diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4269 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.3,
+  "eval_steps": 1000,
+  "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": 31.81644058227539,
+      "learning_rate": 8e-08,
+      "loss": 1.5083,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 16.249370574951172,
+      "learning_rate": 1.7000000000000001e-07,
+      "loss": 1.2724,
+      "step": 20
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 16.85301971435547,
+      "learning_rate": 2.7e-07,
+      "loss": 1.2571,
+      "step": 30
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 12.382726669311523,
+      "learning_rate": 3.7e-07,
+      "loss": 1.101,
+      "step": 40
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 43.00498962402344,
+      "learning_rate": 4.6999999999999995e-07,
+      "loss": 1.2847,
+      "step": 50
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 16.686534881591797,
+      "learning_rate": 5.6e-07,
+      "loss": 0.8594,
+      "step": 60
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 23.844993591308594,
+      "learning_rate": 6.6e-07,
+      "loss": 0.7645,
+      "step": 70
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 15.943971633911133,
+      "learning_rate": 7.599999999999999e-07,
+      "loss": 0.7681,
+      "step": 80
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 12.707146644592285,
+      "learning_rate": 8.599999999999999e-07,
+      "loss": 0.8728,
+      "step": 90
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 13.066338539123535,
+      "learning_rate": 9.6e-07,
+      "loss": 0.7508,
+      "step": 100
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 21.36861801147461,
+      "learning_rate": 9.993939393939394e-07,
+      "loss": 0.999,
+      "step": 110
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 23.451967239379883,
+      "learning_rate": 9.983838383838383e-07,
+      "loss": 0.8396,
+      "step": 120
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 14.662145614624023,
+      "learning_rate": 9.973737373737373e-07,
+      "loss": 1.0884,
+      "step": 130
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 9.530464172363281,
+      "learning_rate": 9.963636363636362e-07,
+      "loss": 0.8505,
+      "step": 140
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 30.15620231628418,
+      "learning_rate": 9.953535353535352e-07,
+      "loss": 0.971,
+      "step": 150
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 43.50494384765625,
+      "learning_rate": 9.943434343434343e-07,
+      "loss": 0.9758,
+      "step": 160
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 18.704565048217773,
+      "learning_rate": 9.933333333333333e-07,
+      "loss": 1.1095,
+      "step": 170
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 51.077552795410156,
+      "learning_rate": 9.923232323232322e-07,
+      "loss": 1.0251,
+      "step": 180
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 32.405906677246094,
+      "learning_rate": 9.913131313131314e-07,
+      "loss": 0.8793,
+      "step": 190
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 35.2808952331543,
+      "learning_rate": 9.903030303030303e-07,
+      "loss": 0.6785,
+      "step": 200
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 17.738399505615234,
+      "learning_rate": 9.892929292929293e-07,
+      "loss": 0.8258,
+      "step": 210
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 8.202994346618652,
+      "learning_rate": 9.882828282828282e-07,
+      "loss": 0.8756,
+      "step": 220
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 16.26487159729004,
+      "learning_rate": 9.872727272727272e-07,
+      "loss": 1.0467,
+      "step": 230
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 32.37431335449219,
+      "learning_rate": 9.862626262626263e-07,
+      "loss": 0.9119,
+      "step": 240
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.586151599884033,
+      "learning_rate": 9.852525252525253e-07,
+      "loss": 0.8556,
+      "step": 250
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 16.249475479125977,
+      "learning_rate": 9.842424242424242e-07,
+      "loss": 0.7945,
+      "step": 260
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 60.965965270996094,
+      "learning_rate": 9.832323232323231e-07,
+      "loss": 1.0668,
+      "step": 270
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 19.940616607666016,
+      "learning_rate": 9.82222222222222e-07,
+      "loss": 0.6501,
+      "step": 280
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 26.526460647583008,
+      "learning_rate": 9.812121212121212e-07,
+      "loss": 1.0423,
+      "step": 290
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 27.455795288085938,
+      "learning_rate": 9.802020202020202e-07,
+      "loss": 0.6693,
+      "step": 300
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 18.113489151000977,
+      "learning_rate": 9.791919191919191e-07,
+      "loss": 0.4784,
+      "step": 310
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 29.75452423095703,
+      "learning_rate": 9.78181818181818e-07,
+      "loss": 0.5004,
+      "step": 320
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 24.82709503173828,
+      "learning_rate": 9.77171717171717e-07,
+      "loss": 0.8543,
+      "step": 330
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 22.23027801513672,
+      "learning_rate": 9.761616161616162e-07,
+      "loss": 0.796,
+      "step": 340
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 9.593890190124512,
+      "learning_rate": 9.751515151515151e-07,
+      "loss": 0.672,
+      "step": 350
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 19.550216674804688,
+      "learning_rate": 9.74141414141414e-07,
+      "loss": 0.5915,
+      "step": 360
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 18.101367950439453,
+      "learning_rate": 9.731313131313132e-07,
+      "loss": 0.8967,
+      "step": 370
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.973883152008057,
+      "learning_rate": 9.721212121212122e-07,
+      "loss": 0.7841,
+      "step": 380
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 18.076982498168945,
+      "learning_rate": 9.711111111111111e-07,
+      "loss": 0.7783,
+      "step": 390
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.5223013162612915,
+      "learning_rate": 9.7010101010101e-07,
+      "loss": 0.8647,
+      "step": 400
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 14.444634437561035,
+      "learning_rate": 9.69090909090909e-07,
+      "loss": 0.6745,
+      "step": 410
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 41.90908432006836,
+      "learning_rate": 9.680808080808082e-07,
+      "loss": 1.0749,
+      "step": 420
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 27.912317276000977,
+      "learning_rate": 9.67070707070707e-07,
+      "loss": 0.5588,
+      "step": 430
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 22.31749725341797,
+      "learning_rate": 9.66060606060606e-07,
+      "loss": 0.7683,
+      "step": 440
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 57.88829040527344,
+      "learning_rate": 9.65050505050505e-07,
+      "loss": 0.5442,
+      "step": 450
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 43.25718688964844,
+      "learning_rate": 9.64040404040404e-07,
+      "loss": 0.8028,
+      "step": 460
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 21.001134872436523,
+      "learning_rate": 9.630303030303029e-07,
+      "loss": 0.7472,
+      "step": 470
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 13.919650077819824,
+      "learning_rate": 9.62020202020202e-07,
+      "loss": 0.5318,
+      "step": 480
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 26.845335006713867,
+      "learning_rate": 9.61010101010101e-07,
+      "loss": 1.012,
+      "step": 490
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 20.0955867767334,
+      "learning_rate": 9.6e-07,
+      "loss": 0.9174,
+      "step": 500
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 9.6511812210083,
+      "learning_rate": 9.589898989898989e-07,
+      "loss": 0.691,
+      "step": 510
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 12.875068664550781,
+      "learning_rate": 9.579797979797978e-07,
+      "loss": 0.7781,
+      "step": 520
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 28.99612045288086,
+      "learning_rate": 9.56969696969697e-07,
+      "loss": 0.7791,
+      "step": 530
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 47.6656379699707,
+      "learning_rate": 9.55959595959596e-07,
+      "loss": 0.6166,
+      "step": 540
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 59.9251708984375,
+      "learning_rate": 9.549494949494948e-07,
+      "loss": 0.6836,
+      "step": 550
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 34.26139831542969,
+      "learning_rate": 9.53939393939394e-07,
+      "loss": 0.4317,
+      "step": 560
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 47.75581359863281,
+      "learning_rate": 9.529292929292929e-07,
+      "loss": 0.5685,
+      "step": 570
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 63.76834487915039,
+      "learning_rate": 9.519191919191919e-07,
+      "loss": 0.8805,
+      "step": 580
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 22.971323013305664,
+      "learning_rate": 9.509090909090908e-07,
+      "loss": 0.8103,
+      "step": 590
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 42.151161193847656,
+      "learning_rate": 9.498989898989899e-07,
+      "loss": 0.7563,
+      "step": 600
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 13.12915325164795,
+      "learning_rate": 9.488888888888888e-07,
+      "loss": 0.6844,
+      "step": 610
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 37.901973724365234,
+      "learning_rate": 9.478787878787879e-07,
+      "loss": 0.7467,
+      "step": 620
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 29.114883422851562,
+      "learning_rate": 9.468686868686868e-07,
+      "loss": 0.7759,
+      "step": 630
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 37.42153549194336,
+      "learning_rate": 9.458585858585858e-07,
+      "loss": 0.7221,
+      "step": 640
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 46.287559509277344,
+      "learning_rate": 9.448484848484848e-07,
+      "loss": 0.7219,
+      "step": 650
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 76.34626770019531,
+      "learning_rate": 9.438383838383838e-07,
+      "loss": 0.6938,
+      "step": 660
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 20.480300903320312,
+      "learning_rate": 9.428282828282827e-07,
+      "loss": 0.8135,
+      "step": 670
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 9.508841514587402,
+      "learning_rate": 9.418181818181818e-07,
+      "loss": 0.6024,
+      "step": 680
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 29.385066986083984,
+      "learning_rate": 9.408080808080807e-07,
+      "loss": 0.8891,
+      "step": 690
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 13.06400203704834,
+      "learning_rate": 9.397979797979797e-07,
+      "loss": 0.6723,
+      "step": 700
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 36.94926071166992,
+      "learning_rate": 9.387878787878788e-07,
+      "loss": 0.5682,
+      "step": 710
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 30.849035263061523,
+      "learning_rate": 9.377777777777777e-07,
+      "loss": 0.6693,
+      "step": 720
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 27.781005859375,
+      "learning_rate": 9.367676767676768e-07,
+      "loss": 0.7705,
+      "step": 730
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.268486976623535,
+      "learning_rate": 9.357575757575757e-07,
+      "loss": 0.7672,
+      "step": 740
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 70.79887390136719,
+      "learning_rate": 9.347474747474747e-07,
+      "loss": 0.7617,
+      "step": 750
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 71.6796875,
+      "learning_rate": 9.337373737373737e-07,
+      "loss": 0.5425,
+      "step": 760
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 21.205890655517578,
+      "learning_rate": 9.327272727272727e-07,
+      "loss": 0.6768,
+      "step": 770
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 38.276634216308594,
+      "learning_rate": 9.317171717171717e-07,
+      "loss": 0.8681,
+      "step": 780
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 17.904993057250977,
+      "learning_rate": 9.307070707070707e-07,
+      "loss": 0.7128,
+      "step": 790
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 32.66667938232422,
+      "learning_rate": 9.296969696969696e-07,
+      "loss": 0.8662,
+      "step": 800
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.020589828491211,
+      "learning_rate": 9.286868686868687e-07,
+      "loss": 0.7344,
+      "step": 810
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 33.822608947753906,
+      "learning_rate": 9.276767676767676e-07,
+      "loss": 0.7825,
+      "step": 820
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 21.147552490234375,
+      "learning_rate": 9.266666666666665e-07,
+      "loss": 0.7716,
+      "step": 830
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 22.41349983215332,
+      "learning_rate": 9.256565656565656e-07,
+      "loss": 0.4673,
+      "step": 840
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 24.813737869262695,
+      "learning_rate": 9.246464646464645e-07,
+      "loss": 0.7137,
+      "step": 850
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 31.02164649963379,
+      "learning_rate": 9.236363636363636e-07,
+      "loss": 1.0195,
+      "step": 860
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 19.22023582458496,
+      "learning_rate": 9.226262626262625e-07,
+      "loss": 0.8201,
+      "step": 870
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 21.087970733642578,
+      "learning_rate": 9.216161616161616e-07,
+      "loss": 0.6296,
+      "step": 880
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 8.970023155212402,
+      "learning_rate": 9.206060606060606e-07,
+      "loss": 0.3712,
+      "step": 890
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 28.498889923095703,
+      "learning_rate": 9.195959595959596e-07,
+      "loss": 0.7953,
+      "step": 900
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 21.259479522705078,
+      "learning_rate": 9.185858585858585e-07,
+      "loss": 0.7346,
+      "step": 910
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 13.871064186096191,
+      "learning_rate": 9.175757575757576e-07,
+      "loss": 0.6644,
+      "step": 920
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 44.91512680053711,
+      "learning_rate": 9.165656565656565e-07,
+      "loss": 0.5467,
+      "step": 930
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 11.775423049926758,
+      "learning_rate": 9.155555555555556e-07,
+      "loss": 0.5149,
+      "step": 940
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 35.62663650512695,
+      "learning_rate": 9.145454545454545e-07,
+      "loss": 0.6002,
+      "step": 950
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 32.12046813964844,
+      "learning_rate": 9.135353535353535e-07,
+      "loss": 0.8875,
+      "step": 960
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 37.11963653564453,
+      "learning_rate": 9.126262626262626e-07,
+      "loss": 0.9515,
+      "step": 970
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 21.641096115112305,
+      "learning_rate": 9.116161616161616e-07,
+      "loss": 0.7849,
+      "step": 980
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 29.253921508789062,
+      "learning_rate": 9.106060606060606e-07,
+      "loss": 0.5926,
+      "step": 990
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 27.01972198486328,
+      "learning_rate": 9.095959595959596e-07,
+      "loss": 0.7061,
+      "step": 1000
+    },
+    {
+      "epoch": 0.05,
+      "eval_loss": 0.725902259349823,
+      "eval_runtime": 274.8847,
+      "eval_samples_per_second": 3.638,
+      "eval_steps_per_second": 3.638,
+      "step": 1000
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 16.84354019165039,
+      "learning_rate": 9.085858585858586e-07,
+      "loss": 0.6186,
+      "step": 1010
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 36.384700775146484,
+      "learning_rate": 9.075757575757576e-07,
+      "loss": 0.8362,
+      "step": 1020
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 37.640892028808594,
+      "learning_rate": 9.065656565656565e-07,
+      "loss": 0.7901,
+      "step": 1030
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 44.735076904296875,
+      "learning_rate": 9.055555555555556e-07,
+      "loss": 0.5116,
+      "step": 1040
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 46.5770263671875,
+      "learning_rate": 9.045454545454545e-07,
+      "loss": 0.7274,
+      "step": 1050
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.10406364500522614,
+      "learning_rate": 9.035353535353534e-07,
+      "loss": 0.4903,
+      "step": 1060
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 18.263145446777344,
+      "learning_rate": 9.025252525252525e-07,
+      "loss": 0.7371,
+      "step": 1070
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 33.42209243774414,
+      "learning_rate": 9.015151515151514e-07,
+      "loss": 0.8111,
+      "step": 1080
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 48.237525939941406,
+      "learning_rate": 9.005050505050504e-07,
+      "loss": 0.6233,
+      "step": 1090
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.5421327352523804,
+      "learning_rate": 8.994949494949494e-07,
+      "loss": 0.6609,
+      "step": 1100
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 29.761442184448242,
+      "learning_rate": 8.984848484848484e-07,
+      "loss": 0.4407,
+      "step": 1110
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 31.435007095336914,
+      "learning_rate": 8.974747474747474e-07,
+      "loss": 0.7207,
+      "step": 1120
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 28.20315170288086,
+      "learning_rate": 8.964646464646465e-07,
+      "loss": 0.6879,
+      "step": 1130
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 19.431041717529297,
+      "learning_rate": 8.954545454545454e-07,
+      "loss": 0.7292,
+      "step": 1140
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 40.8221435546875,
+      "learning_rate": 8.944444444444445e-07,
+      "loss": 0.9112,
+      "step": 1150
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.42428731918335,
+      "learning_rate": 8.934343434343434e-07,
+      "loss": 0.484,
+      "step": 1160
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 24.146841049194336,
+      "learning_rate": 8.924242424242425e-07,
+      "loss": 0.4881,
+      "step": 1170
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 52.02534866333008,
+      "learning_rate": 8.914141414141414e-07,
+      "loss": 0.8042,
+      "step": 1180
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 26.81230926513672,
+      "learning_rate": 8.904040404040404e-07,
+      "loss": 0.6995,
+      "step": 1190
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 9.775701522827148,
+      "learning_rate": 8.893939393939394e-07,
+      "loss": 0.8203,
+      "step": 1200
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 20.65981674194336,
+      "learning_rate": 8.883838383838383e-07,
+      "loss": 0.9154,
+      "step": 1210
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 20.086986541748047,
+      "learning_rate": 8.873737373737373e-07,
+      "loss": 0.4731,
+      "step": 1220
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 20.6555233001709,
+      "learning_rate": 8.863636363636363e-07,
+      "loss": 0.7798,
+      "step": 1230
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 13.60240364074707,
+      "learning_rate": 8.853535353535353e-07,
+      "loss": 0.6382,
+      "step": 1240
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 20.39061737060547,
+      "learning_rate": 8.843434343434343e-07,
+      "loss": 0.6077,
+      "step": 1250
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 14.251206398010254,
+      "learning_rate": 8.833333333333333e-07,
+      "loss": 0.5736,
+      "step": 1260
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 63.768611907958984,
+      "learning_rate": 8.823232323232322e-07,
+      "loss": 0.7059,
+      "step": 1270
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 44.81045150756836,
+      "learning_rate": 8.813131313131313e-07,
+      "loss": 0.6352,
+      "step": 1280
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 15.952017784118652,
+      "learning_rate": 8.803030303030302e-07,
+      "loss": 0.7523,
+      "step": 1290
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 27.234148025512695,
+      "learning_rate": 8.792929292929293e-07,
+      "loss": 0.4924,
+      "step": 1300
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 44.29439163208008,
+      "learning_rate": 8.782828282828283e-07,
+      "loss": 0.6422,
+      "step": 1310
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 35.106658935546875,
+      "learning_rate": 8.772727272727273e-07,
+      "loss": 0.655,
+      "step": 1320
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.1618399620056152,
+      "learning_rate": 8.762626262626263e-07,
+      "loss": 0.6463,
+      "step": 1330
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 36.30659866333008,
+      "learning_rate": 8.752525252525253e-07,
+      "loss": 0.5618,
+      "step": 1340
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 18.705991744995117,
+      "learning_rate": 8.742424242424242e-07,
+      "loss": 0.5887,
+      "step": 1350
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.828080654144287,
+      "learning_rate": 8.732323232323232e-07,
+      "loss": 0.6611,
+      "step": 1360
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 47.979820251464844,
+      "learning_rate": 8.722222222222222e-07,
+      "loss": 0.4975,
+      "step": 1370
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 25.263946533203125,
+      "learning_rate": 8.712121212121211e-07,
+      "loss": 0.761,
+      "step": 1380
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.21978527307510376,
+      "learning_rate": 8.702020202020202e-07,
+      "loss": 0.5013,
+      "step": 1390
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 15.600090026855469,
+      "learning_rate": 8.691919191919191e-07,
+      "loss": 0.5375,
+      "step": 1400
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 25.815698623657227,
+      "learning_rate": 8.681818181818182e-07,
+      "loss": 0.8176,
+      "step": 1410
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 44.83120346069336,
+      "learning_rate": 8.671717171717171e-07,
+      "loss": 0.5207,
+      "step": 1420
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 20.984037399291992,
+      "learning_rate": 8.661616161616161e-07,
+      "loss": 0.5024,
+      "step": 1430
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 26.290699005126953,
+      "learning_rate": 8.651515151515151e-07,
+      "loss": 0.7017,
+      "step": 1440
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 23.3577880859375,
+      "learning_rate": 8.641414141414141e-07,
+      "loss": 0.6419,
+      "step": 1450
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 21.74049949645996,
+      "learning_rate": 8.63131313131313e-07,
+      "loss": 0.5426,
+      "step": 1460
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.5753927230834961,
+      "learning_rate": 8.62121212121212e-07,
+      "loss": 0.6417,
+      "step": 1470
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 36.96406936645508,
+      "learning_rate": 8.611111111111111e-07,
+      "loss": 0.6999,
+      "step": 1480
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.05758470296859741,
+      "learning_rate": 8.601010101010102e-07,
+      "loss": 0.816,
+      "step": 1490
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 51.20552062988281,
+      "learning_rate": 8.590909090909091e-07,
+      "loss": 0.7171,
+      "step": 1500
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 34.43236541748047,
+      "learning_rate": 8.58080808080808e-07,
+      "loss": 0.7448,
+      "step": 1510
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 89.9273681640625,
+      "learning_rate": 8.570707070707071e-07,
+      "loss": 0.6324,
+      "step": 1520
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 9.697078704833984,
+      "learning_rate": 8.56060606060606e-07,
+      "loss": 0.7951,
+      "step": 1530
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 28.903240203857422,
+      "learning_rate": 8.55050505050505e-07,
+      "loss": 0.6224,
+      "step": 1540
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 19.746826171875,
+      "learning_rate": 8.54040404040404e-07,
+      "loss": 0.4143,
+      "step": 1550
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 33.706146240234375,
+      "learning_rate": 8.53030303030303e-07,
+      "loss": 0.6803,
+      "step": 1560
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 24.144351959228516,
+      "learning_rate": 8.52020202020202e-07,
+      "loss": 0.6015,
+      "step": 1570
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 34.93477249145508,
+      "learning_rate": 8.51010101010101e-07,
+      "loss": 0.5599,
+      "step": 1580
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 31.863859176635742,
+      "learning_rate": 8.499999999999999e-07,
+      "loss": 0.7127,
+      "step": 1590
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 48.534549713134766,
+      "learning_rate": 8.48989898989899e-07,
+      "loss": 0.7777,
+      "step": 1600
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 35.08165740966797,
+      "learning_rate": 8.479797979797979e-07,
+      "loss": 0.6571,
+      "step": 1610
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 33.532325744628906,
+      "learning_rate": 8.469696969696968e-07,
+      "loss": 0.7386,
+      "step": 1620
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.014142200350761414,
+      "learning_rate": 8.459595959595959e-07,
+      "loss": 0.5505,
+      "step": 1630
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 29.99919319152832,
+      "learning_rate": 8.449494949494948e-07,
+      "loss": 0.4274,
+      "step": 1640
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 13.206497192382812,
+      "learning_rate": 8.43939393939394e-07,
+      "loss": 0.4399,
+      "step": 1650
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 17.408052444458008,
+      "learning_rate": 8.429292929292929e-07,
+      "loss": 0.5262,
+      "step": 1660
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 14.05802059173584,
+      "learning_rate": 8.419191919191919e-07,
+      "loss": 0.5503,
+      "step": 1670
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 13.401651382446289,
+      "learning_rate": 8.409090909090909e-07,
+      "loss": 0.813,
+      "step": 1680
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 28.40484619140625,
+      "learning_rate": 8.398989898989899e-07,
+      "loss": 0.7321,
+      "step": 1690
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 39.133670806884766,
+      "learning_rate": 8.388888888888888e-07,
+      "loss": 0.7129,
+      "step": 1700
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 26.99220848083496,
+      "learning_rate": 8.378787878787879e-07,
+      "loss": 0.645,
+      "step": 1710
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 22.875778198242188,
+      "learning_rate": 8.368686868686868e-07,
+      "loss": 0.6779,
+      "step": 1720
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 37.771705627441406,
+      "learning_rate": 8.358585858585859e-07,
+      "loss": 0.6104,
+      "step": 1730
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 17.48222541809082,
+      "learning_rate": 8.348484848484848e-07,
+      "loss": 0.5982,
+      "step": 1740
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 13.396077156066895,
+      "learning_rate": 8.338383838383838e-07,
+      "loss": 0.6295,
+      "step": 1750
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 48.97739791870117,
+      "learning_rate": 8.328282828282828e-07,
+      "loss": 0.8439,
+      "step": 1760
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 74.33853149414062,
+      "learning_rate": 8.318181818181817e-07,
+      "loss": 1.0566,
+      "step": 1770
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 7.2910051345825195,
+      "learning_rate": 8.308080808080807e-07,
+      "loss": 0.5226,
+      "step": 1780
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 27.389278411865234,
+      "learning_rate": 8.297979797979797e-07,
+      "loss": 1.1226,
+      "step": 1790
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 9.284111976623535,
+      "learning_rate": 8.287878787878787e-07,
+      "loss": 0.4101,
+      "step": 1800
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 8.962563514709473,
+      "learning_rate": 8.277777777777777e-07,
+      "loss": 0.5349,
+      "step": 1810
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 47.56216049194336,
+      "learning_rate": 8.267676767676768e-07,
+      "loss": 0.6863,
+      "step": 1820
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 31.547109603881836,
+      "learning_rate": 8.257575757575757e-07,
+      "loss": 0.7484,
+      "step": 1830
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 8.890266418457031,
+      "learning_rate": 8.247474747474748e-07,
+      "loss": 0.7518,
+      "step": 1840
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 25.44080352783203,
+      "learning_rate": 8.237373737373737e-07,
+      "loss": 0.8552,
+      "step": 1850
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 19.411556243896484,
+      "learning_rate": 8.227272727272727e-07,
+      "loss": 0.6285,
+      "step": 1860
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.5483574867248535,
+      "learning_rate": 8.217171717171717e-07,
+      "loss": 0.5478,
+      "step": 1870
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 22.145336151123047,
+      "learning_rate": 8.207070707070707e-07,
+      "loss": 0.7499,
+      "step": 1880
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 30.043731689453125,
+      "learning_rate": 8.196969696969697e-07,
+      "loss": 0.7252,
+      "step": 1890
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 39.9935417175293,
+      "learning_rate": 8.186868686868687e-07,
+      "loss": 0.7348,
+      "step": 1900
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.8807525634765625,
+      "learning_rate": 8.176767676767676e-07,
+      "loss": 0.381,
+      "step": 1910
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 16.082597732543945,
+      "learning_rate": 8.166666666666666e-07,
+      "loss": 0.6413,
+      "step": 1920
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 15.713370323181152,
+      "learning_rate": 8.156565656565656e-07,
+      "loss": 0.4765,
+      "step": 1930
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 43.45647048950195,
+      "learning_rate": 8.146464646464645e-07,
+      "loss": 0.6653,
+      "step": 1940
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 51.137046813964844,
+      "learning_rate": 8.136363636363636e-07,
+      "loss": 0.6915,
+      "step": 1950
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 33.2936897277832,
+      "learning_rate": 8.126262626262625e-07,
+      "loss": 0.4993,
+      "step": 1960
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 10.87304973602295,
+      "learning_rate": 8.116161616161616e-07,
+      "loss": 0.7893,
+      "step": 1970
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.774446427822113,
+      "learning_rate": 8.106060606060605e-07,
+      "loss": 0.5946,
+      "step": 1980
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 26.221094131469727,
+      "learning_rate": 8.095959595959596e-07,
+      "loss": 0.8534,
+      "step": 1990
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 46.51724624633789,
+      "learning_rate": 8.085858585858586e-07,
+      "loss": 0.4862,
+      "step": 2000
+    },
+    {
+      "epoch": 0.1,
+      "eval_loss": 0.6236868500709534,
+      "eval_runtime": 274.2428,
+      "eval_samples_per_second": 3.646,
+      "eval_steps_per_second": 3.646,
+      "step": 2000
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 15.500929832458496,
+      "learning_rate": 8.075757575757576e-07,
+      "loss": 0.6197,
+      "step": 2010
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 7.338476657867432,
+      "learning_rate": 8.065656565656565e-07,
+      "loss": 0.591,
+      "step": 2020
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 28.497867584228516,
+      "learning_rate": 8.055555555555556e-07,
+      "loss": 0.5593,
+      "step": 2030
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 15.371612548828125,
+      "learning_rate": 8.045454545454545e-07,
+      "loss": 0.53,
+      "step": 2040
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 38.33600616455078,
+      "learning_rate": 8.035353535353536e-07,
+      "loss": 1.0264,
+      "step": 2050
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 17.20992088317871,
+      "learning_rate": 8.025252525252525e-07,
+      "loss": 0.4638,
+      "step": 2060
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 58.250579833984375,
+      "learning_rate": 8.015151515151514e-07,
+      "loss": 0.5415,
+      "step": 2070
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 20.2476806640625,
+      "learning_rate": 8.005050505050505e-07,
+      "loss": 0.5162,
+      "step": 2080
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 27.523902893066406,
+      "learning_rate": 7.994949494949494e-07,
+      "loss": 0.4223,
+      "step": 2090
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 25.663818359375,
+      "learning_rate": 7.984848484848484e-07,
+      "loss": 0.6016,
+      "step": 2100
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 25.809614181518555,
+      "learning_rate": 7.974747474747474e-07,
+      "loss": 1.0505,
+      "step": 2110
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 52.98808670043945,
+      "learning_rate": 7.964646464646464e-07,
+      "loss": 1.1373,
+      "step": 2120
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 15.879914283752441,
+      "learning_rate": 7.954545454545454e-07,
+      "loss": 0.5527,
+      "step": 2130
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 16.313142776489258,
+      "learning_rate": 7.944444444444444e-07,
+      "loss": 0.4782,
+      "step": 2140
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 39.388511657714844,
+      "learning_rate": 7.934343434343433e-07,
+      "loss": 0.9302,
+      "step": 2150
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 27.657390594482422,
+      "learning_rate": 7.924242424242425e-07,
+      "loss": 0.4342,
+      "step": 2160
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.85066819190979,
+      "learning_rate": 7.914141414141414e-07,
+      "loss": 0.7398,
+      "step": 2170
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 20.403371810913086,
+      "learning_rate": 7.904040404040404e-07,
+      "loss": 0.7303,
+      "step": 2180
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 19.42263412475586,
+      "learning_rate": 7.893939393939394e-07,
+      "loss": 0.7445,
+      "step": 2190
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 21.36237144470215,
+      "learning_rate": 7.883838383838383e-07,
+      "loss": 0.6936,
+      "step": 2200
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 27.883874893188477,
+      "learning_rate": 7.873737373737374e-07,
+      "loss": 0.6665,
+      "step": 2210
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 16.943950653076172,
+      "learning_rate": 7.863636363636363e-07,
+      "loss": 0.6188,
+      "step": 2220
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 117.27821350097656,
+      "learning_rate": 7.853535353535353e-07,
+      "loss": 0.4547,
+      "step": 2230
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 16.697284698486328,
+      "learning_rate": 7.843434343434343e-07,
+      "loss": 0.3462,
+      "step": 2240
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 9.313762664794922,
+      "learning_rate": 7.833333333333333e-07,
+      "loss": 0.5729,
+      "step": 2250
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 43.60643005371094,
+      "learning_rate": 7.823232323232322e-07,
+      "loss": 0.8304,
+      "step": 2260
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 22.356792449951172,
+      "learning_rate": 7.813131313131313e-07,
+      "loss": 0.5437,
+      "step": 2270
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 32.53278732299805,
+      "learning_rate": 7.803030303030302e-07,
+      "loss": 1.0315,
+      "step": 2280
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 46.504573822021484,
+      "learning_rate": 7.792929292929293e-07,
+      "loss": 0.6465,
+      "step": 2290
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 7.453057765960693,
+      "learning_rate": 7.782828282828282e-07,
+      "loss": 0.6023,
+      "step": 2300
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 73.96940612792969,
+      "learning_rate": 7.772727272727272e-07,
+      "loss": 0.7558,
+      "step": 2310
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 31.267314910888672,
+      "learning_rate": 7.762626262626262e-07,
+      "loss": 0.527,
+      "step": 2320
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 41.19007873535156,
+      "learning_rate": 7.752525252525253e-07,
+      "loss": 0.7043,
+      "step": 2330
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 21.09952735900879,
+      "learning_rate": 7.742424242424243e-07,
+      "loss": 0.9584,
+      "step": 2340
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 42.14522933959961,
+      "learning_rate": 7.732323232323232e-07,
+      "loss": 0.5388,
+      "step": 2350
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 13.435091972351074,
+      "learning_rate": 7.722222222222222e-07,
+      "loss": 0.6348,
+      "step": 2360
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 26.72304916381836,
+      "learning_rate": 7.712121212121212e-07,
+      "loss": 0.3811,
+      "step": 2370
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 43.16832733154297,
+      "learning_rate": 7.702020202020202e-07,
+      "loss": 0.7426,
+      "step": 2380
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.5628132820129395,
+      "learning_rate": 7.691919191919191e-07,
+      "loss": 0.4845,
+      "step": 2390
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 52.14829635620117,
+      "learning_rate": 7.681818181818182e-07,
+      "loss": 0.6397,
+      "step": 2400
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 33.835205078125,
+      "learning_rate": 7.671717171717171e-07,
+      "loss": 0.6927,
+      "step": 2410
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 33.98677444458008,
+      "learning_rate": 7.661616161616161e-07,
+      "loss": 0.6525,
+      "step": 2420
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 23.454181671142578,
+      "learning_rate": 7.651515151515151e-07,
+      "loss": 0.7971,
+      "step": 2430
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 39.06905746459961,
+      "learning_rate": 7.641414141414141e-07,
+      "loss": 0.516,
+      "step": 2440
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 36.33759307861328,
+      "learning_rate": 7.631313131313131e-07,
+      "loss": 0.6733,
+      "step": 2450
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 39.87421417236328,
+      "learning_rate": 7.621212121212121e-07,
+      "loss": 0.827,
+      "step": 2460
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 27.0250244140625,
+      "learning_rate": 7.61111111111111e-07,
+      "loss": 0.7167,
+      "step": 2470
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 42.34687042236328,
+      "learning_rate": 7.6010101010101e-07,
+      "loss": 0.8127,
+      "step": 2480
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 20.025238037109375,
+      "learning_rate": 7.59090909090909e-07,
+      "loss": 0.5936,
+      "step": 2490
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.185176372528076,
+      "learning_rate": 7.580808080808081e-07,
+      "loss": 0.4552,
+      "step": 2500
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 10.323554039001465,
+      "learning_rate": 7.570707070707071e-07,
+      "loss": 0.6821,
+      "step": 2510
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 59.452205657958984,
+      "learning_rate": 7.56060606060606e-07,
+      "loss": 0.8375,
+      "step": 2520
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.1424858570098877,
+      "learning_rate": 7.550505050505051e-07,
+      "loss": 0.3893,
+      "step": 2530
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 19.907209396362305,
+      "learning_rate": 7.54040404040404e-07,
+      "loss": 0.6082,
+      "step": 2540
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.8457748889923096,
+      "learning_rate": 7.53030303030303e-07,
+      "loss": 0.7132,
+      "step": 2550
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 12.082324028015137,
+      "learning_rate": 7.52020202020202e-07,
+      "loss": 0.7951,
+      "step": 2560
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 24.198322296142578,
+      "learning_rate": 7.51010101010101e-07,
+      "loss": 0.5836,
+      "step": 2570
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 25.63511085510254,
+      "learning_rate": 7.5e-07,
+      "loss": 0.6352,
+      "step": 2580
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 24.314882278442383,
+      "learning_rate": 7.48989898989899e-07,
+      "loss": 0.7092,
+      "step": 2590
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 10.859787940979004,
+      "learning_rate": 7.479797979797979e-07,
+      "loss": 0.5931,
+      "step": 2600
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 13.863005638122559,
+      "learning_rate": 7.46969696969697e-07,
+      "loss": 0.5174,
+      "step": 2610
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 22.458776473999023,
+      "learning_rate": 7.459595959595959e-07,
+      "loss": 0.4968,
+      "step": 2620
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 49.36175537109375,
+      "learning_rate": 7.449494949494948e-07,
+      "loss": 0.6513,
+      "step": 2630
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 7.616428375244141,
+      "learning_rate": 7.439393939393939e-07,
+      "loss": 0.4914,
+      "step": 2640
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.416635036468506,
+      "learning_rate": 7.429292929292928e-07,
+      "loss": 0.2537,
+      "step": 2650
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 16.390323638916016,
+      "learning_rate": 7.419191919191918e-07,
+      "loss": 0.7159,
+      "step": 2660
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.848669052124023,
+      "learning_rate": 7.409090909090909e-07,
+      "loss": 0.4026,
+      "step": 2670
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 26.498138427734375,
+      "learning_rate": 7.398989898989899e-07,
+      "loss": 0.5316,
+      "step": 2680
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 20.28061866760254,
+      "learning_rate": 7.388888888888889e-07,
+      "loss": 0.8416,
+      "step": 2690
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 29.66952133178711,
+      "learning_rate": 7.378787878787879e-07,
+      "loss": 0.6887,
+      "step": 2700
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 18.624435424804688,
+      "learning_rate": 7.368686868686868e-07,
+      "loss": 0.4513,
+      "step": 2710
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 22.85711669921875,
+      "learning_rate": 7.358585858585859e-07,
+      "loss": 0.894,
+      "step": 2720
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 30.069570541381836,
+      "learning_rate": 7.348484848484848e-07,
+      "loss": 0.8273,
+      "step": 2730
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 52.604408264160156,
+      "learning_rate": 7.338383838383839e-07,
+      "loss": 0.6306,
+      "step": 2740
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 30.587234497070312,
+      "learning_rate": 7.328282828282828e-07,
+      "loss": 0.7653,
+      "step": 2750
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 25.937252044677734,
+      "learning_rate": 7.318181818181818e-07,
+      "loss": 0.7205,
+      "step": 2760
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 18.365734100341797,
+      "learning_rate": 7.308080808080808e-07,
+      "loss": 0.4825,
+      "step": 2770
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 17.263690948486328,
+      "learning_rate": 7.297979797979797e-07,
+      "loss": 0.7032,
+      "step": 2780
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 34.739078521728516,
+      "learning_rate": 7.287878787878787e-07,
+      "loss": 0.463,
+      "step": 2790
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 10.641708374023438,
+      "learning_rate": 7.277777777777777e-07,
+      "loss": 0.4351,
+      "step": 2800
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 75.48184967041016,
+      "learning_rate": 7.267676767676767e-07,
+      "loss": 0.7079,
+      "step": 2810
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 45.05419921875,
+      "learning_rate": 7.257575757575756e-07,
+      "loss": 0.5934,
+      "step": 2820
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 29.237016677856445,
+      "learning_rate": 7.247474747474747e-07,
+      "loss": 0.4162,
+      "step": 2830
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.004866959527134895,
+      "learning_rate": 7.237373737373737e-07,
+      "loss": 0.7245,
+      "step": 2840
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 128.69998168945312,
+      "learning_rate": 7.227272727272728e-07,
+      "loss": 0.4543,
+      "step": 2850
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.031988680362701416,
+      "learning_rate": 7.217171717171717e-07,
+      "loss": 0.6369,
+      "step": 2860
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 23.29071807861328,
+      "learning_rate": 7.207070707070707e-07,
+      "loss": 0.8627,
+      "step": 2870
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 16.787782669067383,
+      "learning_rate": 7.196969696969697e-07,
+      "loss": 0.7068,
+      "step": 2880
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 27.934904098510742,
+      "learning_rate": 7.186868686868687e-07,
+      "loss": 0.6029,
+      "step": 2890
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 28.06117057800293,
+      "learning_rate": 7.176767676767677e-07,
+      "loss": 0.4879,
+      "step": 2900
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 13.214853286743164,
+      "learning_rate": 7.166666666666667e-07,
+      "loss": 0.4323,
+      "step": 2910
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.088837146759033,
+      "learning_rate": 7.156565656565656e-07,
+      "loss": 0.5254,
+      "step": 2920
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 29.167570114135742,
+      "learning_rate": 7.146464646464646e-07,
+      "loss": 0.7426,
+      "step": 2930
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 55.35223388671875,
+      "learning_rate": 7.136363636363636e-07,
+      "loss": 0.5928,
+      "step": 2940
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 46.83303451538086,
+      "learning_rate": 7.126262626262625e-07,
+      "loss": 0.5281,
+      "step": 2950
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 62.05622482299805,
+      "learning_rate": 7.116161616161616e-07,
+      "loss": 0.6006,
+      "step": 2960
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 47.03242874145508,
+      "learning_rate": 7.106060606060605e-07,
+      "loss": 0.5264,
+      "step": 2970
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 41.88718032836914,
+      "learning_rate": 7.095959595959596e-07,
+      "loss": 0.5463,
+      "step": 2980
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 10.629190444946289,
+      "learning_rate": 7.085858585858585e-07,
+      "loss": 0.4873,
+      "step": 2990
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 34.29298400878906,
+      "learning_rate": 7.075757575757575e-07,
+      "loss": 0.5327,
+      "step": 3000
+    },
+    {
+      "epoch": 0.15,
+      "eval_loss": 0.6308945417404175,
+      "eval_runtime": 273.2852,
+      "eval_samples_per_second": 3.659,
+      "eval_steps_per_second": 3.659,
+      "step": 3000
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 57.80482482910156,
+      "learning_rate": 7.065656565656566e-07,
+      "loss": 0.5131,
+      "step": 3010
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 15.977564811706543,
+      "learning_rate": 7.055555555555556e-07,
+      "loss": 0.7116,
+      "step": 3020
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 24.813589096069336,
+      "learning_rate": 7.045454545454545e-07,
+      "loss": 0.8132,
+      "step": 3030
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 22.984556198120117,
+      "learning_rate": 7.035353535353536e-07,
+      "loss": 0.9015,
+      "step": 3040
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 15.78228759765625,
+      "learning_rate": 7.025252525252525e-07,
+      "loss": 0.5522,
+      "step": 3050
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 19.144140243530273,
+      "learning_rate": 7.015151515151516e-07,
+      "loss": 0.5387,
+      "step": 3060
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 47.44633483886719,
+      "learning_rate": 7.005050505050505e-07,
+      "loss": 0.5754,
+      "step": 3070
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 36.51036834716797,
+      "learning_rate": 6.994949494949494e-07,
+      "loss": 0.8067,
+      "step": 3080
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.8666948080062866,
+      "learning_rate": 6.984848484848485e-07,
+      "loss": 0.6058,
+      "step": 3090
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 23.022371292114258,
+      "learning_rate": 6.974747474747474e-07,
+      "loss": 0.6043,
+      "step": 3100
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 42.4169807434082,
+      "learning_rate": 6.964646464646464e-07,
+      "loss": 0.5526,
+      "step": 3110
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 65.37395477294922,
+      "learning_rate": 6.954545454545454e-07,
+      "loss": 0.5219,
+      "step": 3120
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 18.157527923583984,
+      "learning_rate": 6.944444444444444e-07,
+      "loss": 0.6028,
+      "step": 3130
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 52.82014846801758,
+      "learning_rate": 6.934343434343434e-07,
+      "loss": 0.7391,
+      "step": 3140
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 19.03993034362793,
+      "learning_rate": 6.924242424242424e-07,
+      "loss": 0.9328,
+      "step": 3150
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 38.07152557373047,
+      "learning_rate": 6.914141414141413e-07,
+      "loss": 0.677,
+      "step": 3160
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 10.974771499633789,
+      "learning_rate": 6.904040404040404e-07,
+      "loss": 0.638,
+      "step": 3170
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 33.91596221923828,
+      "learning_rate": 6.894949494949494e-07,
+      "loss": 0.9438,
+      "step": 3180
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 48.037166595458984,
+      "learning_rate": 6.884848484848485e-07,
+      "loss": 0.4394,
+      "step": 3190
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 28.13571548461914,
+      "learning_rate": 6.874747474747474e-07,
+      "loss": 0.4743,
+      "step": 3200
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 35.86332702636719,
+      "learning_rate": 6.864646464646464e-07,
+      "loss": 0.5854,
+      "step": 3210
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 28.57064437866211,
+      "learning_rate": 6.854545454545454e-07,
+      "loss": 0.9211,
+      "step": 3220
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 41.90704345703125,
+      "learning_rate": 6.844444444444444e-07,
+      "loss": 0.6961,
+      "step": 3230
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 11.663514137268066,
+      "learning_rate": 6.834343434343434e-07,
+      "loss": 0.6082,
+      "step": 3240
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.3548143804073334,
+      "learning_rate": 6.824242424242424e-07,
+      "loss": 0.5743,
+      "step": 3250
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 49.9235954284668,
+      "learning_rate": 6.814141414141413e-07,
+      "loss": 0.8361,
+      "step": 3260
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 63.49104309082031,
+      "learning_rate": 6.804040404040405e-07,
+      "loss": 0.6426,
+      "step": 3270
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 34.379417419433594,
+      "learning_rate": 6.793939393939394e-07,
+      "loss": 0.6631,
+      "step": 3280
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 38.65358352661133,
+      "learning_rate": 6.783838383838383e-07,
+      "loss": 0.7235,
+      "step": 3290
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 65.52446746826172,
+      "learning_rate": 6.773737373737374e-07,
+      "loss": 0.718,
+      "step": 3300
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 11.010351181030273,
+      "learning_rate": 6.763636363636363e-07,
+      "loss": 0.7492,
+      "step": 3310
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.8809446096420288,
+      "learning_rate": 6.753535353535354e-07,
+      "loss": 0.685,
+      "step": 3320
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 18.255809783935547,
+      "learning_rate": 6.743434343434343e-07,
+      "loss": 0.5533,
+      "step": 3330
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 26.685049057006836,
+      "learning_rate": 6.733333333333333e-07,
+      "loss": 0.5235,
+      "step": 3340
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 249.94290161132812,
+      "learning_rate": 6.723232323232323e-07,
+      "loss": 0.6984,
+      "step": 3350
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 45.482479095458984,
+      "learning_rate": 6.713131313131313e-07,
+      "loss": 0.4921,
+      "step": 3360
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 19.996156692504883,
+      "learning_rate": 6.703030303030302e-07,
+      "loss": 0.4737,
+      "step": 3370
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 155.62306213378906,
+      "learning_rate": 6.692929292929293e-07,
+      "loss": 0.6733,
+      "step": 3380
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 35.31591033935547,
+      "learning_rate": 6.682828282828282e-07,
+      "loss": 0.5598,
+      "step": 3390
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 32.30759048461914,
+      "learning_rate": 6.672727272727273e-07,
+      "loss": 0.7611,
+      "step": 3400
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 30.439285278320312,
+      "learning_rate": 6.662626262626262e-07,
+      "loss": 0.8177,
+      "step": 3410
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 39.084266662597656,
+      "learning_rate": 6.652525252525251e-07,
+      "loss": 0.683,
+      "step": 3420
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 28.271728515625,
+      "learning_rate": 6.642424242424242e-07,
+      "loss": 0.5019,
+      "step": 3430
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.6252301931381226,
+      "learning_rate": 6.632323232323232e-07,
+      "loss": 0.4204,
+      "step": 3440
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 39.03291702270508,
+      "learning_rate": 6.622222222222222e-07,
+      "loss": 0.4508,
+      "step": 3450
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.38857364654541,
+      "learning_rate": 6.612121212121212e-07,
+      "loss": 0.5938,
+      "step": 3460
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 24.514738082885742,
+      "learning_rate": 6.602020202020202e-07,
+      "loss": 0.7128,
+      "step": 3470
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 45.85287857055664,
+      "learning_rate": 6.591919191919192e-07,
+      "loss": 0.7125,
+      "step": 3480
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.022573018446564674,
+      "learning_rate": 6.581818181818182e-07,
+      "loss": 0.4128,
+      "step": 3490
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 21.102893829345703,
+      "learning_rate": 6.571717171717171e-07,
+      "loss": 0.5341,
+      "step": 3500
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 27.609329223632812,
+      "learning_rate": 6.561616161616162e-07,
+      "loss": 0.4808,
+      "step": 3510
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 28.081892013549805,
+      "learning_rate": 6.551515151515151e-07,
+      "loss": 0.4426,
+      "step": 3520
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 35.72859573364258,
+      "learning_rate": 6.541414141414141e-07,
+      "loss": 0.5072,
+      "step": 3530
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 31.80050277709961,
+      "learning_rate": 6.531313131313131e-07,
+      "loss": 0.5102,
+      "step": 3540
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.7258970737457275,
+      "learning_rate": 6.52121212121212e-07,
+      "loss": 0.4568,
+      "step": 3550
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 23.297748565673828,
+      "learning_rate": 6.511111111111111e-07,
+      "loss": 0.5721,
+      "step": 3560
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 63.63188171386719,
+      "learning_rate": 6.5010101010101e-07,
+      "loss": 0.6361,
+      "step": 3570
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 38.485416412353516,
+      "learning_rate": 6.49090909090909e-07,
+      "loss": 0.4438,
+      "step": 3580
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 40.09241485595703,
+      "learning_rate": 6.48080808080808e-07,
+      "loss": 0.732,
+      "step": 3590
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 49.900917053222656,
+      "learning_rate": 6.47070707070707e-07,
+      "loss": 0.6085,
+      "step": 3600
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 26.581472396850586,
+      "learning_rate": 6.46060606060606e-07,
+      "loss": 0.7013,
+      "step": 3610
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 9.770761489868164,
+      "learning_rate": 6.450505050505051e-07,
+      "loss": 0.4664,
+      "step": 3620
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 31.70376205444336,
+      "learning_rate": 6.44040404040404e-07,
+      "loss": 0.6208,
+      "step": 3630
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 37.74061965942383,
+      "learning_rate": 6.430303030303031e-07,
+      "loss": 0.6137,
+      "step": 3640
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 33.57075119018555,
+      "learning_rate": 6.42020202020202e-07,
+      "loss": 0.591,
+      "step": 3650
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 52.23899841308594,
+      "learning_rate": 6.41010101010101e-07,
+      "loss": 0.8013,
+      "step": 3660
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 62.8396110534668,
+      "learning_rate": 6.4e-07,
+      "loss": 0.7005,
+      "step": 3670
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 36.724098205566406,
+      "learning_rate": 6.38989898989899e-07,
+      "loss": 0.8353,
+      "step": 3680
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.098162889480591,
+      "learning_rate": 6.379797979797979e-07,
+      "loss": 0.5307,
+      "step": 3690
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 41.13184356689453,
+      "learning_rate": 6.36969696969697e-07,
+      "loss": 0.5198,
+      "step": 3700
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 46.46861267089844,
+      "learning_rate": 6.359595959595959e-07,
+      "loss": 0.4319,
+      "step": 3710
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 65.68028259277344,
+      "learning_rate": 6.349494949494949e-07,
+      "loss": 0.5386,
+      "step": 3720
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.740784168243408,
+      "learning_rate": 6.339393939393939e-07,
+      "loss": 0.6834,
+      "step": 3730
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 15.918974876403809,
+      "learning_rate": 6.329292929292928e-07,
+      "loss": 0.6303,
+      "step": 3740
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.351881980895996,
+      "learning_rate": 6.319191919191919e-07,
+      "loss": 0.3967,
+      "step": 3750
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 76.57394409179688,
+      "learning_rate": 6.309090909090908e-07,
+      "loss": 0.5537,
+      "step": 3760
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 17.062864303588867,
+      "learning_rate": 6.298989898989898e-07,
+      "loss": 0.6029,
+      "step": 3770
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 39.9703254699707,
+      "learning_rate": 6.288888888888889e-07,
+      "loss": 0.3583,
+      "step": 3780
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 61.81090545654297,
+      "learning_rate": 6.278787878787879e-07,
+      "loss": 0.5372,
+      "step": 3790
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 25.4251708984375,
+      "learning_rate": 6.268686868686869e-07,
+      "loss": 0.4157,
+      "step": 3800
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 3.276371479034424,
+      "learning_rate": 6.258585858585859e-07,
+      "loss": 0.6531,
+      "step": 3810
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 47.67403030395508,
+      "learning_rate": 6.248484848484848e-07,
+      "loss": 0.6813,
+      "step": 3820
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 13.192293167114258,
+      "learning_rate": 6.238383838383839e-07,
+      "loss": 0.4612,
+      "step": 3830
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.580405056476593,
+      "learning_rate": 6.228282828282828e-07,
+      "loss": 0.3817,
+      "step": 3840
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 17.856048583984375,
+      "learning_rate": 6.218181818181817e-07,
+      "loss": 0.7929,
+      "step": 3850
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 33.620033264160156,
+      "learning_rate": 6.208080808080808e-07,
+      "loss": 0.5922,
+      "step": 3860
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 33.94918441772461,
+      "learning_rate": 6.197979797979797e-07,
+      "loss": 0.4963,
+      "step": 3870
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 15.053759574890137,
+      "learning_rate": 6.187878787878788e-07,
+      "loss": 0.6975,
+      "step": 3880
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 44.79494857788086,
+      "learning_rate": 6.177777777777777e-07,
+      "loss": 0.6962,
+      "step": 3890
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 9.810379028320312,
+      "learning_rate": 6.167676767676767e-07,
+      "loss": 0.3916,
+      "step": 3900
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 54.83355712890625,
+      "learning_rate": 6.157575757575757e-07,
+      "loss": 0.6929,
+      "step": 3910
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 20.908220291137695,
+      "learning_rate": 6.147474747474747e-07,
+      "loss": 0.7343,
+      "step": 3920
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 48.0946044921875,
+      "learning_rate": 6.137373737373736e-07,
+      "loss": 0.5088,
+      "step": 3930
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 37.02779006958008,
+      "learning_rate": 6.127272727272727e-07,
+      "loss": 0.5441,
+      "step": 3940
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 16.095773696899414,
+      "learning_rate": 6.117171717171717e-07,
+      "loss": 0.7542,
+      "step": 3950
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 48.33674621582031,
+      "learning_rate": 6.107070707070708e-07,
+      "loss": 0.4938,
+      "step": 3960
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 33.02685546875,
+      "learning_rate": 6.096969696969697e-07,
+      "loss": 0.6449,
+      "step": 3970
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 18.588003158569336,
+      "learning_rate": 6.086868686868687e-07,
+      "loss": 0.3599,
+      "step": 3980
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 46.80269241333008,
+      "learning_rate": 6.076767676767677e-07,
+      "loss": 0.5536,
+      "step": 3990
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 29.706817626953125,
+      "learning_rate": 6.066666666666666e-07,
+      "loss": 0.6733,
+      "step": 4000
+    },
+    {
+      "epoch": 0.2,
+      "eval_loss": 0.5796153545379639,
+      "eval_runtime": 272.8278,
+      "eval_samples_per_second": 3.665,
+      "eval_steps_per_second": 3.665,
+      "step": 4000
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 35.10763168334961,
+      "learning_rate": 6.056565656565657e-07,
+      "loss": 0.4982,
+      "step": 4010
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 36.585968017578125,
+      "learning_rate": 6.046464646464646e-07,
+      "loss": 0.5096,
+      "step": 4020
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 26.760963439941406,
+      "learning_rate": 6.036363636363636e-07,
+      "loss": 0.4632,
+      "step": 4030
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 65.30345916748047,
+      "learning_rate": 6.026262626262626e-07,
+      "loss": 0.673,
+      "step": 4040
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 34.205318450927734,
+      "learning_rate": 6.016161616161616e-07,
+      "loss": 0.5949,
+      "step": 4050
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 8.336762428283691,
+      "learning_rate": 6.006060606060605e-07,
+      "loss": 0.5725,
+      "step": 4060
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 23.612590789794922,
+      "learning_rate": 5.995959595959596e-07,
+      "loss": 0.5402,
+      "step": 4070
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.146045207977295,
+      "learning_rate": 5.985858585858585e-07,
+      "loss": 0.5696,
+      "step": 4080
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 18.502138137817383,
+      "learning_rate": 5.975757575757575e-07,
+      "loss": 0.551,
+      "step": 4090
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 48.672119140625,
+      "learning_rate": 5.965656565656565e-07,
+      "loss": 0.713,
+      "step": 4100
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.011800971813499928,
+      "learning_rate": 5.955555555555555e-07,
+      "loss": 0.5925,
+      "step": 4110
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 48.50733947753906,
+      "learning_rate": 5.945454545454546e-07,
+      "loss": 0.6687,
+      "step": 4120
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.01216745562851429,
+      "learning_rate": 5.935353535353536e-07,
+      "loss": 0.6491,
+      "step": 4130
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 34.22660446166992,
+      "learning_rate": 5.925252525252525e-07,
+      "loss": 0.6262,
+      "step": 4140
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 32.8122444152832,
+      "learning_rate": 5.915151515151515e-07,
+      "loss": 0.8408,
+      "step": 4150
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 23.037616729736328,
+      "learning_rate": 5.905050505050505e-07,
+      "loss": 0.6201,
+      "step": 4160
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.8174920082092285,
+      "learning_rate": 5.894949494949495e-07,
+      "loss": 0.6822,
+      "step": 4170
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.4964330196380615,
+      "learning_rate": 5.884848484848485e-07,
+      "loss": 0.6287,
+      "step": 4180
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 25.050561904907227,
+      "learning_rate": 5.874747474747474e-07,
+      "loss": 0.7394,
+      "step": 4190
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 40.42102813720703,
+      "learning_rate": 5.864646464646465e-07,
+      "loss": 0.7008,
+      "step": 4200
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 18.322233200073242,
+      "learning_rate": 5.854545454545454e-07,
+      "loss": 0.4913,
+      "step": 4210
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 19.939029693603516,
+      "learning_rate": 5.844444444444444e-07,
+      "loss": 0.6533,
+      "step": 4220
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 24.60533332824707,
+      "learning_rate": 5.834343434343434e-07,
+      "loss": 0.623,
+      "step": 4230
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 20.365371704101562,
+      "learning_rate": 5.824242424242424e-07,
+      "loss": 0.7523,
+      "step": 4240
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 50.654842376708984,
+      "learning_rate": 5.814141414141414e-07,
+      "loss": 0.5586,
+      "step": 4250
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 28.067501068115234,
+      "learning_rate": 5.804040404040404e-07,
+      "loss": 0.8267,
+      "step": 4260
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 64.40141296386719,
+      "learning_rate": 5.793939393939393e-07,
+      "loss": 0.5589,
+      "step": 4270
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 58.74480438232422,
+      "learning_rate": 5.783838383838383e-07,
+      "loss": 0.508,
+      "step": 4280
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 21.915971755981445,
+      "learning_rate": 5.773737373737374e-07,
+      "loss": 0.2656,
+      "step": 4290
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 20.83277130126953,
+      "learning_rate": 5.763636363636363e-07,
+      "loss": 0.591,
+      "step": 4300
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 21.14466094970703,
+      "learning_rate": 5.753535353535354e-07,
+      "loss": 0.5395,
+      "step": 4310
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.9965362548828125,
+      "learning_rate": 5.743434343434343e-07,
+      "loss": 0.4719,
+      "step": 4320
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 59.90948486328125,
+      "learning_rate": 5.733333333333334e-07,
+      "loss": 0.4387,
+      "step": 4330
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 30.61522674560547,
+      "learning_rate": 5.723232323232323e-07,
+      "loss": 0.8272,
+      "step": 4340
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 42.94061279296875,
+      "learning_rate": 5.713131313131313e-07,
+      "loss": 0.6145,
+      "step": 4350
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 32.636966705322266,
+      "learning_rate": 5.703030303030303e-07,
+      "loss": 0.6995,
+      "step": 4360
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 17.99488067626953,
+      "learning_rate": 5.692929292929293e-07,
+      "loss": 0.7309,
+      "step": 4370
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 119.16413116455078,
+      "learning_rate": 5.682828282828282e-07,
+      "loss": 0.3749,
+      "step": 4380
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 42.35875701904297,
+      "learning_rate": 5.672727272727273e-07,
+      "loss": 0.5476,
+      "step": 4390
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 50.271358489990234,
+      "learning_rate": 5.662626262626262e-07,
+      "loss": 0.4938,
+      "step": 4400
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 66.81938171386719,
+      "learning_rate": 5.652525252525253e-07,
+      "loss": 0.5849,
+      "step": 4410
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.026270627975464,
+      "learning_rate": 5.642424242424242e-07,
+      "loss": 0.6664,
+      "step": 4420
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 46.648406982421875,
+      "learning_rate": 5.632323232323231e-07,
+      "loss": 0.4565,
+      "step": 4430
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 16.63812828063965,
+      "learning_rate": 5.622222222222222e-07,
+      "loss": 0.6162,
+      "step": 4440
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 38.91732406616211,
+      "learning_rate": 5.612121212121211e-07,
+      "loss": 0.5521,
+      "step": 4450
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 60.01820373535156,
+      "learning_rate": 5.602020202020202e-07,
+      "loss": 0.5168,
+      "step": 4460
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 41.567169189453125,
+      "learning_rate": 5.591919191919192e-07,
+      "loss": 0.793,
+      "step": 4470
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 28.133264541625977,
+      "learning_rate": 5.581818181818182e-07,
+      "loss": 0.3861,
+      "step": 4480
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 62.19641876220703,
+      "learning_rate": 5.571717171717172e-07,
+      "loss": 0.6352,
+      "step": 4490
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 57.30091857910156,
+      "learning_rate": 5.561616161616162e-07,
+      "loss": 0.69,
+      "step": 4500
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 43.78639602661133,
+      "learning_rate": 5.551515151515151e-07,
+      "loss": 0.7222,
+      "step": 4510
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 32.01860809326172,
+      "learning_rate": 5.541414141414142e-07,
+      "loss": 0.7186,
+      "step": 4520
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 23.97758674621582,
+      "learning_rate": 5.531313131313131e-07,
+      "loss": 0.6255,
+      "step": 4530
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 50.89016342163086,
+      "learning_rate": 5.521212121212121e-07,
+      "loss": 0.5228,
+      "step": 4540
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 39.55519485473633,
+      "learning_rate": 5.511111111111111e-07,
+      "loss": 0.5833,
+      "step": 4550
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 25.74945640563965,
+      "learning_rate": 5.5010101010101e-07,
+      "loss": 0.5973,
+      "step": 4560
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 31.337507247924805,
+      "learning_rate": 5.490909090909091e-07,
+      "loss": 0.5308,
+      "step": 4570
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 54.57097625732422,
+      "learning_rate": 5.48080808080808e-07,
+      "loss": 0.7696,
+      "step": 4580
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 49.550113677978516,
+      "learning_rate": 5.47070707070707e-07,
+      "loss": 0.6118,
+      "step": 4590
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 50.0933723449707,
+      "learning_rate": 5.46060606060606e-07,
+      "loss": 0.5375,
+      "step": 4600
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 10.150845527648926,
+      "learning_rate": 5.45050505050505e-07,
+      "loss": 0.5641,
+      "step": 4610
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 31.044645309448242,
+      "learning_rate": 5.440404040404039e-07,
+      "loss": 0.795,
+      "step": 4620
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.17512640357017517,
+      "learning_rate": 5.430303030303031e-07,
+      "loss": 0.6226,
+      "step": 4630
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 9.986194610595703,
+      "learning_rate": 5.42020202020202e-07,
+      "loss": 0.5268,
+      "step": 4640
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 11.464401245117188,
+      "learning_rate": 5.410101010101011e-07,
+      "loss": 0.6265,
+      "step": 4650
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 23.791200637817383,
+      "learning_rate": 5.4e-07,
+      "loss": 0.4169,
+      "step": 4660
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 15.029647827148438,
+      "learning_rate": 5.38989898989899e-07,
+      "loss": 0.8112,
+      "step": 4670
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 45.67643356323242,
+      "learning_rate": 5.37979797979798e-07,
+      "loss": 0.6152,
+      "step": 4680
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 21.600475311279297,
+      "learning_rate": 5.36969696969697e-07,
+      "loss": 0.3425,
+      "step": 4690
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 86.80847930908203,
+      "learning_rate": 5.359595959595959e-07,
+      "loss": 0.6447,
+      "step": 4700
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 28.537168502807617,
+      "learning_rate": 5.34949494949495e-07,
+      "loss": 0.9382,
+      "step": 4710
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 7.84658145904541,
+      "learning_rate": 5.339393939393939e-07,
+      "loss": 0.4908,
+      "step": 4720
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 137.72291564941406,
+      "learning_rate": 5.329292929292929e-07,
+      "loss": 0.5919,
+      "step": 4730
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 28.025161743164062,
+      "learning_rate": 5.319191919191919e-07,
+      "loss": 0.45,
+      "step": 4740
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 23.230506896972656,
+      "learning_rate": 5.309090909090908e-07,
+      "loss": 0.5548,
+      "step": 4750
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.3400743007659912,
+      "learning_rate": 5.298989898989899e-07,
+      "loss": 0.3073,
+      "step": 4760
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 61.306583404541016,
+      "learning_rate": 5.288888888888888e-07,
+      "loss": 0.6393,
+      "step": 4770
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 57.7855339050293,
+      "learning_rate": 5.278787878787878e-07,
+      "loss": 1.1223,
+      "step": 4780
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 31.280473709106445,
+      "learning_rate": 5.268686868686868e-07,
+      "loss": 0.5766,
+      "step": 4790
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 43.74176788330078,
+      "learning_rate": 5.258585858585859e-07,
+      "loss": 0.656,
+      "step": 4800
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 9.490304946899414,
+      "learning_rate": 5.248484848484849e-07,
+      "loss": 0.6825,
+      "step": 4810
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 13.078840255737305,
+      "learning_rate": 5.238383838383839e-07,
+      "loss": 0.3547,
+      "step": 4820
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.858570575714111,
+      "learning_rate": 5.228282828282828e-07,
+      "loss": 0.4382,
+      "step": 4830
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 40.836524963378906,
+      "learning_rate": 5.218181818181819e-07,
+      "loss": 0.737,
+      "step": 4840
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 21.203798294067383,
+      "learning_rate": 5.208080808080808e-07,
+      "loss": 0.717,
+      "step": 4850
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 25.54405403137207,
+      "learning_rate": 5.197979797979797e-07,
+      "loss": 0.5552,
+      "step": 4860
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 24.849031448364258,
+      "learning_rate": 5.187878787878788e-07,
+      "loss": 0.6207,
+      "step": 4870
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 59.22574234008789,
+      "learning_rate": 5.177777777777777e-07,
+      "loss": 0.4548,
+      "step": 4880
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 49.025386810302734,
+      "learning_rate": 5.167676767676768e-07,
+      "loss": 0.4617,
+      "step": 4890
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 18.983474731445312,
+      "learning_rate": 5.157575757575757e-07,
+      "loss": 0.6261,
+      "step": 4900
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 24.251028060913086,
+      "learning_rate": 5.147474747474747e-07,
+      "loss": 0.7819,
+      "step": 4910
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 33.367332458496094,
+      "learning_rate": 5.137373737373737e-07,
+      "loss": 0.7485,
+      "step": 4920
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 16.02358627319336,
+      "learning_rate": 5.127272727272727e-07,
+      "loss": 0.4568,
+      "step": 4930
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 21.30974578857422,
+      "learning_rate": 5.117171717171716e-07,
+      "loss": 0.7856,
+      "step": 4940
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 6.685999393463135,
+      "learning_rate": 5.107070707070707e-07,
+      "loss": 0.2796,
+      "step": 4950
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 9.83161735534668,
+      "learning_rate": 5.096969696969696e-07,
+      "loss": 0.4994,
+      "step": 4960
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 28.18449592590332,
+      "learning_rate": 5.086868686868687e-07,
+      "loss": 0.4694,
+      "step": 4970
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 12.509058952331543,
+      "learning_rate": 5.076767676767677e-07,
+      "loss": 0.5545,
+      "step": 4980
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 10.555543899536133,
+      "learning_rate": 5.066666666666667e-07,
+      "loss": 0.3085,
+      "step": 4990
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 30.451934814453125,
+      "learning_rate": 5.056565656565657e-07,
+      "loss": 0.4184,
+      "step": 5000
+    },
+    {
+      "epoch": 0.25,
+      "eval_loss": 0.6317839622497559,
+      "eval_runtime": 273.6525,
+      "eval_samples_per_second": 3.654,
+      "eval_steps_per_second": 3.654,
+      "step": 5000
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 16.75673484802246,
+      "learning_rate": 5.046464646464646e-07,
+      "loss": 0.6444,
+      "step": 5010
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 11.304834365844727,
+      "learning_rate": 5.036363636363636e-07,
+      "loss": 0.422,
+      "step": 5020
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 18.923059463500977,
+      "learning_rate": 5.026262626262626e-07,
+      "loss": 0.744,
+      "step": 5030
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 26.73674964904785,
+      "learning_rate": 5.016161616161616e-07,
+      "loss": 0.3386,
+      "step": 5040
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.0594570636749268,
+      "learning_rate": 5.006060606060606e-07,
+      "loss": 0.6157,
+      "step": 5050
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 20.98480796813965,
+      "learning_rate": 4.995959595959596e-07,
+      "loss": 0.4157,
+      "step": 5060
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 47.005348205566406,
+      "learning_rate": 4.985858585858585e-07,
+      "loss": 0.7854,
+      "step": 5070
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 18.866960525512695,
+      "learning_rate": 4.975757575757576e-07,
+      "loss": 0.6145,
+      "step": 5080
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 12.678258895874023,
+      "learning_rate": 4.965656565656565e-07,
+      "loss": 0.3665,
+      "step": 5090
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 56.87771987915039,
+      "learning_rate": 4.955555555555556e-07,
+      "loss": 0.6198,
+      "step": 5100
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 7.225916862487793,
+      "learning_rate": 4.945454545454545e-07,
+      "loss": 0.519,
+      "step": 5110
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 23.896451950073242,
+      "learning_rate": 4.935353535353536e-07,
+      "loss": 0.5161,
+      "step": 5120
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.715981483459473,
+      "learning_rate": 4.925252525252525e-07,
+      "loss": 0.4298,
+      "step": 5130
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 41.35715866088867,
+      "learning_rate": 4.915151515151514e-07,
+      "loss": 0.83,
+      "step": 5140
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 16.183509826660156,
+      "learning_rate": 4.905050505050505e-07,
+      "loss": 0.5437,
+      "step": 5150
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 35.6097412109375,
+      "learning_rate": 4.894949494949494e-07,
+      "loss": 0.7177,
+      "step": 5160
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 40.16917037963867,
+      "learning_rate": 4.884848484848485e-07,
+      "loss": 0.7704,
+      "step": 5170
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 44.37532424926758,
+      "learning_rate": 4.874747474747474e-07,
+      "loss": 0.7467,
+      "step": 5180
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 48.34253692626953,
+      "learning_rate": 4.864646464646465e-07,
+      "loss": 0.5483,
+      "step": 5190
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 24.17354965209961,
+      "learning_rate": 4.854545454545454e-07,
+      "loss": 0.4913,
+      "step": 5200
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 35.74309539794922,
+      "learning_rate": 4.844444444444445e-07,
+      "loss": 0.5399,
+      "step": 5210
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 27.378297805786133,
+      "learning_rate": 4.834343434343434e-07,
+      "loss": 0.6222,
+      "step": 5220
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 46.51097869873047,
+      "learning_rate": 4.824242424242424e-07,
+      "loss": 0.4005,
+      "step": 5230
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 53.12395477294922,
+      "learning_rate": 4.814141414141414e-07,
+      "loss": 0.6066,
+      "step": 5240
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 16.78499412536621,
+      "learning_rate": 4.804040404040404e-07,
+      "loss": 0.6476,
+      "step": 5250
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 24.126611709594727,
+      "learning_rate": 4.793939393939393e-07,
+      "loss": 0.5924,
+      "step": 5260
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 93.89410400390625,
+      "learning_rate": 4.783838383838384e-07,
+      "loss": 0.596,
+      "step": 5270
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 9.708739280700684,
+      "learning_rate": 4.773737373737374e-07,
+      "loss": 0.4209,
+      "step": 5280
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 13.41931438446045,
+      "learning_rate": 4.7636363636363635e-07,
+      "loss": 0.5598,
+      "step": 5290
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 41.988311767578125,
+      "learning_rate": 4.7535353535353534e-07,
+      "loss": 0.7958,
+      "step": 5300
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 61.960548400878906,
+      "learning_rate": 4.7434343434343434e-07,
+      "loss": 0.643,
+      "step": 5310
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 23.325374603271484,
+      "learning_rate": 4.733333333333333e-07,
+      "loss": 0.7438,
+      "step": 5320
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 32.468814849853516,
+      "learning_rate": 4.723232323232323e-07,
+      "loss": 0.4068,
+      "step": 5330
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.005862310528755188,
+      "learning_rate": 4.713131313131313e-07,
+      "loss": 0.5128,
+      "step": 5340
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 4.65062141418457,
+      "learning_rate": 4.703030303030303e-07,
+      "loss": 0.3018,
+      "step": 5350
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 18.1551456451416,
+      "learning_rate": 4.6929292929292927e-07,
+      "loss": 0.5412,
+      "step": 5360
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 21.883535385131836,
+      "learning_rate": 4.6828282828282827e-07,
+      "loss": 0.3498,
+      "step": 5370
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 3.3442208766937256,
+      "learning_rate": 4.6727272727272727e-07,
+      "loss": 0.3586,
+      "step": 5380
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 10.725494384765625,
+      "learning_rate": 4.6626262626262626e-07,
+      "loss": 0.5572,
+      "step": 5390
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 54.03162384033203,
+      "learning_rate": 4.652525252525252e-07,
+      "loss": 0.699,
+      "step": 5400
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 66.64715576171875,
+      "learning_rate": 4.642424242424242e-07,
+      "loss": 0.6599,
+      "step": 5410
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 60.05572509765625,
+      "learning_rate": 4.632323232323232e-07,
+      "loss": 0.8139,
+      "step": 5420
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 81.59327697753906,
+      "learning_rate": 4.622222222222222e-07,
+      "loss": 0.6494,
+      "step": 5430
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 10.984243392944336,
+      "learning_rate": 4.6121212121212114e-07,
+      "loss": 0.3774,
+      "step": 5440
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 9.627668380737305,
+      "learning_rate": 4.602020202020202e-07,
+      "loss": 0.4846,
+      "step": 5450
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 34.14588928222656,
+      "learning_rate": 4.591919191919192e-07,
+      "loss": 0.6254,
+      "step": 5460
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 23.76165199279785,
+      "learning_rate": 4.581818181818182e-07,
+      "loss": 0.6575,
+      "step": 5470
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 21.314247131347656,
+      "learning_rate": 4.5717171717171713e-07,
+      "loss": 0.666,
+      "step": 5480
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 16.340349197387695,
+      "learning_rate": 4.561616161616161e-07,
+      "loss": 0.7248,
+      "step": 5490
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 12.984785079956055,
+      "learning_rate": 4.551515151515151e-07,
+      "loss": 0.5881,
+      "step": 5500
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 32.25109100341797,
+      "learning_rate": 4.541414141414141e-07,
+      "loss": 0.6264,
+      "step": 5510
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 58.724952697753906,
+      "learning_rate": 4.5313131313131306e-07,
+      "loss": 0.558,
+      "step": 5520
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 42.230377197265625,
+      "learning_rate": 4.521212121212121e-07,
+      "loss": 0.7187,
+      "step": 5530
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 48.40780258178711,
+      "learning_rate": 4.511111111111111e-07,
+      "loss": 0.6857,
+      "step": 5540
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 16.174091339111328,
+      "learning_rate": 4.501010101010101e-07,
+      "loss": 0.4787,
+      "step": 5550
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 19.75393295288086,
+      "learning_rate": 4.4909090909090905e-07,
+      "loss": 0.4903,
+      "step": 5560
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 22.77608299255371,
+      "learning_rate": 4.4808080808080805e-07,
+      "loss": 0.8591,
+      "step": 5570
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 28.29513931274414,
+      "learning_rate": 4.4707070707070704e-07,
+      "loss": 0.3831,
+      "step": 5580
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.01570923626422882,
+      "learning_rate": 4.4606060606060604e-07,
+      "loss": 0.4993,
+      "step": 5590
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 42.494590759277344,
+      "learning_rate": 4.45050505050505e-07,
+      "loss": 0.6979,
+      "step": 5600
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 11.708600044250488,
+      "learning_rate": 4.44040404040404e-07,
+      "loss": 0.4699,
+      "step": 5610
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 84.38910675048828,
+      "learning_rate": 4.4303030303030303e-07,
+      "loss": 0.5278,
+      "step": 5620
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 20.805999755859375,
+      "learning_rate": 4.4202020202020203e-07,
+      "loss": 0.5134,
+      "step": 5630
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 21.2529354095459,
+      "learning_rate": 4.4101010101010097e-07,
+      "loss": 0.5041,
+      "step": 5640
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 66.35508728027344,
+      "learning_rate": 4.3999999999999997e-07,
+      "loss": 0.5179,
+      "step": 5650
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 87.20722961425781,
+      "learning_rate": 4.3898989898989897e-07,
+      "loss": 0.743,
+      "step": 5660
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 61.29191589355469,
+      "learning_rate": 4.3797979797979796e-07,
+      "loss": 0.6709,
+      "step": 5670
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 14.595233917236328,
+      "learning_rate": 4.369696969696969e-07,
+      "loss": 0.4567,
+      "step": 5680
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 49.612754821777344,
+      "learning_rate": 4.359595959595959e-07,
+      "loss": 0.6162,
+      "step": 5690
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 64.8648681640625,
+      "learning_rate": 4.3494949494949495e-07,
+      "loss": 0.4786,
+      "step": 5700
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 19.547813415527344,
+      "learning_rate": 4.3393939393939395e-07,
+      "loss": 0.5041,
+      "step": 5710
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 19.075836181640625,
+      "learning_rate": 4.3292929292929295e-07,
+      "loss": 0.4912,
+      "step": 5720
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 2.2977592945098877,
+      "learning_rate": 4.319191919191919e-07,
+      "loss": 0.6704,
+      "step": 5730
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 13.33616828918457,
+      "learning_rate": 4.309090909090909e-07,
+      "loss": 0.3042,
+      "step": 5740
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 45.46590042114258,
+      "learning_rate": 4.298989898989899e-07,
+      "loss": 0.8667,
+      "step": 5750
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 26.082550048828125,
+      "learning_rate": 4.2888888888888883e-07,
+      "loss": 0.5301,
+      "step": 5760
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 18.556835174560547,
+      "learning_rate": 4.278787878787878e-07,
+      "loss": 0.5641,
+      "step": 5770
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 33.3296012878418,
+      "learning_rate": 4.268686868686868e-07,
+      "loss": 0.9849,
+      "step": 5780
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 8.51718807220459,
+      "learning_rate": 4.2585858585858587e-07,
+      "loss": 0.678,
+      "step": 5790
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 48.53915786743164,
+      "learning_rate": 4.2484848484848487e-07,
+      "loss": 0.9045,
+      "step": 5800
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 25.81182098388672,
+      "learning_rate": 4.238383838383838e-07,
+      "loss": 0.7647,
+      "step": 5810
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 19.481019973754883,
+      "learning_rate": 4.228282828282828e-07,
+      "loss": 0.6162,
+      "step": 5820
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.070164203643799,
+      "learning_rate": 4.218181818181818e-07,
+      "loss": 0.4276,
+      "step": 5830
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 45.53562545776367,
+      "learning_rate": 4.208080808080808e-07,
+      "loss": 0.5479,
+      "step": 5840
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 36.35445022583008,
+      "learning_rate": 4.1979797979797975e-07,
+      "loss": 0.4958,
+      "step": 5850
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.0024314168840646744,
+      "learning_rate": 4.1878787878787875e-07,
+      "loss": 0.5411,
+      "step": 5860
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 29.073923110961914,
+      "learning_rate": 4.177777777777778e-07,
+      "loss": 0.588,
+      "step": 5870
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 27.702180862426758,
+      "learning_rate": 4.167676767676768e-07,
+      "loss": 0.5949,
+      "step": 5880
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 31.89980697631836,
+      "learning_rate": 4.1575757575757574e-07,
+      "loss": 0.9579,
+      "step": 5890
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 16.54688262939453,
+      "learning_rate": 4.1474747474747473e-07,
+      "loss": 0.483,
+      "step": 5900
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 17.019641876220703,
+      "learning_rate": 4.1373737373737373e-07,
+      "loss": 0.5992,
+      "step": 5910
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 60.383583068847656,
+      "learning_rate": 4.1272727272727273e-07,
+      "loss": 0.6877,
+      "step": 5920
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 31.626358032226562,
+      "learning_rate": 4.1171717171717167e-07,
+      "loss": 0.3786,
+      "step": 5930
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 41.01628112792969,
+      "learning_rate": 4.1070707070707067e-07,
+      "loss": 1.0079,
+      "step": 5940
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.4029645919799805,
+      "learning_rate": 4.0969696969696966e-07,
+      "loss": 0.2839,
+      "step": 5950
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 25.453466415405273,
+      "learning_rate": 4.086868686868687e-07,
+      "loss": 0.3568,
+      "step": 5960
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 37.09002685546875,
+      "learning_rate": 4.0767676767676766e-07,
+      "loss": 0.6424,
+      "step": 5970
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 72.52601623535156,
+      "learning_rate": 4.0666666666666666e-07,
+      "loss": 0.5851,
+      "step": 5980
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 24.996829986572266,
+      "learning_rate": 4.0565656565656565e-07,
+      "loss": 0.3083,
+      "step": 5990
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.386436939239502,
+      "learning_rate": 4.0464646464646465e-07,
+      "loss": 0.5214,
+      "step": 6000
+    },
+    {
+      "epoch": 0.3,
+      "eval_loss": 0.5321834683418274,
+      "eval_runtime": 272.7035,
+      "eval_samples_per_second": 3.667,
+      "eval_steps_per_second": 3.667,
+      "step": 6000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 2000,
+  "total_flos": 2.82846296014848e+17,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}