diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5653 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.8689259458490226,
+  "eval_steps": 2000,
+  "global_step": 8000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.9492120146751404,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 0.9769,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.344916582107544,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 1.0718,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.6680417060852051,
+      "learning_rate": 3e-06,
+      "loss": 0.8901,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.8369797468185425,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 0.9398,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.5738756656646729,
+      "learning_rate": 5e-06,
+      "loss": 0.9152,
+      "step": 50
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.028475284576416,
+      "learning_rate": 6e-06,
+      "loss": 0.849,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.2893372774124146,
+      "learning_rate": 7e-06,
+      "loss": 0.7312,
+      "step": 70
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.8779547810554504,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 0.6707,
+      "step": 80
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.8972748517990112,
+      "learning_rate": 9e-06,
+      "loss": 0.6413,
+      "step": 90
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.505399227142334,
+      "learning_rate": 1e-05,
+      "loss": 0.6129,
+      "step": 100
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.6442121267318726,
+      "learning_rate": 9.989898989898991e-06,
+      "loss": 0.5579,
+      "step": 110
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.623290479183197,
+      "learning_rate": 9.97979797979798e-06,
+      "loss": 0.6022,
+      "step": 120
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.7216657996177673,
+      "learning_rate": 9.96969696969697e-06,
+      "loss": 0.5481,
+      "step": 130
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.7031328678131104,
+      "learning_rate": 9.95959595959596e-06,
+      "loss": 0.5594,
+      "step": 140
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.7411965727806091,
+      "learning_rate": 9.94949494949495e-06,
+      "loss": 0.6241,
+      "step": 150
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.8735277652740479,
+      "learning_rate": 9.939393939393939e-06,
+      "loss": 0.498,
+      "step": 160
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.1062073707580566,
+      "learning_rate": 9.92929292929293e-06,
+      "loss": 0.6463,
+      "step": 170
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.7930333614349365,
+      "learning_rate": 9.91919191919192e-06,
+      "loss": 0.6231,
+      "step": 180
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.8774802684783936,
+      "learning_rate": 9.90909090909091e-06,
+      "loss": 0.5895,
+      "step": 190
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.929542064666748,
+      "learning_rate": 9.8989898989899e-06,
+      "loss": 0.632,
+      "step": 200
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.6024735569953918,
+      "learning_rate": 9.88888888888889e-06,
+      "loss": 0.5682,
+      "step": 210
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.5887441635131836,
+      "learning_rate": 9.87878787878788e-06,
+      "loss": 0.5192,
+      "step": 220
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.7813496589660645,
+      "learning_rate": 9.86868686868687e-06,
+      "loss": 0.5841,
+      "step": 230
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.5949407815933228,
+      "learning_rate": 9.85858585858586e-06,
+      "loss": 0.5263,
+      "step": 240
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.8114556670188904,
+      "learning_rate": 9.84848484848485e-06,
+      "loss": 0.5513,
+      "step": 250
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.7692855596542358,
+      "learning_rate": 9.838383838383839e-06,
+      "loss": 0.5217,
+      "step": 260
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.8319826126098633,
+      "learning_rate": 9.828282828282829e-06,
+      "loss": 0.4386,
+      "step": 270
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.6725042462348938,
+      "learning_rate": 9.81818181818182e-06,
+      "loss": 0.4919,
+      "step": 280
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.779315173625946,
+      "learning_rate": 9.80808080808081e-06,
+      "loss": 0.5281,
+      "step": 290
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.8005223274230957,
+      "learning_rate": 9.797979797979798e-06,
+      "loss": 0.5206,
+      "step": 300
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.6395801901817322,
+      "learning_rate": 9.787878787878788e-06,
+      "loss": 0.474,
+      "step": 310
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.1016992330551147,
+      "learning_rate": 9.777777777777779e-06,
+      "loss": 0.5403,
+      "step": 320
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.395189881324768,
+      "learning_rate": 9.767676767676767e-06,
+      "loss": 0.5181,
+      "step": 330
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.801499605178833,
+      "learning_rate": 9.757575757575758e-06,
+      "loss": 0.5296,
+      "step": 340
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.7606889605522156,
+      "learning_rate": 9.747474747474748e-06,
+      "loss": 0.5406,
+      "step": 350
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.580170750617981,
+      "learning_rate": 9.737373737373738e-06,
+      "loss": 0.541,
+      "step": 360
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.6730871796607971,
+      "learning_rate": 9.727272727272728e-06,
+      "loss": 0.5665,
+      "step": 370
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.240430474281311,
+      "learning_rate": 9.717171717171719e-06,
+      "loss": 0.5148,
+      "step": 380
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.9439683556556702,
+      "learning_rate": 9.707070707070709e-06,
+      "loss": 0.513,
+      "step": 390
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.6674547791481018,
+      "learning_rate": 9.696969696969698e-06,
+      "loss": 0.5202,
+      "step": 400
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.8516045808792114,
+      "learning_rate": 9.686868686868688e-06,
+      "loss": 0.5162,
+      "step": 410
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.8700432181358337,
+      "learning_rate": 9.676767676767678e-06,
+      "loss": 0.5392,
+      "step": 420
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.5687388777732849,
+      "learning_rate": 9.666666666666667e-06,
+      "loss": 0.5106,
+      "step": 430
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.2382631301879883,
+      "learning_rate": 9.656565656565657e-06,
+      "loss": 0.5074,
+      "step": 440
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.8921974301338196,
+      "learning_rate": 9.646464646464647e-06,
+      "loss": 0.561,
+      "step": 450
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.3508048057556152,
+      "learning_rate": 9.636363636363638e-06,
+      "loss": 0.5484,
+      "step": 460
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.1822205781936646,
+      "learning_rate": 9.626262626262626e-06,
+      "loss": 0.5719,
+      "step": 470
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.6617271304130554,
+      "learning_rate": 9.616161616161616e-06,
+      "loss": 0.535,
+      "step": 480
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.6614571809768677,
+      "learning_rate": 9.606060606060607e-06,
+      "loss": 0.4549,
+      "step": 490
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.8642953038215637,
+      "learning_rate": 9.595959595959597e-06,
+      "loss": 0.4789,
+      "step": 500
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.614743173122406,
+      "learning_rate": 9.585858585858586e-06,
+      "loss": 0.4854,
+      "step": 510
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.970829427242279,
+      "learning_rate": 9.575757575757576e-06,
+      "loss": 0.5196,
+      "step": 520
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.7311980128288269,
+      "learning_rate": 9.565656565656566e-06,
+      "loss": 0.5106,
+      "step": 530
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.765849769115448,
+      "learning_rate": 9.555555555555556e-06,
+      "loss": 0.5782,
+      "step": 540
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.0889312028884888,
+      "learning_rate": 9.545454545454547e-06,
+      "loss": 0.5824,
+      "step": 550
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.7402384877204895,
+      "learning_rate": 9.535353535353537e-06,
+      "loss": 0.5005,
+      "step": 560
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.707028329372406,
+      "learning_rate": 9.525252525252526e-06,
+      "loss": 0.5233,
+      "step": 570
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.8338315486907959,
+      "learning_rate": 9.515151515151516e-06,
+      "loss": 0.4694,
+      "step": 580
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.7450662851333618,
+      "learning_rate": 9.505050505050506e-06,
+      "loss": 0.4762,
+      "step": 590
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.7595840692520142,
+      "learning_rate": 9.494949494949497e-06,
+      "loss": 0.5018,
+      "step": 600
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.5880123376846313,
+      "learning_rate": 9.484848484848485e-06,
+      "loss": 0.5228,
+      "step": 610
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.9635146260261536,
+      "learning_rate": 9.474747474747475e-06,
+      "loss": 0.4987,
+      "step": 620
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.1274373531341553,
+      "learning_rate": 9.464646464646466e-06,
+      "loss": 0.5085,
+      "step": 630
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.1324831247329712,
+      "learning_rate": 9.454545454545456e-06,
+      "loss": 0.504,
+      "step": 640
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.5410157442092896,
+      "learning_rate": 9.444444444444445e-06,
+      "loss": 0.4619,
+      "step": 650
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.7583281993865967,
+      "learning_rate": 9.434343434343435e-06,
+      "loss": 0.5162,
+      "step": 660
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.6546668410301208,
+      "learning_rate": 9.424242424242425e-06,
+      "loss": 0.4969,
+      "step": 670
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.6070376634597778,
+      "learning_rate": 9.414141414141414e-06,
+      "loss": 0.4805,
+      "step": 680
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.0108693838119507,
+      "learning_rate": 9.404040404040404e-06,
+      "loss": 0.4808,
+      "step": 690
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.8799183368682861,
+      "learning_rate": 9.393939393939396e-06,
+      "loss": 0.5082,
+      "step": 700
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.3432070016860962,
+      "learning_rate": 9.383838383838385e-06,
+      "loss": 0.4353,
+      "step": 710
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.6518195271492004,
+      "learning_rate": 9.373737373737375e-06,
+      "loss": 0.4933,
+      "step": 720
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.6736329793930054,
+      "learning_rate": 9.363636363636365e-06,
+      "loss": 0.5342,
+      "step": 730
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.776785135269165,
+      "learning_rate": 9.353535353535354e-06,
+      "loss": 0.5162,
+      "step": 740
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.9443957805633545,
+      "learning_rate": 9.343434343434344e-06,
+      "loss": 0.4486,
+      "step": 750
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.8882728815078735,
+      "learning_rate": 9.333333333333334e-06,
+      "loss": 0.4774,
+      "step": 760
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.819239616394043,
+      "learning_rate": 9.323232323232325e-06,
+      "loss": 0.5,
+      "step": 770
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.8829065561294556,
+      "learning_rate": 9.313131313131313e-06,
+      "loss": 0.4655,
+      "step": 780
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.1993345022201538,
+      "learning_rate": 9.303030303030303e-06,
+      "loss": 0.5392,
+      "step": 790
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.681409478187561,
+      "learning_rate": 9.292929292929294e-06,
+      "loss": 0.5076,
+      "step": 800
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.075088381767273,
+      "learning_rate": 9.282828282828284e-06,
+      "loss": 0.4953,
+      "step": 810
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.7040195465087891,
+      "learning_rate": 9.272727272727273e-06,
+      "loss": 0.5596,
+      "step": 820
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.1210192441940308,
+      "learning_rate": 9.262626262626263e-06,
+      "loss": 0.5299,
+      "step": 830
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.583011269569397,
+      "learning_rate": 9.252525252525253e-06,
+      "loss": 0.497,
+      "step": 840
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.6583887338638306,
+      "learning_rate": 9.242424242424244e-06,
+      "loss": 0.5556,
+      "step": 850
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.8040810227394104,
+      "learning_rate": 9.232323232323232e-06,
+      "loss": 0.5257,
+      "step": 860
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.9269919991493225,
+      "learning_rate": 9.222222222222224e-06,
+      "loss": 0.4421,
+      "step": 870
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.9947918653488159,
+      "learning_rate": 9.212121212121213e-06,
+      "loss": 0.5297,
+      "step": 880
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.6900811791419983,
+      "learning_rate": 9.202020202020203e-06,
+      "loss": 0.4833,
+      "step": 890
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.8033557534217834,
+      "learning_rate": 9.191919191919193e-06,
+      "loss": 0.4894,
+      "step": 900
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.635124683380127,
+      "learning_rate": 9.181818181818184e-06,
+      "loss": 0.4554,
+      "step": 910
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.7293840646743774,
+      "learning_rate": 9.171717171717172e-06,
+      "loss": 0.4693,
+      "step": 920
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.7628031373023987,
+      "learning_rate": 9.161616161616162e-06,
+      "loss": 0.5181,
+      "step": 930
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.0783181190490723,
+      "learning_rate": 9.151515151515153e-06,
+      "loss": 0.4632,
+      "step": 940
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.5340379476547241,
+      "learning_rate": 9.141414141414143e-06,
+      "loss": 0.4664,
+      "step": 950
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.9029551148414612,
+      "learning_rate": 9.131313131313132e-06,
+      "loss": 0.5333,
+      "step": 960
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.7257616519927979,
+      "learning_rate": 9.121212121212122e-06,
+      "loss": 0.5168,
+      "step": 970
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.761325478553772,
+      "learning_rate": 9.111111111111112e-06,
+      "loss": 0.5606,
+      "step": 980
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.8582245707511902,
+      "learning_rate": 9.1010101010101e-06,
+      "loss": 0.4332,
+      "step": 990
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.8598415851593018,
+      "learning_rate": 9.090909090909091e-06,
+      "loss": 0.5884,
+      "step": 1000
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.8292351365089417,
+      "learning_rate": 9.080808080808081e-06,
+      "loss": 0.4848,
+      "step": 1010
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.0559266805648804,
+      "learning_rate": 9.070707070707072e-06,
+      "loss": 0.4588,
+      "step": 1020
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.6693033576011658,
+      "learning_rate": 9.06060606060606e-06,
+      "loss": 0.5333,
+      "step": 1030
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.8114706873893738,
+      "learning_rate": 9.050505050505052e-06,
+      "loss": 0.5166,
+      "step": 1040
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.8659316301345825,
+      "learning_rate": 9.040404040404042e-06,
+      "loss": 0.4504,
+      "step": 1050
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.9083582758903503,
+      "learning_rate": 9.030303030303031e-06,
+      "loss": 0.5611,
+      "step": 1060
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.6691566109657288,
+      "learning_rate": 9.020202020202021e-06,
+      "loss": 0.5192,
+      "step": 1070
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.5889317989349365,
+      "learning_rate": 9.010101010101012e-06,
+      "loss": 0.4515,
+      "step": 1080
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.9215373992919922,
+      "learning_rate": 9e-06,
+      "loss": 0.4776,
+      "step": 1090
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.7439729571342468,
+      "learning_rate": 8.98989898989899e-06,
+      "loss": 0.4656,
+      "step": 1100
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.1780657768249512,
+      "learning_rate": 8.97979797979798e-06,
+      "loss": 0.4933,
+      "step": 1110
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.9686077833175659,
+      "learning_rate": 8.969696969696971e-06,
+      "loss": 0.5167,
+      "step": 1120
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.829994261264801,
+      "learning_rate": 8.95959595959596e-06,
+      "loss": 0.491,
+      "step": 1130
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.6313827633857727,
+      "learning_rate": 8.94949494949495e-06,
+      "loss": 0.4864,
+      "step": 1140
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.596537709236145,
+      "learning_rate": 8.93939393939394e-06,
+      "loss": 0.4807,
+      "step": 1150
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.8714896440505981,
+      "learning_rate": 8.92929292929293e-06,
+      "loss": 0.512,
+      "step": 1160
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.9466399550437927,
+      "learning_rate": 8.919191919191919e-06,
+      "loss": 0.4883,
+      "step": 1170
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.7337993383407593,
+      "learning_rate": 8.90909090909091e-06,
+      "loss": 0.4757,
+      "step": 1180
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.7684504985809326,
+      "learning_rate": 8.8989898989899e-06,
+      "loss": 0.5224,
+      "step": 1190
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.6455299854278564,
+      "learning_rate": 8.888888888888888e-06,
+      "loss": 0.5346,
+      "step": 1200
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.7279661893844604,
+      "learning_rate": 8.87878787878788e-06,
+      "loss": 0.4845,
+      "step": 1210
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.7996425032615662,
+      "learning_rate": 8.86868686868687e-06,
+      "loss": 0.4812,
+      "step": 1220
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.7299336791038513,
+      "learning_rate": 8.85858585858586e-06,
+      "loss": 0.4985,
+      "step": 1230
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.7462379932403564,
+      "learning_rate": 8.84848484848485e-06,
+      "loss": 0.5742,
+      "step": 1240
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.7165307998657227,
+      "learning_rate": 8.83838383838384e-06,
+      "loss": 0.4627,
+      "step": 1250
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.7239411473274231,
+      "learning_rate": 8.82828282828283e-06,
+      "loss": 0.5469,
+      "step": 1260
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.5761345028877258,
+      "learning_rate": 8.818181818181819e-06,
+      "loss": 0.4391,
+      "step": 1270
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.8207817077636719,
+      "learning_rate": 8.808080808080809e-06,
+      "loss": 0.5477,
+      "step": 1280
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.9331930875778198,
+      "learning_rate": 8.7979797979798e-06,
+      "loss": 0.4975,
+      "step": 1290
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.0401968955993652,
+      "learning_rate": 8.787878787878788e-06,
+      "loss": 0.5149,
+      "step": 1300
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.7848596572875977,
+      "learning_rate": 8.777777777777778e-06,
+      "loss": 0.5448,
+      "step": 1310
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.5979989767074585,
+      "learning_rate": 8.767676767676768e-06,
+      "loss": 0.4861,
+      "step": 1320
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.7734145522117615,
+      "learning_rate": 8.757575757575759e-06,
+      "loss": 0.5955,
+      "step": 1330
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.9506089687347412,
+      "learning_rate": 8.747474747474747e-06,
+      "loss": 0.5212,
+      "step": 1340
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.7118907570838928,
+      "learning_rate": 8.737373737373738e-06,
+      "loss": 0.474,
+      "step": 1350
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.0045005083084106,
+      "learning_rate": 8.727272727272728e-06,
+      "loss": 0.509,
+      "step": 1360
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.1283302307128906,
+      "learning_rate": 8.717171717171718e-06,
+      "loss": 0.5096,
+      "step": 1370
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.3884085416793823,
+      "learning_rate": 8.707070707070707e-06,
+      "loss": 0.5008,
+      "step": 1380
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.7452566027641296,
+      "learning_rate": 8.696969696969699e-06,
+      "loss": 0.4792,
+      "step": 1390
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.3434414863586426,
+      "learning_rate": 8.686868686868687e-06,
+      "loss": 0.5385,
+      "step": 1400
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.9600369930267334,
+      "learning_rate": 8.676767676767678e-06,
+      "loss": 0.4743,
+      "step": 1410
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.6895599961280823,
+      "learning_rate": 8.666666666666668e-06,
+      "loss": 0.5639,
+      "step": 1420
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.9460670948028564,
+      "learning_rate": 8.656565656565658e-06,
+      "loss": 0.5957,
+      "step": 1430
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.6181186437606812,
+      "learning_rate": 8.646464646464647e-06,
+      "loss": 0.5032,
+      "step": 1440
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.6992838382720947,
+      "learning_rate": 8.636363636363637e-06,
+      "loss": 0.5371,
+      "step": 1450
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.6007195711135864,
+      "learning_rate": 8.626262626262627e-06,
+      "loss": 0.4446,
+      "step": 1460
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.8171485662460327,
+      "learning_rate": 8.616161616161618e-06,
+      "loss": 0.4835,
+      "step": 1470
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.9369902610778809,
+      "learning_rate": 8.606060606060606e-06,
+      "loss": 0.5437,
+      "step": 1480
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.1894652843475342,
+      "learning_rate": 8.595959595959596e-06,
+      "loss": 0.5151,
+      "step": 1490
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.694837212562561,
+      "learning_rate": 8.585858585858587e-06,
+      "loss": 0.488,
+      "step": 1500
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.8813522458076477,
+      "learning_rate": 8.575757575757575e-06,
+      "loss": 0.5129,
+      "step": 1510
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.0121350288391113,
+      "learning_rate": 8.565656565656566e-06,
+      "loss": 0.5165,
+      "step": 1520
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.7935439944267273,
+      "learning_rate": 8.555555555555556e-06,
+      "loss": 0.4736,
+      "step": 1530
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.7670463919639587,
+      "learning_rate": 8.545454545454546e-06,
+      "loss": 0.5214,
+      "step": 1540
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.212927222251892,
+      "learning_rate": 8.535353535353535e-06,
+      "loss": 0.5125,
+      "step": 1550
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.7966919541358948,
+      "learning_rate": 8.525252525252527e-06,
+      "loss": 0.4823,
+      "step": 1560
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.0880494117736816,
+      "learning_rate": 8.515151515151517e-06,
+      "loss": 0.478,
+      "step": 1570
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.0308737754821777,
+      "learning_rate": 8.505050505050506e-06,
+      "loss": 0.5368,
+      "step": 1580
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.7291275262832642,
+      "learning_rate": 8.494949494949496e-06,
+      "loss": 0.4838,
+      "step": 1590
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.6764214038848877,
+      "learning_rate": 8.484848484848486e-06,
+      "loss": 0.4882,
+      "step": 1600
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.051628828048706,
+      "learning_rate": 8.474747474747475e-06,
+      "loss": 0.4564,
+      "step": 1610
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.8614614605903625,
+      "learning_rate": 8.464646464646465e-06,
+      "loss": 0.5632,
+      "step": 1620
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.1045228242874146,
+      "learning_rate": 8.454545454545455e-06,
+      "loss": 0.4535,
+      "step": 1630
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.8160364031791687,
+      "learning_rate": 8.444444444444446e-06,
+      "loss": 0.4964,
+      "step": 1640
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.7776429653167725,
+      "learning_rate": 8.434343434343434e-06,
+      "loss": 0.4554,
+      "step": 1650
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.7589672207832336,
+      "learning_rate": 8.424242424242425e-06,
+      "loss": 0.4458,
+      "step": 1660
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.825233519077301,
+      "learning_rate": 8.414141414141415e-06,
+      "loss": 0.4814,
+      "step": 1670
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.8226912617683411,
+      "learning_rate": 8.404040404040405e-06,
+      "loss": 0.4892,
+      "step": 1680
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.9273412823677063,
+      "learning_rate": 8.393939393939394e-06,
+      "loss": 0.5143,
+      "step": 1690
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.9043828248977661,
+      "learning_rate": 8.383838383838384e-06,
+      "loss": 0.4598,
+      "step": 1700
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 2.2596805095672607,
+      "learning_rate": 8.373737373737374e-06,
+      "loss": 0.5415,
+      "step": 1710
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.0041202306747437,
+      "learning_rate": 8.363636363636365e-06,
+      "loss": 0.5899,
+      "step": 1720
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.9188370704650879,
+      "learning_rate": 8.353535353535355e-06,
+      "loss": 0.5119,
+      "step": 1730
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.7778961062431335,
+      "learning_rate": 8.343434343434345e-06,
+      "loss": 0.5237,
+      "step": 1740
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.7438649535179138,
+      "learning_rate": 8.333333333333334e-06,
+      "loss": 0.4999,
+      "step": 1750
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.5649489760398865,
+      "learning_rate": 8.323232323232324e-06,
+      "loss": 0.552,
+      "step": 1760
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.5625451803207397,
+      "learning_rate": 8.313131313131314e-06,
+      "loss": 0.4549,
+      "step": 1770
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.3711755275726318,
+      "learning_rate": 8.303030303030305e-06,
+      "loss": 0.445,
+      "step": 1780
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.4339165687561035,
+      "learning_rate": 8.292929292929293e-06,
+      "loss": 0.4975,
+      "step": 1790
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.8113200068473816,
+      "learning_rate": 8.282828282828283e-06,
+      "loss": 0.5288,
+      "step": 1800
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.1567124128341675,
+      "learning_rate": 8.272727272727274e-06,
+      "loss": 0.4669,
+      "step": 1810
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.7966761589050293,
+      "learning_rate": 8.262626262626264e-06,
+      "loss": 0.4856,
+      "step": 1820
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.0181186199188232,
+      "learning_rate": 8.252525252525253e-06,
+      "loss": 0.4576,
+      "step": 1830
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.611566960811615,
+      "learning_rate": 8.242424242424243e-06,
+      "loss": 0.496,
+      "step": 1840
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.6482832431793213,
+      "learning_rate": 8.232323232323233e-06,
+      "loss": 0.4601,
+      "step": 1850
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.7550622820854187,
+      "learning_rate": 8.222222222222222e-06,
+      "loss": 0.5036,
+      "step": 1860
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.7835694551467896,
+      "learning_rate": 8.212121212121212e-06,
+      "loss": 0.5617,
+      "step": 1870
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.7926068305969238,
+      "learning_rate": 8.202020202020202e-06,
+      "loss": 0.4327,
+      "step": 1880
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.786851167678833,
+      "learning_rate": 8.191919191919193e-06,
+      "loss": 0.4654,
+      "step": 1890
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.9023171663284302,
+      "learning_rate": 8.181818181818183e-06,
+      "loss": 0.5426,
+      "step": 1900
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.0345401763916016,
+      "learning_rate": 8.171717171717173e-06,
+      "loss": 0.52,
+      "step": 1910
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.945004940032959,
+      "learning_rate": 8.161616161616162e-06,
+      "loss": 0.5512,
+      "step": 1920
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.658362090587616,
+      "learning_rate": 8.151515151515152e-06,
+      "loss": 0.513,
+      "step": 1930
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.6390058398246765,
+      "learning_rate": 8.141414141414142e-06,
+      "loss": 0.5652,
+      "step": 1940
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.7705880403518677,
+      "learning_rate": 8.131313131313133e-06,
+      "loss": 0.4929,
+      "step": 1950
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.5400047302246094,
+      "learning_rate": 8.121212121212121e-06,
+      "loss": 0.4932,
+      "step": 1960
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.9128320217132568,
+      "learning_rate": 8.111111111111112e-06,
+      "loss": 0.5085,
+      "step": 1970
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.0019017457962036,
+      "learning_rate": 8.101010101010102e-06,
+      "loss": 0.4552,
+      "step": 1980
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.818148136138916,
+      "learning_rate": 8.090909090909092e-06,
+      "loss": 0.5517,
+      "step": 1990
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.8848174810409546,
+      "learning_rate": 8.08080808080808e-06,
+      "loss": 0.5094,
+      "step": 2000
+    },
+    {
+      "epoch": 0.72,
+      "eval_loss": 0.6485620141029358,
+      "eval_runtime": 340.5095,
+      "eval_samples_per_second": 2.937,
+      "eval_steps_per_second": 2.937,
+      "step": 2000
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.9389908313751221,
+      "learning_rate": 8.070707070707071e-06,
+      "loss": 0.5228,
+      "step": 2010
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.901942789554596,
+      "learning_rate": 8.060606060606061e-06,
+      "loss": 0.4322,
+      "step": 2020
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.6482081413269043,
+      "learning_rate": 8.050505050505052e-06,
+      "loss": 0.4123,
+      "step": 2030
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.188251256942749,
+      "learning_rate": 8.04040404040404e-06,
+      "loss": 0.3863,
+      "step": 2040
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.6995498538017273,
+      "learning_rate": 8.03030303030303e-06,
+      "loss": 0.4834,
+      "step": 2050
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.7236158847808838,
+      "learning_rate": 8.02020202020202e-06,
+      "loss": 0.5279,
+      "step": 2060
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.6699327826499939,
+      "learning_rate": 8.010101010101011e-06,
+      "loss": 0.5294,
+      "step": 2070
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.7172369956970215,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 0.4882,
+      "step": 2080
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.7741474509239197,
+      "learning_rate": 7.989898989898992e-06,
+      "loss": 0.4928,
+      "step": 2090
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.7598722577095032,
+      "learning_rate": 7.97979797979798e-06,
+      "loss": 0.477,
+      "step": 2100
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.7849257588386536,
+      "learning_rate": 7.96969696969697e-06,
+      "loss": 0.497,
+      "step": 2110
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.7920640707015991,
+      "learning_rate": 7.95959595959596e-06,
+      "loss": 0.4974,
+      "step": 2120
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.7092992067337036,
+      "learning_rate": 7.949494949494951e-06,
+      "loss": 0.5234,
+      "step": 2130
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.8268713355064392,
+      "learning_rate": 7.93939393939394e-06,
+      "loss": 0.3966,
+      "step": 2140
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.1046198606491089,
+      "learning_rate": 7.92929292929293e-06,
+      "loss": 0.452,
+      "step": 2150
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.8234682083129883,
+      "learning_rate": 7.91919191919192e-06,
+      "loss": 0.5364,
+      "step": 2160
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.8524590730667114,
+      "learning_rate": 7.909090909090909e-06,
+      "loss": 0.3922,
+      "step": 2170
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.964320182800293,
+      "learning_rate": 7.898989898989899e-06,
+      "loss": 0.4991,
+      "step": 2180
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.2167562246322632,
+      "learning_rate": 7.88888888888889e-06,
+      "loss": 0.4237,
+      "step": 2190
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.893385648727417,
+      "learning_rate": 7.87878787878788e-06,
+      "loss": 0.503,
+      "step": 2200
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.4781544208526611,
+      "learning_rate": 7.868686868686868e-06,
+      "loss": 0.4084,
+      "step": 2210
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.0282255411148071,
+      "learning_rate": 7.858585858585859e-06,
+      "loss": 0.5531,
+      "step": 2220
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.5956897139549255,
+      "learning_rate": 7.848484848484849e-06,
+      "loss": 0.421,
+      "step": 2230
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.674526572227478,
+      "learning_rate": 7.838383838383839e-06,
+      "loss": 0.5221,
+      "step": 2240
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.7602831125259399,
+      "learning_rate": 7.82828282828283e-06,
+      "loss": 0.4826,
+      "step": 2250
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.637804388999939,
+      "learning_rate": 7.81818181818182e-06,
+      "loss": 0.4965,
+      "step": 2260
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.0468482971191406,
+      "learning_rate": 7.808080808080808e-06,
+      "loss": 0.4352,
+      "step": 2270
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.8232371807098389,
+      "learning_rate": 7.797979797979799e-06,
+      "loss": 0.4314,
+      "step": 2280
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.2416061162948608,
+      "learning_rate": 7.787878787878789e-06,
+      "loss": 0.4559,
+      "step": 2290
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.7591462135314941,
+      "learning_rate": 7.77777777777778e-06,
+      "loss": 0.5443,
+      "step": 2300
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.7769295573234558,
+      "learning_rate": 7.767676767676768e-06,
+      "loss": 0.4915,
+      "step": 2310
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.7007648944854736,
+      "learning_rate": 7.757575757575758e-06,
+      "loss": 0.4102,
+      "step": 2320
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.8512477874755859,
+      "learning_rate": 7.747474747474748e-06,
+      "loss": 0.488,
+      "step": 2330
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.0806859731674194,
+      "learning_rate": 7.737373737373739e-06,
+      "loss": 0.4482,
+      "step": 2340
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.8377964496612549,
+      "learning_rate": 7.727272727272727e-06,
+      "loss": 0.4714,
+      "step": 2350
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.9962327480316162,
+      "learning_rate": 7.717171717171717e-06,
+      "loss": 0.4354,
+      "step": 2360
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.7444832921028137,
+      "learning_rate": 7.707070707070708e-06,
+      "loss": 0.5651,
+      "step": 2370
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.7137871384620667,
+      "learning_rate": 7.696969696969696e-06,
+      "loss": 0.4773,
+      "step": 2380
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.8521376252174377,
+      "learning_rate": 7.686868686868687e-06,
+      "loss": 0.4961,
+      "step": 2390
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.9442594647407532,
+      "learning_rate": 7.676767676767677e-06,
+      "loss": 0.4627,
+      "step": 2400
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.169360876083374,
+      "learning_rate": 7.666666666666667e-06,
+      "loss": 0.4601,
+      "step": 2410
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.8181536197662354,
+      "learning_rate": 7.656565656565658e-06,
+      "loss": 0.4635,
+      "step": 2420
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.9113516211509705,
+      "learning_rate": 7.646464646464648e-06,
+      "loss": 0.4219,
+      "step": 2430
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.26584792137146,
+      "learning_rate": 7.636363636363638e-06,
+      "loss": 0.5165,
+      "step": 2440
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.6445989608764648,
+      "learning_rate": 7.6262626262626275e-06,
+      "loss": 0.4755,
+      "step": 2450
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.7551100850105286,
+      "learning_rate": 7.616161616161617e-06,
+      "loss": 0.4976,
+      "step": 2460
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.0578736066818237,
+      "learning_rate": 7.606060606060606e-06,
+      "loss": 0.524,
+      "step": 2470
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.0422261953353882,
+      "learning_rate": 7.595959595959597e-06,
+      "loss": 0.5136,
+      "step": 2480
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.0524308681488037,
+      "learning_rate": 7.585858585858586e-06,
+      "loss": 0.5058,
+      "step": 2490
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.8334051966667175,
+      "learning_rate": 7.5757575757575764e-06,
+      "loss": 0.4487,
+      "step": 2500
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.9613906145095825,
+      "learning_rate": 7.565656565656566e-06,
+      "loss": 0.5055,
+      "step": 2510
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.9347293972969055,
+      "learning_rate": 7.555555555555556e-06,
+      "loss": 0.4672,
+      "step": 2520
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.0498868227005005,
+      "learning_rate": 7.545454545454546e-06,
+      "loss": 0.4336,
+      "step": 2530
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.0005099773406982,
+      "learning_rate": 7.535353535353536e-06,
+      "loss": 0.523,
+      "step": 2540
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.027875542640686,
+      "learning_rate": 7.525252525252525e-06,
+      "loss": 0.5088,
+      "step": 2550
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.7960171103477478,
+      "learning_rate": 7.515151515151516e-06,
+      "loss": 0.4636,
+      "step": 2560
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.1945937871932983,
+      "learning_rate": 7.505050505050505e-06,
+      "loss": 0.4736,
+      "step": 2570
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.9849948287010193,
+      "learning_rate": 7.494949494949496e-06,
+      "loss": 0.4766,
+      "step": 2580
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.7381343841552734,
+      "learning_rate": 7.484848484848486e-06,
+      "loss": 0.4427,
+      "step": 2590
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.964435875415802,
+      "learning_rate": 7.474747474747476e-06,
+      "loss": 0.5179,
+      "step": 2600
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.0832308530807495,
+      "learning_rate": 7.464646464646465e-06,
+      "loss": 0.5179,
+      "step": 2610
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.8275842070579529,
+      "learning_rate": 7.454545454545456e-06,
+      "loss": 0.3957,
+      "step": 2620
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.7618038654327393,
+      "learning_rate": 7.444444444444445e-06,
+      "loss": 0.4626,
+      "step": 2630
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.9667460918426514,
+      "learning_rate": 7.434343434343435e-06,
+      "loss": 0.4338,
+      "step": 2640
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.8243513107299805,
+      "learning_rate": 7.424242424242425e-06,
+      "loss": 0.496,
+      "step": 2650
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.8097424507141113,
+      "learning_rate": 7.414141414141415e-06,
+      "loss": 0.4911,
+      "step": 2660
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.219351053237915,
+      "learning_rate": 7.4040404040404045e-06,
+      "loss": 0.5163,
+      "step": 2670
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.7272489666938782,
+      "learning_rate": 7.393939393939395e-06,
+      "loss": 0.4783,
+      "step": 2680
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.9799202084541321,
+      "learning_rate": 7.383838383838384e-06,
+      "loss": 0.4638,
+      "step": 2690
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.1537474393844604,
+      "learning_rate": 7.373737373737374e-06,
+      "loss": 0.4184,
+      "step": 2700
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.750483512878418,
+      "learning_rate": 7.363636363636364e-06,
+      "loss": 0.4544,
+      "step": 2710
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.915256679058075,
+      "learning_rate": 7.353535353535353e-06,
+      "loss": 0.4579,
+      "step": 2720
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.0220776796340942,
+      "learning_rate": 7.343434343434344e-06,
+      "loss": 0.4568,
+      "step": 2730
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.0540516376495361,
+      "learning_rate": 7.333333333333333e-06,
+      "loss": 0.5228,
+      "step": 2740
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 2.014601230621338,
+      "learning_rate": 7.323232323232324e-06,
+      "loss": 0.4482,
+      "step": 2750
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.9005866646766663,
+      "learning_rate": 7.3131313131313146e-06,
+      "loss": 0.4868,
+      "step": 2760
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.0538028478622437,
+      "learning_rate": 7.303030303030304e-06,
+      "loss": 0.4884,
+      "step": 2770
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.9115027785301208,
+      "learning_rate": 7.2929292929292934e-06,
+      "loss": 0.4421,
+      "step": 2780
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 1.5715361833572388,
+      "learning_rate": 7.282828282828284e-06,
+      "loss": 0.4442,
+      "step": 2790
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.8492701649665833,
+      "learning_rate": 7.272727272727273e-06,
+      "loss": 0.4828,
+      "step": 2800
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.7459146976470947,
+      "learning_rate": 7.2626262626262635e-06,
+      "loss": 0.4649,
+      "step": 2810
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.6268482804298401,
+      "learning_rate": 7.252525252525253e-06,
+      "loss": 0.4521,
+      "step": 2820
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.9125592112541199,
+      "learning_rate": 7.242424242424243e-06,
+      "loss": 0.57,
+      "step": 2830
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.8162791132926941,
+      "learning_rate": 7.232323232323233e-06,
+      "loss": 0.4746,
+      "step": 2840
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.9345506429672241,
+      "learning_rate": 7.222222222222223e-06,
+      "loss": 0.4777,
+      "step": 2850
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.6382574439048767,
+      "learning_rate": 7.212121212121212e-06,
+      "loss": 0.4832,
+      "step": 2860
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.9409840703010559,
+      "learning_rate": 7.202020202020203e-06,
+      "loss": 0.4659,
+      "step": 2870
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 1.0053200721740723,
+      "learning_rate": 7.191919191919192e-06,
+      "loss": 0.4041,
+      "step": 2880
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.8394672870635986,
+      "learning_rate": 7.181818181818182e-06,
+      "loss": 0.4477,
+      "step": 2890
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.8685116171836853,
+      "learning_rate": 7.171717171717172e-06,
+      "loss": 0.4155,
+      "step": 2900
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.8150361776351929,
+      "learning_rate": 7.161616161616162e-06,
+      "loss": 0.3901,
+      "step": 2910
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.7799510955810547,
+      "learning_rate": 7.151515151515152e-06,
+      "loss": 0.4273,
+      "step": 2920
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 1.1197009086608887,
+      "learning_rate": 7.141414141414143e-06,
+      "loss": 0.447,
+      "step": 2930
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.8753387331962585,
+      "learning_rate": 7.131313131313132e-06,
+      "loss": 0.3847,
+      "step": 2940
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.9633245468139648,
+      "learning_rate": 7.121212121212122e-06,
+      "loss": 0.4985,
+      "step": 2950
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 1.0391559600830078,
+      "learning_rate": 7.111111111111112e-06,
+      "loss": 0.4476,
+      "step": 2960
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.9432191252708435,
+      "learning_rate": 7.101010101010102e-06,
+      "loss": 0.4669,
+      "step": 2970
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 1.078410267829895,
+      "learning_rate": 7.0909090909090916e-06,
+      "loss": 0.401,
+      "step": 2980
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.9405683875083923,
+      "learning_rate": 7.080808080808082e-06,
+      "loss": 0.4744,
+      "step": 2990
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 1.4189893007278442,
+      "learning_rate": 7.070707070707071e-06,
+      "loss": 0.4334,
+      "step": 3000
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.9169256091117859,
+      "learning_rate": 7.060606060606061e-06,
+      "loss": 0.4234,
+      "step": 3010
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.8117393851280212,
+      "learning_rate": 7.050505050505051e-06,
+      "loss": 0.4676,
+      "step": 3020
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.967890739440918,
+      "learning_rate": 7.0404040404040404e-06,
+      "loss": 0.4497,
+      "step": 3030
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 1.0548313856124878,
+      "learning_rate": 7.030303030303031e-06,
+      "loss": 0.4429,
+      "step": 3040
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 1.071303367614746,
+      "learning_rate": 7.02020202020202e-06,
+      "loss": 0.4251,
+      "step": 3050
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 1.1772490739822388,
+      "learning_rate": 7.0101010101010105e-06,
+      "loss": 0.4897,
+      "step": 3060
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 1.3903322219848633,
+      "learning_rate": 7e-06,
+      "loss": 0.4953,
+      "step": 3070
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.8463280200958252,
+      "learning_rate": 6.98989898989899e-06,
+      "loss": 0.3923,
+      "step": 3080
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 1.146888256072998,
+      "learning_rate": 6.979797979797981e-06,
+      "loss": 0.4723,
+      "step": 3090
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 1.0877312421798706,
+      "learning_rate": 6.969696969696971e-06,
+      "loss": 0.4822,
+      "step": 3100
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.7444977164268494,
+      "learning_rate": 6.95959595959596e-06,
+      "loss": 0.4591,
+      "step": 3110
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.7259607911109924,
+      "learning_rate": 6.9494949494949505e-06,
+      "loss": 0.4564,
+      "step": 3120
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.9535048604011536,
+      "learning_rate": 6.93939393939394e-06,
+      "loss": 0.4402,
+      "step": 3130
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.9428088068962097,
+      "learning_rate": 6.92929292929293e-06,
+      "loss": 0.4039,
+      "step": 3140
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.9858347177505493,
+      "learning_rate": 6.91919191919192e-06,
+      "loss": 0.4334,
+      "step": 3150
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 1.0837122201919556,
+      "learning_rate": 6.90909090909091e-06,
+      "loss": 0.3957,
+      "step": 3160
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 1.0605705976486206,
+      "learning_rate": 6.898989898989899e-06,
+      "loss": 0.4195,
+      "step": 3170
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 1.0856525897979736,
+      "learning_rate": 6.88888888888889e-06,
+      "loss": 0.4286,
+      "step": 3180
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 1.02509605884552,
+      "learning_rate": 6.878787878787879e-06,
+      "loss": 0.4413,
+      "step": 3190
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.8847616314888,
+      "learning_rate": 6.868686868686869e-06,
+      "loss": 0.4117,
+      "step": 3200
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 1.1464060544967651,
+      "learning_rate": 6.858585858585859e-06,
+      "loss": 0.44,
+      "step": 3210
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.9589990973472595,
+      "learning_rate": 6.848484848484849e-06,
+      "loss": 0.4896,
+      "step": 3220
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 1.080541968345642,
+      "learning_rate": 6.8383838383838386e-06,
+      "loss": 0.4203,
+      "step": 3230
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 1.3150702714920044,
+      "learning_rate": 6.828282828282828e-06,
+      "loss": 0.4213,
+      "step": 3240
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 1.0854105949401855,
+      "learning_rate": 6.818181818181818e-06,
+      "loss": 0.4538,
+      "step": 3250
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.7211188673973083,
+      "learning_rate": 6.808080808080809e-06,
+      "loss": 0.4332,
+      "step": 3260
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 1.0323004722595215,
+      "learning_rate": 6.797979797979799e-06,
+      "loss": 0.465,
+      "step": 3270
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.9176574349403381,
+      "learning_rate": 6.787878787878789e-06,
+      "loss": 0.4898,
+      "step": 3280
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 1.2033518552780151,
+      "learning_rate": 6.777777777777779e-06,
+      "loss": 0.4652,
+      "step": 3290
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.6069207191467285,
+      "learning_rate": 6.767676767676769e-06,
+      "loss": 0.4663,
+      "step": 3300
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.9627435803413391,
+      "learning_rate": 6.757575757575758e-06,
+      "loss": 0.4992,
+      "step": 3310
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.7333098649978638,
+      "learning_rate": 6.747474747474749e-06,
+      "loss": 0.4876,
+      "step": 3320
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.6735036969184875,
+      "learning_rate": 6.737373737373738e-06,
+      "loss": 0.4797,
+      "step": 3330
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.8416649699211121,
+      "learning_rate": 6.7272727272727275e-06,
+      "loss": 0.3889,
+      "step": 3340
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 1.227149486541748,
+      "learning_rate": 6.717171717171718e-06,
+      "loss": 0.4677,
+      "step": 3350
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 1.5126155614852905,
+      "learning_rate": 6.707070707070707e-06,
+      "loss": 0.4444,
+      "step": 3360
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 1.1622529029846191,
+      "learning_rate": 6.6969696969696975e-06,
+      "loss": 0.4372,
+      "step": 3370
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.975615918636322,
+      "learning_rate": 6.686868686868687e-06,
+      "loss": 0.4037,
+      "step": 3380
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.8992089033126831,
+      "learning_rate": 6.676767676767677e-06,
+      "loss": 0.3785,
+      "step": 3390
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 1.4439970254898071,
+      "learning_rate": 6.666666666666667e-06,
+      "loss": 0.5169,
+      "step": 3400
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.9601142406463623,
+      "learning_rate": 6.656565656565657e-06,
+      "loss": 0.4396,
+      "step": 3410
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 1.0457966327667236,
+      "learning_rate": 6.646464646464646e-06,
+      "loss": 0.4065,
+      "step": 3420
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 1.0014325380325317,
+      "learning_rate": 6.6363636363636375e-06,
+      "loss": 0.4616,
+      "step": 3430
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 1.3321492671966553,
+      "learning_rate": 6.626262626262627e-06,
+      "loss": 0.4164,
+      "step": 3440
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.9500308036804199,
+      "learning_rate": 6.616161616161617e-06,
+      "loss": 0.4156,
+      "step": 3450
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 1.1907135248184204,
+      "learning_rate": 6.606060606060607e-06,
+      "loss": 0.4353,
+      "step": 3460
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 1.7826789617538452,
+      "learning_rate": 6.595959595959597e-06,
+      "loss": 0.4764,
+      "step": 3470
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 1.0160069465637207,
+      "learning_rate": 6.585858585858586e-06,
+      "loss": 0.4594,
+      "step": 3480
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 1.0795533657073975,
+      "learning_rate": 6.575757575757577e-06,
+      "loss": 0.4557,
+      "step": 3490
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 1.2484716176986694,
+      "learning_rate": 6.565656565656566e-06,
+      "loss": 0.5057,
+      "step": 3500
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 1.6144826412200928,
+      "learning_rate": 6.555555555555556e-06,
+      "loss": 0.4554,
+      "step": 3510
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 1.3575657606124878,
+      "learning_rate": 6.545454545454546e-06,
+      "loss": 0.3373,
+      "step": 3520
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.9021044373512268,
+      "learning_rate": 6.535353535353536e-06,
+      "loss": 0.4765,
+      "step": 3530
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 1.0576077699661255,
+      "learning_rate": 6.525252525252526e-06,
+      "loss": 0.4344,
+      "step": 3540
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 1.0687928199768066,
+      "learning_rate": 6.515151515151516e-06,
+      "loss": 0.5022,
+      "step": 3550
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.9124734401702881,
+      "learning_rate": 6.505050505050505e-06,
+      "loss": 0.4133,
+      "step": 3560
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.8715507984161377,
+      "learning_rate": 6.494949494949495e-06,
+      "loss": 0.454,
+      "step": 3570
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 1.0889540910720825,
+      "learning_rate": 6.484848484848485e-06,
+      "loss": 0.443,
+      "step": 3580
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.9323639273643494,
+      "learning_rate": 6.4747474747474745e-06,
+      "loss": 0.4983,
+      "step": 3590
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.6946113109588623,
+      "learning_rate": 6.464646464646466e-06,
+      "loss": 0.4364,
+      "step": 3600
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 1.1879271268844604,
+      "learning_rate": 6.454545454545456e-06,
+      "loss": 0.4816,
+      "step": 3610
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 1.337172031402588,
+      "learning_rate": 6.444444444444445e-06,
+      "loss": 0.4963,
+      "step": 3620
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.9384005069732666,
+      "learning_rate": 6.434343434343436e-06,
+      "loss": 0.5185,
+      "step": 3630
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 1.2666473388671875,
+      "learning_rate": 6.424242424242425e-06,
+      "loss": 0.389,
+      "step": 3640
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 1.037668228149414,
+      "learning_rate": 6.4141414141414145e-06,
+      "loss": 0.4529,
+      "step": 3650
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 1.4652140140533447,
+      "learning_rate": 6.404040404040405e-06,
+      "loss": 0.3915,
+      "step": 3660
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 1.1823006868362427,
+      "learning_rate": 6.393939393939394e-06,
+      "loss": 0.5123,
+      "step": 3670
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 1.0706424713134766,
+      "learning_rate": 6.3838383838383845e-06,
+      "loss": 0.4989,
+      "step": 3680
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.9993262887001038,
+      "learning_rate": 6.373737373737374e-06,
+      "loss": 0.4268,
+      "step": 3690
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.9947906732559204,
+      "learning_rate": 6.363636363636364e-06,
+      "loss": 0.4246,
+      "step": 3700
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.8563311696052551,
+      "learning_rate": 6.353535353535354e-06,
+      "loss": 0.4439,
+      "step": 3710
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 1.2519429922103882,
+      "learning_rate": 6.343434343434344e-06,
+      "loss": 0.3973,
+      "step": 3720
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 1.2925148010253906,
+      "learning_rate": 6.333333333333333e-06,
+      "loss": 0.4179,
+      "step": 3730
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 1.1731756925582886,
+      "learning_rate": 6.323232323232324e-06,
+      "loss": 0.4748,
+      "step": 3740
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 1.030536413192749,
+      "learning_rate": 6.313131313131313e-06,
+      "loss": 0.4131,
+      "step": 3750
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 1.1279093027114868,
+      "learning_rate": 6.303030303030303e-06,
+      "loss": 0.4143,
+      "step": 3760
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 1.1602801084518433,
+      "learning_rate": 6.292929292929294e-06,
+      "loss": 0.4718,
+      "step": 3770
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 1.5314010381698608,
+      "learning_rate": 6.282828282828284e-06,
+      "loss": 0.5463,
+      "step": 3780
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 1.6124476194381714,
+      "learning_rate": 6.2727272727272734e-06,
+      "loss": 0.4857,
+      "step": 3790
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 1.185355305671692,
+      "learning_rate": 6.262626262626264e-06,
+      "loss": 0.4167,
+      "step": 3800
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 1.1900933980941772,
+      "learning_rate": 6.252525252525253e-06,
+      "loss": 0.4658,
+      "step": 3810
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 1.2930880784988403,
+      "learning_rate": 6.2424242424242434e-06,
+      "loss": 0.4912,
+      "step": 3820
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 1.1819838285446167,
+      "learning_rate": 6.232323232323233e-06,
+      "loss": 0.4129,
+      "step": 3830
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 1.315292477607727,
+      "learning_rate": 6.222222222222223e-06,
+      "loss": 0.477,
+      "step": 3840
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 1.007524013519287,
+      "learning_rate": 6.212121212121213e-06,
+      "loss": 0.414,
+      "step": 3850
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 1.1948471069335938,
+      "learning_rate": 6.202020202020203e-06,
+      "loss": 0.4576,
+      "step": 3860
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 1.3339303731918335,
+      "learning_rate": 6.191919191919192e-06,
+      "loss": 0.5547,
+      "step": 3870
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 1.2568715810775757,
+      "learning_rate": 6.181818181818182e-06,
+      "loss": 0.3941,
+      "step": 3880
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 1.2084232568740845,
+      "learning_rate": 6.171717171717172e-06,
+      "loss": 0.462,
+      "step": 3890
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 1.0328818559646606,
+      "learning_rate": 6.1616161616161615e-06,
+      "loss": 0.4484,
+      "step": 3900
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 1.9673748016357422,
+      "learning_rate": 6.151515151515152e-06,
+      "loss": 0.4543,
+      "step": 3910
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 1.4105424880981445,
+      "learning_rate": 6.141414141414141e-06,
+      "loss": 0.4964,
+      "step": 3920
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.7082942128181458,
+      "learning_rate": 6.1313131313131315e-06,
+      "loss": 0.5297,
+      "step": 3930
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 1.2604327201843262,
+      "learning_rate": 6.121212121212121e-06,
+      "loss": 0.4207,
+      "step": 3940
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.999332070350647,
+      "learning_rate": 6.111111111111112e-06,
+      "loss": 0.4791,
+      "step": 3950
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 1.139615535736084,
+      "learning_rate": 6.1010101010101015e-06,
+      "loss": 0.4281,
+      "step": 3960
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 1.1164727210998535,
+      "learning_rate": 6.090909090909092e-06,
+      "loss": 0.3807,
+      "step": 3970
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 1.0829392671585083,
+      "learning_rate": 6.080808080808081e-06,
+      "loss": 0.4996,
+      "step": 3980
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.8551573753356934,
+      "learning_rate": 6.0707070707070715e-06,
+      "loss": 0.4759,
+      "step": 3990
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 1.224134087562561,
+      "learning_rate": 6.060606060606061e-06,
+      "loss": 0.4244,
+      "step": 4000
+    },
+    {
+      "epoch": 1.43,
+      "eval_loss": 0.5992664694786072,
+      "eval_runtime": 334.8169,
+      "eval_samples_per_second": 2.987,
+      "eval_steps_per_second": 2.987,
+      "step": 4000
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 1.1042165756225586,
+      "learning_rate": 6.050505050505051e-06,
+      "loss": 0.4815,
+      "step": 4010
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 1.087483525276184,
+      "learning_rate": 6.040404040404041e-06,
+      "loss": 0.4918,
+      "step": 4020
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 1.0040034055709839,
+      "learning_rate": 6.030303030303031e-06,
+      "loss": 0.4666,
+      "step": 4030
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 1.2772231101989746,
+      "learning_rate": 6.0202020202020204e-06,
+      "loss": 0.4847,
+      "step": 4040
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 1.217619776725769,
+      "learning_rate": 6.010101010101011e-06,
+      "loss": 0.4138,
+      "step": 4050
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 1.109666347503662,
+      "learning_rate": 6e-06,
+      "loss": 0.4799,
+      "step": 4060
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 1.2436999082565308,
+      "learning_rate": 5.9898989898989904e-06,
+      "loss": 0.4218,
+      "step": 4070
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 1.252548098564148,
+      "learning_rate": 5.97979797979798e-06,
+      "loss": 0.4585,
+      "step": 4080
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 1.2544574737548828,
+      "learning_rate": 5.96969696969697e-06,
+      "loss": 0.4633,
+      "step": 4090
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 1.144337773323059,
+      "learning_rate": 5.95959595959596e-06,
+      "loss": 0.4408,
+      "step": 4100
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 1.067790150642395,
+      "learning_rate": 5.949494949494949e-06,
+      "loss": 0.4349,
+      "step": 4110
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.8417388200759888,
+      "learning_rate": 5.93939393939394e-06,
+      "loss": 0.4859,
+      "step": 4120
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 1.5285634994506836,
+      "learning_rate": 5.9292929292929305e-06,
+      "loss": 0.4556,
+      "step": 4130
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 1.1098241806030273,
+      "learning_rate": 5.91919191919192e-06,
+      "loss": 0.5067,
+      "step": 4140
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.8364139795303345,
+      "learning_rate": 5.90909090909091e-06,
+      "loss": 0.4781,
+      "step": 4150
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.8289257287979126,
+      "learning_rate": 5.8989898989899e-06,
+      "loss": 0.3966,
+      "step": 4160
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.9829197525978088,
+      "learning_rate": 5.88888888888889e-06,
+      "loss": 0.4382,
+      "step": 4170
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.7760851383209229,
+      "learning_rate": 5.878787878787879e-06,
+      "loss": 0.3977,
+      "step": 4180
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 1.0614337921142578,
+      "learning_rate": 5.868686868686869e-06,
+      "loss": 0.4382,
+      "step": 4190
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 1.098379135131836,
+      "learning_rate": 5.858585858585859e-06,
+      "loss": 0.3688,
+      "step": 4200
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 1.2058717012405396,
+      "learning_rate": 5.8484848484848485e-06,
+      "loss": 0.4048,
+      "step": 4210
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.9828083515167236,
+      "learning_rate": 5.838383838383839e-06,
+      "loss": 0.4997,
+      "step": 4220
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 1.327776551246643,
+      "learning_rate": 5.828282828282828e-06,
+      "loss": 0.4621,
+      "step": 4230
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 1.2171368598937988,
+      "learning_rate": 5.8181818181818185e-06,
+      "loss": 0.4177,
+      "step": 4240
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 1.4939155578613281,
+      "learning_rate": 5.808080808080808e-06,
+      "loss": 0.4821,
+      "step": 4250
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 1.523369550704956,
+      "learning_rate": 5.797979797979798e-06,
+      "loss": 0.4116,
+      "step": 4260
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 1.3256771564483643,
+      "learning_rate": 5.787878787878788e-06,
+      "loss": 0.4948,
+      "step": 4270
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 1.5732948780059814,
+      "learning_rate": 5.777777777777778e-06,
+      "loss": 0.5275,
+      "step": 4280
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 1.1372538805007935,
+      "learning_rate": 5.767676767676768e-06,
+      "loss": 0.463,
+      "step": 4290
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.9834716320037842,
+      "learning_rate": 5.7575757575757586e-06,
+      "loss": 0.4383,
+      "step": 4300
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 1.443036437034607,
+      "learning_rate": 5.747474747474748e-06,
+      "loss": 0.4781,
+      "step": 4310
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 1.2556253671646118,
+      "learning_rate": 5.737373737373738e-06,
+      "loss": 0.4159,
+      "step": 4320
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 1.272565245628357,
+      "learning_rate": 5.727272727272728e-06,
+      "loss": 0.433,
+      "step": 4330
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 1.078231930732727,
+      "learning_rate": 5.717171717171718e-06,
+      "loss": 0.4644,
+      "step": 4340
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 1.1723682880401611,
+      "learning_rate": 5.7070707070707075e-06,
+      "loss": 0.4555,
+      "step": 4350
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 1.2048708200454712,
+      "learning_rate": 5.696969696969698e-06,
+      "loss": 0.4618,
+      "step": 4360
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.8913971781730652,
+      "learning_rate": 5.686868686868687e-06,
+      "loss": 0.4015,
+      "step": 4370
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 1.1153560876846313,
+      "learning_rate": 5.6767676767676775e-06,
+      "loss": 0.5233,
+      "step": 4380
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 1.055448293685913,
+      "learning_rate": 5.666666666666667e-06,
+      "loss": 0.398,
+      "step": 4390
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 1.2389256954193115,
+      "learning_rate": 5.656565656565657e-06,
+      "loss": 0.453,
+      "step": 4400
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.9692911505699158,
+      "learning_rate": 5.646464646464647e-06,
+      "loss": 0.3773,
+      "step": 4410
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.9325522780418396,
+      "learning_rate": 5.636363636363636e-06,
+      "loss": 0.4471,
+      "step": 4420
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 1.0931851863861084,
+      "learning_rate": 5.626262626262626e-06,
+      "loss": 0.452,
+      "step": 4430
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.8405607342720032,
+      "learning_rate": 5.616161616161616e-06,
+      "loss": 0.4743,
+      "step": 4440
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 1.1500225067138672,
+      "learning_rate": 5.606060606060606e-06,
+      "loss": 0.4135,
+      "step": 4450
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 1.0774083137512207,
+      "learning_rate": 5.595959595959597e-06,
+      "loss": 0.3775,
+      "step": 4460
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 1.1228448152542114,
+      "learning_rate": 5.585858585858587e-06,
+      "loss": 0.5094,
+      "step": 4470
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.7777533531188965,
+      "learning_rate": 5.575757575757577e-06,
+      "loss": 0.4158,
+      "step": 4480
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 1.1433957815170288,
+      "learning_rate": 5.565656565656566e-06,
+      "loss": 0.4466,
+      "step": 4490
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 1.3501386642456055,
+      "learning_rate": 5.555555555555557e-06,
+      "loss": 0.5255,
+      "step": 4500
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 1.4286872148513794,
+      "learning_rate": 5.545454545454546e-06,
+      "loss": 0.5021,
+      "step": 4510
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.9643106460571289,
+      "learning_rate": 5.5353535353535355e-06,
+      "loss": 0.4004,
+      "step": 4520
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.8871945142745972,
+      "learning_rate": 5.525252525252526e-06,
+      "loss": 0.3657,
+      "step": 4530
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 1.7113863229751587,
+      "learning_rate": 5.515151515151515e-06,
+      "loss": 0.374,
+      "step": 4540
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 1.3388639688491821,
+      "learning_rate": 5.5050505050505056e-06,
+      "loss": 0.4632,
+      "step": 4550
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 1.1603946685791016,
+      "learning_rate": 5.494949494949495e-06,
+      "loss": 0.4569,
+      "step": 4560
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 1.250316858291626,
+      "learning_rate": 5.484848484848485e-06,
+      "loss": 0.3804,
+      "step": 4570
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 1.1444791555404663,
+      "learning_rate": 5.474747474747475e-06,
+      "loss": 0.4521,
+      "step": 4580
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.804335355758667,
+      "learning_rate": 5.464646464646465e-06,
+      "loss": 0.3925,
+      "step": 4590
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 1.2074679136276245,
+      "learning_rate": 5.4545454545454545e-06,
+      "loss": 0.3929,
+      "step": 4600
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 1.3282480239868164,
+      "learning_rate": 5.444444444444445e-06,
+      "loss": 0.4577,
+      "step": 4610
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 1.075721263885498,
+      "learning_rate": 5.434343434343434e-06,
+      "loss": 0.4422,
+      "step": 4620
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 1.4774171113967896,
+      "learning_rate": 5.424242424242425e-06,
+      "loss": 0.4462,
+      "step": 4630
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.7713523507118225,
+      "learning_rate": 5.414141414141415e-06,
+      "loss": 0.4863,
+      "step": 4640
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.9707368016242981,
+      "learning_rate": 5.404040404040405e-06,
+      "loss": 0.4625,
+      "step": 4650
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 1.6415660381317139,
+      "learning_rate": 5.3939393939393945e-06,
+      "loss": 0.5686,
+      "step": 4660
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.9006029367446899,
+      "learning_rate": 5.383838383838385e-06,
+      "loss": 0.4107,
+      "step": 4670
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 1.580544114112854,
+      "learning_rate": 5.373737373737374e-06,
+      "loss": 0.4864,
+      "step": 4680
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 2.001404285430908,
+      "learning_rate": 5.3636363636363645e-06,
+      "loss": 0.4311,
+      "step": 4690
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.8837793469429016,
+      "learning_rate": 5.353535353535354e-06,
+      "loss": 0.5013,
+      "step": 4700
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 1.3012306690216064,
+      "learning_rate": 5.343434343434344e-06,
+      "loss": 0.4551,
+      "step": 4710
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.9690237641334534,
+      "learning_rate": 5.333333333333334e-06,
+      "loss": 0.4975,
+      "step": 4720
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 1.2548136711120605,
+      "learning_rate": 5.323232323232324e-06,
+      "loss": 0.3947,
+      "step": 4730
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 1.3401527404785156,
+      "learning_rate": 5.313131313131313e-06,
+      "loss": 0.4532,
+      "step": 4740
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 1.481223225593567,
+      "learning_rate": 5.303030303030303e-06,
+      "loss": 0.4406,
+      "step": 4750
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 1.0860910415649414,
+      "learning_rate": 5.292929292929293e-06,
+      "loss": 0.4261,
+      "step": 4760
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 1.8275758028030396,
+      "learning_rate": 5.2828282828282825e-06,
+      "loss": 0.4282,
+      "step": 4770
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.8796372413635254,
+      "learning_rate": 5.272727272727273e-06,
+      "loss": 0.4372,
+      "step": 4780
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 1.7672245502471924,
+      "learning_rate": 5.262626262626262e-06,
+      "loss": 0.4625,
+      "step": 4790
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 1.187982439994812,
+      "learning_rate": 5.252525252525253e-06,
+      "loss": 0.4431,
+      "step": 4800
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 1.0611240863800049,
+      "learning_rate": 5.242424242424244e-06,
+      "loss": 0.4097,
+      "step": 4810
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 1.259895920753479,
+      "learning_rate": 5.232323232323233e-06,
+      "loss": 0.42,
+      "step": 4820
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 1.0518171787261963,
+      "learning_rate": 5.2222222222222226e-06,
+      "loss": 0.4103,
+      "step": 4830
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 1.1404093503952026,
+      "learning_rate": 5.212121212121213e-06,
+      "loss": 0.4167,
+      "step": 4840
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 1.3545587062835693,
+      "learning_rate": 5.202020202020202e-06,
+      "loss": 0.4231,
+      "step": 4850
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 1.3752548694610596,
+      "learning_rate": 5.191919191919193e-06,
+      "loss": 0.4385,
+      "step": 4860
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 1.0795814990997314,
+      "learning_rate": 5.181818181818182e-06,
+      "loss": 0.5562,
+      "step": 4870
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 1.176770806312561,
+      "learning_rate": 5.171717171717172e-06,
+      "loss": 0.3841,
+      "step": 4880
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 1.0362882614135742,
+      "learning_rate": 5.161616161616162e-06,
+      "loss": 0.412,
+      "step": 4890
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.854006826877594,
+      "learning_rate": 5.151515151515152e-06,
+      "loss": 0.4833,
+      "step": 4900
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 1.257077693939209,
+      "learning_rate": 5.1414141414141415e-06,
+      "loss": 0.4173,
+      "step": 4910
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 1.0905293226242065,
+      "learning_rate": 5.131313131313132e-06,
+      "loss": 0.4145,
+      "step": 4920
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 1.1781564950942993,
+      "learning_rate": 5.121212121212121e-06,
+      "loss": 0.4619,
+      "step": 4930
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.746283233165741,
+      "learning_rate": 5.1111111111111115e-06,
+      "loss": 0.4203,
+      "step": 4940
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 1.3257133960723877,
+      "learning_rate": 5.101010101010101e-06,
+      "loss": 0.442,
+      "step": 4950
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.8805459141731262,
+      "learning_rate": 5.090909090909091e-06,
+      "loss": 0.4587,
+      "step": 4960
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 1.0892632007598877,
+      "learning_rate": 5.0808080808080815e-06,
+      "loss": 0.4005,
+      "step": 4970
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 1.2787294387817383,
+      "learning_rate": 5.070707070707072e-06,
+      "loss": 0.4589,
+      "step": 4980
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.8316894173622131,
+      "learning_rate": 5.060606060606061e-06,
+      "loss": 0.4888,
+      "step": 4990
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 1.487030029296875,
+      "learning_rate": 5.0505050505050515e-06,
+      "loss": 0.3977,
+      "step": 5000
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.9546489119529724,
+      "learning_rate": 5.040404040404041e-06,
+      "loss": 0.5509,
+      "step": 5010
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 1.015142560005188,
+      "learning_rate": 5.030303030303031e-06,
+      "loss": 0.3927,
+      "step": 5020
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 1.2369190454483032,
+      "learning_rate": 5.020202020202021e-06,
+      "loss": 0.4477,
+      "step": 5030
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 1.1981439590454102,
+      "learning_rate": 5.010101010101011e-06,
+      "loss": 0.4566,
+      "step": 5040
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 1.507062554359436,
+      "learning_rate": 5e-06,
+      "loss": 0.3958,
+      "step": 5050
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 1.2192580699920654,
+      "learning_rate": 4.98989898989899e-06,
+      "loss": 0.4423,
+      "step": 5060
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 1.2157553434371948,
+      "learning_rate": 4.97979797979798e-06,
+      "loss": 0.4268,
+      "step": 5070
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 1.0647870302200317,
+      "learning_rate": 4.9696969696969696e-06,
+      "loss": 0.4907,
+      "step": 5080
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 1.071527123451233,
+      "learning_rate": 4.95959595959596e-06,
+      "loss": 0.4471,
+      "step": 5090
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 1.6718330383300781,
+      "learning_rate": 4.94949494949495e-06,
+      "loss": 0.4899,
+      "step": 5100
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 1.4382431507110596,
+      "learning_rate": 4.93939393939394e-06,
+      "loss": 0.4316,
+      "step": 5110
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 1.5574830770492554,
+      "learning_rate": 4.92929292929293e-06,
+      "loss": 0.4293,
+      "step": 5120
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 1.417203664779663,
+      "learning_rate": 4.919191919191919e-06,
+      "loss": 0.3967,
+      "step": 5130
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.9984129667282104,
+      "learning_rate": 4.90909090909091e-06,
+      "loss": 0.4927,
+      "step": 5140
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 1.204757809638977,
+      "learning_rate": 4.898989898989899e-06,
+      "loss": 0.5291,
+      "step": 5150
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.959578812122345,
+      "learning_rate": 4.888888888888889e-06,
+      "loss": 0.4275,
+      "step": 5160
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 1.5679115056991577,
+      "learning_rate": 4.878787878787879e-06,
+      "loss": 0.4559,
+      "step": 5170
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.9421458840370178,
+      "learning_rate": 4.868686868686869e-06,
+      "loss": 0.4496,
+      "step": 5180
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 1.3694195747375488,
+      "learning_rate": 4.858585858585859e-06,
+      "loss": 0.4168,
+      "step": 5190
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.9521112442016602,
+      "learning_rate": 4.848484848484849e-06,
+      "loss": 0.469,
+      "step": 5200
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 1.3057090044021606,
+      "learning_rate": 4.838383838383839e-06,
+      "loss": 0.5104,
+      "step": 5210
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 1.2285308837890625,
+      "learning_rate": 4.8282828282828285e-06,
+      "loss": 0.499,
+      "step": 5220
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 1.2023276090621948,
+      "learning_rate": 4.818181818181819e-06,
+      "loss": 0.4013,
+      "step": 5230
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 1.4066063165664673,
+      "learning_rate": 4.808080808080808e-06,
+      "loss": 0.4593,
+      "step": 5240
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 1.085758090019226,
+      "learning_rate": 4.7979797979797985e-06,
+      "loss": 0.3793,
+      "step": 5250
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 1.192099928855896,
+      "learning_rate": 4.787878787878788e-06,
+      "loss": 0.4223,
+      "step": 5260
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 1.4430789947509766,
+      "learning_rate": 4.777777777777778e-06,
+      "loss": 0.407,
+      "step": 5270
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 1.2750104665756226,
+      "learning_rate": 4.7676767676767685e-06,
+      "loss": 0.435,
+      "step": 5280
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.9866845607757568,
+      "learning_rate": 4.757575757575758e-06,
+      "loss": 0.4411,
+      "step": 5290
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 1.2444196939468384,
+      "learning_rate": 4.747474747474748e-06,
+      "loss": 0.4571,
+      "step": 5300
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 1.0778636932373047,
+      "learning_rate": 4.737373737373738e-06,
+      "loss": 0.3838,
+      "step": 5310
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 1.4677258729934692,
+      "learning_rate": 4.727272727272728e-06,
+      "loss": 0.4906,
+      "step": 5320
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 1.2698739767074585,
+      "learning_rate": 4.717171717171717e-06,
+      "loss": 0.4368,
+      "step": 5330
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 1.499267578125,
+      "learning_rate": 4.707070707070707e-06,
+      "loss": 0.4075,
+      "step": 5340
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 1.106027364730835,
+      "learning_rate": 4.696969696969698e-06,
+      "loss": 0.482,
+      "step": 5350
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.8394500017166138,
+      "learning_rate": 4.6868686868686874e-06,
+      "loss": 0.4472,
+      "step": 5360
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 1.4100090265274048,
+      "learning_rate": 4.676767676767677e-06,
+      "loss": 0.4038,
+      "step": 5370
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 1.072623610496521,
+      "learning_rate": 4.666666666666667e-06,
+      "loss": 0.4491,
+      "step": 5380
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 1.120101809501648,
+      "learning_rate": 4.656565656565657e-06,
+      "loss": 0.4316,
+      "step": 5390
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 1.4536925554275513,
+      "learning_rate": 4.646464646464647e-06,
+      "loss": 0.425,
+      "step": 5400
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 1.1416407823562622,
+      "learning_rate": 4.636363636363636e-06,
+      "loss": 0.4577,
+      "step": 5410
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.91941237449646,
+      "learning_rate": 4.626262626262627e-06,
+      "loss": 0.4237,
+      "step": 5420
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 1.3551466464996338,
+      "learning_rate": 4.616161616161616e-06,
+      "loss": 0.4854,
+      "step": 5430
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 1.127760648727417,
+      "learning_rate": 4.606060606060606e-06,
+      "loss": 0.4674,
+      "step": 5440
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 1.0127209424972534,
+      "learning_rate": 4.595959595959597e-06,
+      "loss": 0.347,
+      "step": 5450
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 1.1834722757339478,
+      "learning_rate": 4.585858585858586e-06,
+      "loss": 0.4954,
+      "step": 5460
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 1.0718438625335693,
+      "learning_rate": 4.575757575757576e-06,
+      "loss": 0.4073,
+      "step": 5470
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 1.146881341934204,
+      "learning_rate": 4.565656565656566e-06,
+      "loss": 0.4052,
+      "step": 5480
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 1.426000714302063,
+      "learning_rate": 4.555555555555556e-06,
+      "loss": 0.4552,
+      "step": 5490
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 1.4062806367874146,
+      "learning_rate": 4.5454545454545455e-06,
+      "loss": 0.3671,
+      "step": 5500
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 1.276281714439392,
+      "learning_rate": 4.535353535353536e-06,
+      "loss": 0.3792,
+      "step": 5510
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 1.0640497207641602,
+      "learning_rate": 4.525252525252526e-06,
+      "loss": 0.5246,
+      "step": 5520
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 1.0342713594436646,
+      "learning_rate": 4.5151515151515155e-06,
+      "loss": 0.5189,
+      "step": 5530
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 2.512465000152588,
+      "learning_rate": 4.505050505050506e-06,
+      "loss": 0.4245,
+      "step": 5540
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 1.3084385395050049,
+      "learning_rate": 4.494949494949495e-06,
+      "loss": 0.462,
+      "step": 5550
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 1.2058395147323608,
+      "learning_rate": 4.4848484848484855e-06,
+      "loss": 0.466,
+      "step": 5560
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 1.2008490562438965,
+      "learning_rate": 4.474747474747475e-06,
+      "loss": 0.4181,
+      "step": 5570
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.923298716545105,
+      "learning_rate": 4.464646464646465e-06,
+      "loss": 0.408,
+      "step": 5580
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 1.1701871156692505,
+      "learning_rate": 4.454545454545455e-06,
+      "loss": 0.4696,
+      "step": 5590
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 1.1120387315750122,
+      "learning_rate": 4.444444444444444e-06,
+      "loss": 0.432,
+      "step": 5600
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 1.445493221282959,
+      "learning_rate": 4.434343434343435e-06,
+      "loss": 0.4397,
+      "step": 5610
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.978252649307251,
+      "learning_rate": 4.424242424242425e-06,
+      "loss": 0.4448,
+      "step": 5620
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 1.1885161399841309,
+      "learning_rate": 4.414141414141415e-06,
+      "loss": 0.3415,
+      "step": 5630
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 1.4562844038009644,
+      "learning_rate": 4.4040404040404044e-06,
+      "loss": 0.4312,
+      "step": 5640
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 1.5203760862350464,
+      "learning_rate": 4.393939393939394e-06,
+      "loss": 0.4395,
+      "step": 5650
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 1.4762277603149414,
+      "learning_rate": 4.383838383838384e-06,
+      "loss": 0.3605,
+      "step": 5660
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 2.0301284790039062,
+      "learning_rate": 4.373737373737374e-06,
+      "loss": 0.3548,
+      "step": 5670
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 1.2977855205535889,
+      "learning_rate": 4.363636363636364e-06,
+      "loss": 0.4309,
+      "step": 5680
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 1.4677530527114868,
+      "learning_rate": 4.353535353535353e-06,
+      "loss": 0.4431,
+      "step": 5690
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 1.1909211874008179,
+      "learning_rate": 4.343434343434344e-06,
+      "loss": 0.3571,
+      "step": 5700
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 1.2588609457015991,
+      "learning_rate": 4.333333333333334e-06,
+      "loss": 0.4226,
+      "step": 5710
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 1.4346741437911987,
+      "learning_rate": 4.323232323232323e-06,
+      "loss": 0.3029,
+      "step": 5720
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 1.5790741443634033,
+      "learning_rate": 4.313131313131314e-06,
+      "loss": 0.4236,
+      "step": 5730
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.9452656507492065,
+      "learning_rate": 4.303030303030303e-06,
+      "loss": 0.3965,
+      "step": 5740
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 1.3782027959823608,
+      "learning_rate": 4.292929292929293e-06,
+      "loss": 0.3925,
+      "step": 5750
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 1.4818460941314697,
+      "learning_rate": 4.282828282828283e-06,
+      "loss": 0.4207,
+      "step": 5760
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 1.8508193492889404,
+      "learning_rate": 4.272727272727273e-06,
+      "loss": 0.3512,
+      "step": 5770
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.9073811173439026,
+      "learning_rate": 4.262626262626263e-06,
+      "loss": 0.3946,
+      "step": 5780
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 1.285936951637268,
+      "learning_rate": 4.252525252525253e-06,
+      "loss": 0.4387,
+      "step": 5790
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 1.3050516843795776,
+      "learning_rate": 4.242424242424243e-06,
+      "loss": 0.5095,
+      "step": 5800
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 1.601285696029663,
+      "learning_rate": 4.2323232323232325e-06,
+      "loss": 0.3595,
+      "step": 5810
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 1.4098488092422485,
+      "learning_rate": 4.222222222222223e-06,
+      "loss": 0.3726,
+      "step": 5820
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 1.3759149312973022,
+      "learning_rate": 4.212121212121212e-06,
+      "loss": 0.3576,
+      "step": 5830
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 1.1540992259979248,
+      "learning_rate": 4.2020202020202026e-06,
+      "loss": 0.4564,
+      "step": 5840
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 1.2275508642196655,
+      "learning_rate": 4.191919191919192e-06,
+      "loss": 0.4272,
+      "step": 5850
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 2.126412868499756,
+      "learning_rate": 4.181818181818182e-06,
+      "loss": 0.403,
+      "step": 5860
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 1.4939548969268799,
+      "learning_rate": 4.1717171717171726e-06,
+      "loss": 0.4235,
+      "step": 5870
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 1.2192233800888062,
+      "learning_rate": 4.161616161616162e-06,
+      "loss": 0.4342,
+      "step": 5880
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 1.251043438911438,
+      "learning_rate": 4.151515151515152e-06,
+      "loss": 0.3835,
+      "step": 5890
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 1.4634102582931519,
+      "learning_rate": 4.141414141414142e-06,
+      "loss": 0.424,
+      "step": 5900
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 1.5133440494537354,
+      "learning_rate": 4.131313131313132e-06,
+      "loss": 0.4223,
+      "step": 5910
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.8985018730163574,
+      "learning_rate": 4.1212121212121215e-06,
+      "loss": 0.5013,
+      "step": 5920
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 1.4917223453521729,
+      "learning_rate": 4.111111111111111e-06,
+      "loss": 0.4636,
+      "step": 5930
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 1.6001038551330566,
+      "learning_rate": 4.101010101010101e-06,
+      "loss": 0.3814,
+      "step": 5940
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 1.5513330698013306,
+      "learning_rate": 4.0909090909090915e-06,
+      "loss": 0.4455,
+      "step": 5950
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 1.8656607866287231,
+      "learning_rate": 4.080808080808081e-06,
+      "loss": 0.4245,
+      "step": 5960
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 1.4433702230453491,
+      "learning_rate": 4.070707070707071e-06,
+      "loss": 0.4544,
+      "step": 5970
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 1.2587636709213257,
+      "learning_rate": 4.060606060606061e-06,
+      "loss": 0.4406,
+      "step": 5980
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 1.6164227724075317,
+      "learning_rate": 4.050505050505051e-06,
+      "loss": 0.3538,
+      "step": 5990
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 1.4971836805343628,
+      "learning_rate": 4.04040404040404e-06,
+      "loss": 0.4548,
+      "step": 6000
+    },
+    {
+      "epoch": 2.15,
+      "eval_loss": 0.6277722716331482,
+      "eval_runtime": 334.0785,
+      "eval_samples_per_second": 2.993,
+      "eval_steps_per_second": 2.993,
+      "step": 6000
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 1.5129014253616333,
+      "learning_rate": 4.030303030303031e-06,
+      "loss": 0.4389,
+      "step": 6010
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 1.4870412349700928,
+      "learning_rate": 4.02020202020202e-06,
+      "loss": 0.3601,
+      "step": 6020
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 1.3821163177490234,
+      "learning_rate": 4.01010101010101e-06,
+      "loss": 0.3857,
+      "step": 6030
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 1.152169942855835,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 0.3656,
+      "step": 6040
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 1.6697760820388794,
+      "learning_rate": 3.98989898989899e-06,
+      "loss": 0.389,
+      "step": 6050
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 1.7103252410888672,
+      "learning_rate": 3.97979797979798e-06,
+      "loss": 0.3876,
+      "step": 6060
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 1.7297860383987427,
+      "learning_rate": 3.96969696969697e-06,
+      "loss": 0.4479,
+      "step": 6070
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 1.400660753250122,
+      "learning_rate": 3.95959595959596e-06,
+      "loss": 0.3743,
+      "step": 6080
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 1.217431664466858,
+      "learning_rate": 3.9494949494949496e-06,
+      "loss": 0.4568,
+      "step": 6090
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 1.5356212854385376,
+      "learning_rate": 3.93939393939394e-06,
+      "loss": 0.4324,
+      "step": 6100
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 1.227851390838623,
+      "learning_rate": 3.929292929292929e-06,
+      "loss": 0.3789,
+      "step": 6110
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 1.499538779258728,
+      "learning_rate": 3.9191919191919196e-06,
+      "loss": 0.4569,
+      "step": 6120
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 1.25396728515625,
+      "learning_rate": 3.90909090909091e-06,
+      "loss": 0.3727,
+      "step": 6130
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 1.3319553136825562,
+      "learning_rate": 3.898989898989899e-06,
+      "loss": 0.4564,
+      "step": 6140
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 2.0902342796325684,
+      "learning_rate": 3.88888888888889e-06,
+      "loss": 0.4177,
+      "step": 6150
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 1.4777964353561401,
+      "learning_rate": 3.878787878787879e-06,
+      "loss": 0.431,
+      "step": 6160
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 1.1954903602600098,
+      "learning_rate": 3.868686868686869e-06,
+      "loss": 0.411,
+      "step": 6170
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 1.538866400718689,
+      "learning_rate": 3.858585858585859e-06,
+      "loss": 0.3584,
+      "step": 6180
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 1.0525565147399902,
+      "learning_rate": 3.848484848484848e-06,
+      "loss": 0.3909,
+      "step": 6190
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 1.3541011810302734,
+      "learning_rate": 3.8383838383838385e-06,
+      "loss": 0.358,
+      "step": 6200
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 1.3592562675476074,
+      "learning_rate": 3.828282828282829e-06,
+      "loss": 0.3781,
+      "step": 6210
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 1.8746424913406372,
+      "learning_rate": 3.818181818181819e-06,
+      "loss": 0.5072,
+      "step": 6220
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 1.6377208232879639,
+      "learning_rate": 3.8080808080808085e-06,
+      "loss": 0.5005,
+      "step": 6230
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 1.8624753952026367,
+      "learning_rate": 3.7979797979797984e-06,
+      "loss": 0.3831,
+      "step": 6240
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 1.2943841218948364,
+      "learning_rate": 3.7878787878787882e-06,
+      "loss": 0.4214,
+      "step": 6250
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 1.8572790622711182,
+      "learning_rate": 3.777777777777778e-06,
+      "loss": 0.3869,
+      "step": 6260
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 1.3550498485565186,
+      "learning_rate": 3.767676767676768e-06,
+      "loss": 0.4073,
+      "step": 6270
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 1.6371854543685913,
+      "learning_rate": 3.757575757575758e-06,
+      "loss": 0.4122,
+      "step": 6280
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 1.6021220684051514,
+      "learning_rate": 3.747474747474748e-06,
+      "loss": 0.4068,
+      "step": 6290
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 1.9373170137405396,
+      "learning_rate": 3.737373737373738e-06,
+      "loss": 0.4103,
+      "step": 6300
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 1.2740237712860107,
+      "learning_rate": 3.727272727272728e-06,
+      "loss": 0.4246,
+      "step": 6310
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 1.4819579124450684,
+      "learning_rate": 3.7171717171717177e-06,
+      "loss": 0.4298,
+      "step": 6320
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 1.417238712310791,
+      "learning_rate": 3.7070707070707075e-06,
+      "loss": 0.3961,
+      "step": 6330
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 1.6524369716644287,
+      "learning_rate": 3.6969696969696974e-06,
+      "loss": 0.3793,
+      "step": 6340
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 1.2900643348693848,
+      "learning_rate": 3.686868686868687e-06,
+      "loss": 0.3608,
+      "step": 6350
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 1.5816043615341187,
+      "learning_rate": 3.6767676767676767e-06,
+      "loss": 0.3703,
+      "step": 6360
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 1.6086316108703613,
+      "learning_rate": 3.6666666666666666e-06,
+      "loss": 0.3597,
+      "step": 6370
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 1.3897535800933838,
+      "learning_rate": 3.6565656565656573e-06,
+      "loss": 0.4305,
+      "step": 6380
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 2.2776060104370117,
+      "learning_rate": 3.6464646464646467e-06,
+      "loss": 0.3946,
+      "step": 6390
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 1.1567403078079224,
+      "learning_rate": 3.6363636363636366e-06,
+      "loss": 0.3988,
+      "step": 6400
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 1.0619862079620361,
+      "learning_rate": 3.6262626262626264e-06,
+      "loss": 0.3561,
+      "step": 6410
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 2.857037305831909,
+      "learning_rate": 3.6161616161616163e-06,
+      "loss": 0.4394,
+      "step": 6420
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 1.4043927192687988,
+      "learning_rate": 3.606060606060606e-06,
+      "loss": 0.3804,
+      "step": 6430
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 1.1000697612762451,
+      "learning_rate": 3.595959595959596e-06,
+      "loss": 0.3943,
+      "step": 6440
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 1.880038857460022,
+      "learning_rate": 3.585858585858586e-06,
+      "loss": 0.425,
+      "step": 6450
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 1.408372163772583,
+      "learning_rate": 3.575757575757576e-06,
+      "loss": 0.5019,
+      "step": 6460
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 1.4338643550872803,
+      "learning_rate": 3.565656565656566e-06,
+      "loss": 0.4515,
+      "step": 6470
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 2.0675547122955322,
+      "learning_rate": 3.555555555555556e-06,
+      "loss": 0.4414,
+      "step": 6480
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 1.9026405811309814,
+      "learning_rate": 3.5454545454545458e-06,
+      "loss": 0.4789,
+      "step": 6490
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 1.708087682723999,
+      "learning_rate": 3.5353535353535356e-06,
+      "loss": 0.3745,
+      "step": 6500
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 1.4911117553710938,
+      "learning_rate": 3.5252525252525255e-06,
+      "loss": 0.4387,
+      "step": 6510
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 2.0985512733459473,
+      "learning_rate": 3.5151515151515154e-06,
+      "loss": 0.4337,
+      "step": 6520
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 1.6386370658874512,
+      "learning_rate": 3.5050505050505052e-06,
+      "loss": 0.3987,
+      "step": 6530
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 1.5496248006820679,
+      "learning_rate": 3.494949494949495e-06,
+      "loss": 0.4323,
+      "step": 6540
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 1.8171871900558472,
+      "learning_rate": 3.4848484848484854e-06,
+      "loss": 0.4434,
+      "step": 6550
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 1.2652944326400757,
+      "learning_rate": 3.4747474747474752e-06,
+      "loss": 0.3703,
+      "step": 6560
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 1.0056226253509521,
+      "learning_rate": 3.464646464646465e-06,
+      "loss": 0.3524,
+      "step": 6570
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 1.5017718076705933,
+      "learning_rate": 3.454545454545455e-06,
+      "loss": 0.3776,
+      "step": 6580
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 1.4463821649551392,
+      "learning_rate": 3.444444444444445e-06,
+      "loss": 0.4926,
+      "step": 6590
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 1.4280391931533813,
+      "learning_rate": 3.4343434343434347e-06,
+      "loss": 0.4056,
+      "step": 6600
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 1.3990509510040283,
+      "learning_rate": 3.4242424242424246e-06,
+      "loss": 0.3657,
+      "step": 6610
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 2.033510446548462,
+      "learning_rate": 3.414141414141414e-06,
+      "loss": 0.3977,
+      "step": 6620
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 1.3510665893554688,
+      "learning_rate": 3.4040404040404047e-06,
+      "loss": 0.4227,
+      "step": 6630
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 1.2315497398376465,
+      "learning_rate": 3.3939393939393946e-06,
+      "loss": 0.395,
+      "step": 6640
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 1.7533644437789917,
+      "learning_rate": 3.3838383838383844e-06,
+      "loss": 0.4927,
+      "step": 6650
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 1.7972692251205444,
+      "learning_rate": 3.3737373737373743e-06,
+      "loss": 0.3826,
+      "step": 6660
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 1.2529045343399048,
+      "learning_rate": 3.3636363636363637e-06,
+      "loss": 0.3399,
+      "step": 6670
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 1.9954489469528198,
+      "learning_rate": 3.3535353535353536e-06,
+      "loss": 0.3751,
+      "step": 6680
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 1.2907588481903076,
+      "learning_rate": 3.3434343434343435e-06,
+      "loss": 0.3789,
+      "step": 6690
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 1.9876916408538818,
+      "learning_rate": 3.3333333333333333e-06,
+      "loss": 0.41,
+      "step": 6700
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 2.368973970413208,
+      "learning_rate": 3.323232323232323e-06,
+      "loss": 0.4146,
+      "step": 6710
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 1.4498718976974487,
+      "learning_rate": 3.3131313131313135e-06,
+      "loss": 0.4255,
+      "step": 6720
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 1.1825134754180908,
+      "learning_rate": 3.3030303030303033e-06,
+      "loss": 0.3758,
+      "step": 6730
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 1.837396502494812,
+      "learning_rate": 3.292929292929293e-06,
+      "loss": 0.4129,
+      "step": 6740
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 1.2423911094665527,
+      "learning_rate": 3.282828282828283e-06,
+      "loss": 0.4081,
+      "step": 6750
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 2.317641258239746,
+      "learning_rate": 3.272727272727273e-06,
+      "loss": 0.4525,
+      "step": 6760
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 1.5409480333328247,
+      "learning_rate": 3.262626262626263e-06,
+      "loss": 0.3919,
+      "step": 6770
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 1.5040165185928345,
+      "learning_rate": 3.2525252525252527e-06,
+      "loss": 0.3644,
+      "step": 6780
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 1.073905110359192,
+      "learning_rate": 3.2424242424242425e-06,
+      "loss": 0.4178,
+      "step": 6790
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 1.8809980154037476,
+      "learning_rate": 3.232323232323233e-06,
+      "loss": 0.4324,
+      "step": 6800
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 1.6550686359405518,
+      "learning_rate": 3.2222222222222227e-06,
+      "loss": 0.41,
+      "step": 6810
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 1.2928190231323242,
+      "learning_rate": 3.2121212121212125e-06,
+      "loss": 0.4178,
+      "step": 6820
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 2.924306869506836,
+      "learning_rate": 3.2020202020202024e-06,
+      "loss": 0.4388,
+      "step": 6830
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 1.3699476718902588,
+      "learning_rate": 3.1919191919191923e-06,
+      "loss": 0.3703,
+      "step": 6840
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 1.1735562086105347,
+      "learning_rate": 3.181818181818182e-06,
+      "loss": 0.4546,
+      "step": 6850
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 2.5108041763305664,
+      "learning_rate": 3.171717171717172e-06,
+      "loss": 0.4331,
+      "step": 6860
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 1.8626171350479126,
+      "learning_rate": 3.161616161616162e-06,
+      "loss": 0.4169,
+      "step": 6870
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 1.476062297821045,
+      "learning_rate": 3.1515151515151517e-06,
+      "loss": 0.4041,
+      "step": 6880
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 1.1613414287567139,
+      "learning_rate": 3.141414141414142e-06,
+      "loss": 0.3558,
+      "step": 6890
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 1.7106380462646484,
+      "learning_rate": 3.131313131313132e-06,
+      "loss": 0.4175,
+      "step": 6900
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 1.3592215776443481,
+      "learning_rate": 3.1212121212121217e-06,
+      "loss": 0.3388,
+      "step": 6910
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 1.7043094635009766,
+      "learning_rate": 3.1111111111111116e-06,
+      "loss": 0.4189,
+      "step": 6920
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 1.5782170295715332,
+      "learning_rate": 3.1010101010101014e-06,
+      "loss": 0.434,
+      "step": 6930
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 1.4425175189971924,
+      "learning_rate": 3.090909090909091e-06,
+      "loss": 0.4013,
+      "step": 6940
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 1.5864651203155518,
+      "learning_rate": 3.0808080808080807e-06,
+      "loss": 0.4702,
+      "step": 6950
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 1.678449034690857,
+      "learning_rate": 3.0707070707070706e-06,
+      "loss": 0.4722,
+      "step": 6960
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 1.8242285251617432,
+      "learning_rate": 3.0606060606060605e-06,
+      "loss": 0.4264,
+      "step": 6970
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 2.037673234939575,
+      "learning_rate": 3.0505050505050508e-06,
+      "loss": 0.4268,
+      "step": 6980
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 1.530069351196289,
+      "learning_rate": 3.0404040404040406e-06,
+      "loss": 0.4355,
+      "step": 6990
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 2.4778358936309814,
+      "learning_rate": 3.0303030303030305e-06,
+      "loss": 0.4214,
+      "step": 7000
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 2.0160224437713623,
+      "learning_rate": 3.0202020202020203e-06,
+      "loss": 0.3948,
+      "step": 7010
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 1.18585205078125,
+      "learning_rate": 3.0101010101010102e-06,
+      "loss": 0.4199,
+      "step": 7020
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 1.516297698020935,
+      "learning_rate": 3e-06,
+      "loss": 0.4519,
+      "step": 7030
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 1.7270208597183228,
+      "learning_rate": 2.98989898989899e-06,
+      "loss": 0.4673,
+      "step": 7040
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 1.7389216423034668,
+      "learning_rate": 2.97979797979798e-06,
+      "loss": 0.4164,
+      "step": 7050
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 1.3944525718688965,
+      "learning_rate": 2.96969696969697e-06,
+      "loss": 0.3907,
+      "step": 7060
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 1.219854712486267,
+      "learning_rate": 2.95959595959596e-06,
+      "loss": 0.3551,
+      "step": 7070
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 2.0792431831359863,
+      "learning_rate": 2.94949494949495e-06,
+      "loss": 0.4021,
+      "step": 7080
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 1.2674493789672852,
+      "learning_rate": 2.9393939393939397e-06,
+      "loss": 0.346,
+      "step": 7090
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 1.7033464908599854,
+      "learning_rate": 2.9292929292929295e-06,
+      "loss": 0.422,
+      "step": 7100
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 3.0683019161224365,
+      "learning_rate": 2.9191919191919194e-06,
+      "loss": 0.3551,
+      "step": 7110
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 1.7624000310897827,
+      "learning_rate": 2.9090909090909093e-06,
+      "loss": 0.4915,
+      "step": 7120
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 1.8647630214691162,
+      "learning_rate": 2.898989898989899e-06,
+      "loss": 0.3403,
+      "step": 7130
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 1.8980121612548828,
+      "learning_rate": 2.888888888888889e-06,
+      "loss": 0.4502,
+      "step": 7140
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 1.0902882814407349,
+      "learning_rate": 2.8787878787878793e-06,
+      "loss": 0.3302,
+      "step": 7150
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 1.4546654224395752,
+      "learning_rate": 2.868686868686869e-06,
+      "loss": 0.3765,
+      "step": 7160
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 1.2355202436447144,
+      "learning_rate": 2.858585858585859e-06,
+      "loss": 0.3574,
+      "step": 7170
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 1.8936539888381958,
+      "learning_rate": 2.848484848484849e-06,
+      "loss": 0.4014,
+      "step": 7180
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 1.3609634637832642,
+      "learning_rate": 2.8383838383838387e-06,
+      "loss": 0.4422,
+      "step": 7190
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 1.4429062604904175,
+      "learning_rate": 2.8282828282828286e-06,
+      "loss": 0.3675,
+      "step": 7200
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 1.6361467838287354,
+      "learning_rate": 2.818181818181818e-06,
+      "loss": 0.3673,
+      "step": 7210
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 1.495545744895935,
+      "learning_rate": 2.808080808080808e-06,
+      "loss": 0.3771,
+      "step": 7220
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 1.4525377750396729,
+      "learning_rate": 2.7979797979797986e-06,
+      "loss": 0.4315,
+      "step": 7230
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 1.5371636152267456,
+      "learning_rate": 2.7878787878787885e-06,
+      "loss": 0.3852,
+      "step": 7240
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 1.2793852090835571,
+      "learning_rate": 2.7777777777777783e-06,
+      "loss": 0.4267,
+      "step": 7250
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 1.440574288368225,
+      "learning_rate": 2.7676767676767678e-06,
+      "loss": 0.4398,
+      "step": 7260
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 1.8655972480773926,
+      "learning_rate": 2.7575757575757576e-06,
+      "loss": 0.3594,
+      "step": 7270
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 1.0927013158798218,
+      "learning_rate": 2.7474747474747475e-06,
+      "loss": 0.3555,
+      "step": 7280
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 2.1925394535064697,
+      "learning_rate": 2.7373737373737374e-06,
+      "loss": 0.3849,
+      "step": 7290
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 2.2034037113189697,
+      "learning_rate": 2.7272727272727272e-06,
+      "loss": 0.4148,
+      "step": 7300
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 1.269071102142334,
+      "learning_rate": 2.717171717171717e-06,
+      "loss": 0.3587,
+      "step": 7310
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 1.5352184772491455,
+      "learning_rate": 2.7070707070707074e-06,
+      "loss": 0.3811,
+      "step": 7320
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 2.0000691413879395,
+      "learning_rate": 2.6969696969696972e-06,
+      "loss": 0.4263,
+      "step": 7330
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 1.831752896308899,
+      "learning_rate": 2.686868686868687e-06,
+      "loss": 0.3788,
+      "step": 7340
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 1.6060395240783691,
+      "learning_rate": 2.676767676767677e-06,
+      "loss": 0.4195,
+      "step": 7350
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 1.7236618995666504,
+      "learning_rate": 2.666666666666667e-06,
+      "loss": 0.3733,
+      "step": 7360
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 1.2738254070281982,
+      "learning_rate": 2.6565656565656567e-06,
+      "loss": 0.3988,
+      "step": 7370
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 1.5250334739685059,
+      "learning_rate": 2.6464646464646466e-06,
+      "loss": 0.4551,
+      "step": 7380
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 2.97196364402771,
+      "learning_rate": 2.6363636363636364e-06,
+      "loss": 0.4457,
+      "step": 7390
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 1.76936674118042,
+      "learning_rate": 2.6262626262626267e-06,
+      "loss": 0.364,
+      "step": 7400
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 1.852996587753296,
+      "learning_rate": 2.6161616161616166e-06,
+      "loss": 0.3847,
+      "step": 7410
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 1.663035273551941,
+      "learning_rate": 2.6060606060606064e-06,
+      "loss": 0.3657,
+      "step": 7420
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 1.9585185050964355,
+      "learning_rate": 2.5959595959595963e-06,
+      "loss": 0.4104,
+      "step": 7430
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 1.87462317943573,
+      "learning_rate": 2.585858585858586e-06,
+      "loss": 0.3836,
+      "step": 7440
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 1.6113030910491943,
+      "learning_rate": 2.575757575757576e-06,
+      "loss": 0.508,
+      "step": 7450
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 1.8767685890197754,
+      "learning_rate": 2.565656565656566e-06,
+      "loss": 0.4736,
+      "step": 7460
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 1.8729183673858643,
+      "learning_rate": 2.5555555555555557e-06,
+      "loss": 0.4945,
+      "step": 7470
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 1.2904634475708008,
+      "learning_rate": 2.5454545454545456e-06,
+      "loss": 0.3822,
+      "step": 7480
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 1.2536317110061646,
+      "learning_rate": 2.535353535353536e-06,
+      "loss": 0.4009,
+      "step": 7490
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 1.9122370481491089,
+      "learning_rate": 2.5252525252525258e-06,
+      "loss": 0.48,
+      "step": 7500
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 1.3834714889526367,
+      "learning_rate": 2.5151515151515156e-06,
+      "loss": 0.448,
+      "step": 7510
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 1.2706272602081299,
+      "learning_rate": 2.5050505050505055e-06,
+      "loss": 0.3774,
+      "step": 7520
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 1.407159686088562,
+      "learning_rate": 2.494949494949495e-06,
+      "loss": 0.4177,
+      "step": 7530
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 1.6914201974868774,
+      "learning_rate": 2.4848484848484848e-06,
+      "loss": 0.4274,
+      "step": 7540
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 1.6943836212158203,
+      "learning_rate": 2.474747474747475e-06,
+      "loss": 0.3993,
+      "step": 7550
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 2.1789302825927734,
+      "learning_rate": 2.464646464646465e-06,
+      "loss": 0.4064,
+      "step": 7560
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 1.6597346067428589,
+      "learning_rate": 2.454545454545455e-06,
+      "loss": 0.3905,
+      "step": 7570
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 1.2498160600662231,
+      "learning_rate": 2.4444444444444447e-06,
+      "loss": 0.4245,
+      "step": 7580
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 2.0250113010406494,
+      "learning_rate": 2.4343434343434345e-06,
+      "loss": 0.3716,
+      "step": 7590
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 1.5236455202102661,
+      "learning_rate": 2.4242424242424244e-06,
+      "loss": 0.4595,
+      "step": 7600
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 1.778412103652954,
+      "learning_rate": 2.4141414141414143e-06,
+      "loss": 0.385,
+      "step": 7610
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 1.3730769157409668,
+      "learning_rate": 2.404040404040404e-06,
+      "loss": 0.3449,
+      "step": 7620
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 1.288809061050415,
+      "learning_rate": 2.393939393939394e-06,
+      "loss": 0.4099,
+      "step": 7630
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 1.6752142906188965,
+      "learning_rate": 2.3838383838383843e-06,
+      "loss": 0.4574,
+      "step": 7640
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 1.304540753364563,
+      "learning_rate": 2.373737373737374e-06,
+      "loss": 0.4122,
+      "step": 7650
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 1.5244289636611938,
+      "learning_rate": 2.363636363636364e-06,
+      "loss": 0.4547,
+      "step": 7660
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 1.4657320976257324,
+      "learning_rate": 2.3535353535353534e-06,
+      "loss": 0.3845,
+      "step": 7670
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 1.4184165000915527,
+      "learning_rate": 2.3434343434343437e-06,
+      "loss": 0.3815,
+      "step": 7680
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 1.4459203481674194,
+      "learning_rate": 2.3333333333333336e-06,
+      "loss": 0.3597,
+      "step": 7690
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 1.182944893836975,
+      "learning_rate": 2.3232323232323234e-06,
+      "loss": 0.3873,
+      "step": 7700
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 1.7956231832504272,
+      "learning_rate": 2.3131313131313133e-06,
+      "loss": 0.4001,
+      "step": 7710
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 1.7619251012802124,
+      "learning_rate": 2.303030303030303e-06,
+      "loss": 0.4178,
+      "step": 7720
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 1.4194896221160889,
+      "learning_rate": 2.292929292929293e-06,
+      "loss": 0.3489,
+      "step": 7730
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 1.4569259881973267,
+      "learning_rate": 2.282828282828283e-06,
+      "loss": 0.4014,
+      "step": 7740
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 1.6338679790496826,
+      "learning_rate": 2.2727272727272728e-06,
+      "loss": 0.3425,
+      "step": 7750
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 1.6008449792861938,
+      "learning_rate": 2.262626262626263e-06,
+      "loss": 0.4129,
+      "step": 7760
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 2.567551851272583,
+      "learning_rate": 2.252525252525253e-06,
+      "loss": 0.4304,
+      "step": 7770
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 1.5594419240951538,
+      "learning_rate": 2.2424242424242428e-06,
+      "loss": 0.4809,
+      "step": 7780
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 2.100327253341675,
+      "learning_rate": 2.2323232323232326e-06,
+      "loss": 0.3663,
+      "step": 7790
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 1.9268852472305298,
+      "learning_rate": 2.222222222222222e-06,
+      "loss": 0.485,
+      "step": 7800
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 1.378103494644165,
+      "learning_rate": 2.2121212121212124e-06,
+      "loss": 0.3487,
+      "step": 7810
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 1.3824481964111328,
+      "learning_rate": 2.2020202020202022e-06,
+      "loss": 0.4487,
+      "step": 7820
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 1.4353455305099487,
+      "learning_rate": 2.191919191919192e-06,
+      "loss": 0.3761,
+      "step": 7830
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 1.4255388975143433,
+      "learning_rate": 2.181818181818182e-06,
+      "loss": 0.4219,
+      "step": 7840
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 1.2923437356948853,
+      "learning_rate": 2.171717171717172e-06,
+      "loss": 0.4413,
+      "step": 7850
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 1.5469424724578857,
+      "learning_rate": 2.1616161616161617e-06,
+      "loss": 0.3907,
+      "step": 7860
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 1.5933393239974976,
+      "learning_rate": 2.1515151515151515e-06,
+      "loss": 0.3958,
+      "step": 7870
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 1.3640625476837158,
+      "learning_rate": 2.1414141414141414e-06,
+      "loss": 0.3836,
+      "step": 7880
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 1.3010022640228271,
+      "learning_rate": 2.1313131313131317e-06,
+      "loss": 0.4202,
+      "step": 7890
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 2.231398344039917,
+      "learning_rate": 2.1212121212121216e-06,
+      "loss": 0.4527,
+      "step": 7900
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 1.9403585195541382,
+      "learning_rate": 2.1111111111111114e-06,
+      "loss": 0.3446,
+      "step": 7910
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 3.341773748397827,
+      "learning_rate": 2.1010101010101013e-06,
+      "loss": 0.3637,
+      "step": 7920
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 1.5845195055007935,
+      "learning_rate": 2.090909090909091e-06,
+      "loss": 0.379,
+      "step": 7930
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 1.9772624969482422,
+      "learning_rate": 2.080808080808081e-06,
+      "loss": 0.3954,
+      "step": 7940
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 1.5883660316467285,
+      "learning_rate": 2.070707070707071e-06,
+      "loss": 0.4589,
+      "step": 7950
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 1.7022806406021118,
+      "learning_rate": 2.0606060606060607e-06,
+      "loss": 0.3767,
+      "step": 7960
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 1.246584177017212,
+      "learning_rate": 2.0505050505050506e-06,
+      "loss": 0.4161,
+      "step": 7970
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 2.11494517326355,
+      "learning_rate": 2.0404040404040405e-06,
+      "loss": 0.3635,
+      "step": 7980
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 2.0264945030212402,
+      "learning_rate": 2.0303030303030303e-06,
+      "loss": 0.3698,
+      "step": 7990
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 1.9246190786361694,
+      "learning_rate": 2.02020202020202e-06,
+      "loss": 0.3972,
+      "step": 8000
+    },
+    {
+      "epoch": 2.87,
+      "eval_loss": 0.6264104843139648,
+      "eval_runtime": 335.1608,
+      "eval_samples_per_second": 2.984,
+      "eval_steps_per_second": 2.984,
+      "step": 8000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 2000,
+  "total_flos": 6.52822009872384e+17,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}