diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4245 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.3,
+  "eval_steps": 2000,
+  "global_step": 6000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 3.9070186614990234,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 1.0857,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 1.667642593383789,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 0.6733,
+      "step": 20
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 3.0051660537719727,
+      "learning_rate": 3e-06,
+      "loss": 0.6534,
+      "step": 30
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 1.0736082792282104,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 0.5221,
+      "step": 40
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 4.684818744659424,
+      "learning_rate": 5e-06,
+      "loss": 0.6807,
+      "step": 50
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 2.5355663299560547,
+      "learning_rate": 6e-06,
+      "loss": 0.4381,
+      "step": 60
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 2.204728126525879,
+      "learning_rate": 7e-06,
+      "loss": 0.4578,
+      "step": 70
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 1.8668341636657715,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 0.362,
+      "step": 80
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 1.7530344724655151,
+      "learning_rate": 9e-06,
+      "loss": 0.6272,
+      "step": 90
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.6746323108673096,
+      "learning_rate": 1e-05,
+      "loss": 0.5589,
+      "step": 100
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.4799275398254395,
+      "learning_rate": 9.989898989898991e-06,
+      "loss": 0.6067,
+      "step": 110
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.678624153137207,
+      "learning_rate": 9.97979797979798e-06,
+      "loss": 0.5582,
+      "step": 120
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.9827778339385986,
+      "learning_rate": 9.96969696969697e-06,
+      "loss": 0.4943,
+      "step": 130
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.8628324270248413,
+      "learning_rate": 9.95959595959596e-06,
+      "loss": 0.6706,
+      "step": 140
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.0590109825134277,
+      "learning_rate": 9.94949494949495e-06,
+      "loss": 0.4861,
+      "step": 150
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.6400913596153259,
+      "learning_rate": 9.939393939393939e-06,
+      "loss": 0.6373,
+      "step": 160
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.803528070449829,
+      "learning_rate": 9.92929292929293e-06,
+      "loss": 0.6073,
+      "step": 170
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.7043955326080322,
+      "learning_rate": 9.91919191919192e-06,
+      "loss": 0.6351,
+      "step": 180
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.0071933269500732,
+      "learning_rate": 9.90909090909091e-06,
+      "loss": 0.6624,
+      "step": 190
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.8767812252044678,
+      "learning_rate": 9.8989898989899e-06,
+      "loss": 0.5892,
+      "step": 200
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.037977695465088,
+      "learning_rate": 9.88888888888889e-06,
+      "loss": 0.5099,
+      "step": 210
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.31269109249115,
+      "learning_rate": 9.87878787878788e-06,
+      "loss": 0.5048,
+      "step": 220
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.604134559631348,
+      "learning_rate": 9.86868686868687e-06,
+      "loss": 0.7424,
+      "step": 230
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.378148078918457,
+      "learning_rate": 9.85858585858586e-06,
+      "loss": 0.6627,
+      "step": 240
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.8570343852043152,
+      "learning_rate": 9.84848484848485e-06,
+      "loss": 0.5931,
+      "step": 250
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.617653489112854,
+      "learning_rate": 9.838383838383839e-06,
+      "loss": 0.4749,
+      "step": 260
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.7800276279449463,
+      "learning_rate": 9.828282828282829e-06,
+      "loss": 0.8069,
+      "step": 270
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.901750564575195,
+      "learning_rate": 9.81818181818182e-06,
+      "loss": 0.4457,
+      "step": 280
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.434395790100098,
+      "learning_rate": 9.80808080808081e-06,
+      "loss": 0.7167,
+      "step": 290
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.92600154876709,
+      "learning_rate": 9.797979797979798e-06,
+      "loss": 0.3975,
+      "step": 300
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.07546558231115341,
+      "learning_rate": 9.787878787878788e-06,
+      "loss": 0.3655,
+      "step": 310
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.365915775299072,
+      "learning_rate": 9.777777777777779e-06,
+      "loss": 0.5224,
+      "step": 320
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.0982255935668945,
+      "learning_rate": 9.767676767676767e-06,
+      "loss": 0.6165,
+      "step": 330
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.7056803703308105,
+      "learning_rate": 9.757575757575758e-06,
+      "loss": 0.5539,
+      "step": 340
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.2261226177215576,
+      "learning_rate": 9.747474747474748e-06,
+      "loss": 0.4769,
+      "step": 350
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.9531151056289673,
+      "learning_rate": 9.737373737373738e-06,
+      "loss": 0.3609,
+      "step": 360
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.2107353210449219,
+      "learning_rate": 9.727272727272728e-06,
+      "loss": 0.5399,
+      "step": 370
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.582059860229492,
+      "learning_rate": 9.717171717171719e-06,
+      "loss": 0.6101,
+      "step": 380
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.4467086791992188,
+      "learning_rate": 9.707070707070709e-06,
+      "loss": 0.5436,
+      "step": 390
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.19772769510746002,
+      "learning_rate": 9.696969696969698e-06,
+      "loss": 0.6105,
+      "step": 400
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.347663164138794,
+      "learning_rate": 9.686868686868688e-06,
+      "loss": 0.4353,
+      "step": 410
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 5.073924541473389,
+      "learning_rate": 9.676767676767678e-06,
+      "loss": 0.6866,
+      "step": 420
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.099503517150879,
+      "learning_rate": 9.666666666666667e-06,
+      "loss": 0.5744,
+      "step": 430
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.2135908603668213,
+      "learning_rate": 9.656565656565657e-06,
+      "loss": 0.5677,
+      "step": 440
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.5008598566055298,
+      "learning_rate": 9.646464646464647e-06,
+      "loss": 0.3873,
+      "step": 450
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.6704565286636353,
+      "learning_rate": 9.636363636363638e-06,
+      "loss": 0.6814,
+      "step": 460
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.4024767875671387,
+      "learning_rate": 9.626262626262626e-06,
+      "loss": 0.5843,
+      "step": 470
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.00467598857358098,
+      "learning_rate": 9.616161616161616e-06,
+      "loss": 0.4113,
+      "step": 480
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.787539958953857,
+      "learning_rate": 9.606060606060607e-06,
+      "loss": 0.7154,
+      "step": 490
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.0733729600906372,
+      "learning_rate": 9.595959595959597e-06,
+      "loss": 0.6492,
+      "step": 500
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.1921634674072266,
+      "learning_rate": 9.585858585858586e-06,
+      "loss": 0.5533,
+      "step": 510
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.963071584701538,
+      "learning_rate": 9.575757575757576e-06,
+      "loss": 0.6422,
+      "step": 520
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.266462326049805,
+      "learning_rate": 9.565656565656566e-06,
+      "loss": 0.4266,
+      "step": 530
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.4488701820373535,
+      "learning_rate": 9.555555555555556e-06,
+      "loss": 0.4274,
+      "step": 540
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.9400691986083984,
+      "learning_rate": 9.545454545454547e-06,
+      "loss": 0.5337,
+      "step": 550
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.224618434906006,
+      "learning_rate": 9.535353535353537e-06,
+      "loss": 0.5675,
+      "step": 560
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.3731207847595215,
+      "learning_rate": 9.525252525252526e-06,
+      "loss": 0.362,
+      "step": 570
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.9596872329711914,
+      "learning_rate": 9.515151515151516e-06,
+      "loss": 0.5264,
+      "step": 580
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.6627348065376282,
+      "learning_rate": 9.505050505050506e-06,
+      "loss": 0.435,
+      "step": 590
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 7.430358409881592,
+      "learning_rate": 9.494949494949497e-06,
+      "loss": 0.5424,
+      "step": 600
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.306655406951904,
+      "learning_rate": 9.484848484848485e-06,
+      "loss": 0.4153,
+      "step": 610
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.6357769966125488,
+      "learning_rate": 9.474747474747475e-06,
+      "loss": 0.5251,
+      "step": 620
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 7.038285732269287,
+      "learning_rate": 9.464646464646466e-06,
+      "loss": 0.5593,
+      "step": 630
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.054030895233154,
+      "learning_rate": 9.454545454545456e-06,
+      "loss": 0.582,
+      "step": 640
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.8453586101531982,
+      "learning_rate": 9.444444444444445e-06,
+      "loss": 0.5627,
+      "step": 650
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.760792851448059,
+      "learning_rate": 9.434343434343435e-06,
+      "loss": 0.4531,
+      "step": 660
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.7284159660339355,
+      "learning_rate": 9.424242424242425e-06,
+      "loss": 0.4397,
+      "step": 670
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.4066224694252014,
+      "learning_rate": 9.414141414141414e-06,
+      "loss": 0.4056,
+      "step": 680
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.083108425140381,
+      "learning_rate": 9.404040404040404e-06,
+      "loss": 0.6092,
+      "step": 690
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.98835563659668,
+      "learning_rate": 9.393939393939396e-06,
+      "loss": 0.4545,
+      "step": 700
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.849491119384766,
+      "learning_rate": 9.383838383838385e-06,
+      "loss": 0.4038,
+      "step": 710
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.9796783924102783,
+      "learning_rate": 9.373737373737375e-06,
+      "loss": 0.6097,
+      "step": 720
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.3015940189361572,
+      "learning_rate": 9.363636363636365e-06,
+      "loss": 0.4561,
+      "step": 730
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.0265772342681885,
+      "learning_rate": 9.353535353535354e-06,
+      "loss": 0.5972,
+      "step": 740
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.158243656158447,
+      "learning_rate": 9.343434343434344e-06,
+      "loss": 0.4375,
+      "step": 750
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.6322413682937622,
+      "learning_rate": 9.333333333333334e-06,
+      "loss": 0.4394,
+      "step": 760
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.8402198553085327,
+      "learning_rate": 9.323232323232325e-06,
+      "loss": 0.4942,
+      "step": 770
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.9348700046539307,
+      "learning_rate": 9.313131313131313e-06,
+      "loss": 0.6713,
+      "step": 780
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.633822917938232,
+      "learning_rate": 9.303030303030303e-06,
+      "loss": 0.5727,
+      "step": 790
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.5603723526000977,
+      "learning_rate": 9.292929292929294e-06,
+      "loss": 0.6393,
+      "step": 800
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.2063639163970947,
+      "learning_rate": 9.282828282828284e-06,
+      "loss": 0.6278,
+      "step": 810
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.9794161319732666,
+      "learning_rate": 9.272727272727273e-06,
+      "loss": 0.5332,
+      "step": 820
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.441140055656433,
+      "learning_rate": 9.262626262626263e-06,
+      "loss": 0.5951,
+      "step": 830
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.256943941116333,
+      "learning_rate": 9.252525252525253e-06,
+      "loss": 0.3266,
+      "step": 840
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.15118904411792755,
+      "learning_rate": 9.242424242424244e-06,
+      "loss": 0.5511,
+      "step": 850
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.8574471473693848,
+      "learning_rate": 9.232323232323232e-06,
+      "loss": 0.8058,
+      "step": 860
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.2425239086151123,
+      "learning_rate": 9.222222222222224e-06,
+      "loss": 0.5866,
+      "step": 870
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.476180911064148,
+      "learning_rate": 9.212121212121213e-06,
+      "loss": 0.5133,
+      "step": 880
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.133812665939331,
+      "learning_rate": 9.202020202020203e-06,
+      "loss": 0.3262,
+      "step": 890
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.0041747093200684,
+      "learning_rate": 9.191919191919193e-06,
+      "loss": 0.6861,
+      "step": 900
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.9870094060897827,
+      "learning_rate": 9.181818181818184e-06,
+      "loss": 0.4655,
+      "step": 910
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.3034363985061646,
+      "learning_rate": 9.171717171717172e-06,
+      "loss": 0.6274,
+      "step": 920
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.1342576742172241,
+      "learning_rate": 9.161616161616162e-06,
+      "loss": 0.404,
+      "step": 930
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.735161781311035,
+      "learning_rate": 9.151515151515153e-06,
+      "loss": 0.4047,
+      "step": 940
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.926325798034668,
+      "learning_rate": 9.141414141414143e-06,
+      "loss": 0.3813,
+      "step": 950
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.1905412673950195,
+      "learning_rate": 9.131313131313132e-06,
+      "loss": 0.7336,
+      "step": 960
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.6212941408157349,
+      "learning_rate": 9.121212121212122e-06,
+      "loss": 0.6218,
+      "step": 970
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.1676979064941406,
+      "learning_rate": 9.111111111111112e-06,
+      "loss": 0.4894,
+      "step": 980
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.414123773574829,
+      "learning_rate": 9.1010101010101e-06,
+      "loss": 0.505,
+      "step": 990
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.8107662200927734,
+      "learning_rate": 9.090909090909091e-06,
+      "loss": 0.3938,
+      "step": 1000
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.8321704864501953,
+      "learning_rate": 9.080808080808081e-06,
+      "loss": 0.504,
+      "step": 1010
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.217884540557861,
+      "learning_rate": 9.070707070707072e-06,
+      "loss": 0.5721,
+      "step": 1020
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.3424525260925293,
+      "learning_rate": 9.06060606060606e-06,
+      "loss": 0.7381,
+      "step": 1030
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.54730749130249,
+      "learning_rate": 9.050505050505052e-06,
+      "loss": 0.4286,
+      "step": 1040
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.27701061964035034,
+      "learning_rate": 9.040404040404042e-06,
+      "loss": 0.4992,
+      "step": 1050
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.010037928819656372,
+      "learning_rate": 9.030303030303031e-06,
+      "loss": 0.3576,
+      "step": 1060
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.300331711769104,
+      "learning_rate": 9.020202020202021e-06,
+      "loss": 0.4664,
+      "step": 1070
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.9744607210159302,
+      "learning_rate": 9.010101010101012e-06,
+      "loss": 0.6967,
+      "step": 1080
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 11.244192123413086,
+      "learning_rate": 9e-06,
+      "loss": 0.5748,
+      "step": 1090
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.7657531499862671,
+      "learning_rate": 8.98989898989899e-06,
+      "loss": 0.5886,
+      "step": 1100
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.8190665245056152,
+      "learning_rate": 8.97979797979798e-06,
+      "loss": 0.4612,
+      "step": 1110
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.700965404510498,
+      "learning_rate": 8.969696969696971e-06,
+      "loss": 0.4551,
+      "step": 1120
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.6943060755729675,
+      "learning_rate": 8.95959595959596e-06,
+      "loss": 0.4099,
+      "step": 1130
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.1452864408493042,
+      "learning_rate": 8.94949494949495e-06,
+      "loss": 0.5128,
+      "step": 1140
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.24422715604305267,
+      "learning_rate": 8.93939393939394e-06,
+      "loss": 0.6912,
+      "step": 1150
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.50047492980957,
+      "learning_rate": 8.92929292929293e-06,
+      "loss": 0.4031,
+      "step": 1160
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.4065107107162476,
+      "learning_rate": 8.919191919191919e-06,
+      "loss": 0.4238,
+      "step": 1170
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.607456922531128,
+      "learning_rate": 8.90909090909091e-06,
+      "loss": 0.5757,
+      "step": 1180
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.915406584739685,
+      "learning_rate": 8.8989898989899e-06,
+      "loss": 0.5571,
+      "step": 1190
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.252658843994141,
+      "learning_rate": 8.888888888888888e-06,
+      "loss": 0.6749,
+      "step": 1200
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.4744722843170166,
+      "learning_rate": 8.87878787878788e-06,
+      "loss": 0.6021,
+      "step": 1210
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.5909355878829956,
+      "learning_rate": 8.86868686868687e-06,
+      "loss": 0.3381,
+      "step": 1220
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.02662992477417,
+      "learning_rate": 8.85858585858586e-06,
+      "loss": 0.5778,
+      "step": 1230
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.4688441753387451,
+      "learning_rate": 8.84848484848485e-06,
+      "loss": 0.512,
+      "step": 1240
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.9729368090629578,
+      "learning_rate": 8.83838383838384e-06,
+      "loss": 0.5566,
+      "step": 1250
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.964110255241394,
+      "learning_rate": 8.82828282828283e-06,
+      "loss": 0.3959,
+      "step": 1260
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.288043022155762,
+      "learning_rate": 8.818181818181819e-06,
+      "loss": 0.5016,
+      "step": 1270
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.6841915845870972,
+      "learning_rate": 8.808080808080809e-06,
+      "loss": 0.598,
+      "step": 1280
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.1980228424072266,
+      "learning_rate": 8.7979797979798e-06,
+      "loss": 0.5803,
+      "step": 1290
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.8826773166656494,
+      "learning_rate": 8.787878787878788e-06,
+      "loss": 0.3286,
+      "step": 1300
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.622875690460205,
+      "learning_rate": 8.777777777777778e-06,
+      "loss": 0.4741,
+      "step": 1310
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.7105564475059509,
+      "learning_rate": 8.767676767676768e-06,
+      "loss": 0.515,
+      "step": 1320
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.3432900607585907,
+      "learning_rate": 8.757575757575759e-06,
+      "loss": 0.7004,
+      "step": 1330
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.025210857391357,
+      "learning_rate": 8.747474747474747e-06,
+      "loss": 0.3985,
+      "step": 1340
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.137606382369995,
+      "learning_rate": 8.737373737373738e-06,
+      "loss": 0.4918,
+      "step": 1350
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.4660451114177704,
+      "learning_rate": 8.727272727272728e-06,
+      "loss": 0.6669,
+      "step": 1360
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.9560216069221497,
+      "learning_rate": 8.717171717171718e-06,
+      "loss": 0.3283,
+      "step": 1370
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.245961904525757,
+      "learning_rate": 8.707070707070707e-06,
+      "loss": 0.7393,
+      "step": 1380
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.01912454143166542,
+      "learning_rate": 8.696969696969699e-06,
+      "loss": 0.418,
+      "step": 1390
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.0197285413742065,
+      "learning_rate": 8.686868686868687e-06,
+      "loss": 0.492,
+      "step": 1400
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.5338358879089355,
+      "learning_rate": 8.676767676767678e-06,
+      "loss": 0.6951,
+      "step": 1410
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.198877334594727,
+      "learning_rate": 8.666666666666668e-06,
+      "loss": 0.3249,
+      "step": 1420
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.8914008140563965,
+      "learning_rate": 8.656565656565658e-06,
+      "loss": 0.3945,
+      "step": 1430
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.6320083141326904,
+      "learning_rate": 8.646464646464647e-06,
+      "loss": 0.6922,
+      "step": 1440
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.5630173683166504,
+      "learning_rate": 8.636363636363637e-06,
+      "loss": 0.4658,
+      "step": 1450
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.9083809852600098,
+      "learning_rate": 8.626262626262627e-06,
+      "loss": 0.4208,
+      "step": 1460
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.0006579883047379553,
+      "learning_rate": 8.616161616161618e-06,
+      "loss": 0.4293,
+      "step": 1470
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.7547767162323,
+      "learning_rate": 8.606060606060606e-06,
+      "loss": 0.5838,
+      "step": 1480
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.998627662658691,
+      "learning_rate": 8.595959595959596e-06,
+      "loss": 0.6267,
+      "step": 1490
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.166638374328613,
+      "learning_rate": 8.585858585858587e-06,
+      "loss": 0.7075,
+      "step": 1500
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.313971519470215,
+      "learning_rate": 8.575757575757575e-06,
+      "loss": 0.6543,
+      "step": 1510
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.2389867305755615,
+      "learning_rate": 8.565656565656566e-06,
+      "loss": 0.5254,
+      "step": 1520
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.2064523696899414,
+      "learning_rate": 8.555555555555556e-06,
+      "loss": 0.4022,
+      "step": 1530
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.1788508892059326,
+      "learning_rate": 8.545454545454546e-06,
+      "loss": 0.4377,
+      "step": 1540
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.40915897488594055,
+      "learning_rate": 8.535353535353535e-06,
+      "loss": 0.5123,
+      "step": 1550
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.2502992153167725,
+      "learning_rate": 8.525252525252527e-06,
+      "loss": 0.5134,
+      "step": 1560
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.067376136779785,
+      "learning_rate": 8.515151515151517e-06,
+      "loss": 0.7065,
+      "step": 1570
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.364538669586182,
+      "learning_rate": 8.505050505050506e-06,
+      "loss": 0.5286,
+      "step": 1580
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.8198277950286865,
+      "learning_rate": 8.494949494949496e-06,
+      "loss": 0.7122,
+      "step": 1590
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.0054919719696045,
+      "learning_rate": 8.484848484848486e-06,
+      "loss": 0.6057,
+      "step": 1600
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.340848684310913,
+      "learning_rate": 8.474747474747475e-06,
+      "loss": 0.586,
+      "step": 1610
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.649854898452759,
+      "learning_rate": 8.464646464646465e-06,
+      "loss": 0.6599,
+      "step": 1620
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.0032835814636200666,
+      "learning_rate": 8.454545454545455e-06,
+      "loss": 0.5668,
+      "step": 1630
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.295167326927185,
+      "learning_rate": 8.444444444444446e-06,
+      "loss": 0.4962,
+      "step": 1640
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.4222893714904785,
+      "learning_rate": 8.434343434343434e-06,
+      "loss": 0.303,
+      "step": 1650
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.70741868019104,
+      "learning_rate": 8.424242424242425e-06,
+      "loss": 0.3325,
+      "step": 1660
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.6272850632667542,
+      "learning_rate": 8.414141414141415e-06,
+      "loss": 0.3552,
+      "step": 1670
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.39682525396347046,
+      "learning_rate": 8.404040404040405e-06,
+      "loss": 0.5188,
+      "step": 1680
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.233819484710693,
+      "learning_rate": 8.393939393939394e-06,
+      "loss": 0.4309,
+      "step": 1690
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.6901202201843262,
+      "learning_rate": 8.383838383838384e-06,
+      "loss": 0.4005,
+      "step": 1700
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.25327730178833,
+      "learning_rate": 8.373737373737374e-06,
+      "loss": 0.5327,
+      "step": 1710
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.563572883605957,
+      "learning_rate": 8.363636363636365e-06,
+      "loss": 0.5132,
+      "step": 1720
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.6525028944015503,
+      "learning_rate": 8.353535353535355e-06,
+      "loss": 0.3818,
+      "step": 1730
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.6655619144439697,
+      "learning_rate": 8.343434343434345e-06,
+      "loss": 0.4418,
+      "step": 1740
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.8698630332946777,
+      "learning_rate": 8.333333333333334e-06,
+      "loss": 0.6928,
+      "step": 1750
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.353645324707031,
+      "learning_rate": 8.323232323232324e-06,
+      "loss": 0.6257,
+      "step": 1760
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 6.349965572357178,
+      "learning_rate": 8.313131313131314e-06,
+      "loss": 0.9007,
+      "step": 1770
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.7333180904388428,
+      "learning_rate": 8.303030303030305e-06,
+      "loss": 0.4963,
+      "step": 1780
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.6360317468643188,
+      "learning_rate": 8.292929292929293e-06,
+      "loss": 0.9028,
+      "step": 1790
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.3366623520851135,
+      "learning_rate": 8.282828282828283e-06,
+      "loss": 0.4281,
+      "step": 1800
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.0651333332061768,
+      "learning_rate": 8.272727272727274e-06,
+      "loss": 0.3646,
+      "step": 1810
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.6993482708930969,
+      "learning_rate": 8.262626262626264e-06,
+      "loss": 0.5423,
+      "step": 1820
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.010565996170044,
+      "learning_rate": 8.252525252525253e-06,
+      "loss": 0.7413,
+      "step": 1830
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.367119312286377,
+      "learning_rate": 8.242424242424243e-06,
+      "loss": 0.6385,
+      "step": 1840
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.1688125133514404,
+      "learning_rate": 8.232323232323233e-06,
+      "loss": 0.5547,
+      "step": 1850
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.5372488498687744,
+      "learning_rate": 8.222222222222222e-06,
+      "loss": 0.5464,
+      "step": 1860
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 7.86212682723999,
+      "learning_rate": 8.212121212121212e-06,
+      "loss": 0.414,
+      "step": 1870
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.5354101657867432,
+      "learning_rate": 8.202020202020202e-06,
+      "loss": 0.6147,
+      "step": 1880
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.8895275592803955,
+      "learning_rate": 8.191919191919193e-06,
+      "loss": 0.4875,
+      "step": 1890
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.0076402425765991,
+      "learning_rate": 8.181818181818183e-06,
+      "loss": 0.553,
+      "step": 1900
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.115602493286133,
+      "learning_rate": 8.171717171717173e-06,
+      "loss": 0.3642,
+      "step": 1910
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.1112468242645264,
+      "learning_rate": 8.161616161616162e-06,
+      "loss": 0.4393,
+      "step": 1920
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.099217891693115,
+      "learning_rate": 8.151515151515152e-06,
+      "loss": 0.3543,
+      "step": 1930
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.719365358352661,
+      "learning_rate": 8.141414141414142e-06,
+      "loss": 0.6467,
+      "step": 1940
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.9949913024902344,
+      "learning_rate": 8.131313131313133e-06,
+      "loss": 0.5151,
+      "step": 1950
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.5979423522949219,
+      "learning_rate": 8.121212121212121e-06,
+      "loss": 0.3655,
+      "step": 1960
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.0186359882354736,
+      "learning_rate": 8.111111111111112e-06,
+      "loss": 0.5278,
+      "step": 1970
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.232735633850098,
+      "learning_rate": 8.101010101010102e-06,
+      "loss": 0.5965,
+      "step": 1980
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.5541656017303467,
+      "learning_rate": 8.090909090909092e-06,
+      "loss": 0.5226,
+      "step": 1990
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.416280269622803,
+      "learning_rate": 8.08080808080808e-06,
+      "loss": 0.3881,
+      "step": 2000
+    },
+    {
+      "epoch": 0.1,
+      "eval_loss": 0.49726012349128723,
+      "eval_runtime": 296.5886,
+      "eval_samples_per_second": 3.372,
+      "eval_steps_per_second": 3.372,
+      "step": 2000
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.3124769926071167,
+      "learning_rate": 8.070707070707071e-06,
+      "loss": 0.6145,
+      "step": 2010
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.8283050060272217,
+      "learning_rate": 8.060606060606061e-06,
+      "loss": 0.5876,
+      "step": 2020
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.924971580505371,
+      "learning_rate": 8.050505050505052e-06,
+      "loss": 0.6135,
+      "step": 2030
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.105405569076538,
+      "learning_rate": 8.04040404040404e-06,
+      "loss": 0.3757,
+      "step": 2040
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.2510244846343994,
+      "learning_rate": 8.03030303030303e-06,
+      "loss": 0.6634,
+      "step": 2050
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.0584206581115723,
+      "learning_rate": 8.02020202020202e-06,
+      "loss": 0.464,
+      "step": 2060
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.1460766792297363,
+      "learning_rate": 8.010101010101011e-06,
+      "loss": 0.4943,
+      "step": 2070
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.760594367980957,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 0.4359,
+      "step": 2080
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.7750723361968994,
+      "learning_rate": 7.989898989898992e-06,
+      "loss": 0.3657,
+      "step": 2090
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.917431116104126,
+      "learning_rate": 7.97979797979798e-06,
+      "loss": 0.3915,
+      "step": 2100
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.772168755531311,
+      "learning_rate": 7.96969696969697e-06,
+      "loss": 0.8333,
+      "step": 2110
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.072344183921814,
+      "learning_rate": 7.95959595959596e-06,
+      "loss": 0.8133,
+      "step": 2120
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.3543601036071777,
+      "learning_rate": 7.949494949494951e-06,
+      "loss": 0.538,
+      "step": 2130
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.6881474256515503,
+      "learning_rate": 7.93939393939394e-06,
+      "loss": 0.3318,
+      "step": 2140
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.6667841672897339,
+      "learning_rate": 7.92929292929293e-06,
+      "loss": 0.6603,
+      "step": 2150
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.7068934440612793,
+      "learning_rate": 7.91919191919192e-06,
+      "loss": 0.3717,
+      "step": 2160
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.8397080898284912,
+      "learning_rate": 7.909090909090909e-06,
+      "loss": 0.5456,
+      "step": 2170
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.438650369644165,
+      "learning_rate": 7.898989898989899e-06,
+      "loss": 0.4993,
+      "step": 2180
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.1788564920425415,
+      "learning_rate": 7.88888888888889e-06,
+      "loss": 0.5379,
+      "step": 2190
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.0841777324676514,
+      "learning_rate": 7.87878787878788e-06,
+      "loss": 0.4348,
+      "step": 2200
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.9438015222549438,
+      "learning_rate": 7.868686868686868e-06,
+      "loss": 0.4776,
+      "step": 2210
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.519500195980072,
+      "learning_rate": 7.858585858585859e-06,
+      "loss": 0.624,
+      "step": 2220
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.4791972637176514,
+      "learning_rate": 7.848484848484849e-06,
+      "loss": 0.491,
+      "step": 2230
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.15498948097229,
+      "learning_rate": 7.838383838383839e-06,
+      "loss": 0.3268,
+      "step": 2240
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.7391984462738037,
+      "learning_rate": 7.82828282828283e-06,
+      "loss": 0.695,
+      "step": 2250
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.0758000612258911,
+      "learning_rate": 7.81818181818182e-06,
+      "loss": 0.614,
+      "step": 2260
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.8113633394241333,
+      "learning_rate": 7.808080808080808e-06,
+      "loss": 0.5223,
+      "step": 2270
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.9470202326774597,
+      "learning_rate": 7.797979797979799e-06,
+      "loss": 0.741,
+      "step": 2280
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.6583170890808105,
+      "learning_rate": 7.787878787878789e-06,
+      "loss": 0.5571,
+      "step": 2290
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.9988512992858887,
+      "learning_rate": 7.77777777777778e-06,
+      "loss": 0.5672,
+      "step": 2300
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 8.29896068572998,
+      "learning_rate": 7.767676767676768e-06,
+      "loss": 0.4828,
+      "step": 2310
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.0121721029281616,
+      "learning_rate": 7.757575757575758e-06,
+      "loss": 0.588,
+      "step": 2320
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.5740251541137695,
+      "learning_rate": 7.747474747474748e-06,
+      "loss": 0.5385,
+      "step": 2330
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.059847593307495,
+      "learning_rate": 7.737373737373739e-06,
+      "loss": 0.7256,
+      "step": 2340
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 30.723230361938477,
+      "learning_rate": 7.727272727272727e-06,
+      "loss": 0.581,
+      "step": 2350
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.9344816207885742,
+      "learning_rate": 7.717171717171717e-06,
+      "loss": 0.5751,
+      "step": 2360
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.999989032745361,
+      "learning_rate": 7.707070707070708e-06,
+      "loss": 0.5531,
+      "step": 2370
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.1106081008911133,
+      "learning_rate": 7.696969696969696e-06,
+      "loss": 0.5414,
+      "step": 2380
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.0552499294281006,
+      "learning_rate": 7.686868686868687e-06,
+      "loss": 0.3083,
+      "step": 2390
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.31422069668769836,
+      "learning_rate": 7.676767676767677e-06,
+      "loss": 0.4043,
+      "step": 2400
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.465919852256775,
+      "learning_rate": 7.666666666666667e-06,
+      "loss": 0.3835,
+      "step": 2410
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.2063610553741455,
+      "learning_rate": 7.656565656565658e-06,
+      "loss": 0.5132,
+      "step": 2420
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.7438418865203857,
+      "learning_rate": 7.646464646464648e-06,
+      "loss": 0.7799,
+      "step": 2430
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.2593297958374023,
+      "learning_rate": 7.636363636363638e-06,
+      "loss": 0.4142,
+      "step": 2440
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.2809324264526367,
+      "learning_rate": 7.6262626262626275e-06,
+      "loss": 0.4862,
+      "step": 2450
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.056776762008667,
+      "learning_rate": 7.616161616161617e-06,
+      "loss": 0.5726,
+      "step": 2460
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.739293098449707,
+      "learning_rate": 7.606060606060606e-06,
+      "loss": 0.4945,
+      "step": 2470
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.824068069458008,
+      "learning_rate": 7.595959595959597e-06,
+      "loss": 0.6431,
+      "step": 2480
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.1176700592041016,
+      "learning_rate": 7.585858585858586e-06,
+      "loss": 0.4511,
+      "step": 2490
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.115619421005249,
+      "learning_rate": 7.5757575757575764e-06,
+      "loss": 0.3236,
+      "step": 2500
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.1324357986450195,
+      "learning_rate": 7.565656565656566e-06,
+      "loss": 0.488,
+      "step": 2510
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.205792427062988,
+      "learning_rate": 7.555555555555556e-06,
+      "loss": 0.6519,
+      "step": 2520
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.29966816306114197,
+      "learning_rate": 7.545454545454546e-06,
+      "loss": 0.2427,
+      "step": 2530
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.695274293422699,
+      "learning_rate": 7.535353535353536e-06,
+      "loss": 0.6302,
+      "step": 2540
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.0804121494293213,
+      "learning_rate": 7.525252525252525e-06,
+      "loss": 0.581,
+      "step": 2550
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.076641798019409,
+      "learning_rate": 7.515151515151516e-06,
+      "loss": 0.5715,
+      "step": 2560
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.8025639057159424,
+      "learning_rate": 7.505050505050505e-06,
+      "loss": 0.5451,
+      "step": 2570
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.8459610342979431,
+      "learning_rate": 7.494949494949496e-06,
+      "loss": 0.6437,
+      "step": 2580
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.0758402347564697,
+      "learning_rate": 7.484848484848486e-06,
+      "loss": 0.5077,
+      "step": 2590
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.3608747720718384,
+      "learning_rate": 7.474747474747476e-06,
+      "loss": 0.5145,
+      "step": 2600
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.9487907886505127,
+      "learning_rate": 7.464646464646465e-06,
+      "loss": 0.4173,
+      "step": 2610
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.6212667226791382,
+      "learning_rate": 7.454545454545456e-06,
+      "loss": 0.4199,
+      "step": 2620
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.783740282058716,
+      "learning_rate": 7.444444444444445e-06,
+      "loss": 0.5639,
+      "step": 2630
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.7923329472541809,
+      "learning_rate": 7.434343434343435e-06,
+      "loss": 0.5934,
+      "step": 2640
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.9525859355926514,
+      "learning_rate": 7.424242424242425e-06,
+      "loss": 0.3554,
+      "step": 2650
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.8090410232543945,
+      "learning_rate": 7.414141414141415e-06,
+      "loss": 0.472,
+      "step": 2660
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.31178975105285645,
+      "learning_rate": 7.4040404040404045e-06,
+      "loss": 0.4174,
+      "step": 2670
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.298391342163086,
+      "learning_rate": 7.393939393939395e-06,
+      "loss": 0.5147,
+      "step": 2680
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.6585010290145874,
+      "learning_rate": 7.383838383838384e-06,
+      "loss": 0.5512,
+      "step": 2690
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.1424102783203125,
+      "learning_rate": 7.373737373737374e-06,
+      "loss": 0.538,
+      "step": 2700
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.008727720007300377,
+      "learning_rate": 7.363636363636364e-06,
+      "loss": 0.3115,
+      "step": 2710
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.3273710012435913,
+      "learning_rate": 7.353535353535353e-06,
+      "loss": 0.602,
+      "step": 2720
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.9286468029022217,
+      "learning_rate": 7.343434343434344e-06,
+      "loss": 0.5314,
+      "step": 2730
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.033441543579102,
+      "learning_rate": 7.333333333333333e-06,
+      "loss": 0.4682,
+      "step": 2740
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.1140573024749756,
+      "learning_rate": 7.323232323232324e-06,
+      "loss": 0.8108,
+      "step": 2750
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.7377740144729614,
+      "learning_rate": 7.3131313131313146e-06,
+      "loss": 0.6938,
+      "step": 2760
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.0510828495025635,
+      "learning_rate": 7.303030303030304e-06,
+      "loss": 0.3355,
+      "step": 2770
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.9355423450469971,
+      "learning_rate": 7.2929292929292934e-06,
+      "loss": 0.6595,
+      "step": 2780
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.6041629314422607,
+      "learning_rate": 7.282828282828284e-06,
+      "loss": 0.3718,
+      "step": 2790
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.9129393696784973,
+      "learning_rate": 7.272727272727273e-06,
+      "loss": 0.42,
+      "step": 2800
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.611159086227417,
+      "learning_rate": 7.2626262626262635e-06,
+      "loss": 0.5017,
+      "step": 2810
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.267125129699707,
+      "learning_rate": 7.252525252525253e-06,
+      "loss": 0.5167,
+      "step": 2820
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.4013969898223877,
+      "learning_rate": 7.242424242424243e-06,
+      "loss": 0.3934,
+      "step": 2830
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.3873810768127441,
+      "learning_rate": 7.232323232323233e-06,
+      "loss": 0.5354,
+      "step": 2840
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.160951018333435,
+      "learning_rate": 7.222222222222223e-06,
+      "loss": 0.442,
+      "step": 2850
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.36929821968078613,
+      "learning_rate": 7.212121212121212e-06,
+      "loss": 0.5383,
+      "step": 2860
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.8486219644546509,
+      "learning_rate": 7.202020202020203e-06,
+      "loss": 0.6933,
+      "step": 2870
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 9.711508750915527,
+      "learning_rate": 7.191919191919192e-06,
+      "loss": 0.5174,
+      "step": 2880
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 6.458278179168701,
+      "learning_rate": 7.181818181818182e-06,
+      "loss": 0.4861,
+      "step": 2890
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.309101104736328,
+      "learning_rate": 7.171717171717172e-06,
+      "loss": 0.3574,
+      "step": 2900
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.2773056030273438,
+      "learning_rate": 7.161616161616162e-06,
+      "loss": 0.3333,
+      "step": 2910
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.31492137908935547,
+      "learning_rate": 7.151515151515152e-06,
+      "loss": 0.4338,
+      "step": 2920
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.8371341228485107,
+      "learning_rate": 7.141414141414143e-06,
+      "loss": 0.659,
+      "step": 2930
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.0017792941071093082,
+      "learning_rate": 7.131313131313132e-06,
+      "loss": 0.4461,
+      "step": 2940
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.9826254844665527,
+      "learning_rate": 7.121212121212122e-06,
+      "loss": 0.4224,
+      "step": 2950
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.41608259081840515,
+      "learning_rate": 7.111111111111112e-06,
+      "loss": 0.4182,
+      "step": 2960
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.108034610748291,
+      "learning_rate": 7.101010101010102e-06,
+      "loss": 0.4149,
+      "step": 2970
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.7766743898391724,
+      "learning_rate": 7.0909090909090916e-06,
+      "loss": 0.4758,
+      "step": 2980
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.912661075592041,
+      "learning_rate": 7.080808080808082e-06,
+      "loss": 0.3912,
+      "step": 2990
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.7931787967681885,
+      "learning_rate": 7.070707070707071e-06,
+      "loss": 0.5228,
+      "step": 3000
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.948633909225464,
+      "learning_rate": 7.060606060606061e-06,
+      "loss": 0.5097,
+      "step": 3010
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.4696679711341858,
+      "learning_rate": 7.050505050505051e-06,
+      "loss": 0.4057,
+      "step": 3020
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.7676633596420288,
+      "learning_rate": 7.0404040404040404e-06,
+      "loss": 0.6008,
+      "step": 3030
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.0048167705535889,
+      "learning_rate": 7.030303030303031e-06,
+      "loss": 0.6355,
+      "step": 3040
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 13.031144142150879,
+      "learning_rate": 7.02020202020202e-06,
+      "loss": 0.5433,
+      "step": 3050
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.508690357208252,
+      "learning_rate": 7.0101010101010105e-06,
+      "loss": 0.5252,
+      "step": 3060
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.6236469745635986,
+      "learning_rate": 7e-06,
+      "loss": 0.3748,
+      "step": 3070
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.831434726715088,
+      "learning_rate": 6.98989898989899e-06,
+      "loss": 0.6294,
+      "step": 3080
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.6510546207427979,
+      "learning_rate": 6.979797979797981e-06,
+      "loss": 0.5122,
+      "step": 3090
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.7243143916130066,
+      "learning_rate": 6.969696969696971e-06,
+      "loss": 0.4985,
+      "step": 3100
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.455175876617432,
+      "learning_rate": 6.95959595959596e-06,
+      "loss": 0.4532,
+      "step": 3110
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.3478859663009644,
+      "learning_rate": 6.9494949494949505e-06,
+      "loss": 0.5427,
+      "step": 3120
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.694676637649536,
+      "learning_rate": 6.93939393939394e-06,
+      "loss": 0.542,
+      "step": 3130
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.880995750427246,
+      "learning_rate": 6.92929292929293e-06,
+      "loss": 0.5949,
+      "step": 3140
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.839517593383789,
+      "learning_rate": 6.91919191919192e-06,
+      "loss": 0.6437,
+      "step": 3150
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.4446146488189697,
+      "learning_rate": 6.90909090909091e-06,
+      "loss": 0.5319,
+      "step": 3160
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.8178725242614746,
+      "learning_rate": 6.898989898989899e-06,
+      "loss": 0.5109,
+      "step": 3170
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.1470345258712769,
+      "learning_rate": 6.88888888888889e-06,
+      "loss": 0.5791,
+      "step": 3180
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.2457770109176636,
+      "learning_rate": 6.878787878787879e-06,
+      "loss": 0.3547,
+      "step": 3190
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.6244733333587646,
+      "learning_rate": 6.868686868686869e-06,
+      "loss": 0.6079,
+      "step": 3200
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.3486361503601074,
+      "learning_rate": 6.858585858585859e-06,
+      "loss": 0.5009,
+      "step": 3210
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 7.4471588134765625,
+      "learning_rate": 6.848484848484849e-06,
+      "loss": 0.7388,
+      "step": 3220
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.836559772491455,
+      "learning_rate": 6.8383838383838386e-06,
+      "loss": 0.5774,
+      "step": 3230
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.352264881134033,
+      "learning_rate": 6.828282828282828e-06,
+      "loss": 0.6538,
+      "step": 3240
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.12757644057273865,
+      "learning_rate": 6.818181818181818e-06,
+      "loss": 0.4133,
+      "step": 3250
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.509770154953003,
+      "learning_rate": 6.808080808080809e-06,
+      "loss": 0.5061,
+      "step": 3260
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.2330329418182373,
+      "learning_rate": 6.797979797979799e-06,
+      "loss": 0.5501,
+      "step": 3270
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.468204975128174,
+      "learning_rate": 6.787878787878789e-06,
+      "loss": 0.5382,
+      "step": 3280
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.865310788154602,
+      "learning_rate": 6.777777777777779e-06,
+      "loss": 0.5676,
+      "step": 3290
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.1490612030029297,
+      "learning_rate": 6.767676767676769e-06,
+      "loss": 0.6547,
+      "step": 3300
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.9897737503051758,
+      "learning_rate": 6.757575757575758e-06,
+      "loss": 0.6402,
+      "step": 3310
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.8899117112159729,
+      "learning_rate": 6.747474747474749e-06,
+      "loss": 0.4865,
+      "step": 3320
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.5953271985054016,
+      "learning_rate": 6.737373737373738e-06,
+      "loss": 0.5601,
+      "step": 3330
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.7100812196731567,
+      "learning_rate": 6.7272727272727275e-06,
+      "loss": 0.4554,
+      "step": 3340
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.1394648551940918,
+      "learning_rate": 6.717171717171718e-06,
+      "loss": 0.4665,
+      "step": 3350
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.531031131744385,
+      "learning_rate": 6.707070707070707e-06,
+      "loss": 0.351,
+      "step": 3360
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.8314118981361389,
+      "learning_rate": 6.6969696969696975e-06,
+      "loss": 0.2993,
+      "step": 3370
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.985590934753418,
+      "learning_rate": 6.686868686868687e-06,
+      "loss": 0.5341,
+      "step": 3380
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.496417284011841,
+      "learning_rate": 6.676767676767677e-06,
+      "loss": 0.3507,
+      "step": 3390
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.9464170932769775,
+      "learning_rate": 6.666666666666667e-06,
+      "loss": 0.4813,
+      "step": 3400
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.9352614283561707,
+      "learning_rate": 6.656565656565657e-06,
+      "loss": 0.7408,
+      "step": 3410
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.2820653915405273,
+      "learning_rate": 6.646464646464646e-06,
+      "loss": 0.5518,
+      "step": 3420
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.6449127793312073,
+      "learning_rate": 6.6363636363636375e-06,
+      "loss": 0.3831,
+      "step": 3430
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.303825855255127,
+      "learning_rate": 6.626262626262627e-06,
+      "loss": 0.3408,
+      "step": 3440
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.5336707830429077,
+      "learning_rate": 6.616161616161617e-06,
+      "loss": 0.4224,
+      "step": 3450
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.874389171600342,
+      "learning_rate": 6.606060606060607e-06,
+      "loss": 0.4829,
+      "step": 3460
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.6338319778442383,
+      "learning_rate": 6.595959595959597e-06,
+      "loss": 0.3992,
+      "step": 3470
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.416031837463379,
+      "learning_rate": 6.585858585858586e-06,
+      "loss": 0.464,
+      "step": 3480
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.0006036623963154852,
+      "learning_rate": 6.575757575757577e-06,
+      "loss": 0.454,
+      "step": 3490
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.059628486633301,
+      "learning_rate": 6.565656565656566e-06,
+      "loss": 0.4677,
+      "step": 3500
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.7535749673843384,
+      "learning_rate": 6.555555555555556e-06,
+      "loss": 0.4647,
+      "step": 3510
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.2723642587661743,
+      "learning_rate": 6.545454545454546e-06,
+      "loss": 0.4257,
+      "step": 3520
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.9197982549667358,
+      "learning_rate": 6.535353535353536e-06,
+      "loss": 0.4922,
+      "step": 3530
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.634425163269043,
+      "learning_rate": 6.525252525252526e-06,
+      "loss": 0.3445,
+      "step": 3540
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.2264757752418518,
+      "learning_rate": 6.515151515151516e-06,
+      "loss": 0.3621,
+      "step": 3550
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.509833574295044,
+      "learning_rate": 6.505050505050505e-06,
+      "loss": 0.4623,
+      "step": 3560
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 9.437053680419922,
+      "learning_rate": 6.494949494949495e-06,
+      "loss": 0.5641,
+      "step": 3570
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.30497670173645,
+      "learning_rate": 6.484848484848485e-06,
+      "loss": 0.3517,
+      "step": 3580
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.4982795715332031,
+      "learning_rate": 6.4747474747474745e-06,
+      "loss": 0.4624,
+      "step": 3590
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.8188090324401855,
+      "learning_rate": 6.464646464646466e-06,
+      "loss": 0.4145,
+      "step": 3600
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 8.528606414794922,
+      "learning_rate": 6.454545454545456e-06,
+      "loss": 0.5919,
+      "step": 3610
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.2666741609573364,
+      "learning_rate": 6.444444444444445e-06,
+      "loss": 0.2912,
+      "step": 3620
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.027946710586548,
+      "learning_rate": 6.434343434343436e-06,
+      "loss": 0.4914,
+      "step": 3630
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.017730474472046,
+      "learning_rate": 6.424242424242425e-06,
+      "loss": 0.7763,
+      "step": 3640
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.3871991634368896,
+      "learning_rate": 6.4141414141414145e-06,
+      "loss": 0.4079,
+      "step": 3650
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.026352882385254,
+      "learning_rate": 6.404040404040405e-06,
+      "loss": 0.6339,
+      "step": 3660
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.719714403152466,
+      "learning_rate": 6.393939393939394e-06,
+      "loss": 0.5748,
+      "step": 3670
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.5980193614959717,
+      "learning_rate": 6.3838383838383845e-06,
+      "loss": 0.7281,
+      "step": 3680
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.5198047161102295,
+      "learning_rate": 6.373737373737374e-06,
+      "loss": 0.6011,
+      "step": 3690
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.3764171600341797,
+      "learning_rate": 6.363636363636364e-06,
+      "loss": 0.4418,
+      "step": 3700
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.8297829627990723,
+      "learning_rate": 6.353535353535354e-06,
+      "loss": 0.4101,
+      "step": 3710
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.657487392425537,
+      "learning_rate": 6.343434343434344e-06,
+      "loss": 0.6139,
+      "step": 3720
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.4731935262680054,
+      "learning_rate": 6.333333333333333e-06,
+      "loss": 0.4851,
+      "step": 3730
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 2.780787229537964,
+      "learning_rate": 6.323232323232324e-06,
+      "loss": 0.5278,
+      "step": 3740
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.002871747827157378,
+      "learning_rate": 6.313131313131313e-06,
+      "loss": 0.3632,
+      "step": 3750
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 3.415860652923584,
+      "learning_rate": 6.303030303030303e-06,
+      "loss": 0.5422,
+      "step": 3760
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 2.642449378967285,
+      "learning_rate": 6.292929292929294e-06,
+      "loss": 0.5575,
+      "step": 3770
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.6322145462036133,
+      "learning_rate": 6.282828282828284e-06,
+      "loss": 0.3176,
+      "step": 3780
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.659850597381592,
+      "learning_rate": 6.2727272727272734e-06,
+      "loss": 0.4444,
+      "step": 3790
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 3.632639169692993,
+      "learning_rate": 6.262626262626264e-06,
+      "loss": 0.4449,
+      "step": 3800
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.3175582885742188,
+      "learning_rate": 6.252525252525253e-06,
+      "loss": 0.4983,
+      "step": 3810
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.5408777594566345,
+      "learning_rate": 6.2424242424242434e-06,
+      "loss": 0.4653,
+      "step": 3820
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.46673011779785156,
+      "learning_rate": 6.232323232323233e-06,
+      "loss": 0.3618,
+      "step": 3830
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.012616774067282677,
+      "learning_rate": 6.222222222222223e-06,
+      "loss": 0.3363,
+      "step": 3840
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.871781826019287,
+      "learning_rate": 6.212121212121213e-06,
+      "loss": 0.5996,
+      "step": 3850
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 2.607459783554077,
+      "learning_rate": 6.202020202020203e-06,
+      "loss": 0.633,
+      "step": 3860
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 2.0456507205963135,
+      "learning_rate": 6.191919191919192e-06,
+      "loss": 0.4236,
+      "step": 3870
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.3647568225860596,
+      "learning_rate": 6.181818181818182e-06,
+      "loss": 0.6016,
+      "step": 3880
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.3835606575012207,
+      "learning_rate": 6.171717171717172e-06,
+      "loss": 0.4672,
+      "step": 3890
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.276355743408203,
+      "learning_rate": 6.1616161616161615e-06,
+      "loss": 0.3073,
+      "step": 3900
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.4117212295532227,
+      "learning_rate": 6.151515151515152e-06,
+      "loss": 0.4799,
+      "step": 3910
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.6188517212867737,
+      "learning_rate": 6.141414141414141e-06,
+      "loss": 0.4852,
+      "step": 3920
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.361952304840088,
+      "learning_rate": 6.1313131313131315e-06,
+      "loss": 0.4593,
+      "step": 3930
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.983630359172821,
+      "learning_rate": 6.121212121212121e-06,
+      "loss": 0.5209,
+      "step": 3940
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.2454302310943604,
+      "learning_rate": 6.111111111111112e-06,
+      "loss": 0.6493,
+      "step": 3950
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.920189619064331,
+      "learning_rate": 6.1010101010101015e-06,
+      "loss": 0.3506,
+      "step": 3960
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.168548107147217,
+      "learning_rate": 6.090909090909092e-06,
+      "loss": 0.4219,
+      "step": 3970
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.7518400549888611,
+      "learning_rate": 6.080808080808081e-06,
+      "loss": 0.2783,
+      "step": 3980
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.4468128681182861,
+      "learning_rate": 6.0707070707070715e-06,
+      "loss": 0.5845,
+      "step": 3990
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.7297825217247009,
+      "learning_rate": 6.060606060606061e-06,
+      "loss": 0.5356,
+      "step": 4000
+    },
+    {
+      "epoch": 0.2,
+      "eval_loss": 0.46820691227912903,
+      "eval_runtime": 298.8864,
+      "eval_samples_per_second": 3.346,
+      "eval_steps_per_second": 3.346,
+      "step": 4000
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.9701703786849976,
+      "learning_rate": 6.050505050505051e-06,
+      "loss": 0.4308,
+      "step": 4010
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.4310851097106934,
+      "learning_rate": 6.040404040404041e-06,
+      "loss": 0.3936,
+      "step": 4020
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.382595539093018,
+      "learning_rate": 6.030303030303031e-06,
+      "loss": 0.4086,
+      "step": 4030
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.4613182544708252,
+      "learning_rate": 6.0202020202020204e-06,
+      "loss": 0.5119,
+      "step": 4040
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.6718851327896118,
+      "learning_rate": 6.010101010101011e-06,
+      "loss": 0.5014,
+      "step": 4050
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.999420166015625,
+      "learning_rate": 6e-06,
+      "loss": 0.4534,
+      "step": 4060
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.6182875633239746,
+      "learning_rate": 5.9898989898989904e-06,
+      "loss": 0.4892,
+      "step": 4070
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.8443717956542969,
+      "learning_rate": 5.97979797979798e-06,
+      "loss": 0.5317,
+      "step": 4080
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.2049109935760498,
+      "learning_rate": 5.96969696969697e-06,
+      "loss": 0.4064,
+      "step": 4090
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.2313249111175537,
+      "learning_rate": 5.95959595959596e-06,
+      "loss": 0.6199,
+      "step": 4100
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.08662452548742294,
+      "learning_rate": 5.949494949494949e-06,
+      "loss": 0.4511,
+      "step": 4110
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.9273214340209961,
+      "learning_rate": 5.93939393939394e-06,
+      "loss": 0.5627,
+      "step": 4120
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.07971806079149246,
+      "learning_rate": 5.9292929292929305e-06,
+      "loss": 0.4257,
+      "step": 4130
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.36241334676742554,
+      "learning_rate": 5.91919191919192e-06,
+      "loss": 0.555,
+      "step": 4140
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 2.043637752532959,
+      "learning_rate": 5.90909090909091e-06,
+      "loss": 0.5169,
+      "step": 4150
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.11596064269542694,
+      "learning_rate": 5.8989898989899e-06,
+      "loss": 0.4313,
+      "step": 4160
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.627316951751709,
+      "learning_rate": 5.88888888888889e-06,
+      "loss": 0.622,
+      "step": 4170
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 12.65475845336914,
+      "learning_rate": 5.878787878787879e-06,
+      "loss": 0.6768,
+      "step": 4180
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 10.650860786437988,
+      "learning_rate": 5.868686868686869e-06,
+      "loss": 0.6316,
+      "step": 4190
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.2469587326049805,
+      "learning_rate": 5.858585858585859e-06,
+      "loss": 0.5398,
+      "step": 4200
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.6116452813148499,
+      "learning_rate": 5.8484848484848485e-06,
+      "loss": 0.3981,
+      "step": 4210
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.8067615032196045,
+      "learning_rate": 5.838383838383839e-06,
+      "loss": 0.5844,
+      "step": 4220
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.0388538837432861,
+      "learning_rate": 5.828282828282828e-06,
+      "loss": 0.4986,
+      "step": 4230
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.2012486457824707,
+      "learning_rate": 5.8181818181818185e-06,
+      "loss": 0.5646,
+      "step": 4240
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.834928035736084,
+      "learning_rate": 5.808080808080808e-06,
+      "loss": 0.4437,
+      "step": 4250
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.3954122066497803,
+      "learning_rate": 5.797979797979798e-06,
+      "loss": 0.8383,
+      "step": 4260
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.624404788017273,
+      "learning_rate": 5.787878787878788e-06,
+      "loss": 0.5036,
+      "step": 4270
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.0964741706848145,
+      "learning_rate": 5.777777777777778e-06,
+      "loss": 0.3223,
+      "step": 4280
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.0320823192596436,
+      "learning_rate": 5.767676767676768e-06,
+      "loss": 0.2767,
+      "step": 4290
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 2.0635337829589844,
+      "learning_rate": 5.7575757575757586e-06,
+      "loss": 0.5158,
+      "step": 4300
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.017931342124939,
+      "learning_rate": 5.747474747474748e-06,
+      "loss": 0.4481,
+      "step": 4310
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.0015388656174764037,
+      "learning_rate": 5.737373737373738e-06,
+      "loss": 0.3405,
+      "step": 4320
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.2430816888809204,
+      "learning_rate": 5.727272727272728e-06,
+      "loss": 0.3931,
+      "step": 4330
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.223113775253296,
+      "learning_rate": 5.717171717171718e-06,
+      "loss": 0.678,
+      "step": 4340
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.3319227695465088,
+      "learning_rate": 5.7070707070707075e-06,
+      "loss": 0.4558,
+      "step": 4350
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.6266989707946777,
+      "learning_rate": 5.696969696969698e-06,
+      "loss": 0.5438,
+      "step": 4360
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.511033058166504,
+      "learning_rate": 5.686868686868687e-06,
+      "loss": 0.5391,
+      "step": 4370
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.7635992169380188,
+      "learning_rate": 5.6767676767676775e-06,
+      "loss": 0.3729,
+      "step": 4380
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.7493975162506104,
+      "learning_rate": 5.666666666666667e-06,
+      "loss": 0.6101,
+      "step": 4390
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 3.5871286392211914,
+      "learning_rate": 5.656565656565657e-06,
+      "loss": 0.3751,
+      "step": 4400
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.9037196636199951,
+      "learning_rate": 5.646464646464647e-06,
+      "loss": 0.6219,
+      "step": 4410
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.273498296737671,
+      "learning_rate": 5.636363636363636e-06,
+      "loss": 0.5239,
+      "step": 4420
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.8304288387298584,
+      "learning_rate": 5.626262626262626e-06,
+      "loss": 0.4935,
+      "step": 4430
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.9731048345565796,
+      "learning_rate": 5.616161616161616e-06,
+      "loss": 0.4508,
+      "step": 4440
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.2921247482299805,
+      "learning_rate": 5.606060606060606e-06,
+      "loss": 0.5789,
+      "step": 4450
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.8995174765586853,
+      "learning_rate": 5.595959595959597e-06,
+      "loss": 0.3663,
+      "step": 4460
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.3782875537872314,
+      "learning_rate": 5.585858585858587e-06,
+      "loss": 0.6008,
+      "step": 4470
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.6055619716644287,
+      "learning_rate": 5.575757575757577e-06,
+      "loss": 0.4031,
+      "step": 4480
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.1836442947387695,
+      "learning_rate": 5.565656565656566e-06,
+      "loss": 0.4526,
+      "step": 4490
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.582188606262207,
+      "learning_rate": 5.555555555555557e-06,
+      "loss": 0.52,
+      "step": 4500
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.0075886249542236,
+      "learning_rate": 5.545454545454546e-06,
+      "loss": 0.4464,
+      "step": 4510
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.0015082359313965,
+      "learning_rate": 5.5353535353535355e-06,
+      "loss": 0.4574,
+      "step": 4520
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.7861164808273315,
+      "learning_rate": 5.525252525252526e-06,
+      "loss": 0.4034,
+      "step": 4530
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.422027587890625,
+      "learning_rate": 5.515151515151515e-06,
+      "loss": 0.4549,
+      "step": 4540
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.0352510213851929,
+      "learning_rate": 5.5050505050505056e-06,
+      "loss": 0.5032,
+      "step": 4550
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.7233294248580933,
+      "learning_rate": 5.494949494949495e-06,
+      "loss": 0.4383,
+      "step": 4560
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.963414669036865,
+      "learning_rate": 5.484848484848485e-06,
+      "loss": 0.2754,
+      "step": 4570
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.6809269189834595,
+      "learning_rate": 5.474747474747475e-06,
+      "loss": 0.768,
+      "step": 4580
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.7645313143730164,
+      "learning_rate": 5.464646464646465e-06,
+      "loss": 0.4667,
+      "step": 4590
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.8193786144256592,
+      "learning_rate": 5.4545454545454545e-06,
+      "loss": 0.4173,
+      "step": 4600
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.8643292188644409,
+      "learning_rate": 5.444444444444445e-06,
+      "loss": 0.5369,
+      "step": 4610
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.9830732941627502,
+      "learning_rate": 5.434343434343434e-06,
+      "loss": 0.5983,
+      "step": 4620
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.591908931732178,
+      "learning_rate": 5.424242424242425e-06,
+      "loss": 0.5857,
+      "step": 4630
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.289383888244629,
+      "learning_rate": 5.414141414141415e-06,
+      "loss": 0.4864,
+      "step": 4640
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 2.4989371299743652,
+      "learning_rate": 5.404040404040405e-06,
+      "loss": 0.6042,
+      "step": 4650
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.448102951049805,
+      "learning_rate": 5.3939393939393945e-06,
+      "loss": 0.3964,
+      "step": 4660
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.414404392242432,
+      "learning_rate": 5.383838383838385e-06,
+      "loss": 0.6371,
+      "step": 4670
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.583644866943359,
+      "learning_rate": 5.373737373737374e-06,
+      "loss": 0.478,
+      "step": 4680
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.100620985031128,
+      "learning_rate": 5.3636363636363645e-06,
+      "loss": 0.2518,
+      "step": 4690
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.0264525413513184,
+      "learning_rate": 5.353535353535354e-06,
+      "loss": 0.5209,
+      "step": 4700
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.579074501991272,
+      "learning_rate": 5.343434343434344e-06,
+      "loss": 0.6922,
+      "step": 4710
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.7735052108764648,
+      "learning_rate": 5.333333333333334e-06,
+      "loss": 0.5191,
+      "step": 4720
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.306868076324463,
+      "learning_rate": 5.323232323232324e-06,
+      "loss": 0.6043,
+      "step": 4730
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.187537908554077,
+      "learning_rate": 5.313131313131313e-06,
+      "loss": 0.419,
+      "step": 4740
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 2.8361172676086426,
+      "learning_rate": 5.303030303030303e-06,
+      "loss": 0.4583,
+      "step": 4750
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.10423949360847473,
+      "learning_rate": 5.292929292929293e-06,
+      "loss": 0.3714,
+      "step": 4760
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.2612528800964355,
+      "learning_rate": 5.2828282828282825e-06,
+      "loss": 0.5742,
+      "step": 4770
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.243372440338135,
+      "learning_rate": 5.272727272727273e-06,
+      "loss": 0.8028,
+      "step": 4780
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.7587629556655884,
+      "learning_rate": 5.262626262626262e-06,
+      "loss": 0.477,
+      "step": 4790
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.8695833683013916,
+      "learning_rate": 5.252525252525253e-06,
+      "loss": 0.4379,
+      "step": 4800
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.1608699560165405,
+      "learning_rate": 5.242424242424244e-06,
+      "loss": 0.4925,
+      "step": 4810
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.7934430837631226,
+      "learning_rate": 5.232323232323233e-06,
+      "loss": 0.3633,
+      "step": 4820
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.8061491250991821,
+      "learning_rate": 5.2222222222222226e-06,
+      "loss": 0.4519,
+      "step": 4830
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.7798353433609009,
+      "learning_rate": 5.212121212121213e-06,
+      "loss": 0.6176,
+      "step": 4840
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 2.436166524887085,
+      "learning_rate": 5.202020202020202e-06,
+      "loss": 0.5651,
+      "step": 4850
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.363344669342041,
+      "learning_rate": 5.191919191919193e-06,
+      "loss": 0.4536,
+      "step": 4860
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 2.3978679180145264,
+      "learning_rate": 5.181818181818182e-06,
+      "loss": 0.5007,
+      "step": 4870
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.5228248834609985,
+      "learning_rate": 5.171717171717172e-06,
+      "loss": 0.5675,
+      "step": 4880
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.1637253761291504,
+      "learning_rate": 5.161616161616162e-06,
+      "loss": 0.5289,
+      "step": 4890
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.7931678891181946,
+      "learning_rate": 5.151515151515152e-06,
+      "loss": 0.4039,
+      "step": 4900
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.720406532287598,
+      "learning_rate": 5.1414141414141415e-06,
+      "loss": 0.663,
+      "step": 4910
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.9488446712493896,
+      "learning_rate": 5.131313131313132e-06,
+      "loss": 0.7561,
+      "step": 4920
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.417807936668396,
+      "learning_rate": 5.121212121212121e-06,
+      "loss": 0.385,
+      "step": 4930
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.4536147117614746,
+      "learning_rate": 5.1111111111111115e-06,
+      "loss": 0.5771,
+      "step": 4940
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 2.030744791030884,
+      "learning_rate": 5.101010101010101e-06,
+      "loss": 0.2427,
+      "step": 4950
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.7181262969970703,
+      "learning_rate": 5.090909090909091e-06,
+      "loss": 0.4993,
+      "step": 4960
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.121447801589966,
+      "learning_rate": 5.0808080808080815e-06,
+      "loss": 0.5866,
+      "step": 4970
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.3056178092956543,
+      "learning_rate": 5.070707070707072e-06,
+      "loss": 0.397,
+      "step": 4980
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 2.310112714767456,
+      "learning_rate": 5.060606060606061e-06,
+      "loss": 0.3125,
+      "step": 4990
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.247709274291992,
+      "learning_rate": 5.0505050505050515e-06,
+      "loss": 0.3879,
+      "step": 5000
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.0834882259368896,
+      "learning_rate": 5.040404040404041e-06,
+      "loss": 0.4569,
+      "step": 5010
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.565437912940979,
+      "learning_rate": 5.030303030303031e-06,
+      "loss": 0.5117,
+      "step": 5020
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.598476529121399,
+      "learning_rate": 5.020202020202021e-06,
+      "loss": 0.3712,
+      "step": 5030
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 2.3510398864746094,
+      "learning_rate": 5.010101010101011e-06,
+      "loss": 0.2965,
+      "step": 5040
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.4666527807712555,
+      "learning_rate": 5e-06,
+      "loss": 0.3273,
+      "step": 5050
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.44954514503479004,
+      "learning_rate": 4.98989898989899e-06,
+      "loss": 0.3681,
+      "step": 5060
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.983162879943848,
+      "learning_rate": 4.97979797979798e-06,
+      "loss": 0.5314,
+      "step": 5070
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.5216596126556396,
+      "learning_rate": 4.9696969696969696e-06,
+      "loss": 0.3539,
+      "step": 5080
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.6397231221199036,
+      "learning_rate": 4.95959595959596e-06,
+      "loss": 0.3995,
+      "step": 5090
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 3.938267230987549,
+      "learning_rate": 4.94949494949495e-06,
+      "loss": 0.6112,
+      "step": 5100
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 2.2102057933807373,
+      "learning_rate": 4.93939393939394e-06,
+      "loss": 0.3688,
+      "step": 5110
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.375343918800354,
+      "learning_rate": 4.92929292929293e-06,
+      "loss": 0.5569,
+      "step": 5120
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.578484058380127,
+      "learning_rate": 4.919191919191919e-06,
+      "loss": 0.5205,
+      "step": 5130
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.618846416473389,
+      "learning_rate": 4.90909090909091e-06,
+      "loss": 0.6503,
+      "step": 5140
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 2.470555067062378,
+      "learning_rate": 4.898989898989899e-06,
+      "loss": 0.4006,
+      "step": 5150
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.930334448814392,
+      "learning_rate": 4.888888888888889e-06,
+      "loss": 0.5905,
+      "step": 5160
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.3920586109161377,
+      "learning_rate": 4.878787878787879e-06,
+      "loss": 0.6443,
+      "step": 5170
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.649546146392822,
+      "learning_rate": 4.868686868686869e-06,
+      "loss": 0.5574,
+      "step": 5180
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.529122829437256,
+      "learning_rate": 4.858585858585859e-06,
+      "loss": 0.4254,
+      "step": 5190
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 2.8728573322296143,
+      "learning_rate": 4.848484848484849e-06,
+      "loss": 0.5545,
+      "step": 5200
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 2.6086132526397705,
+      "learning_rate": 4.838383838383839e-06,
+      "loss": 0.4381,
+      "step": 5210
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.7182550430297852,
+      "learning_rate": 4.8282828282828285e-06,
+      "loss": 0.3298,
+      "step": 5220
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.6570979356765747,
+      "learning_rate": 4.818181818181819e-06,
+      "loss": 0.3376,
+      "step": 5230
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 2.316779375076294,
+      "learning_rate": 4.808080808080808e-06,
+      "loss": 0.467,
+      "step": 5240
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.5528929233551025,
+      "learning_rate": 4.7979797979797985e-06,
+      "loss": 0.4656,
+      "step": 5250
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.1301220655441284,
+      "learning_rate": 4.787878787878788e-06,
+      "loss": 0.4471,
+      "step": 5260
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.5341253280639648,
+      "learning_rate": 4.777777777777778e-06,
+      "loss": 0.2958,
+      "step": 5270
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.7554420232772827,
+      "learning_rate": 4.7676767676767685e-06,
+      "loss": 0.5693,
+      "step": 5280
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.294740200042725,
+      "learning_rate": 4.757575757575758e-06,
+      "loss": 0.4211,
+      "step": 5290
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.5546411275863647,
+      "learning_rate": 4.747474747474748e-06,
+      "loss": 0.5162,
+      "step": 5300
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 4.636204719543457,
+      "learning_rate": 4.737373737373738e-06,
+      "loss": 0.4967,
+      "step": 5310
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.9998151659965515,
+      "learning_rate": 4.727272727272728e-06,
+      "loss": 0.3829,
+      "step": 5320
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 7.0191216468811035,
+      "learning_rate": 4.717171717171717e-06,
+      "loss": 0.3572,
+      "step": 5330
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.0018032088410109282,
+      "learning_rate": 4.707070707070707e-06,
+      "loss": 0.4327,
+      "step": 5340
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.4587302207946777,
+      "learning_rate": 4.696969696969698e-06,
+      "loss": 0.3761,
+      "step": 5350
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.8359494209289551,
+      "learning_rate": 4.6868686868686874e-06,
+      "loss": 0.4907,
+      "step": 5360
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.3724963665008545,
+      "learning_rate": 4.676767676767677e-06,
+      "loss": 0.4124,
+      "step": 5370
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.6787943840026855,
+      "learning_rate": 4.666666666666667e-06,
+      "loss": 0.2109,
+      "step": 5380
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.7160095572471619,
+      "learning_rate": 4.656565656565657e-06,
+      "loss": 0.4117,
+      "step": 5390
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.2293567657470703,
+      "learning_rate": 4.646464646464647e-06,
+      "loss": 0.6078,
+      "step": 5400
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.7237870693206787,
+      "learning_rate": 4.636363636363636e-06,
+      "loss": 0.5515,
+      "step": 5410
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.36238169670105,
+      "learning_rate": 4.626262626262627e-06,
+      "loss": 0.579,
+      "step": 5420
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.5351545810699463,
+      "learning_rate": 4.616161616161616e-06,
+      "loss": 0.5916,
+      "step": 5430
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.963181257247925,
+      "learning_rate": 4.606060606060606e-06,
+      "loss": 0.335,
+      "step": 5440
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.0211395025253296,
+      "learning_rate": 4.595959595959597e-06,
+      "loss": 0.4902,
+      "step": 5450
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.6900724172592163,
+      "learning_rate": 4.585858585858586e-06,
+      "loss": 0.5184,
+      "step": 5460
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.763094425201416,
+      "learning_rate": 4.575757575757576e-06,
+      "loss": 0.4821,
+      "step": 5470
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.7416090965270996,
+      "learning_rate": 4.565656565656566e-06,
+      "loss": 0.7305,
+      "step": 5480
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.6166847944259644,
+      "learning_rate": 4.555555555555556e-06,
+      "loss": 0.5848,
+      "step": 5490
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.6287155151367188,
+      "learning_rate": 4.5454545454545455e-06,
+      "loss": 0.3352,
+      "step": 5500
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.130851149559021,
+      "learning_rate": 4.535353535353536e-06,
+      "loss": 0.5486,
+      "step": 5510
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.963855504989624,
+      "learning_rate": 4.525252525252526e-06,
+      "loss": 0.4826,
+      "step": 5520
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.296228051185608,
+      "learning_rate": 4.5151515151515155e-06,
+      "loss": 0.5068,
+      "step": 5530
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.7077557444572449,
+      "learning_rate": 4.505050505050506e-06,
+      "loss": 0.4431,
+      "step": 5540
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.7670864462852478,
+      "learning_rate": 4.494949494949495e-06,
+      "loss": 0.4496,
+      "step": 5550
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 3.895642042160034,
+      "learning_rate": 4.4848484848484855e-06,
+      "loss": 0.3879,
+      "step": 5560
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 3.4866116046905518,
+      "learning_rate": 4.474747474747475e-06,
+      "loss": 0.4512,
+      "step": 5570
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.829552173614502,
+      "learning_rate": 4.464646464646465e-06,
+      "loss": 0.3798,
+      "step": 5580
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 2.2859883308410645,
+      "learning_rate": 4.454545454545455e-06,
+      "loss": 0.388,
+      "step": 5590
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.3286418914794922,
+      "learning_rate": 4.444444444444444e-06,
+      "loss": 0.5353,
+      "step": 5600
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.3154336214065552,
+      "learning_rate": 4.434343434343435e-06,
+      "loss": 0.3768,
+      "step": 5610
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.2424612045288086,
+      "learning_rate": 4.424242424242425e-06,
+      "loss": 0.4719,
+      "step": 5620
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.6577768921852112,
+      "learning_rate": 4.414141414141415e-06,
+      "loss": 0.4158,
+      "step": 5630
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.895909309387207,
+      "learning_rate": 4.4040404040404044e-06,
+      "loss": 0.3556,
+      "step": 5640
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 11.399181365966797,
+      "learning_rate": 4.393939393939394e-06,
+      "loss": 0.4352,
+      "step": 5650
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 3.739205837249756,
+      "learning_rate": 4.383838383838384e-06,
+      "loss": 0.571,
+      "step": 5660
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.779272437095642,
+      "learning_rate": 4.373737373737374e-06,
+      "loss": 0.4831,
+      "step": 5670
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.9623099565505981,
+      "learning_rate": 4.363636363636364e-06,
+      "loss": 0.4237,
+      "step": 5680
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.008009433746338,
+      "learning_rate": 4.353535353535353e-06,
+      "loss": 0.3931,
+      "step": 5690
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.515387535095215,
+      "learning_rate": 4.343434343434344e-06,
+      "loss": 0.3997,
+      "step": 5700
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.7498732805252075,
+      "learning_rate": 4.333333333333334e-06,
+      "loss": 0.3813,
+      "step": 5710
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.6171377301216125,
+      "learning_rate": 4.323232323232323e-06,
+      "loss": 0.4299,
+      "step": 5720
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 2.0732200145721436,
+      "learning_rate": 4.313131313131314e-06,
+      "loss": 0.5707,
+      "step": 5730
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.580662727355957,
+      "learning_rate": 4.303030303030303e-06,
+      "loss": 0.4222,
+      "step": 5740
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.307980060577393,
+      "learning_rate": 4.292929292929293e-06,
+      "loss": 0.625,
+      "step": 5750
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.0482300519943237,
+      "learning_rate": 4.282828282828283e-06,
+      "loss": 0.4784,
+      "step": 5760
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.34380489587783813,
+      "learning_rate": 4.272727272727273e-06,
+      "loss": 0.3402,
+      "step": 5770
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.853585124015808,
+      "learning_rate": 4.262626262626263e-06,
+      "loss": 0.6769,
+      "step": 5780
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.7492414712905884,
+      "learning_rate": 4.252525252525253e-06,
+      "loss": 0.5447,
+      "step": 5790
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.7095075845718384,
+      "learning_rate": 4.242424242424243e-06,
+      "loss": 0.8639,
+      "step": 5800
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.8637481927871704,
+      "learning_rate": 4.2323232323232325e-06,
+      "loss": 0.5604,
+      "step": 5810
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 4.048046112060547,
+      "learning_rate": 4.222222222222223e-06,
+      "loss": 0.4485,
+      "step": 5820
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.0044870376586914,
+      "learning_rate": 4.212121212121212e-06,
+      "loss": 0.2267,
+      "step": 5830
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 2.193854808807373,
+      "learning_rate": 4.2020202020202026e-06,
+      "loss": 0.321,
+      "step": 5840
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.7237836122512817,
+      "learning_rate": 4.191919191919192e-06,
+      "loss": 0.4881,
+      "step": 5850
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.00034179864451289177,
+      "learning_rate": 4.181818181818182e-06,
+      "loss": 0.5096,
+      "step": 5860
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.6282609701156616,
+      "learning_rate": 4.1717171717171726e-06,
+      "loss": 0.5055,
+      "step": 5870
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.46074479818344116,
+      "learning_rate": 4.161616161616162e-06,
+      "loss": 0.5646,
+      "step": 5880
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.0498151779174805,
+      "learning_rate": 4.151515151515152e-06,
+      "loss": 0.5992,
+      "step": 5890
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 7.670099258422852,
+      "learning_rate": 4.141414141414142e-06,
+      "loss": 0.2634,
+      "step": 5900
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.9730322957038879,
+      "learning_rate": 4.131313131313132e-06,
+      "loss": 0.4204,
+      "step": 5910
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 3.937800884246826,
+      "learning_rate": 4.1212121212121215e-06,
+      "loss": 0.5122,
+      "step": 5920
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.5764747858047485,
+      "learning_rate": 4.111111111111111e-06,
+      "loss": 0.3836,
+      "step": 5930
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.066262722015381,
+      "learning_rate": 4.101010101010101e-06,
+      "loss": 0.6885,
+      "step": 5940
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 2.2596259117126465,
+      "learning_rate": 4.0909090909090915e-06,
+      "loss": 0.429,
+      "step": 5950
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.8355130553245544,
+      "learning_rate": 4.080808080808081e-06,
+      "loss": 0.3883,
+      "step": 5960
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.9566431045532227,
+      "learning_rate": 4.070707070707071e-06,
+      "loss": 0.4395,
+      "step": 5970
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 2.293888568878174,
+      "learning_rate": 4.060606060606061e-06,
+      "loss": 0.4456,
+      "step": 5980
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.1133636236190796,
+      "learning_rate": 4.050505050505051e-06,
+      "loss": 0.4765,
+      "step": 5990
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.47316935658454895,
+      "learning_rate": 4.04040404040404e-06,
+      "loss": 0.4218,
+      "step": 6000
+    },
+    {
+      "epoch": 0.3,
+      "eval_loss": 0.49143511056900024,
+      "eval_runtime": 296.3065,
+      "eval_samples_per_second": 3.375,
+      "eval_steps_per_second": 3.375,
+      "step": 6000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 2000,
+  "total_flos": 2.44808253702144e+17,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}