diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,7101 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.5861574323112784,
+  "eval_steps": 1000,
+  "global_step": 10000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 10.134367942810059,
+      "learning_rate": 9e-08,
+      "loss": 1.2511,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 18.544782638549805,
+      "learning_rate": 1.8e-07,
+      "loss": 1.4001,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.087899684906006,
+      "learning_rate": 2.8e-07,
+      "loss": 1.1501,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.356815814971924,
+      "learning_rate": 3.7999999999999996e-07,
+      "loss": 1.2225,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.9461255073547363,
+      "learning_rate": 4.8e-07,
+      "loss": 1.2241,
+      "step": 50
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.9660773277282715,
+      "learning_rate": 5.8e-07,
+      "loss": 1.2118,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.57827615737915,
+      "learning_rate": 6.800000000000001e-07,
+      "loss": 1.0416,
+      "step": 70
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.070159435272217,
+      "learning_rate": 7.799999999999999e-07,
+      "loss": 1.1224,
+      "step": 80
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.661293029785156,
+      "learning_rate": 8.799999999999999e-07,
+      "loss": 1.0603,
+      "step": 90
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.765249252319336,
+      "learning_rate": 9.8e-07,
+      "loss": 1.0449,
+      "step": 100
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.521543502807617,
+      "learning_rate": 9.991919191919192e-07,
+      "loss": 0.9193,
+      "step": 110
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.420640468597412,
+      "learning_rate": 9.98181818181818e-07,
+      "loss": 0.876,
+      "step": 120
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.311656951904297,
+      "learning_rate": 9.97171717171717e-07,
+      "loss": 0.8012,
+      "step": 130
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.07025146484375,
+      "learning_rate": 9.961616161616162e-07,
+      "loss": 0.826,
+      "step": 140
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.859681129455566,
+      "learning_rate": 9.951515151515151e-07,
+      "loss": 0.8442,
+      "step": 150
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.192190170288086,
+      "learning_rate": 9.94141414141414e-07,
+      "loss": 0.708,
+      "step": 160
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.852901935577393,
+      "learning_rate": 9.93131313131313e-07,
+      "loss": 0.8201,
+      "step": 170
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.076972484588623,
+      "learning_rate": 9.92121212121212e-07,
+      "loss": 0.7973,
+      "step": 180
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.36331844329834,
+      "learning_rate": 9.911111111111111e-07,
+      "loss": 0.753,
+      "step": 190
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.03492546081543,
+      "learning_rate": 9.9010101010101e-07,
+      "loss": 0.7909,
+      "step": 200
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.043544769287109,
+      "learning_rate": 9.89090909090909e-07,
+      "loss": 0.7118,
+      "step": 210
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.2020692825317383,
+      "learning_rate": 9.880808080808082e-07,
+      "loss": 0.6516,
+      "step": 220
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.851771354675293,
+      "learning_rate": 9.870707070707071e-07,
+      "loss": 0.7275,
+      "step": 230
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.2986488342285156,
+      "learning_rate": 9.86060606060606e-07,
+      "loss": 0.6484,
+      "step": 240
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.700368404388428,
+      "learning_rate": 9.85050505050505e-07,
+      "loss": 0.7258,
+      "step": 250
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 6.46458625793457,
+      "learning_rate": 9.84040404040404e-07,
+      "loss": 0.6416,
+      "step": 260
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.012514114379883,
+      "learning_rate": 9.830303030303029e-07,
+      "loss": 0.5861,
+      "step": 270
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.328000068664551,
+      "learning_rate": 9.82020202020202e-07,
+      "loss": 0.6239,
+      "step": 280
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.837636947631836,
+      "learning_rate": 9.81010101010101e-07,
+      "loss": 0.6718,
+      "step": 290
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.840264797210693,
+      "learning_rate": 9.8e-07,
+      "loss": 0.6623,
+      "step": 300
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.741188049316406,
+      "learning_rate": 9.789898989898989e-07,
+      "loss": 0.6066,
+      "step": 310
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.170821666717529,
+      "learning_rate": 9.779797979797978e-07,
+      "loss": 0.6634,
+      "step": 320
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 8.481480598449707,
+      "learning_rate": 9.76969696969697e-07,
+      "loss": 0.6474,
+      "step": 330
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.698042631149292,
+      "learning_rate": 9.75959595959596e-07,
+      "loss": 0.6594,
+      "step": 340
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.990598201751709,
+      "learning_rate": 9.749494949494949e-07,
+      "loss": 0.6588,
+      "step": 350
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.2625105381011963,
+      "learning_rate": 9.73939393939394e-07,
+      "loss": 0.6532,
+      "step": 360
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.604375839233398,
+      "learning_rate": 9.72929292929293e-07,
+      "loss": 0.7057,
+      "step": 370
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.352667808532715,
+      "learning_rate": 9.71919191919192e-07,
+      "loss": 0.6273,
+      "step": 380
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.8649349212646484,
+      "learning_rate": 9.709090909090909e-07,
+      "loss": 0.6366,
+      "step": 390
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.2739667892456055,
+      "learning_rate": 9.698989898989898e-07,
+      "loss": 0.6433,
+      "step": 400
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.892832279205322,
+      "learning_rate": 9.68888888888889e-07,
+      "loss": 0.6504,
+      "step": 410
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.36677360534668,
+      "learning_rate": 9.67878787878788e-07,
+      "loss": 0.6661,
+      "step": 420
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.6905229091644287,
+      "learning_rate": 9.668686868686868e-07,
+      "loss": 0.6334,
+      "step": 430
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.99962854385376,
+      "learning_rate": 9.658585858585858e-07,
+      "loss": 0.6028,
+      "step": 440
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.522456169128418,
+      "learning_rate": 9.648484848484847e-07,
+      "loss": 0.6903,
+      "step": 450
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 7.456121921539307,
+      "learning_rate": 9.638383838383839e-07,
+      "loss": 0.6486,
+      "step": 460
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.989662170410156,
+      "learning_rate": 9.628282828282828e-07,
+      "loss": 0.7085,
+      "step": 470
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 3.5824246406555176,
+      "learning_rate": 9.618181818181818e-07,
+      "loss": 0.6652,
+      "step": 480
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.074733018875122,
+      "learning_rate": 9.608080808080807e-07,
+      "loss": 0.5733,
+      "step": 490
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.964071750640869,
+      "learning_rate": 9.597979797979797e-07,
+      "loss": 0.6319,
+      "step": 500
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.8981716632843018,
+      "learning_rate": 9.587878787878786e-07,
+      "loss": 0.5937,
+      "step": 510
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 7.471535682678223,
+      "learning_rate": 9.577777777777778e-07,
+      "loss": 0.6564,
+      "step": 520
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.468303680419922,
+      "learning_rate": 9.567676767676767e-07,
+      "loss": 0.6339,
+      "step": 530
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 3.9055886268615723,
+      "learning_rate": 9.557575757575759e-07,
+      "loss": 0.7103,
+      "step": 540
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.115898132324219,
+      "learning_rate": 9.547474747474748e-07,
+      "loss": 0.6928,
+      "step": 550
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.248528003692627,
+      "learning_rate": 9.537373737373737e-07,
+      "loss": 0.6084,
+      "step": 560
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.727003335952759,
+      "learning_rate": 9.527272727272727e-07,
+      "loss": 0.6554,
+      "step": 570
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.747541904449463,
+      "learning_rate": 9.517171717171717e-07,
+      "loss": 0.5876,
+      "step": 580
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.913226842880249,
+      "learning_rate": 9.507070707070707e-07,
+      "loss": 0.5888,
+      "step": 590
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.400748252868652,
+      "learning_rate": 9.496969696969696e-07,
+      "loss": 0.6196,
+      "step": 600
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 3.053006649017334,
+      "learning_rate": 9.486868686868687e-07,
+      "loss": 0.6356,
+      "step": 610
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.207491874694824,
+      "learning_rate": 9.476767676767676e-07,
+      "loss": 0.6161,
+      "step": 620
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.389359951019287,
+      "learning_rate": 9.466666666666666e-07,
+      "loss": 0.6172,
+      "step": 630
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.967292308807373,
+      "learning_rate": 9.456565656565656e-07,
+      "loss": 0.6173,
+      "step": 640
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.855407953262329,
+      "learning_rate": 9.446464646464646e-07,
+      "loss": 0.585,
+      "step": 650
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.580738067626953,
+      "learning_rate": 9.436363636363636e-07,
+      "loss": 0.6166,
+      "step": 660
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.6277294158935547,
+      "learning_rate": 9.426262626262626e-07,
+      "loss": 0.5951,
+      "step": 670
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.7501285076141357,
+      "learning_rate": 9.416161616161615e-07,
+      "loss": 0.5807,
+      "step": 680
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.80528450012207,
+      "learning_rate": 9.406060606060605e-07,
+      "loss": 0.6053,
+      "step": 690
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.07316780090332,
+      "learning_rate": 9.395959595959596e-07,
+      "loss": 0.626,
+      "step": 700
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.338922023773193,
+      "learning_rate": 9.385858585858585e-07,
+      "loss": 0.5356,
+      "step": 710
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 3.264220714569092,
+      "learning_rate": 9.375757575757576e-07,
+      "loss": 0.5929,
+      "step": 720
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 3.4565188884735107,
+      "learning_rate": 9.365656565656565e-07,
+      "loss": 0.6379,
+      "step": 730
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 4.202028751373291,
+      "learning_rate": 9.355555555555556e-07,
+      "loss": 0.6143,
+      "step": 740
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.128079891204834,
+      "learning_rate": 9.345454545454545e-07,
+      "loss": 0.5531,
+      "step": 750
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 4.205199718475342,
+      "learning_rate": 9.335353535353535e-07,
+      "loss": 0.5842,
+      "step": 760
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.925571918487549,
+      "learning_rate": 9.325252525252525e-07,
+      "loss": 0.6169,
+      "step": 770
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.256628036499023,
+      "learning_rate": 9.315151515151515e-07,
+      "loss": 0.5899,
+      "step": 780
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 3.2796261310577393,
+      "learning_rate": 9.305050505050504e-07,
+      "loss": 0.6654,
+      "step": 790
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.6715095043182373,
+      "learning_rate": 9.294949494949495e-07,
+      "loss": 0.6272,
+      "step": 800
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.043895721435547,
+      "learning_rate": 9.284848484848484e-07,
+      "loss": 0.6019,
+      "step": 810
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 4.5200300216674805,
+      "learning_rate": 9.274747474747475e-07,
+      "loss": 0.6992,
+      "step": 820
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.557192325592041,
+      "learning_rate": 9.264646464646464e-07,
+      "loss": 0.6657,
+      "step": 830
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 3.2051031589508057,
+      "learning_rate": 9.254545454545453e-07,
+      "loss": 0.6262,
+      "step": 840
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 3.465202808380127,
+      "learning_rate": 9.244444444444444e-07,
+      "loss": 0.6603,
+      "step": 850
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.039762496948242,
+      "learning_rate": 9.234343434343433e-07,
+      "loss": 0.6276,
+      "step": 860
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.317465305328369,
+      "learning_rate": 9.224242424242423e-07,
+      "loss": 0.521,
+      "step": 870
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 6.985171318054199,
+      "learning_rate": 9.214141414141414e-07,
+      "loss": 0.6768,
+      "step": 880
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 3.9759483337402344,
+      "learning_rate": 9.204040404040404e-07,
+      "loss": 0.5982,
+      "step": 890
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 4.207100868225098,
+      "learning_rate": 9.193939393939394e-07,
+      "loss": 0.6117,
+      "step": 900
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 4.389718055725098,
+      "learning_rate": 9.183838383838384e-07,
+      "loss": 0.5767,
+      "step": 910
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 3.659482717514038,
+      "learning_rate": 9.173737373737373e-07,
+      "loss": 0.5957,
+      "step": 920
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 6.660622596740723,
+      "learning_rate": 9.163636363636364e-07,
+      "loss": 0.6403,
+      "step": 930
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.03493070602417,
+      "learning_rate": 9.153535353535353e-07,
+      "loss": 0.5857,
+      "step": 940
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 2.699523687362671,
+      "learning_rate": 9.143434343434343e-07,
+      "loss": 0.5913,
+      "step": 950
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 3.8991754055023193,
+      "learning_rate": 9.133333333333333e-07,
+      "loss": 0.6342,
+      "step": 960
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 4.422178268432617,
+      "learning_rate": 9.123232323232323e-07,
+      "loss": 0.6286,
+      "step": 970
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 4.381803512573242,
+      "learning_rate": 9.113131313131313e-07,
+      "loss": 0.6899,
+      "step": 980
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 4.75442361831665,
+      "learning_rate": 9.103030303030302e-07,
+      "loss": 0.5462,
+      "step": 990
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 3.6305062770843506,
+      "learning_rate": 9.092929292929292e-07,
+      "loss": 0.666,
+      "step": 1000
+    },
+    {
+      "epoch": 0.36,
+      "eval_loss": 0.772658109664917,
+      "eval_runtime": 400.6004,
+      "eval_samples_per_second": 2.496,
+      "eval_steps_per_second": 2.496,
+      "step": 1000
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 4.022682189941406,
+      "learning_rate": 9.082828282828282e-07,
+      "loss": 0.5796,
+      "step": 1010
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.148658752441406,
+      "learning_rate": 9.072727272727272e-07,
+      "loss": 0.5738,
+      "step": 1020
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 3.78167462348938,
+      "learning_rate": 9.062626262626261e-07,
+      "loss": 0.6572,
+      "step": 1030
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 4.6793212890625,
+      "learning_rate": 9.052525252525252e-07,
+      "loss": 0.6466,
+      "step": 1040
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 4.607004165649414,
+      "learning_rate": 9.042424242424242e-07,
+      "loss": 0.5599,
+      "step": 1050
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.730196952819824,
+      "learning_rate": 9.032323232323233e-07,
+      "loss": 0.6559,
+      "step": 1060
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 4.408863544464111,
+      "learning_rate": 9.022222222222222e-07,
+      "loss": 0.641,
+      "step": 1070
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 3.0303945541381836,
+      "learning_rate": 9.012121212121212e-07,
+      "loss": 0.5455,
+      "step": 1080
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 5.44600248336792,
+      "learning_rate": 9.002020202020202e-07,
+      "loss": 0.5672,
+      "step": 1090
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 4.877390384674072,
+      "learning_rate": 8.991919191919192e-07,
+      "loss": 0.5626,
+      "step": 1100
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 4.949882507324219,
+      "learning_rate": 8.981818181818181e-07,
+      "loss": 0.6092,
+      "step": 1110
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 4.810749530792236,
+      "learning_rate": 8.971717171717172e-07,
+      "loss": 0.6486,
+      "step": 1120
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 4.6897053718566895,
+      "learning_rate": 8.961616161616161e-07,
+      "loss": 0.5854,
+      "step": 1130
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 3.393486261367798,
+      "learning_rate": 8.951515151515151e-07,
+      "loss": 0.6071,
+      "step": 1140
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 3.1970651149749756,
+      "learning_rate": 8.941414141414141e-07,
+      "loss": 0.601,
+      "step": 1150
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 4.972078800201416,
+      "learning_rate": 8.93131313131313e-07,
+      "loss": 0.6291,
+      "step": 1160
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 5.2576003074646,
+      "learning_rate": 8.921212121212121e-07,
+      "loss": 0.5856,
+      "step": 1170
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 3.552098035812378,
+      "learning_rate": 8.91111111111111e-07,
+      "loss": 0.5712,
+      "step": 1180
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 4.2522382736206055,
+      "learning_rate": 8.9010101010101e-07,
+      "loss": 0.6128,
+      "step": 1190
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 3.3408310413360596,
+      "learning_rate": 8.89090909090909e-07,
+      "loss": 0.6636,
+      "step": 1200
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 3.3888745307922363,
+      "learning_rate": 8.88080808080808e-07,
+      "loss": 0.5906,
+      "step": 1210
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 4.374114990234375,
+      "learning_rate": 8.870707070707071e-07,
+      "loss": 0.5863,
+      "step": 1220
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 3.318901538848877,
+      "learning_rate": 8.860606060606061e-07,
+      "loss": 0.6232,
+      "step": 1230
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 3.4461617469787598,
+      "learning_rate": 8.85050505050505e-07,
+      "loss": 0.6712,
+      "step": 1240
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 3.6037635803222656,
+      "learning_rate": 8.840404040404041e-07,
+      "loss": 0.5787,
+      "step": 1250
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 4.3236002922058105,
+      "learning_rate": 8.83030303030303e-07,
+      "loss": 0.6723,
+      "step": 1260
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 3.199084997177124,
+      "learning_rate": 8.820202020202019e-07,
+      "loss": 0.5484,
+      "step": 1270
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 3.675905704498291,
+      "learning_rate": 8.81010101010101e-07,
+      "loss": 0.6713,
+      "step": 1280
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 4.667660713195801,
+      "learning_rate": 8.799999999999999e-07,
+      "loss": 0.6161,
+      "step": 1290
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.268735408782959,
+      "learning_rate": 8.78989898989899e-07,
+      "loss": 0.6282,
+      "step": 1300
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 3.6777164936065674,
+      "learning_rate": 8.779797979797979e-07,
+      "loss": 0.662,
+      "step": 1310
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.912172317504883,
+      "learning_rate": 8.769696969696969e-07,
+      "loss": 0.5951,
+      "step": 1320
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.206521034240723,
+      "learning_rate": 8.759595959595959e-07,
+      "loss": 0.704,
+      "step": 1330
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.009490013122559,
+      "learning_rate": 8.749494949494949e-07,
+      "loss": 0.652,
+      "step": 1340
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 3.8058555126190186,
+      "learning_rate": 8.739393939393938e-07,
+      "loss": 0.5743,
+      "step": 1350
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 8.637187957763672,
+      "learning_rate": 8.729292929292929e-07,
+      "loss": 0.6198,
+      "step": 1360
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.059712886810303,
+      "learning_rate": 8.719191919191918e-07,
+      "loss": 0.6171,
+      "step": 1370
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.403115272521973,
+      "learning_rate": 8.709090909090909e-07,
+      "loss": 0.5956,
+      "step": 1380
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 4.084959983825684,
+      "learning_rate": 8.698989898989899e-07,
+      "loss": 0.6023,
+      "step": 1390
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 4.863948345184326,
+      "learning_rate": 8.688888888888889e-07,
+      "loss": 0.6295,
+      "step": 1400
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 5.363975524902344,
+      "learning_rate": 8.678787878787879e-07,
+      "loss": 0.5854,
+      "step": 1410
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 4.450655460357666,
+      "learning_rate": 8.668686868686868e-07,
+      "loss": 0.6579,
+      "step": 1420
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 5.538711071014404,
+      "learning_rate": 8.658585858585859e-07,
+      "loss": 0.726,
+      "step": 1430
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 3.3089439868927,
+      "learning_rate": 8.648484848484848e-07,
+      "loss": 0.6435,
+      "step": 1440
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 3.613719940185547,
+      "learning_rate": 8.638383838383838e-07,
+      "loss": 0.6695,
+      "step": 1450
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 2.9638564586639404,
+      "learning_rate": 8.628282828282828e-07,
+      "loss": 0.5516,
+      "step": 1460
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 3.8651678562164307,
+      "learning_rate": 8.618181818181818e-07,
+      "loss": 0.5883,
+      "step": 1470
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 8.56436824798584,
+      "learning_rate": 8.608080808080807e-07,
+      "loss": 0.6476,
+      "step": 1480
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 6.782663345336914,
+      "learning_rate": 8.597979797979798e-07,
+      "loss": 0.6369,
+      "step": 1490
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 3.181833267211914,
+      "learning_rate": 8.587878787878787e-07,
+      "loss": 0.5885,
+      "step": 1500
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 5.068531513214111,
+      "learning_rate": 8.577777777777777e-07,
+      "loss": 0.6241,
+      "step": 1510
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 5.7847900390625,
+      "learning_rate": 8.567676767676767e-07,
+      "loss": 0.623,
+      "step": 1520
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 3.325660228729248,
+      "learning_rate": 8.557575757575757e-07,
+      "loss": 0.574,
+      "step": 1530
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 3.452211618423462,
+      "learning_rate": 8.547474747474747e-07,
+      "loss": 0.6359,
+      "step": 1540
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 5.663365364074707,
+      "learning_rate": 8.537373737373736e-07,
+      "loss": 0.6132,
+      "step": 1550
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 3.970423460006714,
+      "learning_rate": 8.527272727272727e-07,
+      "loss": 0.5644,
+      "step": 1560
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 4.623054027557373,
+      "learning_rate": 8.517171717171717e-07,
+      "loss": 0.5753,
+      "step": 1570
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 4.318519592285156,
+      "learning_rate": 8.507070707070707e-07,
+      "loss": 0.6335,
+      "step": 1580
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 3.2903385162353516,
+      "learning_rate": 8.496969696969697e-07,
+      "loss": 0.5899,
+      "step": 1590
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 2.847003221511841,
+      "learning_rate": 8.486868686868687e-07,
+      "loss": 0.5834,
+      "step": 1600
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 5.138712406158447,
+      "learning_rate": 8.476767676767676e-07,
+      "loss": 0.5485,
+      "step": 1610
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 5.145106792449951,
+      "learning_rate": 8.466666666666667e-07,
+      "loss": 0.6854,
+      "step": 1620
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 5.585366725921631,
+      "learning_rate": 8.456565656565656e-07,
+      "loss": 0.5595,
+      "step": 1630
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 3.857548952102661,
+      "learning_rate": 8.446464646464646e-07,
+      "loss": 0.5856,
+      "step": 1640
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 3.835942029953003,
+      "learning_rate": 8.436363636363636e-07,
+      "loss": 0.5589,
+      "step": 1650
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 4.407024383544922,
+      "learning_rate": 8.426262626262626e-07,
+      "loss": 0.5358,
+      "step": 1660
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 4.5201826095581055,
+      "learning_rate": 8.416161616161616e-07,
+      "loss": 0.6021,
+      "step": 1670
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 4.167496681213379,
+      "learning_rate": 8.406060606060606e-07,
+      "loss": 0.6007,
+      "step": 1680
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 5.0080695152282715,
+      "learning_rate": 8.395959595959595e-07,
+      "loss": 0.6298,
+      "step": 1690
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 4.7244977951049805,
+      "learning_rate": 8.385858585858585e-07,
+      "loss": 0.5713,
+      "step": 1700
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 5.417051315307617,
+      "learning_rate": 8.375757575757575e-07,
+      "loss": 0.6565,
+      "step": 1710
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 6.2620391845703125,
+      "learning_rate": 8.365656565656564e-07,
+      "loss": 0.6977,
+      "step": 1720
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 5.629709720611572,
+      "learning_rate": 8.355555555555556e-07,
+      "loss": 0.6281,
+      "step": 1730
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 4.182885646820068,
+      "learning_rate": 8.345454545454545e-07,
+      "loss": 0.6431,
+      "step": 1740
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 3.080979824066162,
+      "learning_rate": 8.335353535353536e-07,
+      "loss": 0.5938,
+      "step": 1750
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 2.8959693908691406,
+      "learning_rate": 8.325252525252525e-07,
+      "loss": 0.6677,
+      "step": 1760
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 2.9930922985076904,
+      "learning_rate": 8.315151515151515e-07,
+      "loss": 0.5425,
+      "step": 1770
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 11.59288501739502,
+      "learning_rate": 8.305050505050505e-07,
+      "loss": 0.5816,
+      "step": 1780
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 6.010501384735107,
+      "learning_rate": 8.294949494949495e-07,
+      "loss": 0.5944,
+      "step": 1790
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 3.7763831615448,
+      "learning_rate": 8.284848484848484e-07,
+      "loss": 0.6465,
+      "step": 1800
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 4.849707126617432,
+      "learning_rate": 8.274747474747475e-07,
+      "loss": 0.5679,
+      "step": 1810
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 3.6395959854125977,
+      "learning_rate": 8.264646464646464e-07,
+      "loss": 0.6026,
+      "step": 1820
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 4.717153549194336,
+      "learning_rate": 8.254545454545455e-07,
+      "loss": 0.5637,
+      "step": 1830
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 3.236311674118042,
+      "learning_rate": 8.244444444444444e-07,
+      "loss": 0.5955,
+      "step": 1840
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 4.533290386199951,
+      "learning_rate": 8.234343434343433e-07,
+      "loss": 0.5681,
+      "step": 1850
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 3.0948798656463623,
+      "learning_rate": 8.224242424242424e-07,
+      "loss": 0.6272,
+      "step": 1860
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 3.719708204269409,
+      "learning_rate": 8.214141414141413e-07,
+      "loss": 0.6793,
+      "step": 1870
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 3.228376865386963,
+      "learning_rate": 8.204040404040403e-07,
+      "loss": 0.558,
+      "step": 1880
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 3.4296934604644775,
+      "learning_rate": 8.193939393939393e-07,
+      "loss": 0.5717,
+      "step": 1890
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 5.708250999450684,
+      "learning_rate": 8.184848484848484e-07,
+      "loss": 0.6381,
+      "step": 1900
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 3.8306288719177246,
+      "learning_rate": 8.174747474747474e-07,
+      "loss": 0.6083,
+      "step": 1910
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 4.5576252937316895,
+      "learning_rate": 8.164646464646464e-07,
+      "loss": 0.6676,
+      "step": 1920
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 3.2234649658203125,
+      "learning_rate": 8.154545454545454e-07,
+      "loss": 0.6215,
+      "step": 1930
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 3.4826815128326416,
+      "learning_rate": 8.144444444444444e-07,
+      "loss": 0.6827,
+      "step": 1940
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 3.9034299850463867,
+      "learning_rate": 8.134343434343433e-07,
+      "loss": 0.596,
+      "step": 1950
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 2.7787563800811768,
+      "learning_rate": 8.124242424242424e-07,
+      "loss": 0.6234,
+      "step": 1960
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 4.442075252532959,
+      "learning_rate": 8.114141414141413e-07,
+      "loss": 0.6244,
+      "step": 1970
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 3.8511252403259277,
+      "learning_rate": 8.104040404040403e-07,
+      "loss": 0.5721,
+      "step": 1980
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 3.925318717956543,
+      "learning_rate": 8.093939393939394e-07,
+      "loss": 0.6547,
+      "step": 1990
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 4.262260437011719,
+      "learning_rate": 8.083838383838384e-07,
+      "loss": 0.6155,
+      "step": 2000
+    },
+    {
+      "epoch": 0.72,
+      "eval_loss": 0.7234830260276794,
+      "eval_runtime": 401.6534,
+      "eval_samples_per_second": 2.49,
+      "eval_steps_per_second": 2.49,
+      "step": 2000
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 4.890273571014404,
+      "learning_rate": 8.073737373737374e-07,
+      "loss": 0.6392,
+      "step": 2010
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 4.918805122375488,
+      "learning_rate": 8.063636363636364e-07,
+      "loss": 0.5549,
+      "step": 2020
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 3.605008840560913,
+      "learning_rate": 8.053535353535353e-07,
+      "loss": 0.5152,
+      "step": 2030
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 4.606666088104248,
+      "learning_rate": 8.043434343434344e-07,
+      "loss": 0.4798,
+      "step": 2040
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 4.115630626678467,
+      "learning_rate": 8.033333333333333e-07,
+      "loss": 0.5856,
+      "step": 2050
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 4.37373685836792,
+      "learning_rate": 8.023232323232322e-07,
+      "loss": 0.6467,
+      "step": 2060
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 3.3623104095458984,
+      "learning_rate": 8.013131313131313e-07,
+      "loss": 0.6591,
+      "step": 2070
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 3.1440579891204834,
+      "learning_rate": 8.003030303030302e-07,
+      "loss": 0.6003,
+      "step": 2080
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 3.391533136367798,
+      "learning_rate": 7.992929292929293e-07,
+      "loss": 0.5872,
+      "step": 2090
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 4.677767276763916,
+      "learning_rate": 7.982828282828282e-07,
+      "loss": 0.5803,
+      "step": 2100
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 3.759061336517334,
+      "learning_rate": 7.972727272727272e-07,
+      "loss": 0.6155,
+      "step": 2110
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 3.419386863708496,
+      "learning_rate": 7.962626262626262e-07,
+      "loss": 0.6303,
+      "step": 2120
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 3.4935250282287598,
+      "learning_rate": 7.952525252525252e-07,
+      "loss": 0.6348,
+      "step": 2130
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 4.793978214263916,
+      "learning_rate": 7.942424242424241e-07,
+      "loss": 0.5004,
+      "step": 2140
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 5.078155994415283,
+      "learning_rate": 7.932323232323232e-07,
+      "loss": 0.5583,
+      "step": 2150
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 3.8855032920837402,
+      "learning_rate": 7.922222222222222e-07,
+      "loss": 0.626,
+      "step": 2160
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 4.782063007354736,
+      "learning_rate": 7.912121212121213e-07,
+      "loss": 0.4847,
+      "step": 2170
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 4.760573387145996,
+      "learning_rate": 7.902020202020202e-07,
+      "loss": 0.5923,
+      "step": 2180
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 5.196963310241699,
+      "learning_rate": 7.891919191919192e-07,
+      "loss": 0.5216,
+      "step": 2190
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 5.447847843170166,
+      "learning_rate": 7.881818181818182e-07,
+      "loss": 0.6144,
+      "step": 2200
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 5.5991129875183105,
+      "learning_rate": 7.871717171717171e-07,
+      "loss": 0.5247,
+      "step": 2210
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 4.1100921630859375,
+      "learning_rate": 7.861616161616161e-07,
+      "loss": 0.6581,
+      "step": 2220
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 2.998598098754883,
+      "learning_rate": 7.851515151515151e-07,
+      "loss": 0.5111,
+      "step": 2230
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 3.1241579055786133,
+      "learning_rate": 7.841414141414141e-07,
+      "loss": 0.626,
+      "step": 2240
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 3.804655075073242,
+      "learning_rate": 7.831313131313131e-07,
+      "loss": 0.596,
+      "step": 2250
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 3.636692523956299,
+      "learning_rate": 7.821212121212121e-07,
+      "loss": 0.598,
+      "step": 2260
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 5.374478816986084,
+      "learning_rate": 7.81111111111111e-07,
+      "loss": 0.5155,
+      "step": 2270
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 4.611242294311523,
+      "learning_rate": 7.801010101010101e-07,
+      "loss": 0.5482,
+      "step": 2280
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 6.077275276184082,
+      "learning_rate": 7.79090909090909e-07,
+      "loss": 0.5642,
+      "step": 2290
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 4.110134601593018,
+      "learning_rate": 7.78080808080808e-07,
+      "loss": 0.6482,
+      "step": 2300
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 3.603376865386963,
+      "learning_rate": 7.77070707070707e-07,
+      "loss": 0.5755,
+      "step": 2310
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 4.059467315673828,
+      "learning_rate": 7.76060606060606e-07,
+      "loss": 0.5205,
+      "step": 2320
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 4.656579971313477,
+      "learning_rate": 7.750505050505051e-07,
+      "loss": 0.5975,
+      "step": 2330
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 4.863138675689697,
+      "learning_rate": 7.74040404040404e-07,
+      "loss": 0.5375,
+      "step": 2340
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 4.3313517570495605,
+      "learning_rate": 7.73030303030303e-07,
+      "loss": 0.5795,
+      "step": 2350
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 3.870875120162964,
+      "learning_rate": 7.72020202020202e-07,
+      "loss": 0.5249,
+      "step": 2360
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 3.720283031463623,
+      "learning_rate": 7.71010101010101e-07,
+      "loss": 0.6733,
+      "step": 2370
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 3.779296398162842,
+      "learning_rate": 7.699999999999999e-07,
+      "loss": 0.588,
+      "step": 2380
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 4.588533401489258,
+      "learning_rate": 7.68989898989899e-07,
+      "loss": 0.6044,
+      "step": 2390
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 5.907507419586182,
+      "learning_rate": 7.679797979797979e-07,
+      "loss": 0.5715,
+      "step": 2400
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 5.705809116363525,
+      "learning_rate": 7.66969696969697e-07,
+      "loss": 0.5486,
+      "step": 2410
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 3.8104190826416016,
+      "learning_rate": 7.659595959595959e-07,
+      "loss": 0.602,
+      "step": 2420
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 5.82910680770874,
+      "learning_rate": 7.649494949494949e-07,
+      "loss": 0.5111,
+      "step": 2430
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 5.837489604949951,
+      "learning_rate": 7.639393939393939e-07,
+      "loss": 0.6145,
+      "step": 2440
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 3.5505547523498535,
+      "learning_rate": 7.629292929292929e-07,
+      "loss": 0.5834,
+      "step": 2450
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 4.314844131469727,
+      "learning_rate": 7.619191919191918e-07,
+      "loss": 0.6152,
+      "step": 2460
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 4.205423831939697,
+      "learning_rate": 7.609090909090909e-07,
+      "loss": 0.6285,
+      "step": 2470
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 5.292863368988037,
+      "learning_rate": 7.598989898989898e-07,
+      "loss": 0.6531,
+      "step": 2480
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 5.13388204574585,
+      "learning_rate": 7.588888888888888e-07,
+      "loss": 0.628,
+      "step": 2490
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 2.864743232727051,
+      "learning_rate": 7.578787878787879e-07,
+      "loss": 0.5403,
+      "step": 2500
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 4.284875869750977,
+      "learning_rate": 7.568686868686868e-07,
+      "loss": 0.6097,
+      "step": 2510
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 3.3840548992156982,
+      "learning_rate": 7.558585858585859e-07,
+      "loss": 0.5759,
+      "step": 2520
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 4.531360626220703,
+      "learning_rate": 7.548484848484848e-07,
+      "loss": 0.5581,
+      "step": 2530
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 5.861146926879883,
+      "learning_rate": 7.538383838383838e-07,
+      "loss": 0.6329,
+      "step": 2540
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 5.117300987243652,
+      "learning_rate": 7.528282828282828e-07,
+      "loss": 0.6071,
+      "step": 2550
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 3.9243366718292236,
+      "learning_rate": 7.518181818181818e-07,
+      "loss": 0.5476,
+      "step": 2560
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 6.215851783752441,
+      "learning_rate": 7.508080808080808e-07,
+      "loss": 0.5731,
+      "step": 2570
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 6.189659118652344,
+      "learning_rate": 7.497979797979798e-07,
+      "loss": 0.592,
+      "step": 2580
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 2.877923011779785,
+      "learning_rate": 7.487878787878787e-07,
+      "loss": 0.5435,
+      "step": 2590
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 5.963223934173584,
+      "learning_rate": 7.477777777777778e-07,
+      "loss": 0.6104,
+      "step": 2600
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 4.374766826629639,
+      "learning_rate": 7.467676767676767e-07,
+      "loss": 0.6022,
+      "step": 2610
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 4.16854190826416,
+      "learning_rate": 7.457575757575756e-07,
+      "loss": 0.4931,
+      "step": 2620
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 7.225739479064941,
+      "learning_rate": 7.447474747474747e-07,
+      "loss": 0.5659,
+      "step": 2630
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 4.522436141967773,
+      "learning_rate": 7.437373737373736e-07,
+      "loss": 0.5449,
+      "step": 2640
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 4.935425758361816,
+      "learning_rate": 7.427272727272727e-07,
+      "loss": 0.6093,
+      "step": 2650
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 3.947935104370117,
+      "learning_rate": 7.417171717171716e-07,
+      "loss": 0.5976,
+      "step": 2660
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 5.328210830688477,
+      "learning_rate": 7.407070707070707e-07,
+      "loss": 0.6283,
+      "step": 2670
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 3.0664663314819336,
+      "learning_rate": 7.396969696969697e-07,
+      "loss": 0.5853,
+      "step": 2680
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 5.055959701538086,
+      "learning_rate": 7.386868686868687e-07,
+      "loss": 0.5767,
+      "step": 2690
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 5.318157196044922,
+      "learning_rate": 7.376767676767676e-07,
+      "loss": 0.5319,
+      "step": 2700
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 3.5820364952087402,
+      "learning_rate": 7.366666666666667e-07,
+      "loss": 0.5506,
+      "step": 2710
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 4.367839813232422,
+      "learning_rate": 7.356565656565656e-07,
+      "loss": 0.5524,
+      "step": 2720
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 4.305497169494629,
+      "learning_rate": 7.346464646464647e-07,
+      "loss": 0.5639,
+      "step": 2730
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 6.55611515045166,
+      "learning_rate": 7.336363636363636e-07,
+      "loss": 0.6346,
+      "step": 2740
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 9.09347152709961,
+      "learning_rate": 7.326262626262626e-07,
+      "loss": 0.5574,
+      "step": 2750
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 3.9730687141418457,
+      "learning_rate": 7.316161616161616e-07,
+      "loss": 0.5908,
+      "step": 2760
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 16.01730728149414,
+      "learning_rate": 7.306060606060605e-07,
+      "loss": 0.5936,
+      "step": 2770
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 5.077668190002441,
+      "learning_rate": 7.295959595959595e-07,
+      "loss": 0.5556,
+      "step": 2780
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 4.859917640686035,
+      "learning_rate": 7.285858585858585e-07,
+      "loss": 0.5452,
+      "step": 2790
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 5.687443733215332,
+      "learning_rate": 7.275757575757575e-07,
+      "loss": 0.6583,
+      "step": 2800
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 4.369338035583496,
+      "learning_rate": 7.265656565656565e-07,
+      "loss": 0.5623,
+      "step": 2810
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 4.035587310791016,
+      "learning_rate": 7.255555555555555e-07,
+      "loss": 0.571,
+      "step": 2820
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 4.2013139724731445,
+      "learning_rate": 7.245454545454544e-07,
+      "loss": 0.4768,
+      "step": 2830
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 3.272221565246582,
+      "learning_rate": 7.235353535353536e-07,
+      "loss": 0.581,
+      "step": 2840
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 3.97727370262146,
+      "learning_rate": 7.225252525252525e-07,
+      "loss": 0.5654,
+      "step": 2850
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 4.043779373168945,
+      "learning_rate": 7.215151515151516e-07,
+      "loss": 0.5971,
+      "step": 2860
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 4.015261173248291,
+      "learning_rate": 7.205050505050505e-07,
+      "loss": 0.5921,
+      "step": 2870
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 4.540653705596924,
+      "learning_rate": 7.194949494949495e-07,
+      "loss": 0.6103,
+      "step": 2880
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 4.258879661560059,
+      "learning_rate": 7.184848484848485e-07,
+      "loss": 0.5519,
+      "step": 2890
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 7.0088629722595215,
+      "learning_rate": 7.174747474747475e-07,
+      "loss": 0.6054,
+      "step": 2900
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 3.9803833961486816,
+      "learning_rate": 7.164646464646464e-07,
+      "loss": 0.5557,
+      "step": 2910
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 3.6390326023101807,
+      "learning_rate": 7.154545454545454e-07,
+      "loss": 0.5406,
+      "step": 2920
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 3.998908042907715,
+      "learning_rate": 7.144444444444444e-07,
+      "loss": 0.604,
+      "step": 2930
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 4.304749488830566,
+      "learning_rate": 7.134343434343433e-07,
+      "loss": 0.5859,
+      "step": 2940
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 4.742584228515625,
+      "learning_rate": 7.124242424242424e-07,
+      "loss": 0.5184,
+      "step": 2950
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 5.163829326629639,
+      "learning_rate": 7.114141414141413e-07,
+      "loss": 0.5669,
+      "step": 2960
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 5.633606910705566,
+      "learning_rate": 7.104040404040404e-07,
+      "loss": 0.5967,
+      "step": 2970
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 6.37916898727417,
+      "learning_rate": 7.093939393939393e-07,
+      "loss": 0.5971,
+      "step": 2980
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 2.9268598556518555,
+      "learning_rate": 7.083838383838383e-07,
+      "loss": 0.5342,
+      "step": 2990
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 4.331085205078125,
+      "learning_rate": 7.073737373737373e-07,
+      "loss": 0.5779,
+      "step": 3000
+    },
+    {
+      "epoch": 1.08,
+      "eval_loss": 0.7162447571754456,
+      "eval_runtime": 401.0241,
+      "eval_samples_per_second": 2.494,
+      "eval_steps_per_second": 2.494,
+      "step": 3000
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 2.9604339599609375,
+      "learning_rate": 7.063636363636364e-07,
+      "loss": 0.5723,
+      "step": 3010
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 4.6410393714904785,
+      "learning_rate": 7.053535353535354e-07,
+      "loss": 0.5986,
+      "step": 3020
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 4.530472278594971,
+      "learning_rate": 7.043434343434344e-07,
+      "loss": 0.705,
+      "step": 3030
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 4.028800964355469,
+      "learning_rate": 7.033333333333333e-07,
+      "loss": 0.5135,
+      "step": 3040
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 2.882619619369507,
+      "learning_rate": 7.023232323232324e-07,
+      "loss": 0.5373,
+      "step": 3050
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 5.280877590179443,
+      "learning_rate": 7.013131313131313e-07,
+      "loss": 0.5457,
+      "step": 3060
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 3.983971357345581,
+      "learning_rate": 7.003030303030302e-07,
+      "loss": 0.5678,
+      "step": 3070
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 3.4524683952331543,
+      "learning_rate": 6.992929292929293e-07,
+      "loss": 0.5794,
+      "step": 3080
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 4.039379119873047,
+      "learning_rate": 6.982828282828282e-07,
+      "loss": 0.4778,
+      "step": 3090
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 4.454843521118164,
+      "learning_rate": 6.972727272727273e-07,
+      "loss": 0.5312,
+      "step": 3100
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 4.90205192565918,
+      "learning_rate": 6.962626262626262e-07,
+      "loss": 0.5675,
+      "step": 3110
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 3.076164484024048,
+      "learning_rate": 6.952525252525252e-07,
+      "loss": 0.5949,
+      "step": 3120
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 3.707550287246704,
+      "learning_rate": 6.942424242424242e-07,
+      "loss": 0.6053,
+      "step": 3130
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 4.36667537689209,
+      "learning_rate": 6.932323232323232e-07,
+      "loss": 0.638,
+      "step": 3140
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 4.635988235473633,
+      "learning_rate": 6.922222222222221e-07,
+      "loss": 0.5698,
+      "step": 3150
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 3.769178628921509,
+      "learning_rate": 6.912121212121212e-07,
+      "loss": 0.5563,
+      "step": 3160
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 3.096804618835449,
+      "learning_rate": 6.902020202020201e-07,
+      "loss": 0.5268,
+      "step": 3170
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 5.9978461265563965,
+      "learning_rate": 6.891919191919193e-07,
+      "loss": 0.4878,
+      "step": 3180
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 4.963356971740723,
+      "learning_rate": 6.881818181818182e-07,
+      "loss": 0.5904,
+      "step": 3190
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 4.526729583740234,
+      "learning_rate": 6.871717171717171e-07,
+      "loss": 0.5957,
+      "step": 3200
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 3.4197323322296143,
+      "learning_rate": 6.861616161616162e-07,
+      "loss": 0.5522,
+      "step": 3210
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 3.989546060562134,
+      "learning_rate": 6.851515151515151e-07,
+      "loss": 0.6122,
+      "step": 3220
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 5.137784004211426,
+      "learning_rate": 6.841414141414141e-07,
+      "loss": 0.58,
+      "step": 3230
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 4.193089485168457,
+      "learning_rate": 6.831313131313131e-07,
+      "loss": 0.4918,
+      "step": 3240
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 4.1724162101745605,
+      "learning_rate": 6.821212121212121e-07,
+      "loss": 0.6044,
+      "step": 3250
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 4.983777046203613,
+      "learning_rate": 6.811111111111111e-07,
+      "loss": 0.533,
+      "step": 3260
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 7.2310333251953125,
+      "learning_rate": 6.801010101010101e-07,
+      "loss": 0.5843,
+      "step": 3270
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 4.620791912078857,
+      "learning_rate": 6.79090909090909e-07,
+      "loss": 0.5242,
+      "step": 3280
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 4.3235087394714355,
+      "learning_rate": 6.780808080808081e-07,
+      "loss": 0.5836,
+      "step": 3290
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 4.413332462310791,
+      "learning_rate": 6.77070707070707e-07,
+      "loss": 0.6532,
+      "step": 3300
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 4.286377429962158,
+      "learning_rate": 6.76060606060606e-07,
+      "loss": 0.5527,
+      "step": 3310
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 6.150529861450195,
+      "learning_rate": 6.75050505050505e-07,
+      "loss": 0.5745,
+      "step": 3320
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 4.088225841522217,
+      "learning_rate": 6.74040404040404e-07,
+      "loss": 0.5882,
+      "step": 3330
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 4.4364824295043945,
+      "learning_rate": 6.73030303030303e-07,
+      "loss": 0.513,
+      "step": 3340
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 3.8595519065856934,
+      "learning_rate": 6.72020202020202e-07,
+      "loss": 0.5621,
+      "step": 3350
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 6.183053016662598,
+      "learning_rate": 6.71010101010101e-07,
+      "loss": 0.5318,
+      "step": 3360
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 4.677096366882324,
+      "learning_rate": 6.7e-07,
+      "loss": 0.4965,
+      "step": 3370
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 5.507148742675781,
+      "learning_rate": 6.68989898989899e-07,
+      "loss": 0.569,
+      "step": 3380
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 3.813816547393799,
+      "learning_rate": 6.679797979797979e-07,
+      "loss": 0.5326,
+      "step": 3390
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 5.973514080047607,
+      "learning_rate": 6.66969696969697e-07,
+      "loss": 0.6733,
+      "step": 3400
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 5.510775089263916,
+      "learning_rate": 6.659595959595959e-07,
+      "loss": 0.5978,
+      "step": 3410
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 3.8347678184509277,
+      "learning_rate": 6.64949494949495e-07,
+      "loss": 0.5141,
+      "step": 3420
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 4.097416877746582,
+      "learning_rate": 6.639393939393939e-07,
+      "loss": 0.5503,
+      "step": 3430
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 3.52165150642395,
+      "learning_rate": 6.629292929292929e-07,
+      "loss": 0.5506,
+      "step": 3440
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 4.4569573402404785,
+      "learning_rate": 6.619191919191919e-07,
+      "loss": 0.6121,
+      "step": 3450
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 4.094751834869385,
+      "learning_rate": 6.609090909090909e-07,
+      "loss": 0.5562,
+      "step": 3460
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 4.277998447418213,
+      "learning_rate": 6.598989898989898e-07,
+      "loss": 0.5432,
+      "step": 3470
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 5.9544148445129395,
+      "learning_rate": 6.588888888888889e-07,
+      "loss": 0.5891,
+      "step": 3480
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 5.688138008117676,
+      "learning_rate": 6.578787878787878e-07,
+      "loss": 0.6172,
+      "step": 3490
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 5.771819591522217,
+      "learning_rate": 6.568686868686868e-07,
+      "loss": 0.5362,
+      "step": 3500
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 4.87513542175293,
+      "learning_rate": 6.558585858585858e-07,
+      "loss": 0.5608,
+      "step": 3510
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 3.709472179412842,
+      "learning_rate": 6.548484848484848e-07,
+      "loss": 0.5361,
+      "step": 3520
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 4.465284824371338,
+      "learning_rate": 6.538383838383839e-07,
+      "loss": 0.5435,
+      "step": 3530
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 4.30505895614624,
+      "learning_rate": 6.528282828282828e-07,
+      "loss": 0.5714,
+      "step": 3540
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 4.445021152496338,
+      "learning_rate": 6.518181818181818e-07,
+      "loss": 0.5773,
+      "step": 3550
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 4.173781394958496,
+      "learning_rate": 6.508080808080808e-07,
+      "loss": 0.6223,
+      "step": 3560
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 5.676968574523926,
+      "learning_rate": 6.497979797979798e-07,
+      "loss": 0.5957,
+      "step": 3570
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 3.447176694869995,
+      "learning_rate": 6.487878787878788e-07,
+      "loss": 0.5896,
+      "step": 3580
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 4.1883111000061035,
+      "learning_rate": 6.477777777777778e-07,
+      "loss": 0.5602,
+      "step": 3590
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 4.16115140914917,
+      "learning_rate": 6.467676767676767e-07,
+      "loss": 0.5752,
+      "step": 3600
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 4.55715274810791,
+      "learning_rate": 6.457575757575758e-07,
+      "loss": 0.5344,
+      "step": 3610
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 5.83474588394165,
+      "learning_rate": 6.447474747474747e-07,
+      "loss": 0.5867,
+      "step": 3620
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 2.700615644454956,
+      "learning_rate": 6.437373737373736e-07,
+      "loss": 0.5395,
+      "step": 3630
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 4.691040992736816,
+      "learning_rate": 6.428282828282828e-07,
+      "loss": 0.5509,
+      "step": 3640
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 4.966342449188232,
+      "learning_rate": 6.418181818181818e-07,
+      "loss": 0.6047,
+      "step": 3650
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 3.6059834957122803,
+      "learning_rate": 6.408080808080808e-07,
+      "loss": 0.5338,
+      "step": 3660
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 3.471660852432251,
+      "learning_rate": 6.397979797979798e-07,
+      "loss": 0.5532,
+      "step": 3670
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 5.410092353820801,
+      "learning_rate": 6.387878787878788e-07,
+      "loss": 0.5739,
+      "step": 3680
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 4.829250335693359,
+      "learning_rate": 6.377777777777778e-07,
+      "loss": 0.6064,
+      "step": 3690
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 5.880673408508301,
+      "learning_rate": 6.367676767676767e-07,
+      "loss": 0.575,
+      "step": 3700
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 3.647641658782959,
+      "learning_rate": 6.357575757575757e-07,
+      "loss": 0.594,
+      "step": 3710
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 4.267665386199951,
+      "learning_rate": 6.347474747474747e-07,
+      "loss": 0.574,
+      "step": 3720
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 3.5303924083709717,
+      "learning_rate": 6.337373737373736e-07,
+      "loss": 0.496,
+      "step": 3730
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 3.1564202308654785,
+      "learning_rate": 6.327272727272727e-07,
+      "loss": 0.5196,
+      "step": 3740
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 4.458827495574951,
+      "learning_rate": 6.317171717171716e-07,
+      "loss": 0.5616,
+      "step": 3750
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 4.057519435882568,
+      "learning_rate": 6.307070707070707e-07,
+      "loss": 0.5293,
+      "step": 3760
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 5.7592363357543945,
+      "learning_rate": 6.296969696969696e-07,
+      "loss": 0.5433,
+      "step": 3770
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 5.579089641571045,
+      "learning_rate": 6.286868686868687e-07,
+      "loss": 0.546,
+      "step": 3780
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 3.8911049365997314,
+      "learning_rate": 6.276767676767677e-07,
+      "loss": 0.5991,
+      "step": 3790
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 6.364475727081299,
+      "learning_rate": 6.266666666666667e-07,
+      "loss": 0.5908,
+      "step": 3800
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 5.0971760749816895,
+      "learning_rate": 6.256565656565656e-07,
+      "loss": 0.5915,
+      "step": 3810
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 3.897958755493164,
+      "learning_rate": 6.246464646464647e-07,
+      "loss": 0.5859,
+      "step": 3820
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 6.308644771575928,
+      "learning_rate": 6.236363636363636e-07,
+      "loss": 0.5991,
+      "step": 3830
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 3.0512614250183105,
+      "learning_rate": 6.226262626262627e-07,
+      "loss": 0.5742,
+      "step": 3840
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 4.052495956420898,
+      "learning_rate": 6.216161616161616e-07,
+      "loss": 0.54,
+      "step": 3850
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 4.723029613494873,
+      "learning_rate": 6.206060606060605e-07,
+      "loss": 0.5524,
+      "step": 3860
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 4.705147743225098,
+      "learning_rate": 6.195959595959596e-07,
+      "loss": 0.5924,
+      "step": 3870
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 4.098758697509766,
+      "learning_rate": 6.185858585858585e-07,
+      "loss": 0.5482,
+      "step": 3880
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 2.9597623348236084,
+      "learning_rate": 6.175757575757575e-07,
+      "loss": 0.5521,
+      "step": 3890
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 3.7308359146118164,
+      "learning_rate": 6.165656565656565e-07,
+      "loss": 0.5645,
+      "step": 3900
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 4.260051250457764,
+      "learning_rate": 6.155555555555555e-07,
+      "loss": 0.5471,
+      "step": 3910
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 6.539883136749268,
+      "learning_rate": 6.145454545454545e-07,
+      "loss": 0.5439,
+      "step": 3920
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 6.235405921936035,
+      "learning_rate": 6.135353535353535e-07,
+      "loss": 0.5435,
+      "step": 3930
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 3.3652477264404297,
+      "learning_rate": 6.125252525252524e-07,
+      "loss": 0.5544,
+      "step": 3940
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 3.795522928237915,
+      "learning_rate": 6.115151515151516e-07,
+      "loss": 0.5794,
+      "step": 3950
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 5.496920108795166,
+      "learning_rate": 6.105050505050505e-07,
+      "loss": 0.5133,
+      "step": 3960
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 5.946590423583984,
+      "learning_rate": 6.094949494949495e-07,
+      "loss": 0.6391,
+      "step": 3970
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 3.943232774734497,
+      "learning_rate": 6.084848484848485e-07,
+      "loss": 0.5574,
+      "step": 3980
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 4.132944107055664,
+      "learning_rate": 6.074747474747474e-07,
+      "loss": 0.5014,
+      "step": 3990
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 4.284780979156494,
+      "learning_rate": 6.064646464646465e-07,
+      "loss": 0.5907,
+      "step": 4000
+    },
+    {
+      "epoch": 1.43,
+      "eval_loss": 0.698131263256073,
+      "eval_runtime": 401.1999,
+      "eval_samples_per_second": 2.493,
+      "eval_steps_per_second": 2.493,
+      "step": 4000
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 5.0691447257995605,
+      "learning_rate": 6.054545454545454e-07,
+      "loss": 0.5153,
+      "step": 4010
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 4.563512325286865,
+      "learning_rate": 6.044444444444444e-07,
+      "loss": 0.5323,
+      "step": 4020
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 8.346761703491211,
+      "learning_rate": 6.034343434343434e-07,
+      "loss": 0.6202,
+      "step": 4030
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 4.459041118621826,
+      "learning_rate": 6.024242424242424e-07,
+      "loss": 0.5081,
+      "step": 4040
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 6.5770158767700195,
+      "learning_rate": 6.014141414141413e-07,
+      "loss": 0.568,
+      "step": 4050
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 5.020130634307861,
+      "learning_rate": 6.004040404040404e-07,
+      "loss": 0.5027,
+      "step": 4060
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 6.139995098114014,
+      "learning_rate": 5.993939393939393e-07,
+      "loss": 0.5115,
+      "step": 4070
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 5.601007461547852,
+      "learning_rate": 5.983838383838384e-07,
+      "loss": 0.5331,
+      "step": 4080
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 4.264581680297852,
+      "learning_rate": 5.973737373737373e-07,
+      "loss": 0.5474,
+      "step": 4090
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 3.254601001739502,
+      "learning_rate": 5.963636363636363e-07,
+      "loss": 0.5515,
+      "step": 4100
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 4.111956596374512,
+      "learning_rate": 5.953535353535353e-07,
+      "loss": 0.5486,
+      "step": 4110
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 3.814635992050171,
+      "learning_rate": 5.943434343434344e-07,
+      "loss": 0.5332,
+      "step": 4120
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 4.161899089813232,
+      "learning_rate": 5.933333333333334e-07,
+      "loss": 0.5713,
+      "step": 4130
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 3.490006446838379,
+      "learning_rate": 5.923232323232323e-07,
+      "loss": 0.5522,
+      "step": 4140
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 4.460766792297363,
+      "learning_rate": 5.913131313131313e-07,
+      "loss": 0.515,
+      "step": 4150
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 5.571935176849365,
+      "learning_rate": 5.903030303030303e-07,
+      "loss": 0.5976,
+      "step": 4160
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 4.256524562835693,
+      "learning_rate": 5.892929292929293e-07,
+      "loss": 0.5068,
+      "step": 4170
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 5.909682273864746,
+      "learning_rate": 5.882828282828282e-07,
+      "loss": 0.5473,
+      "step": 4180
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 6.469602584838867,
+      "learning_rate": 5.872727272727273e-07,
+      "loss": 0.5715,
+      "step": 4190
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 4.836788654327393,
+      "learning_rate": 5.862626262626262e-07,
+      "loss": 0.4511,
+      "step": 4200
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 2.996185302734375,
+      "learning_rate": 5.852525252525252e-07,
+      "loss": 0.5782,
+      "step": 4210
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 8.92146110534668,
+      "learning_rate": 5.842424242424242e-07,
+      "loss": 0.5329,
+      "step": 4220
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 6.146700382232666,
+      "learning_rate": 5.832323232323232e-07,
+      "loss": 0.5768,
+      "step": 4230
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 4.365715026855469,
+      "learning_rate": 5.822222222222222e-07,
+      "loss": 0.5804,
+      "step": 4240
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 4.362324237823486,
+      "learning_rate": 5.812121212121212e-07,
+      "loss": 0.5435,
+      "step": 4250
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 4.051389217376709,
+      "learning_rate": 5.802020202020201e-07,
+      "loss": 0.5662,
+      "step": 4260
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 4.774472236633301,
+      "learning_rate": 5.791919191919192e-07,
+      "loss": 0.523,
+      "step": 4270
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 4.41732120513916,
+      "learning_rate": 5.781818181818181e-07,
+      "loss": 0.5926,
+      "step": 4280
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 5.613069534301758,
+      "learning_rate": 5.771717171717172e-07,
+      "loss": 0.5205,
+      "step": 4290
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 4.26994514465332,
+      "learning_rate": 5.761616161616162e-07,
+      "loss": 0.5605,
+      "step": 4300
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 4.853590488433838,
+      "learning_rate": 5.751515151515151e-07,
+      "loss": 0.587,
+      "step": 4310
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 5.43099308013916,
+      "learning_rate": 5.741414141414142e-07,
+      "loss": 0.533,
+      "step": 4320
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 4.218691349029541,
+      "learning_rate": 5.731313131313131e-07,
+      "loss": 0.5373,
+      "step": 4330
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 6.392608642578125,
+      "learning_rate": 5.721212121212121e-07,
+      "loss": 0.5346,
+      "step": 4340
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 4.781940937042236,
+      "learning_rate": 5.711111111111111e-07,
+      "loss": 0.6316,
+      "step": 4350
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 4.0989990234375,
+      "learning_rate": 5.702020202020201e-07,
+      "loss": 0.5863,
+      "step": 4360
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 4.195807456970215,
+      "learning_rate": 5.691919191919191e-07,
+      "loss": 0.6091,
+      "step": 4370
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 3.2372219562530518,
+      "learning_rate": 5.681818181818182e-07,
+      "loss": 0.5939,
+      "step": 4380
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 3.523815631866455,
+      "learning_rate": 5.671717171717172e-07,
+      "loss": 0.5908,
+      "step": 4390
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 4.94608736038208,
+      "learning_rate": 5.661616161616162e-07,
+      "loss": 0.5964,
+      "step": 4400
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 8.730268478393555,
+      "learning_rate": 5.651515151515151e-07,
+      "loss": 0.5355,
+      "step": 4410
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 4.550642490386963,
+      "learning_rate": 5.641414141414142e-07,
+      "loss": 0.5295,
+      "step": 4420
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 4.5829176902771,
+      "learning_rate": 5.631313131313131e-07,
+      "loss": 0.5915,
+      "step": 4430
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 4.118924617767334,
+      "learning_rate": 5.621212121212121e-07,
+      "loss": 0.5589,
+      "step": 4440
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 4.678675174713135,
+      "learning_rate": 5.611111111111111e-07,
+      "loss": 0.6621,
+      "step": 4450
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 4.09968376159668,
+      "learning_rate": 5.601010101010101e-07,
+      "loss": 0.6018,
+      "step": 4460
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 6.312856674194336,
+      "learning_rate": 5.590909090909091e-07,
+      "loss": 0.6402,
+      "step": 4470
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 5.487626552581787,
+      "learning_rate": 5.58080808080808e-07,
+      "loss": 0.5799,
+      "step": 4480
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 4.387622833251953,
+      "learning_rate": 5.57070707070707e-07,
+      "loss": 0.5685,
+      "step": 4490
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 4.692462921142578,
+      "learning_rate": 5.56060606060606e-07,
+      "loss": 0.5935,
+      "step": 4500
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 3.9311652183532715,
+      "learning_rate": 5.55050505050505e-07,
+      "loss": 0.6092,
+      "step": 4510
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 4.879345417022705,
+      "learning_rate": 5.540404040404039e-07,
+      "loss": 0.5754,
+      "step": 4520
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 4.526495933532715,
+      "learning_rate": 5.53030303030303e-07,
+      "loss": 0.6167,
+      "step": 4530
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 5.269490718841553,
+      "learning_rate": 5.520202020202019e-07,
+      "loss": 0.5197,
+      "step": 4540
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 6.770501613616943,
+      "learning_rate": 5.510101010101011e-07,
+      "loss": 0.5389,
+      "step": 4550
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 5.688599109649658,
+      "learning_rate": 5.5e-07,
+      "loss": 0.518,
+      "step": 4560
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 3.049668073654175,
+      "learning_rate": 5.48989898989899e-07,
+      "loss": 0.6034,
+      "step": 4570
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 5.590402126312256,
+      "learning_rate": 5.47979797979798e-07,
+      "loss": 0.6146,
+      "step": 4580
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 4.051222801208496,
+      "learning_rate": 5.46969696969697e-07,
+      "loss": 0.5388,
+      "step": 4590
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 3.68947434425354,
+      "learning_rate": 5.459595959595959e-07,
+      "loss": 0.5049,
+      "step": 4600
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 5.772908687591553,
+      "learning_rate": 5.44949494949495e-07,
+      "loss": 0.5447,
+      "step": 4610
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 3.907376289367676,
+      "learning_rate": 5.439393939393939e-07,
+      "loss": 0.5038,
+      "step": 4620
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 6.761162281036377,
+      "learning_rate": 5.42929292929293e-07,
+      "loss": 0.6141,
+      "step": 4630
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 3.054441452026367,
+      "learning_rate": 5.419191919191919e-07,
+      "loss": 0.6043,
+      "step": 4640
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 5.1576457023620605,
+      "learning_rate": 5.409090909090908e-07,
+      "loss": 0.5588,
+      "step": 4650
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 4.096390724182129,
+      "learning_rate": 5.398989898989899e-07,
+      "loss": 0.6355,
+      "step": 4660
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 3.5117170810699463,
+      "learning_rate": 5.388888888888888e-07,
+      "loss": 0.6043,
+      "step": 4670
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 3.7968451976776123,
+      "learning_rate": 5.378787878787878e-07,
+      "loss": 0.6092,
+      "step": 4680
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 5.3210320472717285,
+      "learning_rate": 5.368686868686868e-07,
+      "loss": 0.5375,
+      "step": 4690
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 3.6537818908691406,
+      "learning_rate": 5.358585858585858e-07,
+      "loss": 0.5013,
+      "step": 4700
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 3.1777806282043457,
+      "learning_rate": 5.348484848484848e-07,
+      "loss": 0.5073,
+      "step": 4710
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 5.203103065490723,
+      "learning_rate": 5.338383838383839e-07,
+      "loss": 0.5694,
+      "step": 4720
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 3.941244602203369,
+      "learning_rate": 5.328282828282828e-07,
+      "loss": 0.5337,
+      "step": 4730
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 3.7557549476623535,
+      "learning_rate": 5.318181818181819e-07,
+      "loss": 0.6047,
+      "step": 4740
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 5.3054351806640625,
+      "learning_rate": 5.308080808080808e-07,
+      "loss": 0.5563,
+      "step": 4750
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 3.023254871368408,
+      "learning_rate": 5.297979797979798e-07,
+      "loss": 0.5877,
+      "step": 4760
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 3.8223531246185303,
+      "learning_rate": 5.287878787878788e-07,
+      "loss": 0.5938,
+      "step": 4770
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 2.889775276184082,
+      "learning_rate": 5.277777777777777e-07,
+      "loss": 0.5628,
+      "step": 4780
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 4.528804302215576,
+      "learning_rate": 5.267676767676768e-07,
+      "loss": 0.5518,
+      "step": 4790
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 4.486887454986572,
+      "learning_rate": 5.257575757575757e-07,
+      "loss": 0.6116,
+      "step": 4800
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 3.2164041996002197,
+      "learning_rate": 5.247474747474747e-07,
+      "loss": 0.5949,
+      "step": 4810
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 6.722822666168213,
+      "learning_rate": 5.237373737373737e-07,
+      "loss": 0.6425,
+      "step": 4820
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 4.429785251617432,
+      "learning_rate": 5.227272727272727e-07,
+      "loss": 0.5447,
+      "step": 4830
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 4.903067111968994,
+      "learning_rate": 5.217171717171716e-07,
+      "loss": 0.5635,
+      "step": 4840
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 6.777737617492676,
+      "learning_rate": 5.207070707070707e-07,
+      "loss": 0.5953,
+      "step": 4850
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 5.027159214019775,
+      "learning_rate": 5.196969696969696e-07,
+      "loss": 0.598,
+      "step": 4860
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 3.1004555225372314,
+      "learning_rate": 5.186868686868687e-07,
+      "loss": 0.5972,
+      "step": 4870
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 6.468344211578369,
+      "learning_rate": 5.176767676767676e-07,
+      "loss": 0.5707,
+      "step": 4880
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 4.483850955963135,
+      "learning_rate": 5.166666666666667e-07,
+      "loss": 0.5488,
+      "step": 4890
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 4.9485602378845215,
+      "learning_rate": 5.156565656565657e-07,
+      "loss": 0.5356,
+      "step": 4900
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 3.1415200233459473,
+      "learning_rate": 5.146464646464647e-07,
+      "loss": 0.5455,
+      "step": 4910
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 4.05121374130249,
+      "learning_rate": 5.136363636363636e-07,
+      "loss": 0.5762,
+      "step": 4920
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 4.803497791290283,
+      "learning_rate": 5.126262626262626e-07,
+      "loss": 0.6569,
+      "step": 4930
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 3.7656466960906982,
+      "learning_rate": 5.116161616161616e-07,
+      "loss": 0.5274,
+      "step": 4940
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 6.404185771942139,
+      "learning_rate": 5.106060606060606e-07,
+      "loss": 0.599,
+      "step": 4950
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 4.116934299468994,
+      "learning_rate": 5.095959595959596e-07,
+      "loss": 0.5658,
+      "step": 4960
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 3.6211934089660645,
+      "learning_rate": 5.085858585858585e-07,
+      "loss": 0.6003,
+      "step": 4970
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 4.541522979736328,
+      "learning_rate": 5.075757575757576e-07,
+      "loss": 0.5854,
+      "step": 4980
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 4.65029764175415,
+      "learning_rate": 5.065656565656565e-07,
+      "loss": 0.5504,
+      "step": 4990
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 4.597330093383789,
+      "learning_rate": 5.055555555555555e-07,
+      "loss": 0.5244,
+      "step": 5000
+    },
+    {
+      "epoch": 1.79,
+      "eval_loss": 0.6935617923736572,
+      "eval_runtime": 401.44,
+      "eval_samples_per_second": 2.491,
+      "eval_steps_per_second": 2.491,
+      "step": 5000
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 4.305992126464844,
+      "learning_rate": 5.045454545454545e-07,
+      "loss": 0.5294,
+      "step": 5010
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 9.865640640258789,
+      "learning_rate": 5.035353535353535e-07,
+      "loss": 0.5933,
+      "step": 5020
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 4.958790302276611,
+      "learning_rate": 5.025252525252525e-07,
+      "loss": 0.515,
+      "step": 5030
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 4.508664131164551,
+      "learning_rate": 5.015151515151515e-07,
+      "loss": 0.5725,
+      "step": 5040
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 5.282695293426514,
+      "learning_rate": 5.005050505050504e-07,
+      "loss": 0.5779,
+      "step": 5050
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 3.441600799560547,
+      "learning_rate": 4.994949494949494e-07,
+      "loss": 0.5832,
+      "step": 5060
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 6.297787666320801,
+      "learning_rate": 4.984848484848485e-07,
+      "loss": 0.5923,
+      "step": 5070
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 4.202699184417725,
+      "learning_rate": 4.974747474747474e-07,
+      "loss": 0.5525,
+      "step": 5080
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 6.49127721786499,
+      "learning_rate": 4.964646464646464e-07,
+      "loss": 0.5443,
+      "step": 5090
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 5.5521345138549805,
+      "learning_rate": 4.954545454545454e-07,
+      "loss": 0.5871,
+      "step": 5100
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 6.344796657562256,
+      "learning_rate": 4.944444444444445e-07,
+      "loss": 0.5371,
+      "step": 5110
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 4.388942718505859,
+      "learning_rate": 4.934343434343434e-07,
+      "loss": 0.4874,
+      "step": 5120
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 5.535308837890625,
+      "learning_rate": 4.924242424242424e-07,
+      "loss": 0.5937,
+      "step": 5130
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 3.587024450302124,
+      "learning_rate": 4.914141414141414e-07,
+      "loss": 0.5893,
+      "step": 5140
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 4.178186416625977,
+      "learning_rate": 4.904040404040404e-07,
+      "loss": 0.569,
+      "step": 5150
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 4.824371337890625,
+      "learning_rate": 4.893939393939393e-07,
+      "loss": 0.5923,
+      "step": 5160
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 3.2746853828430176,
+      "learning_rate": 4.883838383838384e-07,
+      "loss": 0.5618,
+      "step": 5170
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 4.693532943725586,
+      "learning_rate": 4.873737373737373e-07,
+      "loss": 0.5875,
+      "step": 5180
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 4.339977264404297,
+      "learning_rate": 4.863636363636364e-07,
+      "loss": 0.5863,
+      "step": 5190
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 4.782194137573242,
+      "learning_rate": 4.853535353535353e-07,
+      "loss": 0.5649,
+      "step": 5200
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 5.156185626983643,
+      "learning_rate": 4.843434343434343e-07,
+      "loss": 0.5377,
+      "step": 5210
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 4.788945198059082,
+      "learning_rate": 4.833333333333333e-07,
+      "loss": 0.4862,
+      "step": 5220
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 4.9057817459106445,
+      "learning_rate": 4.823232323232323e-07,
+      "loss": 0.5838,
+      "step": 5230
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 5.085701942443848,
+      "learning_rate": 4.813131313131313e-07,
+      "loss": 0.6107,
+      "step": 5240
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 3.601012945175171,
+      "learning_rate": 4.803030303030302e-07,
+      "loss": 0.4859,
+      "step": 5250
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 5.70768404006958,
+      "learning_rate": 4.792929292929293e-07,
+      "loss": 0.5206,
+      "step": 5260
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 5.292908191680908,
+      "learning_rate": 4.782828282828283e-07,
+      "loss": 0.5652,
+      "step": 5270
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 5.359946250915527,
+      "learning_rate": 4.772727272727273e-07,
+      "loss": 0.5807,
+      "step": 5280
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 4.304328441619873,
+      "learning_rate": 4.7626262626262627e-07,
+      "loss": 0.5867,
+      "step": 5290
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 3.993483304977417,
+      "learning_rate": 4.752525252525252e-07,
+      "loss": 0.6,
+      "step": 5300
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 5.294483661651611,
+      "learning_rate": 4.742424242424242e-07,
+      "loss": 0.6168,
+      "step": 5310
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 5.31272554397583,
+      "learning_rate": 4.732323232323232e-07,
+      "loss": 0.5645,
+      "step": 5320
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 3.8329992294311523,
+      "learning_rate": 4.722222222222222e-07,
+      "loss": 0.5978,
+      "step": 5330
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 5.329180717468262,
+      "learning_rate": 4.7121212121212115e-07,
+      "loss": 0.562,
+      "step": 5340
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 9.958372116088867,
+      "learning_rate": 4.7020202020202015e-07,
+      "loss": 0.6556,
+      "step": 5350
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 7.695910453796387,
+      "learning_rate": 4.691919191919192e-07,
+      "loss": 0.4964,
+      "step": 5360
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 3.406792163848877,
+      "learning_rate": 4.681818181818182e-07,
+      "loss": 0.5105,
+      "step": 5370
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 6.779275894165039,
+      "learning_rate": 4.6717171717171714e-07,
+      "loss": 0.4962,
+      "step": 5380
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 6.590269088745117,
+      "learning_rate": 4.6616161616161614e-07,
+      "loss": 0.5726,
+      "step": 5390
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 3.9702072143554688,
+      "learning_rate": 4.6515151515151513e-07,
+      "loss": 0.5568,
+      "step": 5400
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 4.912775039672852,
+      "learning_rate": 4.6414141414141413e-07,
+      "loss": 0.5604,
+      "step": 5410
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 5.167705059051514,
+      "learning_rate": 4.6313131313131307e-07,
+      "loss": 0.5364,
+      "step": 5420
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 5.220252513885498,
+      "learning_rate": 4.6212121212121207e-07,
+      "loss": 0.5841,
+      "step": 5430
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 4.965468883514404,
+      "learning_rate": 4.611111111111111e-07,
+      "loss": 0.5934,
+      "step": 5440
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 6.403870582580566,
+      "learning_rate": 4.601010101010101e-07,
+      "loss": 0.5694,
+      "step": 5450
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 4.846173286437988,
+      "learning_rate": 4.5909090909090906e-07,
+      "loss": 0.5431,
+      "step": 5460
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 4.517751216888428,
+      "learning_rate": 4.5808080808080806e-07,
+      "loss": 0.5966,
+      "step": 5470
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 4.690954208374023,
+      "learning_rate": 4.5707070707070705e-07,
+      "loss": 0.5472,
+      "step": 5480
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 5.523789882659912,
+      "learning_rate": 4.5606060606060605e-07,
+      "loss": 0.5442,
+      "step": 5490
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 3.194629430770874,
+      "learning_rate": 4.55050505050505e-07,
+      "loss": 0.6165,
+      "step": 5500
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 5.22808313369751,
+      "learning_rate": 4.54040404040404e-07,
+      "loss": 0.5074,
+      "step": 5510
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 4.611573219299316,
+      "learning_rate": 4.53030303030303e-07,
+      "loss": 0.4977,
+      "step": 5520
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 3.746086835861206,
+      "learning_rate": 4.5202020202020204e-07,
+      "loss": 0.5699,
+      "step": 5530
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 4.322105884552002,
+      "learning_rate": 4.51010101010101e-07,
+      "loss": 0.5817,
+      "step": 5540
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 4.089654922485352,
+      "learning_rate": 4.5e-07,
+      "loss": 0.5593,
+      "step": 5550
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 4.9758710861206055,
+      "learning_rate": 4.48989898989899e-07,
+      "loss": 0.5716,
+      "step": 5560
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 4.36923885345459,
+      "learning_rate": 4.4797979797979797e-07,
+      "loss": 0.5937,
+      "step": 5570
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 4.592928886413574,
+      "learning_rate": 4.469696969696969e-07,
+      "loss": 0.5471,
+      "step": 5580
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 9.409171104431152,
+      "learning_rate": 4.459595959595959e-07,
+      "loss": 0.7263,
+      "step": 5590
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 8.28042221069336,
+      "learning_rate": 4.449494949494949e-07,
+      "loss": 0.5391,
+      "step": 5600
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 4.366082191467285,
+      "learning_rate": 4.4393939393939396e-07,
+      "loss": 0.5164,
+      "step": 5610
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 3.808896541595459,
+      "learning_rate": 4.429292929292929e-07,
+      "loss": 0.66,
+      "step": 5620
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 4.30920934677124,
+      "learning_rate": 4.419191919191919e-07,
+      "loss": 0.5418,
+      "step": 5630
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 4.720395088195801,
+      "learning_rate": 4.409090909090909e-07,
+      "loss": 0.526,
+      "step": 5640
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 3.699301242828369,
+      "learning_rate": 4.398989898989899e-07,
+      "loss": 0.5309,
+      "step": 5650
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 4.549612522125244,
+      "learning_rate": 4.3888888888888884e-07,
+      "loss": 0.6452,
+      "step": 5660
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 3.252539873123169,
+      "learning_rate": 4.3787878787878784e-07,
+      "loss": 0.5111,
+      "step": 5670
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 4.932587146759033,
+      "learning_rate": 4.3686868686868683e-07,
+      "loss": 0.539,
+      "step": 5680
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 4.650817394256592,
+      "learning_rate": 4.3585858585858583e-07,
+      "loss": 0.6203,
+      "step": 5690
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 6.277070045471191,
+      "learning_rate": 4.3484848484848483e-07,
+      "loss": 0.5435,
+      "step": 5700
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 7.179144859313965,
+      "learning_rate": 4.338383838383838e-07,
+      "loss": 0.5062,
+      "step": 5710
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 3.681550979614258,
+      "learning_rate": 4.328282828282828e-07,
+      "loss": 0.5985,
+      "step": 5720
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 3.818920850753784,
+      "learning_rate": 4.318181818181818e-07,
+      "loss": 0.4808,
+      "step": 5730
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 3.2395308017730713,
+      "learning_rate": 4.3080808080808076e-07,
+      "loss": 0.4888,
+      "step": 5740
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 4.037728786468506,
+      "learning_rate": 4.2979797979797976e-07,
+      "loss": 0.6042,
+      "step": 5750
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 5.200090408325195,
+      "learning_rate": 4.2878787878787876e-07,
+      "loss": 0.5801,
+      "step": 5760
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 6.219593524932861,
+      "learning_rate": 4.2777777777777775e-07,
+      "loss": 0.5453,
+      "step": 5770
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 5.049505233764648,
+      "learning_rate": 4.267676767676767e-07,
+      "loss": 0.5342,
+      "step": 5780
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 4.03887939453125,
+      "learning_rate": 4.2575757575757575e-07,
+      "loss": 0.511,
+      "step": 5790
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 4.79861307144165,
+      "learning_rate": 4.2474747474747474e-07,
+      "loss": 0.5106,
+      "step": 5800
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 4.6482744216918945,
+      "learning_rate": 4.2373737373737374e-07,
+      "loss": 0.5887,
+      "step": 5810
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 4.516157150268555,
+      "learning_rate": 4.227272727272727e-07,
+      "loss": 0.5637,
+      "step": 5820
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 5.392308712005615,
+      "learning_rate": 4.217171717171717e-07,
+      "loss": 0.5485,
+      "step": 5830
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 3.8693811893463135,
+      "learning_rate": 4.207070707070707e-07,
+      "loss": 0.496,
+      "step": 5840
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 4.755632400512695,
+      "learning_rate": 4.196969696969697e-07,
+      "loss": 0.5315,
+      "step": 5850
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 5.732184410095215,
+      "learning_rate": 4.186868686868686e-07,
+      "loss": 0.5608,
+      "step": 5860
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 4.804543495178223,
+      "learning_rate": 4.1767676767676767e-07,
+      "loss": 0.5641,
+      "step": 5870
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 6.783200263977051,
+      "learning_rate": 4.1666666666666667e-07,
+      "loss": 0.4667,
+      "step": 5880
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 4.994887828826904,
+      "learning_rate": 4.1565656565656566e-07,
+      "loss": 0.5759,
+      "step": 5890
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 5.834484100341797,
+      "learning_rate": 4.1464646464646466e-07,
+      "loss": 0.5619,
+      "step": 5900
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 5.325469970703125,
+      "learning_rate": 4.136363636363636e-07,
+      "loss": 0.5464,
+      "step": 5910
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 4.983416557312012,
+      "learning_rate": 4.126262626262626e-07,
+      "loss": 0.5974,
+      "step": 5920
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 4.775305271148682,
+      "learning_rate": 4.116161616161616e-07,
+      "loss": 0.5709,
+      "step": 5930
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 5.547243595123291,
+      "learning_rate": 4.1060606060606054e-07,
+      "loss": 0.5474,
+      "step": 5940
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 3.5384747982025146,
+      "learning_rate": 4.0959595959595954e-07,
+      "loss": 0.4197,
+      "step": 5950
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 5.288349628448486,
+      "learning_rate": 4.085858585858586e-07,
+      "loss": 0.6163,
+      "step": 5960
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 5.531615734100342,
+      "learning_rate": 4.075757575757576e-07,
+      "loss": 0.53,
+      "step": 5970
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 6.349400997161865,
+      "learning_rate": 4.065656565656566e-07,
+      "loss": 0.5457,
+      "step": 5980
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 5.897380828857422,
+      "learning_rate": 4.055555555555555e-07,
+      "loss": 0.5782,
+      "step": 5990
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 5.457867622375488,
+      "learning_rate": 4.045454545454545e-07,
+      "loss": 0.5734,
+      "step": 6000
+    },
+    {
+      "epoch": 2.15,
+      "eval_loss": 0.726588785648346,
+      "eval_runtime": 400.7393,
+      "eval_samples_per_second": 2.495,
+      "eval_steps_per_second": 2.495,
+      "step": 6000
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 5.9234538078308105,
+      "learning_rate": 4.035353535353535e-07,
+      "loss": 0.5843,
+      "step": 6010
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 6.228603363037109,
+      "learning_rate": 4.025252525252525e-07,
+      "loss": 0.5035,
+      "step": 6020
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 4.385622024536133,
+      "learning_rate": 4.0151515151515146e-07,
+      "loss": 0.5226,
+      "step": 6030
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 5.10957670211792,
+      "learning_rate": 4.005050505050505e-07,
+      "loss": 0.5418,
+      "step": 6040
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 3.951185703277588,
+      "learning_rate": 3.994949494949495e-07,
+      "loss": 0.5448,
+      "step": 6050
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 4.83636474609375,
+      "learning_rate": 3.984848484848485e-07,
+      "loss": 0.5483,
+      "step": 6060
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 6.703040599822998,
+      "learning_rate": 3.9747474747474745e-07,
+      "loss": 0.5568,
+      "step": 6070
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 6.855874538421631,
+      "learning_rate": 3.9646464646464644e-07,
+      "loss": 0.5365,
+      "step": 6080
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 4.763612270355225,
+      "learning_rate": 3.9545454545454544e-07,
+      "loss": 0.5672,
+      "step": 6090
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 5.316641330718994,
+      "learning_rate": 3.9444444444444444e-07,
+      "loss": 0.5943,
+      "step": 6100
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 7.093216896057129,
+      "learning_rate": 3.934343434343434e-07,
+      "loss": 0.5548,
+      "step": 6110
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 5.546849250793457,
+      "learning_rate": 3.924242424242424e-07,
+      "loss": 0.5377,
+      "step": 6120
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 6.597376823425293,
+      "learning_rate": 3.9141414141414143e-07,
+      "loss": 0.5512,
+      "step": 6130
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 4.313337326049805,
+      "learning_rate": 3.904040404040404e-07,
+      "loss": 0.498,
+      "step": 6140
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 7.560310363769531,
+      "learning_rate": 3.8939393939393937e-07,
+      "loss": 0.5511,
+      "step": 6150
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 4.503920555114746,
+      "learning_rate": 3.8838383838383837e-07,
+      "loss": 0.5497,
+      "step": 6160
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 5.242457866668701,
+      "learning_rate": 3.8737373737373736e-07,
+      "loss": 0.6372,
+      "step": 6170
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 5.803231716156006,
+      "learning_rate": 3.8636363636363636e-07,
+      "loss": 0.5811,
+      "step": 6180
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 4.9778008460998535,
+      "learning_rate": 3.853535353535353e-07,
+      "loss": 0.5717,
+      "step": 6190
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 3.297873020172119,
+      "learning_rate": 3.843434343434343e-07,
+      "loss": 0.5374,
+      "step": 6200
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 4.822756290435791,
+      "learning_rate": 3.8333333333333335e-07,
+      "loss": 0.5788,
+      "step": 6210
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 6.319406509399414,
+      "learning_rate": 3.8232323232323235e-07,
+      "loss": 0.4811,
+      "step": 6220
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 5.257617950439453,
+      "learning_rate": 3.813131313131313e-07,
+      "loss": 0.6003,
+      "step": 6230
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 5.456783771514893,
+      "learning_rate": 3.803030303030303e-07,
+      "loss": 0.4827,
+      "step": 6240
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 5.607643127441406,
+      "learning_rate": 3.792929292929293e-07,
+      "loss": 0.4954,
+      "step": 6250
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 7.377701282501221,
+      "learning_rate": 3.782828282828283e-07,
+      "loss": 0.6513,
+      "step": 6260
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 5.05380392074585,
+      "learning_rate": 3.7727272727272723e-07,
+      "loss": 0.5108,
+      "step": 6270
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 5.01906681060791,
+      "learning_rate": 3.762626262626262e-07,
+      "loss": 0.5575,
+      "step": 6280
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 4.88652229309082,
+      "learning_rate": 3.752525252525252e-07,
+      "loss": 0.5732,
+      "step": 6290
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 4.311870098114014,
+      "learning_rate": 3.7424242424242427e-07,
+      "loss": 0.5814,
+      "step": 6300
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 5.589522361755371,
+      "learning_rate": 3.732323232323232e-07,
+      "loss": 0.5317,
+      "step": 6310
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 7.32663631439209,
+      "learning_rate": 3.722222222222222e-07,
+      "loss": 0.5446,
+      "step": 6320
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 4.902598857879639,
+      "learning_rate": 3.712121212121212e-07,
+      "loss": 0.5386,
+      "step": 6330
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 5.21954870223999,
+      "learning_rate": 3.702020202020202e-07,
+      "loss": 0.4813,
+      "step": 6340
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 3.8048980236053467,
+      "learning_rate": 3.6919191919191915e-07,
+      "loss": 0.5571,
+      "step": 6350
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 5.75179386138916,
+      "learning_rate": 3.6818181818181815e-07,
+      "loss": 0.6023,
+      "step": 6360
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 6.469977855682373,
+      "learning_rate": 3.6717171717171714e-07,
+      "loss": 0.65,
+      "step": 6370
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 5.1519317626953125,
+      "learning_rate": 3.661616161616162e-07,
+      "loss": 0.5634,
+      "step": 6380
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 4.514432907104492,
+      "learning_rate": 3.6515151515151514e-07,
+      "loss": 0.5638,
+      "step": 6390
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 5.5475029945373535,
+      "learning_rate": 3.6414141414141413e-07,
+      "loss": 0.5922,
+      "step": 6400
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 6.521148204803467,
+      "learning_rate": 3.6313131313131313e-07,
+      "loss": 0.5516,
+      "step": 6410
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 4.127744674682617,
+      "learning_rate": 3.6212121212121213e-07,
+      "loss": 0.5455,
+      "step": 6420
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 5.580835342407227,
+      "learning_rate": 3.6111111111111107e-07,
+      "loss": 0.4762,
+      "step": 6430
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 5.267888069152832,
+      "learning_rate": 3.6010101010101007e-07,
+      "loss": 0.6285,
+      "step": 6440
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 5.960037708282471,
+      "learning_rate": 3.5909090909090907e-07,
+      "loss": 0.5553,
+      "step": 6450
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 3.847745656967163,
+      "learning_rate": 3.5808080808080806e-07,
+      "loss": 0.5603,
+      "step": 6460
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 5.693478107452393,
+      "learning_rate": 3.5707070707070706e-07,
+      "loss": 0.6077,
+      "step": 6470
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 4.48386287689209,
+      "learning_rate": 3.5606060606060606e-07,
+      "loss": 0.5636,
+      "step": 6480
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 6.377192497253418,
+      "learning_rate": 3.5505050505050505e-07,
+      "loss": 0.5089,
+      "step": 6490
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 5.240867614746094,
+      "learning_rate": 3.5404040404040405e-07,
+      "loss": 0.5547,
+      "step": 6500
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 7.864344120025635,
+      "learning_rate": 3.53030303030303e-07,
+      "loss": 0.5254,
+      "step": 6510
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 8.867986679077148,
+      "learning_rate": 3.52020202020202e-07,
+      "loss": 0.5616,
+      "step": 6520
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 4.047301292419434,
+      "learning_rate": 3.51010101010101e-07,
+      "loss": 0.5123,
+      "step": 6530
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 5.919891357421875,
+      "learning_rate": 3.5e-07,
+      "loss": 0.5126,
+      "step": 6540
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 4.536365509033203,
+      "learning_rate": 3.48989898989899e-07,
+      "loss": 0.495,
+      "step": 6550
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 5.329809188842773,
+      "learning_rate": 3.47979797979798e-07,
+      "loss": 0.5118,
+      "step": 6560
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 5.0158867835998535,
+      "learning_rate": 3.46969696969697e-07,
+      "loss": 0.5487,
+      "step": 6570
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 7.028294086456299,
+      "learning_rate": 3.4595959595959597e-07,
+      "loss": 0.5893,
+      "step": 6580
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 4.060366153717041,
+      "learning_rate": 3.449494949494949e-07,
+      "loss": 0.5562,
+      "step": 6590
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 5.36353063583374,
+      "learning_rate": 3.439393939393939e-07,
+      "loss": 0.6158,
+      "step": 6600
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 4.725005626678467,
+      "learning_rate": 3.429292929292929e-07,
+      "loss": 0.4339,
+      "step": 6610
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 4.875920295715332,
+      "learning_rate": 3.419191919191919e-07,
+      "loss": 0.5047,
+      "step": 6620
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 6.4596266746521,
+      "learning_rate": 3.4090909090909085e-07,
+      "loss": 0.5516,
+      "step": 6630
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 6.09486722946167,
+      "learning_rate": 3.398989898989899e-07,
+      "loss": 0.5863,
+      "step": 6640
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 4.232431411743164,
+      "learning_rate": 3.388888888888889e-07,
+      "loss": 0.5221,
+      "step": 6650
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 4.1522111892700195,
+      "learning_rate": 3.378787878787879e-07,
+      "loss": 0.588,
+      "step": 6660
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 5.693157196044922,
+      "learning_rate": 3.3686868686868684e-07,
+      "loss": 0.5966,
+      "step": 6670
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 3.900887966156006,
+      "learning_rate": 3.3585858585858583e-07,
+      "loss": 0.437,
+      "step": 6680
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 5.744729995727539,
+      "learning_rate": 3.3484848484848483e-07,
+      "loss": 0.5483,
+      "step": 6690
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 4.26437520980835,
+      "learning_rate": 3.3383838383838383e-07,
+      "loss": 0.5159,
+      "step": 6700
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 5.422147750854492,
+      "learning_rate": 3.3282828282828277e-07,
+      "loss": 0.4986,
+      "step": 6710
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 4.445804119110107,
+      "learning_rate": 3.318181818181818e-07,
+      "loss": 0.544,
+      "step": 6720
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 5.514401435852051,
+      "learning_rate": 3.308080808080808e-07,
+      "loss": 0.5422,
+      "step": 6730
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 7.159079074859619,
+      "learning_rate": 3.297979797979798e-07,
+      "loss": 0.503,
+      "step": 6740
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 6.815921783447266,
+      "learning_rate": 3.2878787878787876e-07,
+      "loss": 0.4779,
+      "step": 6750
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 4.787651538848877,
+      "learning_rate": 3.2777777777777776e-07,
+      "loss": 0.484,
+      "step": 6760
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 5.437996864318848,
+      "learning_rate": 3.2676767676767675e-07,
+      "loss": 0.5689,
+      "step": 6770
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 2.6372170448303223,
+      "learning_rate": 3.2575757575757575e-07,
+      "loss": 0.6268,
+      "step": 6780
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 6.545858383178711,
+      "learning_rate": 3.247474747474747e-07,
+      "loss": 0.603,
+      "step": 6790
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 3.9520442485809326,
+      "learning_rate": 3.237373737373737e-07,
+      "loss": 0.6243,
+      "step": 6800
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 4.638275623321533,
+      "learning_rate": 3.2272727272727274e-07,
+      "loss": 0.5433,
+      "step": 6810
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 4.08539342880249,
+      "learning_rate": 3.2171717171717174e-07,
+      "loss": 0.576,
+      "step": 6820
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 6.024731636047363,
+      "learning_rate": 3.207070707070707e-07,
+      "loss": 0.61,
+      "step": 6830
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 3.649667263031006,
+      "learning_rate": 3.196969696969697e-07,
+      "loss": 0.4797,
+      "step": 6840
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 7.055994987487793,
+      "learning_rate": 3.186868686868687e-07,
+      "loss": 0.5525,
+      "step": 6850
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 4.7312116622924805,
+      "learning_rate": 3.1767676767676767e-07,
+      "loss": 0.5069,
+      "step": 6860
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 4.483476638793945,
+      "learning_rate": 3.166666666666666e-07,
+      "loss": 0.4884,
+      "step": 6870
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 7.678333282470703,
+      "learning_rate": 3.156565656565656e-07,
+      "loss": 0.4818,
+      "step": 6880
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 4.442699909210205,
+      "learning_rate": 3.1464646464646466e-07,
+      "loss": 0.5771,
+      "step": 6890
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 4.732019424438477,
+      "learning_rate": 3.1363636363636366e-07,
+      "loss": 0.5294,
+      "step": 6900
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 5.647908687591553,
+      "learning_rate": 3.126262626262626e-07,
+      "loss": 0.5496,
+      "step": 6910
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 4.71852970123291,
+      "learning_rate": 3.116161616161616e-07,
+      "loss": 0.503,
+      "step": 6920
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 8.519858360290527,
+      "learning_rate": 3.106060606060606e-07,
+      "loss": 0.547,
+      "step": 6930
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 4.513672351837158,
+      "learning_rate": 3.095959595959596e-07,
+      "loss": 0.5496,
+      "step": 6940
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 4.779497146606445,
+      "learning_rate": 3.0858585858585854e-07,
+      "loss": 0.5193,
+      "step": 6950
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 5.14954948425293,
+      "learning_rate": 3.0757575757575754e-07,
+      "loss": 0.5738,
+      "step": 6960
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 4.722686290740967,
+      "learning_rate": 3.0656565656565653e-07,
+      "loss": 0.5978,
+      "step": 6970
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 4.7376837730407715,
+      "learning_rate": 3.055555555555556e-07,
+      "loss": 0.5476,
+      "step": 6980
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 5.883960247039795,
+      "learning_rate": 3.0454545454545453e-07,
+      "loss": 0.4953,
+      "step": 6990
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 5.1609721183776855,
+      "learning_rate": 3.035353535353535e-07,
+      "loss": 0.5602,
+      "step": 7000
+    },
+    {
+      "epoch": 2.51,
+      "eval_loss": 0.7043054103851318,
+      "eval_runtime": 404.5936,
+      "eval_samples_per_second": 2.472,
+      "eval_steps_per_second": 2.472,
+      "step": 7000
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 4.467599391937256,
+      "learning_rate": 3.025252525252525e-07,
+      "loss": 0.5545,
+      "step": 7010
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 4.831850528717041,
+      "learning_rate": 3.015151515151515e-07,
+      "loss": 0.5765,
+      "step": 7020
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 4.63700008392334,
+      "learning_rate": 3.0050505050505046e-07,
+      "loss": 0.5354,
+      "step": 7030
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 4.822059631347656,
+      "learning_rate": 2.9949494949494946e-07,
+      "loss": 0.5239,
+      "step": 7040
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 4.098064422607422,
+      "learning_rate": 2.9848484848484846e-07,
+      "loss": 0.5775,
+      "step": 7050
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 5.17026948928833,
+      "learning_rate": 2.9747474747474745e-07,
+      "loss": 0.6014,
+      "step": 7060
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 4.5090789794921875,
+      "learning_rate": 2.9646464646464645e-07,
+      "loss": 0.5732,
+      "step": 7070
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 6.120937347412109,
+      "learning_rate": 2.9545454545454545e-07,
+      "loss": 0.4529,
+      "step": 7080
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 5.249495029449463,
+      "learning_rate": 2.9444444444444444e-07,
+      "loss": 0.5804,
+      "step": 7090
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 7.617781639099121,
+      "learning_rate": 2.9343434343434344e-07,
+      "loss": 0.5382,
+      "step": 7100
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 4.722599506378174,
+      "learning_rate": 2.924242424242424e-07,
+      "loss": 0.5419,
+      "step": 7110
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 4.24169921875,
+      "learning_rate": 2.914141414141414e-07,
+      "loss": 0.5464,
+      "step": 7120
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 4.781604766845703,
+      "learning_rate": 2.904040404040404e-07,
+      "loss": 0.4158,
+      "step": 7130
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 5.138591766357422,
+      "learning_rate": 2.893939393939394e-07,
+      "loss": 0.5626,
+      "step": 7140
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 4.901473045349121,
+      "learning_rate": 2.8838383838383837e-07,
+      "loss": 0.5164,
+      "step": 7150
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 3.736123561859131,
+      "learning_rate": 2.8737373737373737e-07,
+      "loss": 0.5159,
+      "step": 7160
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 3.937110662460327,
+      "learning_rate": 2.8636363636363637e-07,
+      "loss": 0.5785,
+      "step": 7170
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 5.2640581130981445,
+      "learning_rate": 2.8535353535353536e-07,
+      "loss": 0.5839,
+      "step": 7180
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 5.652442455291748,
+      "learning_rate": 2.843434343434343e-07,
+      "loss": 0.533,
+      "step": 7190
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 4.787551403045654,
+      "learning_rate": 2.833333333333333e-07,
+      "loss": 0.5114,
+      "step": 7200
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 5.234408855438232,
+      "learning_rate": 2.823232323232323e-07,
+      "loss": 0.5777,
+      "step": 7210
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 2.3309760093688965,
+      "learning_rate": 2.813131313131313e-07,
+      "loss": 0.5085,
+      "step": 7220
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 4.90077543258667,
+      "learning_rate": 2.8030303030303024e-07,
+      "loss": 0.5464,
+      "step": 7230
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 6.651805400848389,
+      "learning_rate": 2.792929292929293e-07,
+      "loss": 0.5143,
+      "step": 7240
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 4.081473350524902,
+      "learning_rate": 2.782828282828283e-07,
+      "loss": 0.5307,
+      "step": 7250
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 8.496968269348145,
+      "learning_rate": 2.772727272727273e-07,
+      "loss": 0.5376,
+      "step": 7260
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 4.46324348449707,
+      "learning_rate": 2.7626262626262623e-07,
+      "loss": 0.5871,
+      "step": 7270
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 5.344108581542969,
+      "learning_rate": 2.752525252525252e-07,
+      "loss": 0.4783,
+      "step": 7280
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 6.3261613845825195,
+      "learning_rate": 2.742424242424242e-07,
+      "loss": 0.4924,
+      "step": 7290
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 4.785797119140625,
+      "learning_rate": 2.732323232323232e-07,
+      "loss": 0.5087,
+      "step": 7300
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 5.484802722930908,
+      "learning_rate": 2.7222222222222216e-07,
+      "loss": 0.5763,
+      "step": 7310
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 5.250958442687988,
+      "learning_rate": 2.712121212121212e-07,
+      "loss": 0.5065,
+      "step": 7320
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 3.2717227935791016,
+      "learning_rate": 2.702020202020202e-07,
+      "loss": 0.4802,
+      "step": 7330
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 5.094733238220215,
+      "learning_rate": 2.691919191919192e-07,
+      "loss": 0.5162,
+      "step": 7340
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 5.360857009887695,
+      "learning_rate": 2.6818181818181815e-07,
+      "loss": 0.5947,
+      "step": 7350
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 5.611083030700684,
+      "learning_rate": 2.6717171717171715e-07,
+      "loss": 0.6174,
+      "step": 7360
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 4.42379093170166,
+      "learning_rate": 2.6616161616161614e-07,
+      "loss": 0.5773,
+      "step": 7370
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 6.272792816162109,
+      "learning_rate": 2.6515151515151514e-07,
+      "loss": 0.6246,
+      "step": 7380
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 5.295745849609375,
+      "learning_rate": 2.641414141414141e-07,
+      "loss": 0.5315,
+      "step": 7390
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 4.078943252563477,
+      "learning_rate": 2.631313131313131e-07,
+      "loss": 0.581,
+      "step": 7400
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 12.75452709197998,
+      "learning_rate": 2.6212121212121213e-07,
+      "loss": 0.5608,
+      "step": 7410
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 5.873865604400635,
+      "learning_rate": 2.6111111111111113e-07,
+      "loss": 0.5588,
+      "step": 7420
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 4.560327529907227,
+      "learning_rate": 2.6010101010101007e-07,
+      "loss": 0.6557,
+      "step": 7430
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 7.971694469451904,
+      "learning_rate": 2.5909090909090907e-07,
+      "loss": 0.5636,
+      "step": 7440
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 4.814733982086182,
+      "learning_rate": 2.5808080808080807e-07,
+      "loss": 0.5851,
+      "step": 7450
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 3.635782241821289,
+      "learning_rate": 2.5707070707070706e-07,
+      "loss": 0.5354,
+      "step": 7460
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 6.356813907623291,
+      "learning_rate": 2.56060606060606e-07,
+      "loss": 0.5934,
+      "step": 7470
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 6.781961441040039,
+      "learning_rate": 2.55050505050505e-07,
+      "loss": 0.5249,
+      "step": 7480
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 3.655015230178833,
+      "learning_rate": 2.5404040404040405e-07,
+      "loss": 0.5198,
+      "step": 7490
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 6.139344692230225,
+      "learning_rate": 2.5303030303030305e-07,
+      "loss": 0.5375,
+      "step": 7500
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 4.679745197296143,
+      "learning_rate": 2.5202020202020205e-07,
+      "loss": 0.5107,
+      "step": 7510
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 5.100747585296631,
+      "learning_rate": 2.51010101010101e-07,
+      "loss": 0.515,
+      "step": 7520
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 3.9256341457366943,
+      "learning_rate": 2.5e-07,
+      "loss": 0.5409,
+      "step": 7530
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 4.619549751281738,
+      "learning_rate": 2.48989898989899e-07,
+      "loss": 0.5024,
+      "step": 7540
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 4.833261489868164,
+      "learning_rate": 2.4797979797979793e-07,
+      "loss": 0.5287,
+      "step": 7550
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 3.5149707794189453,
+      "learning_rate": 2.46969696969697e-07,
+      "loss": 0.5176,
+      "step": 7560
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 6.924714088439941,
+      "learning_rate": 2.45959595959596e-07,
+      "loss": 0.5598,
+      "step": 7570
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 7.813880443572998,
+      "learning_rate": 2.449494949494949e-07,
+      "loss": 0.535,
+      "step": 7580
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 5.943690299987793,
+      "learning_rate": 2.439393939393939e-07,
+      "loss": 0.5327,
+      "step": 7590
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 5.665824890136719,
+      "learning_rate": 2.429292929292929e-07,
+      "loss": 0.501,
+      "step": 7600
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 3.8446450233459473,
+      "learning_rate": 2.419191919191919e-07,
+      "loss": 0.5307,
+      "step": 7610
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 4.556608200073242,
+      "learning_rate": 2.409090909090909e-07,
+      "loss": 0.5244,
+      "step": 7620
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 4.267519950866699,
+      "learning_rate": 2.398989898989899e-07,
+      "loss": 0.512,
+      "step": 7630
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 6.164044380187988,
+      "learning_rate": 2.388888888888889e-07,
+      "loss": 0.4957,
+      "step": 7640
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 5.647671222686768,
+      "learning_rate": 2.3787878787878787e-07,
+      "loss": 0.6485,
+      "step": 7650
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 3.6924054622650146,
+      "learning_rate": 2.3686868686868687e-07,
+      "loss": 0.4993,
+      "step": 7660
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 5.332742691040039,
+      "learning_rate": 2.3585858585858584e-07,
+      "loss": 0.5619,
+      "step": 7670
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 4.201364517211914,
+      "learning_rate": 2.3484848484848486e-07,
+      "loss": 0.5011,
+      "step": 7680
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 4.477077484130859,
+      "learning_rate": 2.3383838383838383e-07,
+      "loss": 0.5369,
+      "step": 7690
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 4.704535961151123,
+      "learning_rate": 2.3282828282828283e-07,
+      "loss": 0.5495,
+      "step": 7700
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 3.2071988582611084,
+      "learning_rate": 2.318181818181818e-07,
+      "loss": 0.5478,
+      "step": 7710
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 4.613890647888184,
+      "learning_rate": 2.308080808080808e-07,
+      "loss": 0.5208,
+      "step": 7720
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 5.759297847747803,
+      "learning_rate": 2.297979797979798e-07,
+      "loss": 0.4989,
+      "step": 7730
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 3.805091619491577,
+      "learning_rate": 2.287878787878788e-07,
+      "loss": 0.521,
+      "step": 7740
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 5.901796817779541,
+      "learning_rate": 2.2777777777777776e-07,
+      "loss": 0.603,
+      "step": 7750
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 4.832998752593994,
+      "learning_rate": 2.2676767676767676e-07,
+      "loss": 0.5698,
+      "step": 7760
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 4.8905439376831055,
+      "learning_rate": 2.2575757575757576e-07,
+      "loss": 0.4658,
+      "step": 7770
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 7.449604511260986,
+      "learning_rate": 2.2474747474747475e-07,
+      "loss": 0.5889,
+      "step": 7780
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 8.905882835388184,
+      "learning_rate": 2.2373737373737372e-07,
+      "loss": 0.5323,
+      "step": 7790
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 4.761772632598877,
+      "learning_rate": 2.2272727272727272e-07,
+      "loss": 0.53,
+      "step": 7800
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 4.811107158660889,
+      "learning_rate": 2.2171717171717172e-07,
+      "loss": 0.5189,
+      "step": 7810
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 2.9238836765289307,
+      "learning_rate": 2.2070707070707071e-07,
+      "loss": 0.5283,
+      "step": 7820
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 5.508478164672852,
+      "learning_rate": 2.1969696969696968e-07,
+      "loss": 0.4561,
+      "step": 7830
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 5.604992866516113,
+      "learning_rate": 2.1868686868686868e-07,
+      "loss": 0.5718,
+      "step": 7840
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 6.870204925537109,
+      "learning_rate": 2.1767676767676765e-07,
+      "loss": 0.4882,
+      "step": 7850
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 4.332894802093506,
+      "learning_rate": 2.1666666666666667e-07,
+      "loss": 0.5755,
+      "step": 7860
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 5.616927623748779,
+      "learning_rate": 2.1565656565656564e-07,
+      "loss": 0.5805,
+      "step": 7870
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 4.477654933929443,
+      "learning_rate": 2.1464646464646464e-07,
+      "loss": 0.582,
+      "step": 7880
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 6.561676025390625,
+      "learning_rate": 2.136363636363636e-07,
+      "loss": 0.4924,
+      "step": 7890
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 5.206873416900635,
+      "learning_rate": 2.1262626262626264e-07,
+      "loss": 0.5878,
+      "step": 7900
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 4.1171345710754395,
+      "learning_rate": 2.116161616161616e-07,
+      "loss": 0.5451,
+      "step": 7910
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 3.9070441722869873,
+      "learning_rate": 2.106060606060606e-07,
+      "loss": 0.5403,
+      "step": 7920
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 4.80625057220459,
+      "learning_rate": 2.0959595959595957e-07,
+      "loss": 0.5552,
+      "step": 7930
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 4.501746654510498,
+      "learning_rate": 2.085858585858586e-07,
+      "loss": 0.4697,
+      "step": 7940
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 3.92985463142395,
+      "learning_rate": 2.0757575757575757e-07,
+      "loss": 0.5565,
+      "step": 7950
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 4.4594807624816895,
+      "learning_rate": 2.0656565656565656e-07,
+      "loss": 0.561,
+      "step": 7960
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 5.731437683105469,
+      "learning_rate": 2.0555555555555553e-07,
+      "loss": 0.5311,
+      "step": 7970
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 4.295395374298096,
+      "learning_rate": 2.0454545454545456e-07,
+      "loss": 0.6082,
+      "step": 7980
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 4.862415313720703,
+      "learning_rate": 2.0353535353535353e-07,
+      "loss": 0.6022,
+      "step": 7990
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 4.604592323303223,
+      "learning_rate": 2.0252525252525253e-07,
+      "loss": 0.5314,
+      "step": 8000
+    },
+    {
+      "epoch": 2.87,
+      "eval_loss": 0.7028300762176514,
+      "eval_runtime": 404.8169,
+      "eval_samples_per_second": 2.47,
+      "eval_steps_per_second": 2.47,
+      "step": 8000
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 4.165442943572998,
+      "learning_rate": 2.015151515151515e-07,
+      "loss": 0.5624,
+      "step": 8010
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 4.097321510314941,
+      "learning_rate": 2.005050505050505e-07,
+      "loss": 0.5041,
+      "step": 8020
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 4.476957321166992,
+      "learning_rate": 1.994949494949495e-07,
+      "loss": 0.4556,
+      "step": 8030
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 4.502745628356934,
+      "learning_rate": 1.9848484848484849e-07,
+      "loss": 0.535,
+      "step": 8040
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 5.133005142211914,
+      "learning_rate": 1.9747474747474746e-07,
+      "loss": 0.4998,
+      "step": 8050
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 5.7199296951293945,
+      "learning_rate": 1.9646464646464645e-07,
+      "loss": 0.6113,
+      "step": 8060
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 4.979962348937988,
+      "learning_rate": 1.9545454545454545e-07,
+      "loss": 0.6419,
+      "step": 8070
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 5.971510410308838,
+      "learning_rate": 1.9444444444444445e-07,
+      "loss": 0.5269,
+      "step": 8080
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 6.548705577850342,
+      "learning_rate": 1.9343434343434342e-07,
+      "loss": 0.456,
+      "step": 8090
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 5.575558185577393,
+      "learning_rate": 1.9242424242424241e-07,
+      "loss": 0.5068,
+      "step": 8100
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 6.25363826751709,
+      "learning_rate": 1.914141414141414e-07,
+      "loss": 0.4749,
+      "step": 8110
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 5.45524787902832,
+      "learning_rate": 1.904040404040404e-07,
+      "loss": 0.5556,
+      "step": 8120
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 5.0804548263549805,
+      "learning_rate": 1.8939393939393938e-07,
+      "loss": 0.5058,
+      "step": 8130
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 6.4303460121154785,
+      "learning_rate": 1.8838383838383838e-07,
+      "loss": 0.553,
+      "step": 8140
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 5.177044868469238,
+      "learning_rate": 1.8737373737373737e-07,
+      "loss": 0.5986,
+      "step": 8150
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 4.609193325042725,
+      "learning_rate": 1.8636363636363637e-07,
+      "loss": 0.5456,
+      "step": 8160
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 5.29266357421875,
+      "learning_rate": 1.8535353535353534e-07,
+      "loss": 0.5806,
+      "step": 8170
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 6.320983409881592,
+      "learning_rate": 1.8434343434343434e-07,
+      "loss": 0.5621,
+      "step": 8180
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 5.851290702819824,
+      "learning_rate": 1.833333333333333e-07,
+      "loss": 0.5907,
+      "step": 8190
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 7.4724555015563965,
+      "learning_rate": 1.8232323232323233e-07,
+      "loss": 0.5415,
+      "step": 8200
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 5.282331943511963,
+      "learning_rate": 1.813131313131313e-07,
+      "loss": 0.4809,
+      "step": 8210
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 5.070183277130127,
+      "learning_rate": 1.803030303030303e-07,
+      "loss": 0.5138,
+      "step": 8220
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 4.092304229736328,
+      "learning_rate": 1.7929292929292927e-07,
+      "loss": 0.4772,
+      "step": 8230
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 5.767254829406738,
+      "learning_rate": 1.782828282828283e-07,
+      "loss": 0.5756,
+      "step": 8240
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 5.086340427398682,
+      "learning_rate": 1.7727272727272726e-07,
+      "loss": 0.6313,
+      "step": 8250
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 7.967188835144043,
+      "learning_rate": 1.7626262626262626e-07,
+      "loss": 0.5307,
+      "step": 8260
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 4.956604480743408,
+      "learning_rate": 1.7525252525252523e-07,
+      "loss": 0.4744,
+      "step": 8270
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 5.134040355682373,
+      "learning_rate": 1.7424242424242425e-07,
+      "loss": 0.5217,
+      "step": 8280
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 4.910244464874268,
+      "learning_rate": 1.7323232323232322e-07,
+      "loss": 0.513,
+      "step": 8290
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 6.457435131072998,
+      "learning_rate": 1.7222222222222222e-07,
+      "loss": 0.5786,
+      "step": 8300
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 5.450909614562988,
+      "learning_rate": 1.712121212121212e-07,
+      "loss": 0.4931,
+      "step": 8310
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 5.564160346984863,
+      "learning_rate": 1.7020202020202021e-07,
+      "loss": 0.5263,
+      "step": 8320
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 6.898198127746582,
+      "learning_rate": 1.6919191919191918e-07,
+      "loss": 0.5814,
+      "step": 8330
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 5.737578392028809,
+      "learning_rate": 1.6818181818181818e-07,
+      "loss": 0.498,
+      "step": 8340
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 4.877408027648926,
+      "learning_rate": 1.6717171717171715e-07,
+      "loss": 0.5743,
+      "step": 8350
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 5.765985012054443,
+      "learning_rate": 1.6616161616161615e-07,
+      "loss": 0.5476,
+      "step": 8360
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 12.480027198791504,
+      "learning_rate": 1.6515151515151515e-07,
+      "loss": 0.5693,
+      "step": 8370
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 8.612441062927246,
+      "learning_rate": 1.6414141414141414e-07,
+      "loss": 0.6035,
+      "step": 8380
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 5.978526592254639,
+      "learning_rate": 1.631313131313131e-07,
+      "loss": 0.5711,
+      "step": 8390
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 6.2041192054748535,
+      "learning_rate": 1.621212121212121e-07,
+      "loss": 0.4838,
+      "step": 8400
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 3.6861565113067627,
+      "learning_rate": 1.611111111111111e-07,
+      "loss": 0.6104,
+      "step": 8410
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 5.080516338348389,
+      "learning_rate": 1.601010101010101e-07,
+      "loss": 0.4688,
+      "step": 8420
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 5.527194976806641,
+      "learning_rate": 1.5909090909090907e-07,
+      "loss": 0.5746,
+      "step": 8430
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 4.398497104644775,
+      "learning_rate": 1.5808080808080807e-07,
+      "loss": 0.4895,
+      "step": 8440
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 5.004568099975586,
+      "learning_rate": 1.5707070707070707e-07,
+      "loss": 0.5808,
+      "step": 8450
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 6.635837554931641,
+      "learning_rate": 1.5606060606060606e-07,
+      "loss": 0.5519,
+      "step": 8460
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 4.912801742553711,
+      "learning_rate": 1.5505050505050504e-07,
+      "loss": 0.5873,
+      "step": 8470
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 5.812663555145264,
+      "learning_rate": 1.5404040404040403e-07,
+      "loss": 0.5286,
+      "step": 8480
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 5.986713409423828,
+      "learning_rate": 1.53030303030303e-07,
+      "loss": 0.5305,
+      "step": 8490
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 5.908451557159424,
+      "learning_rate": 1.5202020202020203e-07,
+      "loss": 0.6125,
+      "step": 8500
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 8.727606773376465,
+      "learning_rate": 1.51010101010101e-07,
+      "loss": 0.5765,
+      "step": 8510
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 5.8582987785339355,
+      "learning_rate": 1.5e-07,
+      "loss": 0.5178,
+      "step": 8520
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 5.690489292144775,
+      "learning_rate": 1.4898989898989896e-07,
+      "loss": 0.4978,
+      "step": 8530
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 5.1287384033203125,
+      "learning_rate": 1.47979797979798e-07,
+      "loss": 0.5135,
+      "step": 8540
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 5.430093765258789,
+      "learning_rate": 1.4696969696969696e-07,
+      "loss": 0.4935,
+      "step": 8550
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 5.307154655456543,
+      "learning_rate": 1.4595959595959595e-07,
+      "loss": 0.4839,
+      "step": 8560
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 4.900834560394287,
+      "learning_rate": 1.4494949494949492e-07,
+      "loss": 0.5637,
+      "step": 8570
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 4.796194553375244,
+      "learning_rate": 1.4393939393939395e-07,
+      "loss": 0.519,
+      "step": 8580
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 6.488938331604004,
+      "learning_rate": 1.4292929292929292e-07,
+      "loss": 0.6017,
+      "step": 8590
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 4.90757417678833,
+      "learning_rate": 1.4191919191919192e-07,
+      "loss": 0.4873,
+      "step": 8600
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 6.44563627243042,
+      "learning_rate": 1.4090909090909089e-07,
+      "loss": 0.5919,
+      "step": 8610
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 3.680788516998291,
+      "learning_rate": 1.398989898989899e-07,
+      "loss": 0.5062,
+      "step": 8620
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 4.880250930786133,
+      "learning_rate": 1.3888888888888888e-07,
+      "loss": 0.5751,
+      "step": 8630
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 4.91787052154541,
+      "learning_rate": 1.3787878787878788e-07,
+      "loss": 0.5604,
+      "step": 8640
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 4.990437984466553,
+      "learning_rate": 1.3686868686868685e-07,
+      "loss": 0.4807,
+      "step": 8650
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 4.082001209259033,
+      "learning_rate": 1.3585858585858584e-07,
+      "loss": 0.4771,
+      "step": 8660
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 3.742047071456909,
+      "learning_rate": 1.3484848484848484e-07,
+      "loss": 0.5773,
+      "step": 8670
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 6.419268608093262,
+      "learning_rate": 1.3383838383838384e-07,
+      "loss": 0.5903,
+      "step": 8680
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 5.8881940841674805,
+      "learning_rate": 1.328282828282828e-07,
+      "loss": 0.5279,
+      "step": 8690
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 4.6417999267578125,
+      "learning_rate": 1.318181818181818e-07,
+      "loss": 0.5368,
+      "step": 8700
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 5.411186218261719,
+      "learning_rate": 1.308080808080808e-07,
+      "loss": 0.5545,
+      "step": 8710
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 4.455116271972656,
+      "learning_rate": 1.297979797979798e-07,
+      "loss": 0.5631,
+      "step": 8720
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 11.706398963928223,
+      "learning_rate": 1.2878787878787877e-07,
+      "loss": 0.5218,
+      "step": 8730
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 6.45891809463501,
+      "learning_rate": 1.2777777777777777e-07,
+      "loss": 0.5383,
+      "step": 8740
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 4.817904949188232,
+      "learning_rate": 1.2676767676767676e-07,
+      "loss": 0.5093,
+      "step": 8750
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 7.290755748748779,
+      "learning_rate": 1.2575757575757576e-07,
+      "loss": 0.4842,
+      "step": 8760
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 3.76725435256958,
+      "learning_rate": 1.2474747474747473e-07,
+      "loss": 0.5476,
+      "step": 8770
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 5.706302642822266,
+      "learning_rate": 1.2373737373737373e-07,
+      "loss": 0.4709,
+      "step": 8780
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 7.164012432098389,
+      "learning_rate": 1.2272727272727272e-07,
+      "loss": 0.5088,
+      "step": 8790
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 5.86743688583374,
+      "learning_rate": 1.2171717171717172e-07,
+      "loss": 0.5361,
+      "step": 8800
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 4.16366720199585,
+      "learning_rate": 1.207070707070707e-07,
+      "loss": 0.536,
+      "step": 8810
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 6.925227642059326,
+      "learning_rate": 1.196969696969697e-07,
+      "loss": 0.5001,
+      "step": 8820
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 3.9916467666625977,
+      "learning_rate": 1.1868686868686869e-07,
+      "loss": 0.5662,
+      "step": 8830
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 7.0623931884765625,
+      "learning_rate": 1.1767676767676767e-07,
+      "loss": 0.5813,
+      "step": 8840
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 5.496065139770508,
+      "learning_rate": 1.1666666666666667e-07,
+      "loss": 0.5751,
+      "step": 8850
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 3.784182071685791,
+      "learning_rate": 1.1565656565656565e-07,
+      "loss": 0.5863,
+      "step": 8860
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 5.068544864654541,
+      "learning_rate": 1.1464646464646465e-07,
+      "loss": 0.5098,
+      "step": 8870
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 5.4960808753967285,
+      "learning_rate": 1.1363636363636363e-07,
+      "loss": 0.6326,
+      "step": 8880
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 4.9256792068481445,
+      "learning_rate": 1.1262626262626263e-07,
+      "loss": 0.4801,
+      "step": 8890
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 4.391407489776611,
+      "learning_rate": 1.1161616161616161e-07,
+      "loss": 0.5866,
+      "step": 8900
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 4.210151195526123,
+      "learning_rate": 1.1060606060606061e-07,
+      "loss": 0.5464,
+      "step": 8910
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 6.131332874298096,
+      "learning_rate": 1.0959595959595959e-07,
+      "loss": 0.4419,
+      "step": 8920
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 5.612468719482422,
+      "learning_rate": 1.0858585858585859e-07,
+      "loss": 0.6155,
+      "step": 8930
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 4.1157097816467285,
+      "learning_rate": 1.0757575757575757e-07,
+      "loss": 0.5514,
+      "step": 8940
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 5.19150972366333,
+      "learning_rate": 1.0656565656565657e-07,
+      "loss": 0.548,
+      "step": 8950
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 4.305140972137451,
+      "learning_rate": 1.0555555555555555e-07,
+      "loss": 0.5313,
+      "step": 8960
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 6.152969837188721,
+      "learning_rate": 1.0454545454545454e-07,
+      "loss": 0.4861,
+      "step": 8970
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 4.60936164855957,
+      "learning_rate": 1.0353535353535353e-07,
+      "loss": 0.6055,
+      "step": 8980
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 9.371330261230469,
+      "learning_rate": 1.0252525252525252e-07,
+      "loss": 0.5255,
+      "step": 8990
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 5.383856296539307,
+      "learning_rate": 1.0151515151515151e-07,
+      "loss": 0.4944,
+      "step": 9000
+    },
+    {
+      "epoch": 3.23,
+      "eval_loss": 0.6957755088806152,
+      "eval_runtime": 404.454,
+      "eval_samples_per_second": 2.472,
+      "eval_steps_per_second": 2.472,
+      "step": 9000
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 6.594298839569092,
+      "learning_rate": 1.005050505050505e-07,
+      "loss": 0.5506,
+      "step": 9010
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 5.5882248878479,
+      "learning_rate": 9.94949494949495e-08,
+      "loss": 0.4821,
+      "step": 9020
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 5.565232753753662,
+      "learning_rate": 9.848484848484848e-08,
+      "loss": 0.5903,
+      "step": 9030
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 4.196237564086914,
+      "learning_rate": 9.747474747474747e-08,
+      "loss": 0.4784,
+      "step": 9040
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 4.757313251495361,
+      "learning_rate": 9.646464646464646e-08,
+      "loss": 0.5256,
+      "step": 9050
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 5.351820945739746,
+      "learning_rate": 9.545454545454546e-08,
+      "loss": 0.4909,
+      "step": 9060
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 5.794235706329346,
+      "learning_rate": 9.444444444444444e-08,
+      "loss": 0.5444,
+      "step": 9070
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 6.105288028717041,
+      "learning_rate": 9.343434343434344e-08,
+      "loss": 0.5256,
+      "step": 9080
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 6.002346992492676,
+      "learning_rate": 9.242424242424242e-08,
+      "loss": 0.5005,
+      "step": 9090
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 5.016416072845459,
+      "learning_rate": 9.141414141414142e-08,
+      "loss": 0.5931,
+      "step": 9100
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 4.596400737762451,
+      "learning_rate": 9.04040404040404e-08,
+      "loss": 0.4898,
+      "step": 9110
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 5.950545310974121,
+      "learning_rate": 8.93939393939394e-08,
+      "loss": 0.6261,
+      "step": 9120
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 6.038389205932617,
+      "learning_rate": 8.838383838383838e-08,
+      "loss": 0.4946,
+      "step": 9130
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 6.918920516967773,
+      "learning_rate": 8.737373737373736e-08,
+      "loss": 0.5929,
+      "step": 9140
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 4.859712600708008,
+      "learning_rate": 8.636363636363636e-08,
+      "loss": 0.5885,
+      "step": 9150
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 8.945250511169434,
+      "learning_rate": 8.535353535353534e-08,
+      "loss": 0.5589,
+      "step": 9160
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 4.006341934204102,
+      "learning_rate": 8.434343434343434e-08,
+      "loss": 0.4796,
+      "step": 9170
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 6.528043746948242,
+      "learning_rate": 8.333333333333333e-08,
+      "loss": 0.5737,
+      "step": 9180
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 7.656773090362549,
+      "learning_rate": 8.232323232323232e-08,
+      "loss": 0.549,
+      "step": 9190
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 4.172515869140625,
+      "learning_rate": 8.13131313131313e-08,
+      "loss": 0.5112,
+      "step": 9200
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 6.6936116218566895,
+      "learning_rate": 8.03030303030303e-08,
+      "loss": 0.5645,
+      "step": 9210
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 7.118635177612305,
+      "learning_rate": 7.929292929292929e-08,
+      "loss": 0.5502,
+      "step": 9220
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 4.658480167388916,
+      "learning_rate": 7.828282828282828e-08,
+      "loss": 0.4883,
+      "step": 9230
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 7.470480442047119,
+      "learning_rate": 7.727272727272727e-08,
+      "loss": 0.5102,
+      "step": 9240
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 4.436034679412842,
+      "learning_rate": 7.626262626262626e-08,
+      "loss": 0.5939,
+      "step": 9250
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 4.717689037322998,
+      "learning_rate": 7.525252525252525e-08,
+      "loss": 0.5976,
+      "step": 9260
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 4.957371711730957,
+      "learning_rate": 7.424242424242424e-08,
+      "loss": 0.5019,
+      "step": 9270
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 4.873682022094727,
+      "learning_rate": 7.323232323232323e-08,
+      "loss": 0.5755,
+      "step": 9280
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 4.161667823791504,
+      "learning_rate": 7.222222222222221e-08,
+      "loss": 0.6371,
+      "step": 9290
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 4.539839744567871,
+      "learning_rate": 7.121212121212121e-08,
+      "loss": 0.6012,
+      "step": 9300
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 5.414936542510986,
+      "learning_rate": 7.020202020202019e-08,
+      "loss": 0.5449,
+      "step": 9310
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 10.87019157409668,
+      "learning_rate": 6.919191919191919e-08,
+      "loss": 0.5272,
+      "step": 9320
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 5.128973484039307,
+      "learning_rate": 6.818181818181817e-08,
+      "loss": 0.5053,
+      "step": 9330
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 4.943745136260986,
+      "learning_rate": 6.717171717171717e-08,
+      "loss": 0.5823,
+      "step": 9340
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 5.1109185218811035,
+      "learning_rate": 6.616161616161615e-08,
+      "loss": 0.5345,
+      "step": 9350
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 7.44246768951416,
+      "learning_rate": 6.515151515151515e-08,
+      "loss": 0.5677,
+      "step": 9360
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 4.263915538787842,
+      "learning_rate": 6.414141414141413e-08,
+      "loss": 0.522,
+      "step": 9370
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 5.933948516845703,
+      "learning_rate": 6.313131313131313e-08,
+      "loss": 0.4782,
+      "step": 9380
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 4.103212356567383,
+      "learning_rate": 6.212121212121211e-08,
+      "loss": 0.5602,
+      "step": 9390
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 5.865060806274414,
+      "learning_rate": 6.111111111111111e-08,
+      "loss": 0.5654,
+      "step": 9400
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 3.8256332874298096,
+      "learning_rate": 6.01010101010101e-08,
+      "loss": 0.5984,
+      "step": 9410
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 6.088188648223877,
+      "learning_rate": 5.9090909090909085e-08,
+      "loss": 0.5782,
+      "step": 9420
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 6.299181938171387,
+      "learning_rate": 5.8080808080808076e-08,
+      "loss": 0.5725,
+      "step": 9430
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 5.62983512878418,
+      "learning_rate": 5.7070707070707066e-08,
+      "loss": 0.5407,
+      "step": 9440
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 5.949933052062988,
+      "learning_rate": 5.6060606060606056e-08,
+      "loss": 0.4399,
+      "step": 9450
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 5.324888229370117,
+      "learning_rate": 5.5050505050505046e-08,
+      "loss": 0.5178,
+      "step": 9460
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 7.061882019042969,
+      "learning_rate": 5.4040404040404037e-08,
+      "loss": 0.5175,
+      "step": 9470
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 4.5119404792785645,
+      "learning_rate": 5.303030303030303e-08,
+      "loss": 0.5282,
+      "step": 9480
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 4.700840473175049,
+      "learning_rate": 5.202020202020202e-08,
+      "loss": 0.5202,
+      "step": 9490
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 5.251560211181641,
+      "learning_rate": 5.101010101010101e-08,
+      "loss": 0.5013,
+      "step": 9500
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 6.137795925140381,
+      "learning_rate": 5e-08,
+      "loss": 0.4514,
+      "step": 9510
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 6.373246669769287,
+      "learning_rate": 4.898989898989899e-08,
+      "loss": 0.5129,
+      "step": 9520
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 6.3960418701171875,
+      "learning_rate": 4.797979797979797e-08,
+      "loss": 0.5055,
+      "step": 9530
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 4.761574745178223,
+      "learning_rate": 4.696969696969696e-08,
+      "loss": 0.5724,
+      "step": 9540
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 4.658292770385742,
+      "learning_rate": 4.595959595959595e-08,
+      "loss": 0.5617,
+      "step": 9550
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 6.56446647644043,
+      "learning_rate": 4.494949494949494e-08,
+      "loss": 0.5495,
+      "step": 9560
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 5.917604446411133,
+      "learning_rate": 4.393939393939393e-08,
+      "loss": 0.5513,
+      "step": 9570
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 6.456116676330566,
+      "learning_rate": 4.292929292929292e-08,
+      "loss": 0.5198,
+      "step": 9580
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 5.084980487823486,
+      "learning_rate": 4.1919191919191913e-08,
+      "loss": 0.5737,
+      "step": 9590
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 3.8530325889587402,
+      "learning_rate": 4.090909090909091e-08,
+      "loss": 0.4901,
+      "step": 9600
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 4.385349750518799,
+      "learning_rate": 3.98989898989899e-08,
+      "loss": 0.5092,
+      "step": 9610
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 3.888326406478882,
+      "learning_rate": 3.888888888888889e-08,
+      "loss": 0.5536,
+      "step": 9620
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 4.46201229095459,
+      "learning_rate": 3.787878787878788e-08,
+      "loss": 0.5697,
+      "step": 9630
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 7.194265842437744,
+      "learning_rate": 3.686868686868687e-08,
+      "loss": 0.4886,
+      "step": 9640
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 6.102540016174316,
+      "learning_rate": 3.585858585858586e-08,
+      "loss": 0.4981,
+      "step": 9650
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 7.469662189483643,
+      "learning_rate": 3.484848484848485e-08,
+      "loss": 0.5294,
+      "step": 9660
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 4.181239128112793,
+      "learning_rate": 3.383838383838384e-08,
+      "loss": 0.5463,
+      "step": 9670
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 5.108190536499023,
+      "learning_rate": 3.282828282828283e-08,
+      "loss": 0.4633,
+      "step": 9680
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 5.501590251922607,
+      "learning_rate": 3.1818181818181816e-08,
+      "loss": 0.5121,
+      "step": 9690
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 6.437745571136475,
+      "learning_rate": 3.0808080808080806e-08,
+      "loss": 0.475,
+      "step": 9700
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 7.1137166023254395,
+      "learning_rate": 2.9797979797979797e-08,
+      "loss": 0.5536,
+      "step": 9710
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 5.652755260467529,
+      "learning_rate": 2.8787878787878787e-08,
+      "loss": 0.5493,
+      "step": 9720
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 4.296222686767578,
+      "learning_rate": 2.7777777777777774e-08,
+      "loss": 0.5323,
+      "step": 9730
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 5.550398826599121,
+      "learning_rate": 2.6767676767676764e-08,
+      "loss": 0.5555,
+      "step": 9740
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 6.785501956939697,
+      "learning_rate": 2.5757575757575755e-08,
+      "loss": 0.5388,
+      "step": 9750
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 7.063258171081543,
+      "learning_rate": 2.4747474747474745e-08,
+      "loss": 0.5742,
+      "step": 9760
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 5.894256114959717,
+      "learning_rate": 2.3737373737373735e-08,
+      "loss": 0.6248,
+      "step": 9770
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 5.81352424621582,
+      "learning_rate": 2.2727272727272725e-08,
+      "loss": 0.4896,
+      "step": 9780
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 5.251163482666016,
+      "learning_rate": 2.1717171717171716e-08,
+      "loss": 0.5633,
+      "step": 9790
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 4.963428020477295,
+      "learning_rate": 2.070707070707071e-08,
+      "loss": 0.4617,
+      "step": 9800
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 4.8795928955078125,
+      "learning_rate": 1.9696969696969696e-08,
+      "loss": 0.5325,
+      "step": 9810
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 5.59514856338501,
+      "learning_rate": 1.8686868686868686e-08,
+      "loss": 0.5535,
+      "step": 9820
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 5.933664321899414,
+      "learning_rate": 1.7676767676767677e-08,
+      "loss": 0.4904,
+      "step": 9830
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 5.348793029785156,
+      "learning_rate": 1.6666666666666667e-08,
+      "loss": 0.543,
+      "step": 9840
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 4.590538024902344,
+      "learning_rate": 1.5656565656565657e-08,
+      "loss": 0.5443,
+      "step": 9850
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 7.6259918212890625,
+      "learning_rate": 1.4646464646464646e-08,
+      "loss": 0.4999,
+      "step": 9860
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 5.3782124519348145,
+      "learning_rate": 1.3636363636363635e-08,
+      "loss": 0.5064,
+      "step": 9870
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 6.371192932128906,
+      "learning_rate": 1.2626262626262625e-08,
+      "loss": 0.5487,
+      "step": 9880
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 5.356314659118652,
+      "learning_rate": 1.1616161616161615e-08,
+      "loss": 0.506,
+      "step": 9890
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 5.2767014503479,
+      "learning_rate": 1.0606060606060607e-08,
+      "loss": 0.5556,
+      "step": 9900
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 4.274359226226807,
+      "learning_rate": 9.595959595959596e-09,
+      "loss": 0.5074,
+      "step": 9910
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 5.464534759521484,
+      "learning_rate": 8.585858585858586e-09,
+      "loss": 0.4612,
+      "step": 9920
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 9.630477905273438,
+      "learning_rate": 7.575757575757576e-09,
+      "loss": 0.6037,
+      "step": 9930
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 3.692291259765625,
+      "learning_rate": 6.565656565656565e-09,
+      "loss": 0.5116,
+      "step": 9940
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 5.269135475158691,
+      "learning_rate": 5.555555555555555e-09,
+      "loss": 0.591,
+      "step": 9950
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 5.141294956207275,
+      "learning_rate": 4.545454545454545e-09,
+      "loss": 0.5726,
+      "step": 9960
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 6.49954080581665,
+      "learning_rate": 3.5353535353535353e-09,
+      "loss": 0.5509,
+      "step": 9970
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 5.085234642028809,
+      "learning_rate": 2.5252525252525255e-09,
+      "loss": 0.5337,
+      "step": 9980
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 5.688786506652832,
+      "learning_rate": 1.515151515151515e-09,
+      "loss": 0.5624,
+      "step": 9990
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 6.861104965209961,
+      "learning_rate": 5.050505050505051e-10,
+      "loss": 0.5167,
+      "step": 10000
+    },
+    {
+      "epoch": 3.59,
+      "eval_loss": 0.711556613445282,
+      "eval_runtime": 405.1832,
+      "eval_samples_per_second": 2.468,
+      "eval_steps_per_second": 2.468,
+      "step": 10000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 2000,
+  "total_flos": 9.4282098671616e+17,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}