate": 3.4120541664565693e-06,
+      "loss": 0.0178,
+      "step": 1326
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.4782731235027313,
+      "learning_rate": 3.3943160612693374e-06,
+      "loss": 0.0573,
+      "step": 1327
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.11758149415254593,
+      "learning_rate": 3.376618299301156e-06,
+      "loss": 0.009,
+      "step": 1328
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.26589435338974,
+      "learning_rate": 3.3589609420722828e-06,
+      "loss": 0.0175,
+      "step": 1329
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.317865252494812,
+      "learning_rate": 3.34134405096252e-06,
+      "loss": 0.0483,
+      "step": 1330
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.8171857595443726,
+      "learning_rate": 3.323767687211006e-06,
+      "loss": 0.0708,
+      "step": 1331
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.26556289196014404,
+      "learning_rate": 3.306231911916005e-06,
+      "loss": 0.0241,
+      "step": 1332
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.3290770351886749,
+      "learning_rate": 3.2887367860346858e-06,
+      "loss": 0.0252,
+      "step": 1333
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.24110348522663116,
+      "learning_rate": 3.271282370382912e-06,
+      "loss": 0.0213,
+      "step": 1334
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.13693028688430786,
+      "learning_rate": 3.2538687256350297e-06,
+      "loss": 0.0139,
+      "step": 1335
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.30453619360923767,
+      "learning_rate": 3.2364959123236575e-06,
+      "loss": 0.0473,
+      "step": 1336
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.3685758709907532,
+      "learning_rate": 3.2191639908394864e-06,
+      "loss": 0.0441,
+      "step": 1337
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.34061896800994873,
+      "learning_rate": 3.2018730214310594e-06,
+      "loss": 0.0299,
+      "step": 1338
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.32021400332450867,
+      "learning_rate": 3.184623064204556e-06,
+      "loss": 0.0303,
+      "step": 1339
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.40669137239456177,
+      "learning_rate": 3.1674141791235887e-06,
+      "loss": 0.0486,
+      "step": 1340
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.12852954864501953,
+      "learning_rate": 3.1502464260090145e-06,
+      "loss": 0.0103,
+      "step": 1341
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.3651524484157562,
+      "learning_rate": 3.1331198645386843e-06,
+      "loss": 0.039,
+      "step": 1342
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.29443657398223877,
+      "learning_rate": 3.116034554247285e-06,
+      "loss": 0.0286,
+      "step": 1343
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.3538154065608978,
+      "learning_rate": 3.0989905545260906e-06,
+      "loss": 0.0389,
+      "step": 1344
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.7326951622962952,
+      "learning_rate": 3.081987924622774e-06,
+      "loss": 0.0803,
+      "step": 1345
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.44199663400650024,
+      "learning_rate": 3.0650267236412107e-06,
+      "loss": 0.0721,
+      "step": 1346
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.30168449878692627,
+      "learning_rate": 3.0481070105412506e-06,
+      "loss": 0.0149,
+      "step": 1347
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.27027618885040283,
+      "learning_rate": 3.031228844138537e-06,
+      "loss": 0.0337,
+      "step": 1348
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.2874445915222168,
+      "learning_rate": 3.01439228310428e-06,
+      "loss": 0.0203,
+      "step": 1349
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.2778574824333191,
+      "learning_rate": 2.997597385965062e-06,
+      "loss": 0.0299,
+      "step": 1350
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.30845001339912415,
+      "learning_rate": 2.980844211102649e-06,
+      "loss": 0.0219,
+      "step": 1351
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.5536012053489685,
+      "learning_rate": 2.9641328167537553e-06,
+      "loss": 0.0478,
+      "step": 1352
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.35103103518486023,
+      "learning_rate": 2.947463261009875e-06,
+      "loss": 0.0331,
+      "step": 1353
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.20268794894218445,
+      "learning_rate": 2.9308356018170567e-06,
+      "loss": 0.0204,
+      "step": 1354
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.1878962516784668,
+      "learning_rate": 2.9142498969757046e-06,
+      "loss": 0.0202,
+      "step": 1355
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.30813878774642944,
+      "learning_rate": 2.8977062041404002e-06,
+      "loss": 0.0205,
+      "step": 1356
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.3760502338409424,
+      "learning_rate": 2.8812045808196625e-06,
+      "loss": 0.0653,
+      "step": 1357
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.40414708852767944,
+      "learning_rate": 2.86474508437579e-06,
+      "loss": 0.0533,
+      "step": 1358
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.11207608878612518,
+      "learning_rate": 2.848327772024628e-06,
+      "loss": 0.0059,
+      "step": 1359
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.21103401482105255,
+      "learning_rate": 2.8319527008353863e-06,
+      "loss": 0.0188,
+      "step": 1360
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.36612969636917114,
+      "learning_rate": 2.8156199277304445e-06,
+      "loss": 0.039,
+      "step": 1361
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.22014780342578888,
+      "learning_rate": 2.799329509485136e-06,
+      "loss": 0.0176,
+      "step": 1362
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.46080338954925537,
+      "learning_rate": 2.7830815027275735e-06,
+      "loss": 0.0557,
+      "step": 1363
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.34078872203826904,
+      "learning_rate": 2.7668759639384315e-06,
+      "loss": 0.0327,
+      "step": 1364
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.568625807762146,
+      "learning_rate": 2.75071294945076e-06,
+      "loss": 0.0528,
+      "step": 1365
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.26577794551849365,
+      "learning_rate": 2.7345925154497915e-06,
+      "loss": 0.0164,
+      "step": 1366
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.41297605633735657,
+      "learning_rate": 2.7185147179727477e-06,
+      "loss": 0.0477,
+      "step": 1367
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.6418724656105042,
+      "learning_rate": 2.702479612908624e-06,
+      "loss": 0.0603,
+      "step": 1368
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.2665959298610687,
+      "learning_rate": 2.6864872559980204e-06,
+      "loss": 0.0292,
+      "step": 1369
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.14747072756290436,
+      "learning_rate": 2.6705377028329314e-06,
+      "loss": 0.0106,
+      "step": 1370
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.2569366693496704,
+      "learning_rate": 2.6546310088565623e-06,
+      "loss": 0.0392,
+      "step": 1371
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.23932015895843506,
+      "learning_rate": 2.6387672293631455e-06,
+      "loss": 0.0167,
+      "step": 1372
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.2998228967189789,
+      "learning_rate": 2.622946419497706e-06,
+      "loss": 0.0346,
+      "step": 1373
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.23906637728214264,
+      "learning_rate": 2.607168634255927e-06,
+      "loss": 0.0286,
+      "step": 1374
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.37534478306770325,
+      "learning_rate": 2.5914339284839143e-06,
+      "loss": 0.0253,
+      "step": 1375
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.2315150499343872,
+      "learning_rate": 2.5757423568780323e-06,
+      "loss": 0.0212,
+      "step": 1376
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.3376396596431732,
+      "learning_rate": 2.56009397398471e-06,
+      "loss": 0.0715,
+      "step": 1377
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.3417412340641022,
+      "learning_rate": 2.54448883420022e-06,
+      "loss": 0.0512,
+      "step": 1378
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.2737486958503723,
+      "learning_rate": 2.528926991770541e-06,
+      "loss": 0.0301,
+      "step": 1379
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.31307896971702576,
+      "learning_rate": 2.513408500791135e-06,
+      "loss": 0.0349,
+      "step": 1380
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.3882101774215698,
+      "learning_rate": 2.4979334152067625e-06,
+      "loss": 0.0675,
+      "step": 1381
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.4506044387817383,
+      "learning_rate": 2.482501788811304e-06,
+      "loss": 0.0628,
+      "step": 1382
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.5073202848434448,
+      "learning_rate": 2.4671136752475647e-06,
+      "loss": 0.0833,
+      "step": 1383
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.19970950484275818,
+      "learning_rate": 2.4517691280070996e-06,
+      "loss": 0.0253,
+      "step": 1384
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.3577216863632202,
+      "learning_rate": 2.4364682004300193e-06,
+      "loss": 0.0384,
+      "step": 1385
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.2306230515241623,
+      "learning_rate": 2.4212109457048004e-06,
+      "loss": 0.0259,
+      "step": 1386
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.321592777967453,
+      "learning_rate": 2.405997416868105e-06,
+      "loss": 0.0503,
+      "step": 1387
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.2956874966621399,
+      "learning_rate": 2.3908276668045977e-06,
+      "loss": 0.0423,
+      "step": 1388
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.4138338267803192,
+      "learning_rate": 2.3757017482467686e-06,
+      "loss": 0.0479,
+      "step": 1389
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.31615543365478516,
+      "learning_rate": 2.3606197137747367e-06,
+      "loss": 0.03,
+      "step": 1390
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.16097970306873322,
+      "learning_rate": 2.345581615816072e-06,
+      "loss": 0.0171,
+      "step": 1391
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.3326115906238556,
+      "learning_rate": 2.330587506645611e-06,
+      "loss": 0.0441,
+      "step": 1392
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.3346329927444458,
+      "learning_rate": 2.3156374383852907e-06,
+      "loss": 0.0591,
+      "step": 1393
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.09273142367601395,
+      "learning_rate": 2.3007314630039384e-06,
+      "loss": 0.0072,
+      "step": 1394
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.2577742636203766,
+      "learning_rate": 2.2858696323171223e-06,
+      "loss": 0.0318,
+      "step": 1395
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.4662782549858093,
+      "learning_rate": 2.271051997986946e-06,
+      "loss": 0.0453,
+      "step": 1396
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.28764113783836365,
+      "learning_rate": 2.2562786115218815e-06,
+      "loss": 0.0268,
+      "step": 1397
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.25518426299095154,
+      "learning_rate": 2.2415495242765958e-06,
+      "loss": 0.0185,
+      "step": 1398
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.23123294115066528,
+      "learning_rate": 2.2268647874517532e-06,
+      "loss": 0.03,
+      "step": 1399
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.30051419138908386,
+      "learning_rate": 2.212224452093859e-06,
+      "loss": 0.0487,
+      "step": 1400
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.24000394344329834,
+      "learning_rate": 2.1976285690950664e-06,
+      "loss": 0.023,
+      "step": 1401
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.3042777180671692,
+      "learning_rate": 2.1830771891930003e-06,
+      "loss": 0.0541,
+      "step": 1402
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.3590790629386902,
+      "learning_rate": 2.1685703629705995e-06,
+      "loss": 0.0322,
+      "step": 1403
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.25271815061569214,
+      "learning_rate": 2.1541081408559142e-06,
+      "loss": 0.0221,
+      "step": 1404
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.21400165557861328,
+      "learning_rate": 2.1396905731219508e-06,
+      "loss": 0.02,
+      "step": 1405
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.39958655834198,
+      "learning_rate": 2.1253177098864886e-06,
+      "loss": 0.0626,
+      "step": 1406
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.17715786397457123,
+      "learning_rate": 2.1109896011118986e-06,
+      "loss": 0.0211,
+      "step": 1407
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.3341400623321533,
+      "learning_rate": 2.0967062966049943e-06,
+      "loss": 0.0452,
+      "step": 1408
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.25962328910827637,
+      "learning_rate": 2.0824678460168246e-06,
+      "loss": 0.0354,
+      "step": 1409
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.40941673517227173,
+      "learning_rate": 2.068274298842537e-06,
+      "loss": 0.0507,
+      "step": 1410
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.22385822236537933,
+      "learning_rate": 2.054125704421174e-06,
+      "loss": 0.0186,
+      "step": 1411
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.43930667638778687,
+      "learning_rate": 2.040022111935516e-06,
+      "loss": 0.0957,
+      "step": 1412
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.17197230458259583,
+      "learning_rate": 2.025963570411919e-06,
+      "loss": 0.0216,
+      "step": 1413
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.37095287442207336,
+      "learning_rate": 2.0119501287201223e-06,
+      "loss": 0.047,
+      "step": 1414
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.29766666889190674,
+      "learning_rate": 1.997981835573102e-06,
+      "loss": 0.051,
+      "step": 1415
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.39576035737991333,
+      "learning_rate": 1.9840587395268855e-06,
+      "loss": 0.0483,
+      "step": 1416
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.2802913784980774,
+      "learning_rate": 1.9701808889803842e-06,
+      "loss": 0.0255,
+      "step": 1417
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.22689564526081085,
+      "learning_rate": 1.9563483321752335e-06,
+      "loss": 0.0118,
+      "step": 1418
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.1750285029411316,
+      "learning_rate": 1.9425611171956254e-06,
+      "loss": 0.019,
+      "step": 1419
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.3820284903049469,
+      "learning_rate": 1.9288192919681276e-06,
+      "loss": 0.0267,
+      "step": 1420
+    },
+    {
+      "epoch": 2.51,
+      "eval_loss": 0.05338440090417862,
+      "eval_runtime": 14.7594,
+      "eval_samples_per_second": 32.251,
+      "eval_steps_per_second": 8.063,
+      "step": 1420
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.2997962236404419,
+      "learning_rate": 1.915122904261527e-06,
+      "loss": 0.0257,
+      "step": 1421
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.2884168326854706,
+      "learning_rate": 1.9014720016866626e-06,
+      "loss": 0.0331,
+      "step": 1422
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.1294047236442566,
+      "learning_rate": 1.8878666316962644e-06,
+      "loss": 0.0101,
+      "step": 1423
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.32022544741630554,
+      "learning_rate": 1.8743068415847875e-06,
+      "loss": 0.0385,
+      "step": 1424
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.2630491554737091,
+      "learning_rate": 1.8607926784882234e-06,
+      "loss": 0.0393,
+      "step": 1425
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.36789068579673767,
+      "learning_rate": 1.8473241893839825e-06,
+      "loss": 0.0335,
+      "step": 1426
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.40261098742485046,
+      "learning_rate": 1.8339014210906858e-06,
+      "loss": 0.0469,
+      "step": 1427
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.3258412182331085,
+      "learning_rate": 1.8205244202680322e-06,
+      "loss": 0.0209,
+      "step": 1428
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.25761544704437256,
+      "learning_rate": 1.8071932334166286e-06,
+      "loss": 0.0333,
+      "step": 1429
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.5692620873451233,
+      "learning_rate": 1.7939079068778075e-06,
+      "loss": 0.0441,
+      "step": 1430
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.3760395646095276,
+      "learning_rate": 1.7806684868335005e-06,
+      "loss": 0.0362,
+      "step": 1431
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.34493476152420044,
+      "learning_rate": 1.7674750193060586e-06,
+      "loss": 0.0483,
+      "step": 1432
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.35743704438209534,
+      "learning_rate": 1.754327550158087e-06,
+      "loss": 0.0344,
+      "step": 1433
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.3065614104270935,
+      "learning_rate": 1.7412261250923007e-06,
+      "loss": 0.0514,
+      "step": 1434
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.3237569332122803,
+      "learning_rate": 1.7281707896513476e-06,
+      "loss": 0.0568,
+      "step": 1435
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.3879953920841217,
+      "learning_rate": 1.7151615892176741e-06,
+      "loss": 0.0382,
+      "step": 1436
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.23983269929885864,
+      "learning_rate": 1.7021985690133496e-06,
+      "loss": 0.0199,
+      "step": 1437
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.3829066753387451,
+      "learning_rate": 1.6892817740999084e-06,
+      "loss": 0.0492,
+      "step": 1438
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.5782652497291565,
+      "learning_rate": 1.676411249378202e-06,
+      "loss": 0.0838,
+      "step": 1439
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.3808516263961792,
+      "learning_rate": 1.6635870395882368e-06,
+      "loss": 0.0724,
+      "step": 1440
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.5493273735046387,
+      "learning_rate": 1.6508091893090238e-06,
+      "loss": 0.0628,
+      "step": 1441
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.3587407171726227,
+      "learning_rate": 1.6380777429584254e-06,
+      "loss": 0.0733,
+      "step": 1442
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.30896487832069397,
+      "learning_rate": 1.6253927447929883e-06,
+      "loss": 0.0278,
+      "step": 1443
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.366315096616745,
+      "learning_rate": 1.612754238907797e-06,
+      "loss": 0.0397,
+      "step": 1444
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.27430596947669983,
+      "learning_rate": 1.6001622692363315e-06,
+      "loss": 0.0407,
+      "step": 1445
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.33580029010772705,
+      "learning_rate": 1.5876168795502931e-06,
+      "loss": 0.0292,
+      "step": 1446
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.2335275560617447,
+      "learning_rate": 1.5751181134594744e-06,
+      "loss": 0.0364,
+      "step": 1447
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.23250934481620789,
+      "learning_rate": 1.5626660144115873e-06,
+      "loss": 0.0286,
+      "step": 1448
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.5116204023361206,
+      "learning_rate": 1.550260625692122e-06,
+      "loss": 0.0318,
+      "step": 1449
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.2867363691329956,
+      "learning_rate": 1.5379019904242087e-06,
+      "loss": 0.0443,
+      "step": 1450
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.5294840335845947,
+      "learning_rate": 1.5255901515684373e-06,
+      "loss": 0.0747,
+      "step": 1451
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.31960994005203247,
+      "learning_rate": 1.5133251519227414e-06,
+      "loss": 0.054,
+      "step": 1452
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.16071946918964386,
+      "learning_rate": 1.501107034122225e-06,
+      "loss": 0.0162,
+      "step": 1453
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.19383801519870758,
+      "learning_rate": 1.4889358406390218e-06,
+      "loss": 0.02,
+      "step": 1454
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.23715536296367645,
+      "learning_rate": 1.4768116137821586e-06,
+      "loss": 0.022,
+      "step": 1455
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.324939101934433,
+      "learning_rate": 1.4647343956973869e-06,
+      "loss": 0.0243,
+      "step": 1456
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.18114933371543884,
+      "learning_rate": 1.4527042283670594e-06,
+      "loss": 0.0183,
+      "step": 1457
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.45289382338523865,
+      "learning_rate": 1.4407211536099662e-06,
+      "loss": 0.0732,
+      "step": 1458
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.2325807362794876,
+      "learning_rate": 1.4287852130811934e-06,
+      "loss": 0.0288,
+      "step": 1459
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.37204915285110474,
+      "learning_rate": 1.4168964482719916e-06,
+      "loss": 0.0568,
+      "step": 1460
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.4067043662071228,
+      "learning_rate": 1.4050549005096053e-06,
+      "loss": 0.037,
+      "step": 1461
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.41933298110961914,
+      "learning_rate": 1.3932606109571643e-06,
+      "loss": 0.0386,
+      "step": 1462
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.2180069088935852,
+      "learning_rate": 1.3815136206135065e-06,
+      "loss": 0.0107,
+      "step": 1463
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.44349929690361023,
+      "learning_rate": 1.3698139703130507e-06,
+      "loss": 0.0753,
+      "step": 1464
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.5247210264205933,
+      "learning_rate": 1.3581617007256647e-06,
+      "loss": 0.0342,
+      "step": 1465
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.3748161196708679,
+      "learning_rate": 1.3465568523565008e-06,
+      "loss": 0.0607,
+      "step": 1466
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.5303784608840942,
+      "learning_rate": 1.3349994655458786e-06,
+      "loss": 0.0714,
+      "step": 1467
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.5706108808517456,
+      "learning_rate": 1.3234895804691277e-06,
+      "loss": 0.0558,
+      "step": 1468
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.2966552674770355,
+      "learning_rate": 1.31202723713645e-06,
+      "loss": 0.0202,
+      "step": 1469
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.33207011222839355,
+      "learning_rate": 1.3006124753927945e-06,
+      "loss": 0.0729,
+      "step": 1470
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.37168270349502563,
+      "learning_rate": 1.2892453349177075e-06,
+      "loss": 0.0315,
+      "step": 1471
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.1958441436290741,
+      "learning_rate": 1.2779258552251826e-06,
+      "loss": 0.0164,
+      "step": 1472
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.26135876774787903,
+      "learning_rate": 1.2666540756635546e-06,
+      "loss": 0.034,
+      "step": 1473
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.15171673893928528,
+      "learning_rate": 1.2554300354153325e-06,
+      "loss": 0.0148,
+      "step": 1474
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.29382503032684326,
+      "learning_rate": 1.2442537734970843e-06,
+      "loss": 0.0253,
+      "step": 1475
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.5358589291572571,
+      "learning_rate": 1.2331253287592931e-06,
+      "loss": 0.101,
+      "step": 1476
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.43804746866226196,
+      "learning_rate": 1.2220447398862079e-06,
+      "loss": 0.0559,
+      "step": 1477
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.25578802824020386,
+      "learning_rate": 1.2110120453957407e-06,
+      "loss": 0.0389,
+      "step": 1478
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.43044573068618774,
+      "learning_rate": 1.2000272836393023e-06,
+      "loss": 0.0289,
+      "step": 1479
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.2467694878578186,
+      "learning_rate": 1.1890904928016928e-06,
+      "loss": 0.022,
+      "step": 1480
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.2924247980117798,
+      "learning_rate": 1.1782017109009458e-06,
+      "loss": 0.0324,
+      "step": 1481
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.39909517765045166,
+      "learning_rate": 1.1673609757882152e-06,
+      "loss": 0.0348,
+      "step": 1482
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.2627503573894501,
+      "learning_rate": 1.1565683251476356e-06,
+      "loss": 0.0317,
+      "step": 1483
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.4407021403312683,
+      "learning_rate": 1.1458237964961943e-06,
+      "loss": 0.0641,
+      "step": 1484
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.20949038863182068,
+      "learning_rate": 1.1351274271835948e-06,
+      "loss": 0.0228,
+      "step": 1485
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.08703193068504333,
+      "learning_rate": 1.1244792543921328e-06,
+      "loss": 0.0068,
+      "step": 1486
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.46513450145721436,
+      "learning_rate": 1.113879315136561e-06,
+      "loss": 0.0703,
+      "step": 1487
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.2758978009223938,
+      "learning_rate": 1.1033276462639736e-06,
+      "loss": 0.0223,
+      "step": 1488
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.22822964191436768,
+      "learning_rate": 1.0928242844536646e-06,
+      "loss": 0.026,
+      "step": 1489
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.1549101322889328,
+      "learning_rate": 1.0823692662170015e-06,
+      "loss": 0.0168,
+      "step": 1490
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.3770397901535034,
+      "learning_rate": 1.071962627897306e-06,
+      "loss": 0.0636,
+      "step": 1491
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.3684152364730835,
+      "learning_rate": 1.0616044056697211e-06,
+      "loss": 0.0334,
+      "step": 1492
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.21533618867397308,
+      "learning_rate": 1.051294635541089e-06,
+      "loss": 0.0218,
+      "step": 1493
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.2539013624191284,
+      "learning_rate": 1.0410333533498274e-06,
+      "loss": 0.0238,
+      "step": 1494
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.26467132568359375,
+      "learning_rate": 1.0308205947657977e-06,
+      "loss": 0.0286,
+      "step": 1495
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.3802349865436554,
+      "learning_rate": 1.0206563952901843e-06,
+      "loss": 0.074,
+      "step": 1496
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.2933976650238037,
+      "learning_rate": 1.0105407902553788e-06,
+      "loss": 0.0384,
+      "step": 1497
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.3100675940513611,
+      "learning_rate": 1.0004738148248433e-06,
+      "loss": 0.0364,
+      "step": 1498
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.2545584738254547,
+      "learning_rate": 9.904555039930019e-07,
+      "loss": 0.0245,
+      "step": 1499
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.22060710191726685,
+      "learning_rate": 9.804858925851124e-07,
+      "loss": 0.0204,
+      "step": 1500
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.5579091906547546,
+      "learning_rate": 9.70565015257136e-07,
+      "loss": 0.0838,
+      "step": 1501
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.2895505726337433,
+      "learning_rate": 9.6069290649564e-07,
+      "loss": 0.025,
+      "step": 1502
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.2577086091041565,
+      "learning_rate": 9.508696006176549e-07,
+      "loss": 0.0327,
+      "step": 1503
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.23269417881965637,
+      "learning_rate": 9.410951317705707e-07,
+      "loss": 0.0145,
+      "step": 1504
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.21622733771800995,
+      "learning_rate": 9.313695339320067e-07,
+      "loss": 0.0348,
+      "step": 1505
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.3240235447883606,
+      "learning_rate": 9.216928409097009e-07,
+      "loss": 0.027,
+      "step": 1506
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.2920982241630554,
+      "learning_rate": 9.120650863413926e-07,
+      "loss": 0.0247,
+      "step": 1507
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.2317061871290207,
+      "learning_rate": 9.024863036946973e-07,
+      "loss": 0.0253,
+      "step": 1508
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.328622043132782,
+      "learning_rate": 8.929565262670031e-07,
+      "loss": 0.0247,
+      "step": 1509
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.3147832453250885,
+      "learning_rate": 8.834757871853461e-07,
+      "loss": 0.0467,
+      "step": 1510
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.18437308073043823,
+      "learning_rate": 8.74044119406292e-07,
+      "loss": 0.0272,
+      "step": 1511
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.0949275940656662,
+      "learning_rate": 8.646615557158349e-07,
+      "loss": 0.0077,
+      "step": 1512
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.1925540417432785,
+      "learning_rate": 8.553281287292685e-07,
+      "loss": 0.0288,
+      "step": 1513
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.376468300819397,
+      "learning_rate": 8.460438708910834e-07,
+      "loss": 0.0293,
+      "step": 1514
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.32442381978034973,
+      "learning_rate": 8.368088144748515e-07,
+      "loss": 0.0477,
+      "step": 1515
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.24803201854228973,
+      "learning_rate": 8.276229915831035e-07,
+      "loss": 0.0189,
+      "step": 1516
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.30078190565109253,
+      "learning_rate": 8.18486434147242e-07,
+      "loss": 0.0484,
+      "step": 1517
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.2197287678718567,
+      "learning_rate": 8.093991739273982e-07,
+      "loss": 0.0151,
+      "step": 1518
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.43168503046035767,
+      "learning_rate": 8.003612425123519e-07,
+      "loss": 0.0733,
+      "step": 1519
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.32117024064064026,
+      "learning_rate": 7.91372671319402e-07,
+      "loss": 0.037,
+      "step": 1520
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.24488098919391632,
+      "learning_rate": 7.824334915942593e-07,
+      "loss": 0.0266,
+      "step": 1521
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.3577379286289215,
+      "learning_rate": 7.735437344109492e-07,
+      "loss": 0.044,
+      "step": 1522
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.2761387825012207,
+      "learning_rate": 7.647034306716988e-07,
+      "loss": 0.0298,
+      "step": 1523
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.2340402603149414,
+      "learning_rate": 7.559126111068132e-07,
+      "loss": 0.0151,
+      "step": 1524
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.26258549094200134,
+      "learning_rate": 7.471713062745967e-07,
+      "loss": 0.0252,
+      "step": 1525
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.33766189217567444,
+      "learning_rate": 7.384795465612243e-07,
+      "loss": 0.0591,
+      "step": 1526
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.3977791965007782,
+      "learning_rate": 7.298373621806492e-07,
+      "loss": 0.0495,
+      "step": 1527
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.49230140447616577,
+      "learning_rate": 7.21244783174494e-07,
+      "loss": 0.0393,
+      "step": 1528
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.26052895188331604,
+      "learning_rate": 7.127018394119322e-07,
+      "loss": 0.0172,
+      "step": 1529
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 0.33915144205093384,
+      "learning_rate": 7.042085605896142e-07,
+      "loss": 0.0365,
+      "step": 1530
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 0.46787312626838684,
+      "learning_rate": 6.957649762315365e-07,
+      "loss": 0.0772,
+      "step": 1531
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 0.3195519745349884,
+      "learning_rate": 6.87371115688954e-07,
+      "loss": 0.0623,
+      "step": 1532
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 0.36855751276016235,
+      "learning_rate": 6.790270081402744e-07,
+      "loss": 0.0335,
+      "step": 1533
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 0.3693338632583618,
+      "learning_rate": 6.707326825909476e-07,
+      "loss": 0.0465,
+      "step": 1534
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.1539740115404129,
+      "learning_rate": 6.624881678733852e-07,
+      "loss": 0.0152,
+      "step": 1535
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.1995919644832611,
+      "learning_rate": 6.542934926468436e-07,
+      "loss": 0.0187,
+      "step": 1536
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.18049630522727966,
+      "learning_rate": 6.461486853973264e-07,
+      "loss": 0.0151,
+      "step": 1537
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.2640184462070465,
+      "learning_rate": 6.380537744374892e-07,
+      "loss": 0.0256,
+      "step": 1538
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.1765688955783844,
+      "learning_rate": 6.300087879065397e-07,
+      "loss": 0.0135,
+      "step": 1539
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.17853586375713348,
+      "learning_rate": 6.220137537701459e-07,
+      "loss": 0.0146,
+      "step": 1540
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.41926825046539307,
+      "learning_rate": 6.140686998203266e-07,
+      "loss": 0.0405,
+      "step": 1541
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.43757468461990356,
+      "learning_rate": 6.061736536753643e-07,
+      "loss": 0.0685,
+      "step": 1542
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.20176298916339874,
+      "learning_rate": 5.983286427797058e-07,
+      "loss": 0.0215,
+      "step": 1543
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.3221362829208374,
+      "learning_rate": 5.905336944038652e-07,
+      "loss": 0.0186,
+      "step": 1544
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.3483205735683441,
+      "learning_rate": 5.827888356443362e-07,
+      "loss": 0.0614,
+      "step": 1545
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.374321311712265,
+      "learning_rate": 5.750940934234894e-07,
+      "loss": 0.0405,
+      "step": 1546
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.34720084071159363,
+      "learning_rate": 5.674494944894853e-07,
+      "loss": 0.0613,
+      "step": 1547
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.277129590511322,
+      "learning_rate": 5.598550654161705e-07,
+      "loss": 0.0175,
+      "step": 1548
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.5764405727386475,
+      "learning_rate": 5.52310832603004e-07,
+      "loss": 0.0541,
+      "step": 1549
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.41348522901535034,
+      "learning_rate": 5.448168222749467e-07,
+      "loss": 0.0372,
+      "step": 1550
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.35786178708076477,
+      "learning_rate": 5.373730604823818e-07,
+      "loss": 0.0388,
+      "step": 1551
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.48260074853897095,
+      "learning_rate": 5.299795731010193e-07,
+      "loss": 0.0548,
+      "step": 1552
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.22245685756206512,
+      "learning_rate": 5.22636385831804e-07,
+      "loss": 0.0197,
+      "step": 1553
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.27846255898475647,
+      "learning_rate": 5.153435242008386e-07,
+      "loss": 0.0198,
+      "step": 1554
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.28354784846305847,
+      "learning_rate": 5.081010135592745e-07,
+      "loss": 0.0384,
+      "step": 1555
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.20960010588169098,
+      "learning_rate": 5.009088790832445e-07,
+      "loss": 0.0255,
+      "step": 1556
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.41556763648986816,
+      "learning_rate": 4.937671457737597e-07,
+      "loss": 0.0372,
+      "step": 1557
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.43101227283477783,
+      "learning_rate": 4.8667583845663e-07,
+      "loss": 0.0554,
+      "step": 1558
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.21200565993785858,
+      "learning_rate": 4.796349817823803e-07,
+      "loss": 0.026,
+      "step": 1559
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.28807052969932556,
+      "learning_rate": 4.726446002261542e-07,
+      "loss": 0.0241,
+      "step": 1560
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.43896737694740295,
+      "learning_rate": 4.657047180876406e-07,
+      "loss": 0.0867,
+      "step": 1561
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.29302000999450684,
+      "learning_rate": 4.5881535949098553e-07,
+      "loss": 0.0213,
+      "step": 1562
+    },
+    {
+      "epoch": 2.76,
+      "eval_loss": 0.05274570360779762,
+      "eval_runtime": 14.7419,
+      "eval_samples_per_second": 32.289,
+      "eval_steps_per_second": 8.072,
+      "step": 1562
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.24197795987129211,
+      "learning_rate": 4.519765483847005e-07,
+      "loss": 0.041,
+      "step": 1563
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.29396310448646545,
+      "learning_rate": 4.4518830854159253e-07,
+      "loss": 0.0489,
+      "step": 1564
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.3782874345779419,
+      "learning_rate": 4.3845066355867104e-07,
+      "loss": 0.0802,
+      "step": 1565
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.2967289686203003,
+      "learning_rate": 4.317636368570727e-07,
+      "loss": 0.0504,
+      "step": 1566
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.36526423692703247,
+      "learning_rate": 4.251272516819732e-07,
+      "loss": 0.0494,
+      "step": 1567
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.3400168716907501,
+      "learning_rate": 4.1854153110251093e-07,
+      "loss": 0.0515,
+      "step": 1568
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.1707027405500412,
+      "learning_rate": 4.1200649801170996e-07,
+      "loss": 0.0088,
+      "step": 1569
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.2685607075691223,
+      "learning_rate": 4.0552217512639213e-07,
+      "loss": 0.0278,
+      "step": 1570
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.26374682784080505,
+      "learning_rate": 3.9908858498710176e-07,
+      "loss": 0.0281,
+      "step": 1571
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.30661627650260925,
+      "learning_rate": 3.9270574995803446e-07,
+      "loss": 0.0421,
+      "step": 1572
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.23328940570354462,
+      "learning_rate": 3.863736922269451e-07,
+      "loss": 0.0275,
+      "step": 1573
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.24918845295906067,
+      "learning_rate": 3.8009243380508485e-07,
+      "loss": 0.0294,
+      "step": 1574
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.23008981347084045,
+      "learning_rate": 3.738619965271145e-07,
+      "loss": 0.0278,
+      "step": 1575
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.42664575576782227,
+      "learning_rate": 3.676824020510328e-07,
+      "loss": 0.0465,
+      "step": 1576
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.27271386981010437,
+      "learning_rate": 3.6155367185810137e-07,
+      "loss": 0.0403,
+      "step": 1577
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.3163011372089386,
+      "learning_rate": 3.5547582725277026e-07,
+      "loss": 0.0386,
+      "step": 1578
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.306237131357193,
+      "learning_rate": 3.4944888936260246e-07,
+      "loss": 0.083,
+      "step": 1579
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.1392659693956375,
+      "learning_rate": 3.4347287913819915e-07,
+      "loss": 0.0134,
+      "step": 1580
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.41139236092567444,
+      "learning_rate": 3.3754781735313155e-07,
+      "loss": 0.0321,
+      "step": 1581
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.4805891811847687,
+      "learning_rate": 3.316737246038676e-07,
+      "loss": 0.0392,
+      "step": 1582
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.42521247267723083,
+      "learning_rate": 3.258506213096901e-07,
+      "loss": 0.0284,
+      "step": 1583
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.4072655737400055,
+      "learning_rate": 3.2007852771264555e-07,
+      "loss": 0.0609,
+      "step": 1584
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.2068723887205124,
+      "learning_rate": 3.1435746387745543e-07,
+      "loss": 0.0186,
+      "step": 1585
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.2665616273880005,
+      "learning_rate": 3.086874496914532e-07,
+      "loss": 0.038,
+      "step": 1586
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.2670504152774811,
+      "learning_rate": 3.030685048645193e-07,
+      "loss": 0.0221,
+      "step": 1587
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.1768016368150711,
+      "learning_rate": 2.975006489290094e-07,
+      "loss": 0.0131,
+      "step": 1588
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.5417407155036926,
+      "learning_rate": 2.9198390123968133e-07,
+      "loss": 0.0587,
+      "step": 1589
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.28088387846946716,
+      "learning_rate": 2.8651828097363664e-07,
+      "loss": 0.0532,
+      "step": 1590
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.6214661598205566,
+      "learning_rate": 2.8110380713024406e-07,
+      "loss": 0.0847,
+      "step": 1591
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.198442280292511,
+      "learning_rate": 2.75740498531088e-07,
+      "loss": 0.0192,
+      "step": 1592
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.38998472690582275,
+      "learning_rate": 2.704283738198882e-07,
+      "loss": 0.0421,
+      "step": 1593
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.37996906042099,
+      "learning_rate": 2.6516745146244193e-07,
+      "loss": 0.0687,
+      "step": 1594
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.1230519637465477,
+      "learning_rate": 2.599577497465605e-07,
+      "loss": 0.0104,
+      "step": 1595
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.43874484300613403,
+      "learning_rate": 2.547992867819993e-07,
+      "loss": 0.0586,
+      "step": 1596
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.32249265909194946,
+      "learning_rate": 2.496920805004044e-07,
+      "loss": 0.0304,
+      "step": 1597
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.3167158365249634,
+      "learning_rate": 2.446361486552462e-07,
+      "loss": 0.0334,
+      "step": 1598
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.3586677610874176,
+      "learning_rate": 2.396315088217543e-07,
+      "loss": 0.0583,
+      "step": 1599
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.3030400276184082,
+      "learning_rate": 2.346781783968577e-07,
+      "loss": 0.0362,
+      "step": 1600
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.10451136529445648,
+      "learning_rate": 2.2977617459912782e-07,
+      "loss": 0.0087,
+      "step": 1601
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.27317672967910767,
+      "learning_rate": 2.2492551446871735e-07,
+      "loss": 0.0257,
+      "step": 1602
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.38889819383621216,
+      "learning_rate": 2.2012621486730001e-07,
+      "loss": 0.0356,
+      "step": 1603
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.36270570755004883,
+      "learning_rate": 2.15378292478009e-07,
+      "loss": 0.0621,
+      "step": 1604
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.282429963350296,
+      "learning_rate": 2.1068176380538372e-07,
+      "loss": 0.0396,
+      "step": 1605
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.22032983601093292,
+      "learning_rate": 2.0603664517531318e-07,
+      "loss": 0.0264,
+      "step": 1606
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.26919829845428467,
+      "learning_rate": 2.0144295273497427e-07,
+      "loss": 0.0287,
+      "step": 1607
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.40903326869010925,
+      "learning_rate": 1.9690070245277858e-07,
+      "loss": 0.053,
+      "step": 1608
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.40531808137893677,
+      "learning_rate": 1.9240991011831576e-07,
+      "loss": 0.0379,
+      "step": 1609
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.24042147397994995,
+      "learning_rate": 1.8797059134230188e-07,
+      "loss": 0.0255,
+      "step": 1610
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.39015695452690125,
+      "learning_rate": 1.8358276155651943e-07,
+      "loss": 0.0294,
+      "step": 1611
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.2639334797859192,
+      "learning_rate": 1.7924643601377078e-07,
+      "loss": 0.024,
+      "step": 1612
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.28216591477394104,
+      "learning_rate": 1.7496162978781815e-07,
+      "loss": 0.0444,
+      "step": 1613
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.20502665638923645,
+      "learning_rate": 1.707283577733354e-07,
+      "loss": 0.0163,
+      "step": 1614
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.3280138075351715,
+      "learning_rate": 1.6654663468585296e-07,
+      "loss": 0.0533,
+      "step": 1615
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.4933052361011505,
+      "learning_rate": 1.6241647506171465e-07,
+      "loss": 0.0312,
+      "step": 1616
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.35983291268348694,
+      "learning_rate": 1.583378932580143e-07,
+      "loss": 0.0317,
+      "step": 1617
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.4504224359989166,
+      "learning_rate": 1.5431090345255917e-07,
+      "loss": 0.0645,
+      "step": 1618
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.3992840349674225,
+      "learning_rate": 1.5033551964381164e-07,
+      "loss": 0.0444,
+      "step": 1619
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.23205743730068207,
+      "learning_rate": 1.4641175565084263e-07,
+      "loss": 0.0312,
+      "step": 1620
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.26017385721206665,
+      "learning_rate": 1.4253962511328655e-07,
+      "loss": 0.0372,
+      "step": 1621
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.5536401271820068,
+      "learning_rate": 1.3871914149128805e-07,
+      "loss": 0.0835,
+      "step": 1622
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.26106053590774536,
+      "learning_rate": 1.3495031806546044e-07,
+      "loss": 0.0339,
+      "step": 1623
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.2727290987968445,
+      "learning_rate": 1.3123316793684225e-07,
+      "loss": 0.0343,
+      "step": 1624
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.4134365916252136,
+      "learning_rate": 1.275677040268408e-07,
+      "loss": 0.0455,
+      "step": 1625
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.2648979425430298,
+      "learning_rate": 1.239539390771971e-07,
+      "loss": 0.016,
+      "step": 1626
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.3541440963745117,
+      "learning_rate": 1.2039188564994086e-07,
+      "loss": 0.0389,
+      "step": 1627
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.3559698462486267,
+      "learning_rate": 1.1688155612733565e-07,
+      "loss": 0.0962,
+      "step": 1628
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.23413486778736115,
+      "learning_rate": 1.1342296271185381e-07,
+      "loss": 0.0229,
+      "step": 1629
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.2433459460735321,
+      "learning_rate": 1.1001611742611828e-07,
+      "loss": 0.0309,
+      "step": 1630
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.32962751388549805,
+      "learning_rate": 1.0666103211287248e-07,
+      "loss": 0.0552,
+      "step": 1631
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.26688799262046814,
+      "learning_rate": 1.0335771843493047e-07,
+      "loss": 0.0232,
+      "step": 1632
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.35954660177230835,
+      "learning_rate": 1.0010618787513859e-07,
+      "loss": 0.0376,
+      "step": 1633
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.3948466181755066,
+      "learning_rate": 9.69064517363405e-08,
+      "loss": 0.0285,
+      "step": 1634
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.23170126974582672,
+      "learning_rate": 9.375852114133221e-08,
+      "loss": 0.0197,
+      "step": 1635
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.14744599163532257,
+      "learning_rate": 9.066240703282547e-08,
+      "loss": 0.0114,
+      "step": 1636
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.2714551091194153,
+      "learning_rate": 8.761812017340776e-08,
+      "loss": 0.0417,
+      "step": 1637
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.16728386282920837,
+      "learning_rate": 8.462567114551067e-08,
+      "loss": 0.0135,
+      "step": 1638
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.15777331590652466,
+      "learning_rate": 8.168507035136663e-08,
+      "loss": 0.0121,
+      "step": 1639
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.6663769483566284,
+      "learning_rate": 7.879632801297387e-08,
+      "loss": 0.0544,
+      "step": 1640
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.435634046792984,
+      "learning_rate": 7.595945417206818e-08,
+      "loss": 0.0585,
+      "step": 1641
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.398020476102829,
+      "learning_rate": 7.317445869007288e-08,
+      "loss": 0.0423,
+      "step": 1642
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.21313266456127167,
+      "learning_rate": 7.044135124808227e-08,
+      "loss": 0.0168,
+      "step": 1643
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.3427776098251343,
+      "learning_rate": 6.776014134681652e-08,
+      "loss": 0.0464,
+      "step": 1644
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.16763639450073242,
+      "learning_rate": 6.513083830659016e-08,
+      "loss": 0.0103,
+      "step": 1645
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.28876662254333496,
+      "learning_rate": 6.255345126728373e-08,
+      "loss": 0.0305,
+      "step": 1646
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.19464822113513947,
+      "learning_rate": 6.002798918830543e-08,
+      "loss": 0.03,
+      "step": 1647
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.39675602316856384,
+      "learning_rate": 5.755446084856786e-08,
+      "loss": 0.0424,
+      "step": 1648
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.42407289147377014,
+      "learning_rate": 5.513287484645302e-08,
+      "loss": 0.0288,
+      "step": 1649
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.3897646367549896,
+      "learning_rate": 5.2763239599782357e-08,
+      "loss": 0.0441,
+      "step": 1650
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.39897575974464417,
+      "learning_rate": 5.044556334578676e-08,
+      "loss": 0.0571,
+      "step": 1651
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.5682809352874756,
+      "learning_rate": 4.8179854141079947e-08,
+      "loss": 0.0417,
+      "step": 1652
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.20976997911930084,
+      "learning_rate": 4.5966119861635125e-08,
+      "loss": 0.0354,
+      "step": 1653
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.3976650834083557,
+      "learning_rate": 4.3804368202741695e-08,
+      "loss": 0.0581,
+      "step": 1654
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.16423006355762482,
+      "learning_rate": 4.169460667900027e-08,
+      "loss": 0.0167,
+      "step": 1655
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.3421482741832733,
+      "learning_rate": 3.963684262427769e-08,
+      "loss": 0.0325,
+      "step": 1656
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.43279898166656494,
+      "learning_rate": 3.7631083191692064e-08,
+      "loss": 0.035,
+      "step": 1657
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.16776159405708313,
+      "learning_rate": 3.567733535358608e-08,
+      "loss": 0.0103,
+      "step": 1658
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.2130642831325531,
+      "learning_rate": 3.3775605901498755e-08,
+      "loss": 0.02,
+      "step": 1659
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.28084850311279297,
+      "learning_rate": 3.1925901446148707e-08,
+      "loss": 0.0459,
+      "step": 1660
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.38052040338516235,
+      "learning_rate": 3.012822841740426e-08,
+      "loss": 0.0618,
+      "step": 1661
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.20792850852012634,
+      "learning_rate": 2.838259306426505e-08,
+      "loss": 0.0174,
+      "step": 1662
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.3616114854812622,
+      "learning_rate": 2.6689001454840432e-08,
+      "loss": 0.0327,
+      "step": 1663
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.5600488781929016,
+      "learning_rate": 2.5047459476324475e-08,
+      "loss": 0.0754,
+      "step": 1664
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.3766483962535858,
+      "learning_rate": 2.34579728349843e-08,
+      "loss": 0.0417,
+      "step": 1665
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.1472347378730774,
+      "learning_rate": 2.192054705612678e-08,
+      "loss": 0.011,
+      "step": 1666
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.14081446826457977,
+      "learning_rate": 2.04351874840919e-08,
+      "loss": 0.0086,
+      "step": 1667
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.11916008591651917,
+      "learning_rate": 1.900189928222773e-08,
+      "loss": 0.0109,
+      "step": 1668
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.20517493784427643,
+      "learning_rate": 1.7620687432873818e-08,
+      "loss": 0.021,
+      "step": 1669
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.2512517273426056,
+      "learning_rate": 1.6291556737344505e-08,
+      "loss": 0.0216,
+      "step": 1670
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.49627548456192017,
+      "learning_rate": 1.501451181591229e-08,
+      "loss": 0.0731,
+      "step": 1671
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.5768187642097473,
+      "learning_rate": 1.3789557107787842e-08,
+      "loss": 0.0499,
+      "step": 1672
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.2951315939426422,
+      "learning_rate": 1.2616696871111666e-08,
+      "loss": 0.0371,
+      "step": 1673
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.30968159437179565,
+      "learning_rate": 1.149593518293246e-08,
+      "loss": 0.0331,
+      "step": 1674
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.3872181177139282,
+      "learning_rate": 1.0427275939200453e-08,
+      "loss": 0.0655,
+      "step": 1675
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.23709285259246826,
+      "learning_rate": 9.410722854744092e-09,
+      "loss": 0.0346,
+      "step": 1676
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.47526708245277405,
+      "learning_rate": 8.446279463266704e-09,
+      "loss": 0.0393,
+      "step": 1677
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.1921570599079132,
+      "learning_rate": 7.533949117328187e-09,
+      "loss": 0.0182,
+      "step": 1678
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.32638993859291077,
+      "learning_rate": 6.673734988333347e-09,
+      "loss": 0.0559,
+      "step": 1679
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.27686092257499695,
+      "learning_rate": 5.865640066525235e-09,
+      "loss": 0.0469,
+      "step": 1680
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.3328918218612671,
+      "learning_rate": 5.109667160973497e-09,
+      "loss": 0.0531,
+      "step": 1681
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.26992952823638916,
+      "learning_rate": 4.405818899557712e-09,
+      "loss": 0.0406,
+      "step": 1682
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.42099204659461975,
+      "learning_rate": 3.754097728970729e-09,
+      "loss": 0.0448,
+      "step": 1683
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.5060787200927734,
+      "learning_rate": 3.1545059146970147e-09,
+      "loss": 0.052,
+      "step": 1684
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.3745463788509369,
+      "learning_rate": 2.607045541015984e-09,
+      "loss": 0.0582,
+      "step": 1685
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.2121371030807495,
+      "learning_rate": 2.11171851098535e-09,
+      "loss": 0.0221,
+      "step": 1686
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.17241089046001434,
+      "learning_rate": 1.6685265464444489e-09,
+      "loss": 0.012,
+      "step": 1687
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.32290107011795044,
+      "learning_rate": 1.277471187997592e-09,
+      "loss": 0.0624,
+      "step": 1688
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.3098170757293701,
+      "learning_rate": 9.385537950173938e-10,
+      "loss": 0.0414,
+      "step": 1689
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.4063720107078552,
+      "learning_rate": 6.517755456331154e-10,
+      "loss": 0.0508,
+      "step": 1690
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.3056665360927582,
+      "learning_rate": 4.1713743673232886e-10,
+      "loss": 0.031,
+      "step": 1691
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.26085221767425537,
+      "learning_rate": 2.346402839559225e-10,
+      "loss": 0.0238,
+      "step": 1692
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.48377731442451477,
+      "learning_rate": 1.0428472169310422e-10,
+      "loss": 0.075,
+      "step": 1693
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.42343586683273315,
+      "learning_rate": 2.6071203079736094e-11,
+      "loss": 0.0849,
+      "step": 1694
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.3969852924346924,
+      "learning_rate": 0.0,
+      "loss": 0.0571,
+      "step": 1695
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 1695,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 565,
+  "total_flos": 1.582741075221545e+17,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}