diff --git "a/trainer_state.json" "b/trainer_state.json" new file mode 100644--- /dev/null +++ "b/trainer_state.json" @@ -0,0 +1,35062 @@ +{ + "best_metric": 0.85816470337969, + "best_model_checkpoint": "/m/triton/scratch/elec/puhe/p/palp3/MUCS/indicwav2vec_outputs/pd_warmup_2000/s400_shuff42/checkpoint-2500", + "epoch": 8.0, + "eval_steps": 2500, + "global_step": 5000, + "is_hyper_param_search": false, + "is_local_process_zero": true, + "is_world_process_zero": true, + "log_history": [ + { + "epoch": 0.0016, + "grad_norm": NaN, + "learning_rate": 0.0, + "loss": 88.1636, + "step": 1 + }, + { + "epoch": 0.0032, + "grad_norm": NaN, + "learning_rate": 0.0, + "loss": 31.9774, + "step": 2 + }, + { + "epoch": 0.0048, + "grad_norm": 25.727998733520508, + "learning_rate": 3e-07, + "loss": 28.7213, + "step": 3 + }, + { + "epoch": 0.0064, + "grad_norm": 18.18585205078125, + "learning_rate": 6e-07, + "loss": 24.9668, + "step": 4 + }, + { + "epoch": 0.008, + "grad_norm": 18.16857147216797, + "learning_rate": 9e-07, + "loss": 25.6094, + "step": 5 + }, + { + "epoch": 0.0096, + "grad_norm": 19.88906478881836, + "learning_rate": 1.2e-06, + "loss": 27.1836, + "step": 6 + }, + { + "epoch": 0.0112, + "grad_norm": 17.007699966430664, + "learning_rate": 1.4999999999999998e-06, + "loss": 25.1928, + "step": 7 + }, + { + "epoch": 0.0128, + "grad_norm": 19.113887786865234, + "learning_rate": 1.8e-06, + "loss": 25.8839, + "step": 8 + }, + { + "epoch": 0.0144, + "grad_norm": 15.991517066955566, + "learning_rate": 2.1e-06, + "loss": 22.7303, + "step": 9 + }, + { + "epoch": 0.016, + "grad_norm": 16.13288688659668, + "learning_rate": 2.4e-06, + "loss": 23.029, + "step": 10 + }, + { + "epoch": 0.0176, + "grad_norm": 14.396987915039062, + "learning_rate": 2.6999999999999996e-06, + "loss": 21.1181, + "step": 11 + }, + { + "epoch": 0.0192, + "grad_norm": 13.756827354431152, + "learning_rate": 2.9999999999999997e-06, + "loss": 20.1848, + "step": 12 + }, + { + "epoch": 0.0208, + "grad_norm": 14.755758285522461, + "learning_rate": 3.2999999999999993e-06, + "loss": 21.5943, + "step": 13 + }, + { + "epoch": 0.0224, + "grad_norm": 15.602395057678223, + "learning_rate": 3.6e-06, + "loss": 21.7457, + "step": 14 + }, + { + "epoch": 0.024, + "grad_norm": 14.094840049743652, + "learning_rate": 3.899999999999999e-06, + "loss": 20.5424, + "step": 15 + }, + { + "epoch": 0.0256, + "grad_norm": 15.644570350646973, + "learning_rate": 4.2e-06, + "loss": 22.0262, + "step": 16 + }, + { + "epoch": 0.0272, + "grad_norm": 25.94748878479004, + "learning_rate": 4.499999999999999e-06, + "loss": 26.2437, + "step": 17 + }, + { + "epoch": 0.0288, + "grad_norm": 16.23424530029297, + "learning_rate": 4.8e-06, + "loss": 23.1837, + "step": 18 + }, + { + "epoch": 0.0304, + "grad_norm": 14.502126693725586, + "learning_rate": 5.1e-06, + "loss": 21.0343, + "step": 19 + }, + { + "epoch": 0.032, + "grad_norm": 20.787498474121094, + "learning_rate": 5.399999999999999e-06, + "loss": 26.3419, + "step": 20 + }, + { + "epoch": 0.0336, + "grad_norm": 13.608593940734863, + "learning_rate": 5.7e-06, + "loss": 18.643, + "step": 21 + }, + { + "epoch": 0.0352, + "grad_norm": 12.794872283935547, + "learning_rate": 5.999999999999999e-06, + "loss": 18.1019, + "step": 22 + }, + { + "epoch": 0.0368, + "grad_norm": 14.386597633361816, + "learning_rate": 6.3e-06, + "loss": 20.7248, + "step": 23 + }, + { + "epoch": 0.0384, + "grad_norm": 13.960803031921387, + "learning_rate": 6.599999999999999e-06, + "loss": 19.1159, + "step": 24 + }, + { + "epoch": 0.04, + "grad_norm": 15.56638240814209, + "learning_rate": 6.899999999999999e-06, + "loss": 20.7411, + "step": 25 + }, + { + "epoch": 0.0416, + "grad_norm": 15.256016731262207, + "learning_rate": 7.2e-06, + "loss": 20.6842, + "step": 26 + }, + { + "epoch": 0.0432, + "grad_norm": 21.440200805664062, + "learning_rate": 7.499999999999999e-06, + "loss": 18.8145, + "step": 27 + }, + { + "epoch": 0.0448, + "grad_norm": 16.310436248779297, + "learning_rate": 7.799999999999998e-06, + "loss": 21.4534, + "step": 28 + }, + { + "epoch": 0.0464, + "grad_norm": 17.503496170043945, + "learning_rate": 8.099999999999999e-06, + "loss": 19.9019, + "step": 29 + }, + { + "epoch": 0.048, + "grad_norm": 16.19658660888672, + "learning_rate": 8.4e-06, + "loss": 20.805, + "step": 30 + }, + { + "epoch": 0.0496, + "grad_norm": 17.486013412475586, + "learning_rate": 8.7e-06, + "loss": 19.6898, + "step": 31 + }, + { + "epoch": 0.0512, + "grad_norm": 16.73200225830078, + "learning_rate": 8.999999999999999e-06, + "loss": 21.2597, + "step": 32 + }, + { + "epoch": 0.0528, + "grad_norm": 16.063310623168945, + "learning_rate": 9.299999999999999e-06, + "loss": 20.0767, + "step": 33 + }, + { + "epoch": 0.0544, + "grad_norm": 16.689453125, + "learning_rate": 9.6e-06, + "loss": 20.4581, + "step": 34 + }, + { + "epoch": 0.056, + "grad_norm": 18.49964714050293, + "learning_rate": 9.9e-06, + "loss": 21.9053, + "step": 35 + }, + { + "epoch": 0.0576, + "grad_norm": 17.38431167602539, + "learning_rate": 1.02e-05, + "loss": 20.7011, + "step": 36 + }, + { + "epoch": 0.0592, + "grad_norm": 16.372385025024414, + "learning_rate": 1.05e-05, + "loss": 19.5237, + "step": 37 + }, + { + "epoch": 0.0608, + "grad_norm": 18.400272369384766, + "learning_rate": 1.0799999999999998e-05, + "loss": 21.3056, + "step": 38 + }, + { + "epoch": 0.0624, + "grad_norm": 20.40253257751465, + "learning_rate": 1.1099999999999999e-05, + "loss": 22.7792, + "step": 39 + }, + { + "epoch": 0.064, + "grad_norm": 16.800230026245117, + "learning_rate": 1.14e-05, + "loss": 19.5917, + "step": 40 + }, + { + "epoch": 0.0656, + "grad_norm": 18.398941040039062, + "learning_rate": 1.17e-05, + "loss": 20.6041, + "step": 41 + }, + { + "epoch": 0.0672, + "grad_norm": 14.947763442993164, + "learning_rate": 1.1999999999999999e-05, + "loss": 17.352, + "step": 42 + }, + { + "epoch": 0.0688, + "grad_norm": 18.517053604125977, + "learning_rate": 1.2299999999999999e-05, + "loss": 20.3232, + "step": 43 + }, + { + "epoch": 0.0704, + "grad_norm": 18.336021423339844, + "learning_rate": 1.26e-05, + "loss": 20.1034, + "step": 44 + }, + { + "epoch": 0.072, + "grad_norm": 20.530216217041016, + "learning_rate": 1.2899999999999998e-05, + "loss": 21.4217, + "step": 45 + }, + { + "epoch": 0.0736, + "grad_norm": 20.346179962158203, + "learning_rate": 1.3199999999999997e-05, + "loss": 21.2479, + "step": 46 + }, + { + "epoch": 0.0752, + "grad_norm": 26.894554138183594, + "learning_rate": 1.3499999999999998e-05, + "loss": 25.558, + "step": 47 + }, + { + "epoch": 0.0768, + "grad_norm": 17.661767959594727, + "learning_rate": 1.3799999999999998e-05, + "loss": 18.5759, + "step": 48 + }, + { + "epoch": 0.0784, + "grad_norm": 19.176250457763672, + "learning_rate": 1.4099999999999999e-05, + "loss": 19.8242, + "step": 49 + }, + { + "epoch": 0.08, + "grad_norm": NaN, + "learning_rate": 1.4099999999999999e-05, + "loss": 18.8538, + "step": 50 + }, + { + "epoch": 0.0816, + "grad_norm": NaN, + "learning_rate": 1.4099999999999999e-05, + "loss": 55.0316, + "step": 51 + }, + { + "epoch": 0.0832, + "grad_norm": 35.90843963623047, + "learning_rate": 1.44e-05, + "loss": 27.8855, + "step": 52 + }, + { + "epoch": 0.0848, + "grad_norm": 41.0328483581543, + "learning_rate": 1.47e-05, + "loss": 32.342, + "step": 53 + }, + { + "epoch": 0.0864, + "grad_norm": 32.60612869262695, + "learning_rate": 1.4999999999999999e-05, + "loss": 27.2888, + "step": 54 + }, + { + "epoch": 0.088, + "grad_norm": 36.59461212158203, + "learning_rate": 1.53e-05, + "loss": 28.7579, + "step": 55 + }, + { + "epoch": 0.0896, + "grad_norm": 31.01473617553711, + "learning_rate": 1.5599999999999996e-05, + "loss": 25.3167, + "step": 56 + }, + { + "epoch": 0.0912, + "grad_norm": 29.129039764404297, + "learning_rate": 1.5899999999999997e-05, + "loss": 23.994, + "step": 57 + }, + { + "epoch": 0.0928, + "grad_norm": 26.573955535888672, + "learning_rate": 1.6199999999999997e-05, + "loss": 20.6755, + "step": 58 + }, + { + "epoch": 0.0944, + "grad_norm": 29.24463653564453, + "learning_rate": 1.6499999999999998e-05, + "loss": 23.024, + "step": 59 + }, + { + "epoch": 0.096, + "grad_norm": 26.921775817871094, + "learning_rate": 1.68e-05, + "loss": 21.5849, + "step": 60 + }, + { + "epoch": 0.0976, + "grad_norm": 27.8332462310791, + "learning_rate": 1.71e-05, + "loss": 21.6015, + "step": 61 + }, + { + "epoch": 0.0992, + "grad_norm": 28.124608993530273, + "learning_rate": 1.74e-05, + "loss": 22.0914, + "step": 62 + }, + { + "epoch": 0.1008, + "grad_norm": 30.353343963623047, + "learning_rate": 1.7699999999999997e-05, + "loss": 22.5719, + "step": 63 + }, + { + "epoch": 0.1024, + "grad_norm": 29.733469009399414, + "learning_rate": 1.7999999999999997e-05, + "loss": 21.8033, + "step": 64 + }, + { + "epoch": 0.104, + "grad_norm": 25.040752410888672, + "learning_rate": 1.8299999999999998e-05, + "loss": 19.152, + "step": 65 + }, + { + "epoch": 0.1056, + "grad_norm": 24.005868911743164, + "learning_rate": 1.8599999999999998e-05, + "loss": 18.4214, + "step": 66 + }, + { + "epoch": 0.1072, + "grad_norm": 23.582408905029297, + "learning_rate": 1.89e-05, + "loss": 17.6597, + "step": 67 + }, + { + "epoch": 0.1088, + "grad_norm": 23.405153274536133, + "learning_rate": 1.92e-05, + "loss": 17.4055, + "step": 68 + }, + { + "epoch": 0.1104, + "grad_norm": 24.748991012573242, + "learning_rate": 1.95e-05, + "loss": 17.9656, + "step": 69 + }, + { + "epoch": 0.112, + "grad_norm": 26.633638381958008, + "learning_rate": 1.98e-05, + "loss": 18.421, + "step": 70 + }, + { + "epoch": 0.1136, + "grad_norm": 26.72304344177246, + "learning_rate": 2.01e-05, + "loss": 17.8431, + "step": 71 + }, + { + "epoch": 0.1152, + "grad_norm": 28.973541259765625, + "learning_rate": 2.04e-05, + "loss": 16.8414, + "step": 72 + }, + { + "epoch": 0.1168, + "grad_norm": 27.353349685668945, + "learning_rate": 2.07e-05, + "loss": 17.8068, + "step": 73 + }, + { + "epoch": 0.1184, + "grad_norm": 30.53626823425293, + "learning_rate": 2.1e-05, + "loss": 16.7757, + "step": 74 + }, + { + "epoch": 0.12, + "grad_norm": 27.967914581298828, + "learning_rate": 2.1299999999999996e-05, + "loss": 16.9582, + "step": 75 + }, + { + "epoch": 0.1216, + "grad_norm": 31.028308868408203, + "learning_rate": 2.1599999999999996e-05, + "loss": 18.2253, + "step": 76 + }, + { + "epoch": 0.1232, + "grad_norm": 27.322860717773438, + "learning_rate": 2.1899999999999997e-05, + "loss": 16.2514, + "step": 77 + }, + { + "epoch": 0.1248, + "grad_norm": 27.86654281616211, + "learning_rate": 2.2199999999999998e-05, + "loss": 15.9259, + "step": 78 + }, + { + "epoch": 0.1264, + "grad_norm": 28.512012481689453, + "learning_rate": 2.2499999999999998e-05, + "loss": 16.0654, + "step": 79 + }, + { + "epoch": 0.128, + "grad_norm": 29.90827751159668, + "learning_rate": 2.28e-05, + "loss": 16.1965, + "step": 80 + }, + { + "epoch": 0.1296, + "grad_norm": 30.172544479370117, + "learning_rate": 2.31e-05, + "loss": 15.6856, + "step": 81 + }, + { + "epoch": 0.1312, + "grad_norm": 34.49613571166992, + "learning_rate": 2.34e-05, + "loss": 16.8192, + "step": 82 + }, + { + "epoch": 0.1328, + "grad_norm": 32.905250549316406, + "learning_rate": 2.3699999999999997e-05, + "loss": 16.1658, + "step": 83 + }, + { + "epoch": 0.1344, + "grad_norm": 30.88019371032715, + "learning_rate": 2.3999999999999997e-05, + "loss": 14.9915, + "step": 84 + }, + { + "epoch": 0.136, + "grad_norm": 30.31485366821289, + "learning_rate": 2.4299999999999998e-05, + "loss": 14.3373, + "step": 85 + }, + { + "epoch": 0.1376, + "grad_norm": 41.14937973022461, + "learning_rate": 2.4599999999999998e-05, + "loss": 17.3286, + "step": 86 + }, + { + "epoch": 0.1392, + "grad_norm": 32.182125091552734, + "learning_rate": 2.49e-05, + "loss": 14.0104, + "step": 87 + }, + { + "epoch": 0.1408, + "grad_norm": 40.61635208129883, + "learning_rate": 2.52e-05, + "loss": 15.6781, + "step": 88 + }, + { + "epoch": 0.1424, + "grad_norm": 33.11867141723633, + "learning_rate": 2.55e-05, + "loss": 13.5609, + "step": 89 + }, + { + "epoch": 0.144, + "grad_norm": 34.575050354003906, + "learning_rate": 2.5799999999999997e-05, + "loss": 13.2412, + "step": 90 + }, + { + "epoch": 0.1456, + "grad_norm": 62.21010971069336, + "learning_rate": 2.6099999999999997e-05, + "loss": 14.6794, + "step": 91 + }, + { + "epoch": 0.1472, + "grad_norm": 34.05390167236328, + "learning_rate": 2.6399999999999995e-05, + "loss": 12.4023, + "step": 92 + }, + { + "epoch": 0.1488, + "grad_norm": 49.62013244628906, + "learning_rate": 2.6699999999999995e-05, + "loss": 15.1849, + "step": 93 + }, + { + "epoch": 0.1504, + "grad_norm": 35.7944450378418, + "learning_rate": 2.6999999999999996e-05, + "loss": 12.0157, + "step": 94 + }, + { + "epoch": 0.152, + "grad_norm": 44.946651458740234, + "learning_rate": 2.7299999999999996e-05, + "loss": 13.118, + "step": 95 + }, + { + "epoch": 0.1536, + "grad_norm": 50.83009719848633, + "learning_rate": 2.7599999999999997e-05, + "loss": 13.8204, + "step": 96 + }, + { + "epoch": 0.1552, + "grad_norm": 40.576942443847656, + "learning_rate": 2.7899999999999997e-05, + "loss": 11.8124, + "step": 97 + }, + { + "epoch": 0.1568, + "grad_norm": 53.70853042602539, + "learning_rate": 2.8199999999999998e-05, + "loss": 14.2831, + "step": 98 + }, + { + "epoch": 0.1584, + "grad_norm": 34.88117218017578, + "learning_rate": 2.8499999999999998e-05, + "loss": 10.0583, + "step": 99 + }, + { + "epoch": 0.16, + "grad_norm": 48.1845703125, + "learning_rate": 2.88e-05, + "loss": 11.8564, + "step": 100 + }, + { + "epoch": 0.1616, + "grad_norm": Infinity, + "learning_rate": 2.88e-05, + "loss": 27.1439, + "step": 101 + }, + { + "epoch": 0.1632, + "grad_norm": 108.25672912597656, + "learning_rate": 2.91e-05, + "loss": 14.243, + "step": 102 + }, + { + "epoch": 0.1648, + "grad_norm": 134.4223175048828, + "learning_rate": 2.94e-05, + "loss": 15.1354, + "step": 103 + }, + { + "epoch": 0.1664, + "grad_norm": 99.52935791015625, + "learning_rate": 2.97e-05, + "loss": 16.8895, + "step": 104 + }, + { + "epoch": 0.168, + "grad_norm": NaN, + "learning_rate": 2.97e-05, + "loss": 15.3123, + "step": 105 + }, + { + "epoch": 0.1696, + "grad_norm": 64.19481658935547, + "learning_rate": 2.9999999999999997e-05, + "loss": 11.9997, + "step": 106 + }, + { + "epoch": 0.1712, + "grad_norm": 71.73747253417969, + "learning_rate": 3.0299999999999998e-05, + "loss": 12.6204, + "step": 107 + }, + { + "epoch": 0.1728, + "grad_norm": 53.78099822998047, + "learning_rate": 3.06e-05, + "loss": 10.1112, + "step": 108 + }, + { + "epoch": 0.1744, + "grad_norm": 51.031742095947266, + "learning_rate": 3.09e-05, + "loss": 9.2134, + "step": 109 + }, + { + "epoch": 0.176, + "grad_norm": 47.309574127197266, + "learning_rate": 3.119999999999999e-05, + "loss": 8.8334, + "step": 110 + }, + { + "epoch": 0.1776, + "grad_norm": 49.82463073730469, + "learning_rate": 3.149999999999999e-05, + "loss": 8.8497, + "step": 111 + }, + { + "epoch": 0.1792, + "grad_norm": 51.74225997924805, + "learning_rate": 3.1799999999999994e-05, + "loss": 9.0849, + "step": 112 + }, + { + "epoch": 0.1808, + "grad_norm": 53.916893005371094, + "learning_rate": 3.2099999999999994e-05, + "loss": 9.0814, + "step": 113 + }, + { + "epoch": 0.1824, + "grad_norm": 40.585506439208984, + "learning_rate": 3.2399999999999995e-05, + "loss": 7.6543, + "step": 114 + }, + { + "epoch": 0.184, + "grad_norm": 44.14476013183594, + "learning_rate": 3.2699999999999995e-05, + "loss": 7.5272, + "step": 115 + }, + { + "epoch": 0.1856, + "grad_norm": 86.90267944335938, + "learning_rate": 3.2999999999999996e-05, + "loss": 9.7604, + "step": 116 + }, + { + "epoch": 0.1872, + "grad_norm": 43.009765625, + "learning_rate": 3.3299999999999996e-05, + "loss": 7.457, + "step": 117 + }, + { + "epoch": 0.1888, + "grad_norm": 45.4506721496582, + "learning_rate": 3.36e-05, + "loss": 7.6341, + "step": 118 + }, + { + "epoch": 0.1904, + "grad_norm": 95.28884887695312, + "learning_rate": 3.39e-05, + "loss": 11.2971, + "step": 119 + }, + { + "epoch": 0.192, + "grad_norm": 40.490135192871094, + "learning_rate": 3.42e-05, + "loss": 6.9917, + "step": 120 + }, + { + "epoch": 0.1936, + "grad_norm": 39.02360153198242, + "learning_rate": 3.45e-05, + "loss": 6.8346, + "step": 121 + }, + { + "epoch": 0.1952, + "grad_norm": 36.35051345825195, + "learning_rate": 3.48e-05, + "loss": 6.5797, + "step": 122 + }, + { + "epoch": 0.1968, + "grad_norm": 30.45292854309082, + "learning_rate": 3.51e-05, + "loss": 6.1, + "step": 123 + }, + { + "epoch": 0.1984, + "grad_norm": 35.50273895263672, + "learning_rate": 3.539999999999999e-05, + "loss": 6.3145, + "step": 124 + }, + { + "epoch": 0.2, + "grad_norm": 28.981294631958008, + "learning_rate": 3.5699999999999994e-05, + "loss": 5.9349, + "step": 125 + }, + { + "epoch": 0.2016, + "grad_norm": 24.302417755126953, + "learning_rate": 3.5999999999999994e-05, + "loss": 5.6469, + "step": 126 + }, + { + "epoch": 0.2032, + "grad_norm": 24.433305740356445, + "learning_rate": 3.6299999999999995e-05, + "loss": 5.5679, + "step": 127 + }, + { + "epoch": 0.2048, + "grad_norm": 26.34304428100586, + "learning_rate": 3.6599999999999995e-05, + "loss": 5.6555, + "step": 128 + }, + { + "epoch": 0.2064, + "grad_norm": 27.908578872680664, + "learning_rate": 3.6899999999999996e-05, + "loss": 5.7359, + "step": 129 + }, + { + "epoch": 0.208, + "grad_norm": 22.0655574798584, + "learning_rate": 3.7199999999999996e-05, + "loss": 5.407, + "step": 130 + }, + { + "epoch": 0.2096, + "grad_norm": 21.288557052612305, + "learning_rate": 3.75e-05, + "loss": 5.354, + "step": 131 + }, + { + "epoch": 0.2112, + "grad_norm": 13.6961088180542, + "learning_rate": 3.78e-05, + "loss": 5.0068, + "step": 132 + }, + { + "epoch": 0.2128, + "grad_norm": 15.53318977355957, + "learning_rate": 3.81e-05, + "loss": 5.0376, + "step": 133 + }, + { + "epoch": 0.2144, + "grad_norm": 13.725676536560059, + "learning_rate": 3.84e-05, + "loss": 4.9991, + "step": 134 + }, + { + "epoch": 0.216, + "grad_norm": 10.959695816040039, + "learning_rate": 3.87e-05, + "loss": 4.9169, + "step": 135 + }, + { + "epoch": 0.2176, + "grad_norm": 13.741371154785156, + "learning_rate": 3.9e-05, + "loss": 4.9547, + "step": 136 + }, + { + "epoch": 0.2192, + "grad_norm": 7.589344501495361, + "learning_rate": 3.93e-05, + "loss": 4.7349, + "step": 137 + }, + { + "epoch": 0.2208, + "grad_norm": 9.38797378540039, + "learning_rate": 3.96e-05, + "loss": 4.7968, + "step": 138 + }, + { + "epoch": 0.2224, + "grad_norm": 7.912783145904541, + "learning_rate": 3.99e-05, + "loss": 4.7519, + "step": 139 + }, + { + "epoch": 0.224, + "grad_norm": 4.842776298522949, + "learning_rate": 4.02e-05, + "loss": 4.6283, + "step": 140 + }, + { + "epoch": 0.2256, + "grad_norm": 3.9212424755096436, + "learning_rate": 4.05e-05, + "loss": 4.6736, + "step": 141 + }, + { + "epoch": 0.2272, + "grad_norm": 3.5087013244628906, + "learning_rate": 4.08e-05, + "loss": 4.5967, + "step": 142 + }, + { + "epoch": 0.2288, + "grad_norm": 7.529483318328857, + "learning_rate": 4.11e-05, + "loss": 4.7427, + "step": 143 + }, + { + "epoch": 0.2304, + "grad_norm": 4.376472473144531, + "learning_rate": 4.14e-05, + "loss": 4.6156, + "step": 144 + }, + { + "epoch": 0.232, + "grad_norm": 3.8088197708129883, + "learning_rate": 4.17e-05, + "loss": 4.5466, + "step": 145 + }, + { + "epoch": 0.2336, + "grad_norm": 3.3608834743499756, + "learning_rate": 4.2e-05, + "loss": 4.4873, + "step": 146 + }, + { + "epoch": 0.2352, + "grad_norm": 6.499572277069092, + "learning_rate": 4.229999999999999e-05, + "loss": 4.7131, + "step": 147 + }, + { + "epoch": 0.2368, + "grad_norm": 2.669039011001587, + "learning_rate": 4.259999999999999e-05, + "loss": 4.5439, + "step": 148 + }, + { + "epoch": 0.2384, + "grad_norm": 3.3045642375946045, + "learning_rate": 4.289999999999999e-05, + "loss": 4.5205, + "step": 149 + }, + { + "epoch": 0.24, + "grad_norm": 4.260329246520996, + "learning_rate": 4.319999999999999e-05, + "loss": 4.5319, + "step": 150 + }, + { + "epoch": 0.2416, + "grad_norm": 33.543827056884766, + "learning_rate": 4.3499999999999993e-05, + "loss": 5.1855, + "step": 151 + }, + { + "epoch": 0.2432, + "grad_norm": 12.372523307800293, + "learning_rate": 4.3799999999999994e-05, + "loss": 4.4256, + "step": 152 + }, + { + "epoch": 0.2448, + "grad_norm": 10.853720664978027, + "learning_rate": 4.4099999999999995e-05, + "loss": 4.3393, + "step": 153 + }, + { + "epoch": 0.2464, + "grad_norm": 31.088172912597656, + "learning_rate": 4.4399999999999995e-05, + "loss": 4.3422, + "step": 154 + }, + { + "epoch": 0.248, + "grad_norm": 2.647914171218872, + "learning_rate": 4.4699999999999996e-05, + "loss": 4.1807, + "step": 155 + }, + { + "epoch": 0.2496, + "grad_norm": 8.41458797454834, + "learning_rate": 4.4999999999999996e-05, + "loss": 4.3605, + "step": 156 + }, + { + "epoch": 0.2512, + "grad_norm": 2.505411386489868, + "learning_rate": 4.5299999999999997e-05, + "loss": 4.1126, + "step": 157 + }, + { + "epoch": 0.2528, + "grad_norm": 6.82640266418457, + "learning_rate": 4.56e-05, + "loss": 4.296, + "step": 158 + }, + { + "epoch": 0.2544, + "grad_norm": 4.044186115264893, + "learning_rate": 4.59e-05, + "loss": 4.0944, + "step": 159 + }, + { + "epoch": 0.256, + "grad_norm": 2.907984495162964, + "learning_rate": 4.62e-05, + "loss": 4.0258, + "step": 160 + }, + { + "epoch": 0.2576, + "grad_norm": 1.547238826751709, + "learning_rate": 4.65e-05, + "loss": 4.0542, + "step": 161 + }, + { + "epoch": 0.2592, + "grad_norm": 2.2683465480804443, + "learning_rate": 4.68e-05, + "loss": 4.0014, + "step": 162 + }, + { + "epoch": 0.2608, + "grad_norm": 2.3638885021209717, + "learning_rate": 4.709999999999999e-05, + "loss": 4.0179, + "step": 163 + }, + { + "epoch": 0.2624, + "grad_norm": 1.394767165184021, + "learning_rate": 4.7399999999999993e-05, + "loss": 4.0572, + "step": 164 + }, + { + "epoch": 0.264, + "grad_norm": 1.4333173036575317, + "learning_rate": 4.7699999999999994e-05, + "loss": 3.9437, + "step": 165 + }, + { + "epoch": 0.2656, + "grad_norm": 5.366274356842041, + "learning_rate": 4.7999999999999994e-05, + "loss": 4.0472, + "step": 166 + }, + { + "epoch": 0.2672, + "grad_norm": 1.0423429012298584, + "learning_rate": 4.8299999999999995e-05, + "loss": 3.9291, + "step": 167 + }, + { + "epoch": 0.2688, + "grad_norm": 1.9777644872665405, + "learning_rate": 4.8599999999999995e-05, + "loss": 3.945, + "step": 168 + }, + { + "epoch": 0.2704, + "grad_norm": 1.1812388896942139, + "learning_rate": 4.8899999999999996e-05, + "loss": 3.9491, + "step": 169 + }, + { + "epoch": 0.272, + "grad_norm": 1.822695016860962, + "learning_rate": 4.9199999999999997e-05, + "loss": 3.9102, + "step": 170 + }, + { + "epoch": 0.2736, + "grad_norm": 4.096814155578613, + "learning_rate": 4.95e-05, + "loss": 3.9744, + "step": 171 + }, + { + "epoch": 0.2752, + "grad_norm": 6.548384666442871, + "learning_rate": 4.98e-05, + "loss": 4.0524, + "step": 172 + }, + { + "epoch": 0.2768, + "grad_norm": 0.9052740931510925, + "learning_rate": 5.01e-05, + "loss": 3.8676, + "step": 173 + }, + { + "epoch": 0.2784, + "grad_norm": 1.3036408424377441, + "learning_rate": 5.04e-05, + "loss": 3.9335, + "step": 174 + }, + { + "epoch": 0.28, + "grad_norm": 3.0321638584136963, + "learning_rate": 5.07e-05, + "loss": 3.9693, + "step": 175 + }, + { + "epoch": 0.2816, + "grad_norm": 1.1829465627670288, + "learning_rate": 5.1e-05, + "loss": 3.9222, + "step": 176 + }, + { + "epoch": 0.2832, + "grad_norm": 2.459949254989624, + "learning_rate": 5.13e-05, + "loss": 3.8752, + "step": 177 + }, + { + "epoch": 0.2848, + "grad_norm": 2.6086926460266113, + "learning_rate": 5.1599999999999994e-05, + "loss": 3.8609, + "step": 178 + }, + { + "epoch": 0.2864, + "grad_norm": 0.8744645118713379, + "learning_rate": 5.1899999999999994e-05, + "loss": 3.8512, + "step": 179 + }, + { + "epoch": 0.288, + "grad_norm": 0.8597729802131653, + "learning_rate": 5.2199999999999995e-05, + "loss": 3.91, + "step": 180 + }, + { + "epoch": 0.2896, + "grad_norm": 0.656095564365387, + "learning_rate": 5.2499999999999995e-05, + "loss": 3.8791, + "step": 181 + }, + { + "epoch": 0.2912, + "grad_norm": 1.1524940729141235, + "learning_rate": 5.279999999999999e-05, + "loss": 3.8383, + "step": 182 + }, + { + "epoch": 0.2928, + "grad_norm": 1.2150574922561646, + "learning_rate": 5.309999999999999e-05, + "loss": 3.9343, + "step": 183 + }, + { + "epoch": 0.2944, + "grad_norm": 3.4474704265594482, + "learning_rate": 5.339999999999999e-05, + "loss": 3.8848, + "step": 184 + }, + { + "epoch": 0.296, + "grad_norm": 3.4787559509277344, + "learning_rate": 5.369999999999999e-05, + "loss": 3.8531, + "step": 185 + }, + { + "epoch": 0.2976, + "grad_norm": 1.857994556427002, + "learning_rate": 5.399999999999999e-05, + "loss": 3.82, + "step": 186 + }, + { + "epoch": 0.2992, + "grad_norm": 1.522542953491211, + "learning_rate": 5.429999999999999e-05, + "loss": 3.8566, + "step": 187 + }, + { + "epoch": 0.3008, + "grad_norm": 3.3759677410125732, + "learning_rate": 5.459999999999999e-05, + "loss": 3.9485, + "step": 188 + }, + { + "epoch": 0.3024, + "grad_norm": 0.7935701012611389, + "learning_rate": 5.489999999999999e-05, + "loss": 3.8288, + "step": 189 + }, + { + "epoch": 0.304, + "grad_norm": 0.7905306816101074, + "learning_rate": 5.519999999999999e-05, + "loss": 3.7754, + "step": 190 + }, + { + "epoch": 0.3056, + "grad_norm": 2.006058931350708, + "learning_rate": 5.5499999999999994e-05, + "loss": 3.9019, + "step": 191 + }, + { + "epoch": 0.3072, + "grad_norm": 0.9049056768417358, + "learning_rate": 5.5799999999999994e-05, + "loss": 3.8752, + "step": 192 + }, + { + "epoch": 0.3088, + "grad_norm": 0.7944461703300476, + "learning_rate": 5.6099999999999995e-05, + "loss": 3.9383, + "step": 193 + }, + { + "epoch": 0.3104, + "grad_norm": 2.660388469696045, + "learning_rate": 5.6399999999999995e-05, + "loss": 3.897, + "step": 194 + }, + { + "epoch": 0.312, + "grad_norm": 1.1633754968643188, + "learning_rate": 5.6699999999999996e-05, + "loss": 3.8026, + "step": 195 + }, + { + "epoch": 0.3136, + "grad_norm": 1.0256235599517822, + "learning_rate": 5.6999999999999996e-05, + "loss": 3.8239, + "step": 196 + }, + { + "epoch": 0.3152, + "grad_norm": 0.90212082862854, + "learning_rate": 5.73e-05, + "loss": 3.8742, + "step": 197 + }, + { + "epoch": 0.3168, + "grad_norm": 1.7041653394699097, + "learning_rate": 5.76e-05, + "loss": 3.8631, + "step": 198 + }, + { + "epoch": 0.3184, + "grad_norm": 5.0367889404296875, + "learning_rate": 5.79e-05, + "loss": 3.9776, + "step": 199 + }, + { + "epoch": 0.32, + "grad_norm": 1.7132095098495483, + "learning_rate": 5.82e-05, + "loss": 3.8886, + "step": 200 + }, + { + "epoch": 0.3216, + "grad_norm": 59.95878219604492, + "learning_rate": 5.85e-05, + "loss": 4.4781, + "step": 201 + }, + { + "epoch": 0.3232, + "grad_norm": 20.7462215423584, + "learning_rate": 5.88e-05, + "loss": 4.1558, + "step": 202 + }, + { + "epoch": 0.3248, + "grad_norm": 7.668243408203125, + "learning_rate": 5.91e-05, + "loss": 3.8673, + "step": 203 + }, + { + "epoch": 0.3264, + "grad_norm": 20.701831817626953, + "learning_rate": 5.94e-05, + "loss": 4.2043, + "step": 204 + }, + { + "epoch": 0.328, + "grad_norm": 15.181974411010742, + "learning_rate": 5.97e-05, + "loss": 4.1824, + "step": 205 + }, + { + "epoch": 0.3296, + "grad_norm": 6.447072982788086, + "learning_rate": 5.9999999999999995e-05, + "loss": 3.8739, + "step": 206 + }, + { + "epoch": 0.3312, + "grad_norm": 1.1761301755905151, + "learning_rate": 6.0299999999999995e-05, + "loss": 3.7815, + "step": 207 + }, + { + "epoch": 0.3328, + "grad_norm": 2.660413980484009, + "learning_rate": 6.0599999999999996e-05, + "loss": 3.8145, + "step": 208 + }, + { + "epoch": 0.3344, + "grad_norm": 2.368325710296631, + "learning_rate": 6.0899999999999996e-05, + "loss": 3.8349, + "step": 209 + }, + { + "epoch": 0.336, + "grad_norm": 2.1028411388397217, + "learning_rate": 6.12e-05, + "loss": 3.8754, + "step": 210 + }, + { + "epoch": 0.3376, + "grad_norm": 5.019386291503906, + "learning_rate": 6.149999999999999e-05, + "loss": 4.0562, + "step": 211 + }, + { + "epoch": 0.3392, + "grad_norm": 1.4638348817825317, + "learning_rate": 6.18e-05, + "loss": 3.8762, + "step": 212 + }, + { + "epoch": 0.3408, + "grad_norm": 4.489582061767578, + "learning_rate": 6.209999999999999e-05, + "loss": 3.8016, + "step": 213 + }, + { + "epoch": 0.3424, + "grad_norm": 4.419440269470215, + "learning_rate": 6.239999999999999e-05, + "loss": 3.7819, + "step": 214 + }, + { + "epoch": 0.344, + "grad_norm": 1.0869696140289307, + "learning_rate": 6.269999999999999e-05, + "loss": 3.8609, + "step": 215 + }, + { + "epoch": 0.3456, + "grad_norm": 2.233304738998413, + "learning_rate": 6.299999999999999e-05, + "loss": 3.7453, + "step": 216 + }, + { + "epoch": 0.3472, + "grad_norm": 1.8570388555526733, + "learning_rate": 6.33e-05, + "loss": 3.8022, + "step": 217 + }, + { + "epoch": 0.3488, + "grad_norm": 1.5490303039550781, + "learning_rate": 6.359999999999999e-05, + "loss": 3.8208, + "step": 218 + }, + { + "epoch": 0.3504, + "grad_norm": 1.4915709495544434, + "learning_rate": 6.39e-05, + "loss": 3.7659, + "step": 219 + }, + { + "epoch": 0.352, + "grad_norm": 4.611435890197754, + "learning_rate": 6.419999999999999e-05, + "loss": 3.7905, + "step": 220 + }, + { + "epoch": 0.3536, + "grad_norm": 3.2338948249816895, + "learning_rate": 6.45e-05, + "loss": 3.769, + "step": 221 + }, + { + "epoch": 0.3552, + "grad_norm": 1.7731317281723022, + "learning_rate": 6.479999999999999e-05, + "loss": 3.7948, + "step": 222 + }, + { + "epoch": 0.3568, + "grad_norm": 4.100704669952393, + "learning_rate": 6.51e-05, + "loss": 3.9139, + "step": 223 + }, + { + "epoch": 0.3584, + "grad_norm": 2.826249599456787, + "learning_rate": 6.539999999999999e-05, + "loss": 3.7927, + "step": 224 + }, + { + "epoch": 0.36, + "grad_norm": 2.4306230545043945, + "learning_rate": 6.57e-05, + "loss": 3.7451, + "step": 225 + }, + { + "epoch": 0.3616, + "grad_norm": 2.495347499847412, + "learning_rate": 6.599999999999999e-05, + "loss": 3.7458, + "step": 226 + }, + { + "epoch": 0.3632, + "grad_norm": 1.5291261672973633, + "learning_rate": 6.63e-05, + "loss": 3.7838, + "step": 227 + }, + { + "epoch": 0.3648, + "grad_norm": 1.6629663705825806, + "learning_rate": 6.659999999999999e-05, + "loss": 3.7364, + "step": 228 + }, + { + "epoch": 0.3664, + "grad_norm": 1.6959452629089355, + "learning_rate": 6.69e-05, + "loss": 3.7711, + "step": 229 + }, + { + "epoch": 0.368, + "grad_norm": 2.6228511333465576, + "learning_rate": 6.72e-05, + "loss": 3.7591, + "step": 230 + }, + { + "epoch": 0.3696, + "grad_norm": 7.408218860626221, + "learning_rate": 6.75e-05, + "loss": 3.9036, + "step": 231 + }, + { + "epoch": 0.3712, + "grad_norm": 1.7726941108703613, + "learning_rate": 6.78e-05, + "loss": 3.7256, + "step": 232 + }, + { + "epoch": 0.3728, + "grad_norm": 1.7417742013931274, + "learning_rate": 6.81e-05, + "loss": 3.7479, + "step": 233 + }, + { + "epoch": 0.3744, + "grad_norm": 2.71058988571167, + "learning_rate": 6.84e-05, + "loss": 3.7953, + "step": 234 + }, + { + "epoch": 0.376, + "grad_norm": 1.043075442314148, + "learning_rate": 6.87e-05, + "loss": 3.7091, + "step": 235 + }, + { + "epoch": 0.3776, + "grad_norm": 3.6556646823883057, + "learning_rate": 6.9e-05, + "loss": 3.8753, + "step": 236 + }, + { + "epoch": 0.3792, + "grad_norm": 1.024652361869812, + "learning_rate": 6.93e-05, + "loss": 3.8208, + "step": 237 + }, + { + "epoch": 0.3808, + "grad_norm": 2.1579537391662598, + "learning_rate": 6.96e-05, + "loss": 3.7698, + "step": 238 + }, + { + "epoch": 0.3824, + "grad_norm": 2.586552858352661, + "learning_rate": 6.989999999999999e-05, + "loss": 3.7483, + "step": 239 + }, + { + "epoch": 0.384, + "grad_norm": 2.8830111026763916, + "learning_rate": 7.02e-05, + "loss": 3.7447, + "step": 240 + }, + { + "epoch": 0.3856, + "grad_norm": 1.5939990282058716, + "learning_rate": 7.049999999999999e-05, + "loss": 3.9125, + "step": 241 + }, + { + "epoch": 0.3872, + "grad_norm": 1.9569231271743774, + "learning_rate": 7.079999999999999e-05, + "loss": 3.698, + "step": 242 + }, + { + "epoch": 0.3888, + "grad_norm": 2.1140618324279785, + "learning_rate": 7.11e-05, + "loss": 3.7141, + "step": 243 + }, + { + "epoch": 0.3904, + "grad_norm": 2.5653762817382812, + "learning_rate": 7.139999999999999e-05, + "loss": 3.7429, + "step": 244 + }, + { + "epoch": 0.392, + "grad_norm": 4.119378089904785, + "learning_rate": 7.17e-05, + "loss": 3.8172, + "step": 245 + }, + { + "epoch": 0.3936, + "grad_norm": 1.2246354818344116, + "learning_rate": 7.199999999999999e-05, + "loss": 3.7213, + "step": 246 + }, + { + "epoch": 0.3952, + "grad_norm": 3.104062557220459, + "learning_rate": 7.23e-05, + "loss": 3.8297, + "step": 247 + }, + { + "epoch": 0.3968, + "grad_norm": 2.702347993850708, + "learning_rate": 7.259999999999999e-05, + "loss": 3.8384, + "step": 248 + }, + { + "epoch": 0.3984, + "grad_norm": 1.7538706064224243, + "learning_rate": 7.29e-05, + "loss": 3.8097, + "step": 249 + }, + { + "epoch": 0.4, + "grad_norm": 1.4835972785949707, + "learning_rate": 7.319999999999999e-05, + "loss": 3.8442, + "step": 250 + }, + { + "epoch": 0.4016, + "grad_norm": 121.84357452392578, + "learning_rate": 7.35e-05, + "loss": 6.6492, + "step": 251 + }, + { + "epoch": 0.4032, + "grad_norm": 6.5602216720581055, + "learning_rate": 7.379999999999999e-05, + "loss": 3.7888, + "step": 252 + }, + { + "epoch": 0.4048, + "grad_norm": 6.373597145080566, + "learning_rate": 7.41e-05, + "loss": 3.8137, + "step": 253 + }, + { + "epoch": 0.4064, + "grad_norm": 15.365026473999023, + "learning_rate": 7.439999999999999e-05, + "loss": 4.0826, + "step": 254 + }, + { + "epoch": 0.408, + "grad_norm": 4.488667964935303, + "learning_rate": 7.47e-05, + "loss": 3.8556, + "step": 255 + }, + { + "epoch": 0.4096, + "grad_norm": 1.5894265174865723, + "learning_rate": 7.5e-05, + "loss": 3.785, + "step": 256 + }, + { + "epoch": 0.4112, + "grad_norm": 2.2378318309783936, + "learning_rate": 7.529999999999999e-05, + "loss": 3.7889, + "step": 257 + }, + { + "epoch": 0.4128, + "grad_norm": 2.854876756668091, + "learning_rate": 7.56e-05, + "loss": 3.8141, + "step": 258 + }, + { + "epoch": 0.4144, + "grad_norm": 5.561819076538086, + "learning_rate": 7.589999999999999e-05, + "loss": 3.7821, + "step": 259 + }, + { + "epoch": 0.416, + "grad_norm": 3.3661203384399414, + "learning_rate": 7.62e-05, + "loss": 3.7868, + "step": 260 + }, + { + "epoch": 0.4176, + "grad_norm": 1.3599531650543213, + "learning_rate": 7.649999999999999e-05, + "loss": 3.8347, + "step": 261 + }, + { + "epoch": 0.4192, + "grad_norm": 1.306137204170227, + "learning_rate": 7.68e-05, + "loss": 3.7819, + "step": 262 + }, + { + "epoch": 0.4208, + "grad_norm": 3.109072685241699, + "learning_rate": 7.709999999999999e-05, + "loss": 3.7242, + "step": 263 + }, + { + "epoch": 0.4224, + "grad_norm": 1.8001930713653564, + "learning_rate": 7.74e-05, + "loss": 3.7492, + "step": 264 + }, + { + "epoch": 0.424, + "grad_norm": 0.8629224300384521, + "learning_rate": 7.769999999999999e-05, + "loss": 3.7207, + "step": 265 + }, + { + "epoch": 0.4256, + "grad_norm": 3.7459707260131836, + "learning_rate": 7.8e-05, + "loss": 3.7963, + "step": 266 + }, + { + "epoch": 0.4272, + "grad_norm": 1.336196780204773, + "learning_rate": 7.829999999999999e-05, + "loss": 3.6887, + "step": 267 + }, + { + "epoch": 0.4288, + "grad_norm": 1.0606918334960938, + "learning_rate": 7.86e-05, + "loss": 3.7572, + "step": 268 + }, + { + "epoch": 0.4304, + "grad_norm": 2.4319143295288086, + "learning_rate": 7.89e-05, + "loss": 3.7525, + "step": 269 + }, + { + "epoch": 0.432, + "grad_norm": 1.2231545448303223, + "learning_rate": 7.92e-05, + "loss": 3.7485, + "step": 270 + }, + { + "epoch": 0.4336, + "grad_norm": 0.9850172400474548, + "learning_rate": 7.95e-05, + "loss": 3.7035, + "step": 271 + }, + { + "epoch": 0.4352, + "grad_norm": 0.7243953347206116, + "learning_rate": 7.98e-05, + "loss": 3.7558, + "step": 272 + }, + { + "epoch": 0.4368, + "grad_norm": 2.755706310272217, + "learning_rate": 8.01e-05, + "loss": 3.7261, + "step": 273 + }, + { + "epoch": 0.4384, + "grad_norm": 1.264993667602539, + "learning_rate": 8.04e-05, + "loss": 3.7083, + "step": 274 + }, + { + "epoch": 0.44, + "grad_norm": 0.8790886402130127, + "learning_rate": 8.07e-05, + "loss": 3.7071, + "step": 275 + }, + { + "epoch": 0.4416, + "grad_norm": 3.400172472000122, + "learning_rate": 8.1e-05, + "loss": 3.7469, + "step": 276 + }, + { + "epoch": 0.4432, + "grad_norm": 2.231881856918335, + "learning_rate": 8.13e-05, + "loss": 3.6753, + "step": 277 + }, + { + "epoch": 0.4448, + "grad_norm": 2.1918094158172607, + "learning_rate": 8.16e-05, + "loss": 3.7519, + "step": 278 + }, + { + "epoch": 0.4464, + "grad_norm": 0.9959167242050171, + "learning_rate": 8.19e-05, + "loss": 3.7235, + "step": 279 + }, + { + "epoch": 0.448, + "grad_norm": 1.0441361665725708, + "learning_rate": 8.22e-05, + "loss": 3.6868, + "step": 280 + }, + { + "epoch": 0.4496, + "grad_norm": 1.4733455181121826, + "learning_rate": 8.25e-05, + "loss": 3.6998, + "step": 281 + }, + { + "epoch": 0.4512, + "grad_norm": 0.8500906229019165, + "learning_rate": 8.28e-05, + "loss": 3.7039, + "step": 282 + }, + { + "epoch": 0.4528, + "grad_norm": 1.184400200843811, + "learning_rate": 8.31e-05, + "loss": 3.6784, + "step": 283 + }, + { + "epoch": 0.4544, + "grad_norm": 0.7087394595146179, + "learning_rate": 8.34e-05, + "loss": 3.7611, + "step": 284 + }, + { + "epoch": 0.456, + "grad_norm": 1.7900768518447876, + "learning_rate": 8.37e-05, + "loss": 3.7507, + "step": 285 + }, + { + "epoch": 0.4576, + "grad_norm": 0.7228295803070068, + "learning_rate": 8.4e-05, + "loss": 3.7109, + "step": 286 + }, + { + "epoch": 0.4592, + "grad_norm": 2.9249191284179688, + "learning_rate": 8.43e-05, + "loss": 3.759, + "step": 287 + }, + { + "epoch": 0.4608, + "grad_norm": 0.8148625493049622, + "learning_rate": 8.459999999999998e-05, + "loss": 3.6881, + "step": 288 + }, + { + "epoch": 0.4624, + "grad_norm": 0.9529969692230225, + "learning_rate": 8.489999999999999e-05, + "loss": 3.6867, + "step": 289 + }, + { + "epoch": 0.464, + "grad_norm": 1.3643995523452759, + "learning_rate": 8.519999999999998e-05, + "loss": 3.6983, + "step": 290 + }, + { + "epoch": 0.4656, + "grad_norm": 0.9659025073051453, + "learning_rate": 8.549999999999999e-05, + "loss": 3.6865, + "step": 291 + }, + { + "epoch": 0.4672, + "grad_norm": 4.1877946853637695, + "learning_rate": 8.579999999999998e-05, + "loss": 3.7463, + "step": 292 + }, + { + "epoch": 0.4688, + "grad_norm": 2.1806647777557373, + "learning_rate": 8.609999999999999e-05, + "loss": 3.7738, + "step": 293 + }, + { + "epoch": 0.4704, + "grad_norm": 1.7634769678115845, + "learning_rate": 8.639999999999999e-05, + "loss": 3.7514, + "step": 294 + }, + { + "epoch": 0.472, + "grad_norm": 1.0072358846664429, + "learning_rate": 8.669999999999998e-05, + "loss": 3.684, + "step": 295 + }, + { + "epoch": 0.4736, + "grad_norm": 0.8580794930458069, + "learning_rate": 8.699999999999999e-05, + "loss": 3.7447, + "step": 296 + }, + { + "epoch": 0.4752, + "grad_norm": 1.7601121664047241, + "learning_rate": 8.729999999999998e-05, + "loss": 3.7797, + "step": 297 + }, + { + "epoch": 0.4768, + "grad_norm": 3.091855525970459, + "learning_rate": 8.759999999999999e-05, + "loss": 3.723, + "step": 298 + }, + { + "epoch": 0.4784, + "grad_norm": 1.7820290327072144, + "learning_rate": 8.789999999999998e-05, + "loss": 3.7341, + "step": 299 + }, + { + "epoch": 0.48, + "grad_norm": 4.246450901031494, + "learning_rate": 8.819999999999999e-05, + "loss": 4.0743, + "step": 300 + }, + { + "epoch": 0.4816, + "grad_norm": 72.1017837524414, + "learning_rate": 8.849999999999998e-05, + "loss": 4.3474, + "step": 301 + }, + { + "epoch": 0.4832, + "grad_norm": 7.5009355545043945, + "learning_rate": 8.879999999999999e-05, + "loss": 3.862, + "step": 302 + }, + { + "epoch": 0.4848, + "grad_norm": 11.676750183105469, + "learning_rate": 8.909999999999998e-05, + "loss": 4.0007, + "step": 303 + }, + { + "epoch": 0.4864, + "grad_norm": 2.041074275970459, + "learning_rate": 8.939999999999999e-05, + "loss": 3.737, + "step": 304 + }, + { + "epoch": 0.488, + "grad_norm": 7.677201747894287, + "learning_rate": 8.969999999999998e-05, + "loss": 3.9839, + "step": 305 + }, + { + "epoch": 0.4896, + "grad_norm": 6.7263617515563965, + "learning_rate": 8.999999999999999e-05, + "loss": 4.1026, + "step": 306 + }, + { + "epoch": 0.4912, + "grad_norm": 4.236703395843506, + "learning_rate": 9.029999999999999e-05, + "loss": 3.8479, + "step": 307 + }, + { + "epoch": 0.4928, + "grad_norm": 5.840229511260986, + "learning_rate": 9.059999999999999e-05, + "loss": 3.7812, + "step": 308 + }, + { + "epoch": 0.4944, + "grad_norm": 8.028979301452637, + "learning_rate": 9.089999999999999e-05, + "loss": 3.8126, + "step": 309 + }, + { + "epoch": 0.496, + "grad_norm": 5.721709728240967, + "learning_rate": 9.12e-05, + "loss": 3.8407, + "step": 310 + }, + { + "epoch": 0.4976, + "grad_norm": 2.0302011966705322, + "learning_rate": 9.149999999999999e-05, + "loss": 3.8688, + "step": 311 + }, + { + "epoch": 0.4992, + "grad_norm": 3.8033218383789062, + "learning_rate": 9.18e-05, + "loss": 3.7185, + "step": 312 + }, + { + "epoch": 0.5008, + "grad_norm": 8.654851913452148, + "learning_rate": 9.209999999999999e-05, + "loss": 3.6954, + "step": 313 + }, + { + "epoch": 0.5024, + "grad_norm": 3.6997175216674805, + "learning_rate": 9.24e-05, + "loss": 3.8454, + "step": 314 + }, + { + "epoch": 0.504, + "grad_norm": 5.1004438400268555, + "learning_rate": 9.269999999999999e-05, + "loss": 3.8321, + "step": 315 + }, + { + "epoch": 0.5056, + "grad_norm": 1.3075134754180908, + "learning_rate": 9.3e-05, + "loss": 3.6619, + "step": 316 + }, + { + "epoch": 0.5072, + "grad_norm": 1.137516736984253, + "learning_rate": 9.329999999999999e-05, + "loss": 3.6382, + "step": 317 + }, + { + "epoch": 0.5088, + "grad_norm": 9.24795913696289, + "learning_rate": 9.36e-05, + "loss": 4.1195, + "step": 318 + }, + { + "epoch": 0.5104, + "grad_norm": 1.6506896018981934, + "learning_rate": 9.389999999999999e-05, + "loss": 3.6742, + "step": 319 + }, + { + "epoch": 0.512, + "grad_norm": 1.2333859205245972, + "learning_rate": 9.419999999999999e-05, + "loss": 3.6605, + "step": 320 + }, + { + "epoch": 0.5136, + "grad_norm": 1.074317455291748, + "learning_rate": 9.449999999999999e-05, + "loss": 3.6296, + "step": 321 + }, + { + "epoch": 0.5152, + "grad_norm": 0.8660179376602173, + "learning_rate": 9.479999999999999e-05, + "loss": 3.6996, + "step": 322 + }, + { + "epoch": 0.5168, + "grad_norm": 4.370603084564209, + "learning_rate": 9.51e-05, + "loss": 3.6899, + "step": 323 + }, + { + "epoch": 0.5184, + "grad_norm": 4.468715667724609, + "learning_rate": 9.539999999999999e-05, + "loss": 3.6464, + "step": 324 + }, + { + "epoch": 0.52, + "grad_norm": 3.4510672092437744, + "learning_rate": 9.57e-05, + "loss": 3.6554, + "step": 325 + }, + { + "epoch": 0.5216, + "grad_norm": 4.948469638824463, + "learning_rate": 9.599999999999999e-05, + "loss": 3.6866, + "step": 326 + }, + { + "epoch": 0.5232, + "grad_norm": 1.9473222494125366, + "learning_rate": 9.63e-05, + "loss": 3.6517, + "step": 327 + }, + { + "epoch": 0.5248, + "grad_norm": 0.8309019207954407, + "learning_rate": 9.659999999999999e-05, + "loss": 3.6118, + "step": 328 + }, + { + "epoch": 0.5264, + "grad_norm": 1.0858643054962158, + "learning_rate": 9.69e-05, + "loss": 3.7663, + "step": 329 + }, + { + "epoch": 0.528, + "grad_norm": 2.959810495376587, + "learning_rate": 9.719999999999999e-05, + "loss": 3.6217, + "step": 330 + }, + { + "epoch": 0.5296, + "grad_norm": 3.2029099464416504, + "learning_rate": 9.75e-05, + "loss": 3.6604, + "step": 331 + }, + { + "epoch": 0.5312, + "grad_norm": 2.6211912631988525, + "learning_rate": 9.779999999999999e-05, + "loss": 3.6423, + "step": 332 + }, + { + "epoch": 0.5328, + "grad_norm": 1.876083493232727, + "learning_rate": 9.81e-05, + "loss": 3.656, + "step": 333 + }, + { + "epoch": 0.5344, + "grad_norm": 1.3034205436706543, + "learning_rate": 9.839999999999999e-05, + "loss": 3.6448, + "step": 334 + }, + { + "epoch": 0.536, + "grad_norm": 2.9727463722229004, + "learning_rate": 9.87e-05, + "loss": 3.6883, + "step": 335 + }, + { + "epoch": 0.5376, + "grad_norm": 5.742773532867432, + "learning_rate": 9.9e-05, + "loss": 3.7493, + "step": 336 + }, + { + "epoch": 0.5392, + "grad_norm": 11.241994857788086, + "learning_rate": 9.93e-05, + "loss": 3.886, + "step": 337 + }, + { + "epoch": 0.5408, + "grad_norm": 5.17420768737793, + "learning_rate": 9.96e-05, + "loss": 3.6835, + "step": 338 + }, + { + "epoch": 0.5424, + "grad_norm": 5.169139385223389, + "learning_rate": 9.99e-05, + "loss": 3.6478, + "step": 339 + }, + { + "epoch": 0.544, + "grad_norm": 1.9756124019622803, + "learning_rate": 0.0001002, + "loss": 3.6164, + "step": 340 + }, + { + "epoch": 0.5456, + "grad_norm": 2.4662652015686035, + "learning_rate": 0.0001005, + "loss": 3.6319, + "step": 341 + }, + { + "epoch": 0.5472, + "grad_norm": 2.1132149696350098, + "learning_rate": 0.0001008, + "loss": 3.6865, + "step": 342 + }, + { + "epoch": 0.5488, + "grad_norm": 2.5962605476379395, + "learning_rate": 0.0001011, + "loss": 3.6532, + "step": 343 + }, + { + "epoch": 0.5504, + "grad_norm": 3.534571647644043, + "learning_rate": 0.0001014, + "loss": 3.5703, + "step": 344 + }, + { + "epoch": 0.552, + "grad_norm": 2.0960686206817627, + "learning_rate": 0.00010169999999999999, + "loss": 3.6295, + "step": 345 + }, + { + "epoch": 0.5536, + "grad_norm": 1.5675119161605835, + "learning_rate": 0.000102, + "loss": 3.6976, + "step": 346 + }, + { + "epoch": 0.5552, + "grad_norm": 1.5193946361541748, + "learning_rate": 0.00010229999999999999, + "loss": 3.6302, + "step": 347 + }, + { + "epoch": 0.5568, + "grad_norm": 2.2176620960235596, + "learning_rate": 0.0001026, + "loss": 3.6557, + "step": 348 + }, + { + "epoch": 0.5584, + "grad_norm": 1.6280394792556763, + "learning_rate": 0.0001029, + "loss": 3.5879, + "step": 349 + }, + { + "epoch": 0.56, + "grad_norm": NaN, + "learning_rate": 0.0001029, + "loss": 3.8835, + "step": 350 + }, + { + "epoch": 0.5616, + "grad_norm": 15.113051414489746, + "learning_rate": 0.00010319999999999999, + "loss": 3.9648, + "step": 351 + }, + { + "epoch": 0.5632, + "grad_norm": 15.56413459777832, + "learning_rate": 0.00010349999999999998, + "loss": 3.9607, + "step": 352 + }, + { + "epoch": 0.5648, + "grad_norm": 12.081535339355469, + "learning_rate": 0.00010379999999999999, + "loss": 3.9409, + "step": 353 + }, + { + "epoch": 0.5664, + "grad_norm": 15.150484085083008, + "learning_rate": 0.00010409999999999998, + "loss": 4.0052, + "step": 354 + }, + { + "epoch": 0.568, + "grad_norm": 3.023535966873169, + "learning_rate": 0.00010439999999999999, + "loss": 3.6932, + "step": 355 + }, + { + "epoch": 0.5696, + "grad_norm": 3.2386553287506104, + "learning_rate": 0.00010469999999999998, + "loss": 3.6472, + "step": 356 + }, + { + "epoch": 0.5712, + "grad_norm": 3.457566022872925, + "learning_rate": 0.00010499999999999999, + "loss": 3.7199, + "step": 357 + }, + { + "epoch": 0.5728, + "grad_norm": 7.193448066711426, + "learning_rate": 0.00010529999999999998, + "loss": 3.678, + "step": 358 + }, + { + "epoch": 0.5744, + "grad_norm": 7.7766032218933105, + "learning_rate": 0.00010559999999999998, + "loss": 3.7412, + "step": 359 + }, + { + "epoch": 0.576, + "grad_norm": 7.320962905883789, + "learning_rate": 0.00010589999999999999, + "loss": 3.6939, + "step": 360 + }, + { + "epoch": 0.5776, + "grad_norm": 6.841294288635254, + "learning_rate": 0.00010619999999999998, + "loss": 3.6927, + "step": 361 + }, + { + "epoch": 0.5792, + "grad_norm": 3.793191909790039, + "learning_rate": 0.00010649999999999999, + "loss": 3.9034, + "step": 362 + }, + { + "epoch": 0.5808, + "grad_norm": 6.122262954711914, + "learning_rate": 0.00010679999999999998, + "loss": 3.6405, + "step": 363 + }, + { + "epoch": 0.5824, + "grad_norm": 5.361271381378174, + "learning_rate": 0.00010709999999999999, + "loss": 3.6555, + "step": 364 + }, + { + "epoch": 0.584, + "grad_norm": 3.1824538707733154, + "learning_rate": 0.00010739999999999998, + "loss": 3.6008, + "step": 365 + }, + { + "epoch": 0.5856, + "grad_norm": 2.6935617923736572, + "learning_rate": 0.00010769999999999999, + "loss": 3.6435, + "step": 366 + }, + { + "epoch": 0.5872, + "grad_norm": 3.3396615982055664, + "learning_rate": 0.00010799999999999998, + "loss": 3.6248, + "step": 367 + }, + { + "epoch": 0.5888, + "grad_norm": 4.874233722686768, + "learning_rate": 0.00010829999999999999, + "loss": 3.569, + "step": 368 + }, + { + "epoch": 0.5904, + "grad_norm": 6.270530700683594, + "learning_rate": 0.00010859999999999998, + "loss": 3.5806, + "step": 369 + }, + { + "epoch": 0.592, + "grad_norm": 5.3780837059021, + "learning_rate": 0.00010889999999999999, + "loss": 3.6301, + "step": 370 + }, + { + "epoch": 0.5936, + "grad_norm": 7.1194634437561035, + "learning_rate": 0.00010919999999999998, + "loss": 3.6537, + "step": 371 + }, + { + "epoch": 0.5952, + "grad_norm": 4.226306438446045, + "learning_rate": 0.00010949999999999999, + "loss": 3.6081, + "step": 372 + }, + { + "epoch": 0.5968, + "grad_norm": 1.910267949104309, + "learning_rate": 0.00010979999999999999, + "loss": 3.5842, + "step": 373 + }, + { + "epoch": 0.5984, + "grad_norm": 0.9893857836723328, + "learning_rate": 0.00011009999999999999, + "loss": 3.5371, + "step": 374 + }, + { + "epoch": 0.6, + "grad_norm": 2.1459553241729736, + "learning_rate": 0.00011039999999999999, + "loss": 3.6621, + "step": 375 + }, + { + "epoch": 0.6016, + "grad_norm": 0.9111649394035339, + "learning_rate": 0.0001107, + "loss": 3.7058, + "step": 376 + }, + { + "epoch": 0.6032, + "grad_norm": 1.2992879152297974, + "learning_rate": 0.00011099999999999999, + "loss": 3.5917, + "step": 377 + }, + { + "epoch": 0.6048, + "grad_norm": 1.082714319229126, + "learning_rate": 0.0001113, + "loss": 3.553, + "step": 378 + }, + { + "epoch": 0.6064, + "grad_norm": 1.2292723655700684, + "learning_rate": 0.00011159999999999999, + "loss": 3.5233, + "step": 379 + }, + { + "epoch": 0.608, + "grad_norm": 1.739877462387085, + "learning_rate": 0.0001119, + "loss": 3.4612, + "step": 380 + }, + { + "epoch": 0.6096, + "grad_norm": 1.2230865955352783, + "learning_rate": 0.00011219999999999999, + "loss": 3.5631, + "step": 381 + }, + { + "epoch": 0.6112, + "grad_norm": 2.661285877227783, + "learning_rate": 0.0001125, + "loss": 3.4753, + "step": 382 + }, + { + "epoch": 0.6128, + "grad_norm": 4.004184246063232, + "learning_rate": 0.00011279999999999999, + "loss": 3.5881, + "step": 383 + }, + { + "epoch": 0.6144, + "grad_norm": 2.7191810607910156, + "learning_rate": 0.00011309999999999998, + "loss": 3.4957, + "step": 384 + }, + { + "epoch": 0.616, + "grad_norm": 1.195847988128662, + "learning_rate": 0.00011339999999999999, + "loss": 3.4449, + "step": 385 + }, + { + "epoch": 0.6176, + "grad_norm": 2.1945791244506836, + "learning_rate": 0.00011369999999999999, + "loss": 3.5439, + "step": 386 + }, + { + "epoch": 0.6192, + "grad_norm": 1.4968360662460327, + "learning_rate": 0.00011399999999999999, + "loss": 3.5066, + "step": 387 + }, + { + "epoch": 0.6208, + "grad_norm": 2.2760701179504395, + "learning_rate": 0.00011429999999999999, + "loss": 3.5232, + "step": 388 + }, + { + "epoch": 0.6224, + "grad_norm": 0.8866758942604065, + "learning_rate": 0.0001146, + "loss": 3.5362, + "step": 389 + }, + { + "epoch": 0.624, + "grad_norm": 1.5824236869812012, + "learning_rate": 0.00011489999999999999, + "loss": 3.533, + "step": 390 + }, + { + "epoch": 0.6256, + "grad_norm": 1.9433133602142334, + "learning_rate": 0.0001152, + "loss": 3.5246, + "step": 391 + }, + { + "epoch": 0.6272, + "grad_norm": 1.2729533910751343, + "learning_rate": 0.00011549999999999999, + "loss": 3.5464, + "step": 392 + }, + { + "epoch": 0.6288, + "grad_norm": 1.4902565479278564, + "learning_rate": 0.0001158, + "loss": 3.5223, + "step": 393 + }, + { + "epoch": 0.6304, + "grad_norm": 2.5921518802642822, + "learning_rate": 0.00011609999999999999, + "loss": 3.535, + "step": 394 + }, + { + "epoch": 0.632, + "grad_norm": 1.2368192672729492, + "learning_rate": 0.0001164, + "loss": 3.4233, + "step": 395 + }, + { + "epoch": 0.6336, + "grad_norm": 1.8510901927947998, + "learning_rate": 0.00011669999999999999, + "loss": 3.681, + "step": 396 + }, + { + "epoch": 0.6352, + "grad_norm": 2.9283556938171387, + "learning_rate": 0.000117, + "loss": 3.5726, + "step": 397 + }, + { + "epoch": 0.6368, + "grad_norm": 3.5072672367095947, + "learning_rate": 0.00011729999999999999, + "loss": 3.466, + "step": 398 + }, + { + "epoch": 0.6384, + "grad_norm": 1.411615014076233, + "learning_rate": 0.0001176, + "loss": 3.5429, + "step": 399 + }, + { + "epoch": 0.64, + "grad_norm": NaN, + "learning_rate": 0.0001176, + "loss": 3.791, + "step": 400 + }, + { + "epoch": 0.6416, + "grad_norm": 148.1685028076172, + "learning_rate": 0.00011789999999999999, + "loss": 4.8401, + "step": 401 + }, + { + "epoch": 0.6432, + "grad_norm": 12.178915977478027, + "learning_rate": 0.0001182, + "loss": 3.9567, + "step": 402 + }, + { + "epoch": 0.6448, + "grad_norm": 5.5876264572143555, + "learning_rate": 0.0001185, + "loss": 3.6857, + "step": 403 + }, + { + "epoch": 0.6464, + "grad_norm": 5.527307033538818, + "learning_rate": 0.0001188, + "loss": 3.7207, + "step": 404 + }, + { + "epoch": 0.648, + "grad_norm": 2.253143548965454, + "learning_rate": 0.0001191, + "loss": 3.6194, + "step": 405 + }, + { + "epoch": 0.6496, + "grad_norm": 1.694656491279602, + "learning_rate": 0.0001194, + "loss": 3.7182, + "step": 406 + }, + { + "epoch": 0.6512, + "grad_norm": 4.179173469543457, + "learning_rate": 0.0001197, + "loss": 3.6074, + "step": 407 + }, + { + "epoch": 0.6528, + "grad_norm": 5.340901851654053, + "learning_rate": 0.00011999999999999999, + "loss": 3.5508, + "step": 408 + }, + { + "epoch": 0.6544, + "grad_norm": 7.354061126708984, + "learning_rate": 0.0001203, + "loss": 3.6077, + "step": 409 + }, + { + "epoch": 0.656, + "grad_norm": 1.8508795499801636, + "learning_rate": 0.00012059999999999999, + "loss": 3.5619, + "step": 410 + }, + { + "epoch": 0.6576, + "grad_norm": 4.04713249206543, + "learning_rate": 0.0001209, + "loss": 3.4478, + "step": 411 + }, + { + "epoch": 0.6592, + "grad_norm": 6.989367485046387, + "learning_rate": 0.00012119999999999999, + "loss": 3.5359, + "step": 412 + }, + { + "epoch": 0.6608, + "grad_norm": 2.084543466567993, + "learning_rate": 0.0001215, + "loss": 3.4906, + "step": 413 + }, + { + "epoch": 0.6624, + "grad_norm": 1.3646767139434814, + "learning_rate": 0.00012179999999999999, + "loss": 3.4266, + "step": 414 + }, + { + "epoch": 0.664, + "grad_norm": 1.7721978425979614, + "learning_rate": 0.00012209999999999999, + "loss": 3.5448, + "step": 415 + }, + { + "epoch": 0.6656, + "grad_norm": 1.3269426822662354, + "learning_rate": 0.0001224, + "loss": 3.4261, + "step": 416 + }, + { + "epoch": 0.6672, + "grad_norm": 1.851492166519165, + "learning_rate": 0.00012269999999999997, + "loss": 3.3502, + "step": 417 + }, + { + "epoch": 0.6688, + "grad_norm": 4.740501880645752, + "learning_rate": 0.00012299999999999998, + "loss": 3.3928, + "step": 418 + }, + { + "epoch": 0.6704, + "grad_norm": 1.7381259202957153, + "learning_rate": 0.0001233, + "loss": 3.3864, + "step": 419 + }, + { + "epoch": 0.672, + "grad_norm": 2.608163833618164, + "learning_rate": 0.0001236, + "loss": 3.3182, + "step": 420 + }, + { + "epoch": 0.6736, + "grad_norm": 1.1638840436935425, + "learning_rate": 0.00012389999999999998, + "loss": 3.367, + "step": 421 + }, + { + "epoch": 0.6752, + "grad_norm": 1.1645619869232178, + "learning_rate": 0.00012419999999999998, + "loss": 3.3808, + "step": 422 + }, + { + "epoch": 0.6768, + "grad_norm": 1.3737239837646484, + "learning_rate": 0.0001245, + "loss": 3.3585, + "step": 423 + }, + { + "epoch": 0.6784, + "grad_norm": 42.222755432128906, + "learning_rate": 0.00012479999999999997, + "loss": 3.3515, + "step": 424 + }, + { + "epoch": 0.68, + "grad_norm": 2.465653419494629, + "learning_rate": 0.00012509999999999998, + "loss": 3.3906, + "step": 425 + }, + { + "epoch": 0.6816, + "grad_norm": 1.9028306007385254, + "learning_rate": 0.00012539999999999999, + "loss": 3.4148, + "step": 426 + }, + { + "epoch": 0.6832, + "grad_norm": 1.9519649744033813, + "learning_rate": 0.0001257, + "loss": 3.3085, + "step": 427 + }, + { + "epoch": 0.6848, + "grad_norm": 1.1407490968704224, + "learning_rate": 0.00012599999999999997, + "loss": 3.42, + "step": 428 + }, + { + "epoch": 0.6864, + "grad_norm": 0.9681979417800903, + "learning_rate": 0.00012629999999999998, + "loss": 3.2658, + "step": 429 + }, + { + "epoch": 0.688, + "grad_norm": 1.3614871501922607, + "learning_rate": 0.0001266, + "loss": 3.2536, + "step": 430 + }, + { + "epoch": 0.6896, + "grad_norm": 1.113405704498291, + "learning_rate": 0.0001269, + "loss": 3.171, + "step": 431 + }, + { + "epoch": 0.6912, + "grad_norm": 2.0724241733551025, + "learning_rate": 0.00012719999999999997, + "loss": 3.1676, + "step": 432 + }, + { + "epoch": 0.6928, + "grad_norm": 3.1103081703186035, + "learning_rate": 0.00012749999999999998, + "loss": 3.1817, + "step": 433 + }, + { + "epoch": 0.6944, + "grad_norm": 3.0139989852905273, + "learning_rate": 0.0001278, + "loss": 3.143, + "step": 434 + }, + { + "epoch": 0.696, + "grad_norm": 1.2784980535507202, + "learning_rate": 0.0001281, + "loss": 3.2375, + "step": 435 + }, + { + "epoch": 0.6976, + "grad_norm": 1.6558539867401123, + "learning_rate": 0.00012839999999999998, + "loss": 3.1586, + "step": 436 + }, + { + "epoch": 0.6992, + "grad_norm": 0.9733999967575073, + "learning_rate": 0.00012869999999999998, + "loss": 3.3419, + "step": 437 + }, + { + "epoch": 0.7008, + "grad_norm": 1.0060323476791382, + "learning_rate": 0.000129, + "loss": 3.1615, + "step": 438 + }, + { + "epoch": 0.7024, + "grad_norm": 1.6919910907745361, + "learning_rate": 0.0001293, + "loss": 3.283, + "step": 439 + }, + { + "epoch": 0.704, + "grad_norm": 1.88755202293396, + "learning_rate": 0.00012959999999999998, + "loss": 3.4583, + "step": 440 + }, + { + "epoch": 0.7056, + "grad_norm": 1.442505121231079, + "learning_rate": 0.00012989999999999999, + "loss": 3.0699, + "step": 441 + }, + { + "epoch": 0.7072, + "grad_norm": 1.363010287284851, + "learning_rate": 0.0001302, + "loss": 3.1889, + "step": 442 + }, + { + "epoch": 0.7088, + "grad_norm": 1.2752346992492676, + "learning_rate": 0.0001305, + "loss": 3.152, + "step": 443 + }, + { + "epoch": 0.7104, + "grad_norm": 1.477892518043518, + "learning_rate": 0.00013079999999999998, + "loss": 3.2669, + "step": 444 + }, + { + "epoch": 0.712, + "grad_norm": 1.7384015321731567, + "learning_rate": 0.0001311, + "loss": 3.1291, + "step": 445 + }, + { + "epoch": 0.7136, + "grad_norm": 1.0315364599227905, + "learning_rate": 0.0001314, + "loss": 3.1201, + "step": 446 + }, + { + "epoch": 0.7152, + "grad_norm": 1.4718317985534668, + "learning_rate": 0.00013169999999999998, + "loss": 3.2393, + "step": 447 + }, + { + "epoch": 0.7168, + "grad_norm": 1.4238083362579346, + "learning_rate": 0.00013199999999999998, + "loss": 3.2598, + "step": 448 + }, + { + "epoch": 0.7184, + "grad_norm": 1.4846147298812866, + "learning_rate": 0.0001323, + "loss": 3.3919, + "step": 449 + }, + { + "epoch": 0.72, + "grad_norm": 4.619383811950684, + "learning_rate": 0.0001326, + "loss": 3.6552, + "step": 450 + }, + { + "epoch": 0.7216, + "grad_norm": 30.76251220703125, + "learning_rate": 0.00013289999999999998, + "loss": 4.6791, + "step": 451 + }, + { + "epoch": 0.7232, + "grad_norm": 7.747359275817871, + "learning_rate": 0.00013319999999999999, + "loss": 3.1795, + "step": 452 + }, + { + "epoch": 0.7248, + "grad_norm": 5.558343887329102, + "learning_rate": 0.0001335, + "loss": 3.1374, + "step": 453 + }, + { + "epoch": 0.7264, + "grad_norm": 2.7429940700531006, + "learning_rate": 0.0001338, + "loss": 3.0504, + "step": 454 + }, + { + "epoch": 0.728, + "grad_norm": 1.343422293663025, + "learning_rate": 0.00013409999999999998, + "loss": 3.0727, + "step": 455 + }, + { + "epoch": 0.7296, + "grad_norm": 1.5927385091781616, + "learning_rate": 0.0001344, + "loss": 3.1032, + "step": 456 + }, + { + "epoch": 0.7312, + "grad_norm": 4.210255146026611, + "learning_rate": 0.0001347, + "loss": 3.0035, + "step": 457 + }, + { + "epoch": 0.7328, + "grad_norm": 3.81455135345459, + "learning_rate": 0.000135, + "loss": 2.7915, + "step": 458 + }, + { + "epoch": 0.7344, + "grad_norm": 2.9454805850982666, + "learning_rate": 0.00013529999999999998, + "loss": 2.9141, + "step": 459 + }, + { + "epoch": 0.736, + "grad_norm": 1.501758337020874, + "learning_rate": 0.0001356, + "loss": 2.7512, + "step": 460 + }, + { + "epoch": 0.7376, + "grad_norm": 1.6795105934143066, + "learning_rate": 0.0001359, + "loss": 2.6789, + "step": 461 + }, + { + "epoch": 0.7392, + "grad_norm": 2.2269608974456787, + "learning_rate": 0.0001362, + "loss": 2.7782, + "step": 462 + }, + { + "epoch": 0.7408, + "grad_norm": 2.0078094005584717, + "learning_rate": 0.00013649999999999998, + "loss": 2.6121, + "step": 463 + }, + { + "epoch": 0.7424, + "grad_norm": 2.3875632286071777, + "learning_rate": 0.0001368, + "loss": 2.6247, + "step": 464 + }, + { + "epoch": 0.744, + "grad_norm": 2.9569144248962402, + "learning_rate": 0.0001371, + "loss": 2.9894, + "step": 465 + }, + { + "epoch": 0.7456, + "grad_norm": 1.4559426307678223, + "learning_rate": 0.0001374, + "loss": 2.6496, + "step": 466 + }, + { + "epoch": 0.7472, + "grad_norm": 2.152031421661377, + "learning_rate": 0.00013769999999999999, + "loss": 2.6579, + "step": 467 + }, + { + "epoch": 0.7488, + "grad_norm": 1.1554604768753052, + "learning_rate": 0.000138, + "loss": 2.67, + "step": 468 + }, + { + "epoch": 0.7504, + "grad_norm": 1.7094281911849976, + "learning_rate": 0.0001383, + "loss": 2.5704, + "step": 469 + }, + { + "epoch": 0.752, + "grad_norm": 5.252382278442383, + "learning_rate": 0.0001386, + "loss": 2.8019, + "step": 470 + }, + { + "epoch": 0.7536, + "grad_norm": 3.1910815238952637, + "learning_rate": 0.0001389, + "loss": 2.7668, + "step": 471 + }, + { + "epoch": 0.7552, + "grad_norm": 1.2235437631607056, + "learning_rate": 0.0001392, + "loss": 2.5559, + "step": 472 + }, + { + "epoch": 0.7568, + "grad_norm": 1.8325499296188354, + "learning_rate": 0.0001395, + "loss": 2.5642, + "step": 473 + }, + { + "epoch": 0.7584, + "grad_norm": 1.3871606588363647, + "learning_rate": 0.00013979999999999998, + "loss": 2.6201, + "step": 474 + }, + { + "epoch": 0.76, + "grad_norm": 1.0033754110336304, + "learning_rate": 0.0001401, + "loss": 2.5613, + "step": 475 + }, + { + "epoch": 0.7616, + "grad_norm": 0.9249265193939209, + "learning_rate": 0.0001404, + "loss": 2.3257, + "step": 476 + }, + { + "epoch": 0.7632, + "grad_norm": 1.0387792587280273, + "learning_rate": 0.00014069999999999998, + "loss": 2.6483, + "step": 477 + }, + { + "epoch": 0.7648, + "grad_norm": 1.5178635120391846, + "learning_rate": 0.00014099999999999998, + "loss": 2.4524, + "step": 478 + }, + { + "epoch": 0.7664, + "grad_norm": 4.191971302032471, + "learning_rate": 0.0001413, + "loss": 2.688, + "step": 479 + }, + { + "epoch": 0.768, + "grad_norm": 3.055619955062866, + "learning_rate": 0.00014159999999999997, + "loss": 2.4534, + "step": 480 + }, + { + "epoch": 0.7696, + "grad_norm": 2.205340623855591, + "learning_rate": 0.00014189999999999998, + "loss": 2.4007, + "step": 481 + }, + { + "epoch": 0.7712, + "grad_norm": 0.9651008248329163, + "learning_rate": 0.0001422, + "loss": 2.3937, + "step": 482 + }, + { + "epoch": 0.7728, + "grad_norm": 1.5760127305984497, + "learning_rate": 0.0001425, + "loss": 2.3266, + "step": 483 + }, + { + "epoch": 0.7744, + "grad_norm": 1.0742831230163574, + "learning_rate": 0.00014279999999999997, + "loss": 2.606, + "step": 484 + }, + { + "epoch": 0.776, + "grad_norm": 1.195465087890625, + "learning_rate": 0.00014309999999999998, + "loss": 2.5657, + "step": 485 + }, + { + "epoch": 0.7776, + "grad_norm": 1.0587530136108398, + "learning_rate": 0.0001434, + "loss": 2.2895, + "step": 486 + }, + { + "epoch": 0.7792, + "grad_norm": 1.911810040473938, + "learning_rate": 0.00014369999999999997, + "loss": 2.2882, + "step": 487 + }, + { + "epoch": 0.7808, + "grad_norm": 1.0950274467468262, + "learning_rate": 0.00014399999999999998, + "loss": 2.3568, + "step": 488 + }, + { + "epoch": 0.7824, + "grad_norm": 1.563812494277954, + "learning_rate": 0.00014429999999999998, + "loss": 2.5129, + "step": 489 + }, + { + "epoch": 0.784, + "grad_norm": 0.9567598104476929, + "learning_rate": 0.0001446, + "loss": 2.3818, + "step": 490 + }, + { + "epoch": 0.7856, + "grad_norm": 0.9638680815696716, + "learning_rate": 0.00014489999999999997, + "loss": 2.1914, + "step": 491 + }, + { + "epoch": 0.7872, + "grad_norm": 1.8414331674575806, + "learning_rate": 0.00014519999999999998, + "loss": 2.5106, + "step": 492 + }, + { + "epoch": 0.7888, + "grad_norm": 1.2512167692184448, + "learning_rate": 0.00014549999999999999, + "loss": 2.1566, + "step": 493 + }, + { + "epoch": 0.7904, + "grad_norm": 1.4093244075775146, + "learning_rate": 0.0001458, + "loss": 2.1137, + "step": 494 + }, + { + "epoch": 0.792, + "grad_norm": 1.3009130954742432, + "learning_rate": 0.00014609999999999997, + "loss": 2.4738, + "step": 495 + }, + { + "epoch": 0.7936, + "grad_norm": 1.0954293012619019, + "learning_rate": 0.00014639999999999998, + "loss": 2.7082, + "step": 496 + }, + { + "epoch": 0.7952, + "grad_norm": 1.9157133102416992, + "learning_rate": 0.0001467, + "loss": 2.4871, + "step": 497 + }, + { + "epoch": 0.7968, + "grad_norm": 1.6690477132797241, + "learning_rate": 0.000147, + "loss": 2.5936, + "step": 498 + }, + { + "epoch": 0.7984, + "grad_norm": 1.933651089668274, + "learning_rate": 0.00014729999999999998, + "loss": 2.4619, + "step": 499 + }, + { + "epoch": 0.8, + "grad_norm": NaN, + "learning_rate": 0.00014729999999999998, + "loss": 3.235, + "step": 500 + }, + { + "epoch": 0.8016, + "grad_norm": 7.1725873947143555, + "learning_rate": 0.00014759999999999998, + "loss": 2.7598, + "step": 501 + }, + { + "epoch": 0.8032, + "grad_norm": 5.490133762359619, + "learning_rate": 0.0001479, + "loss": 2.3616, + "step": 502 + }, + { + "epoch": 0.8048, + "grad_norm": 1.2265512943267822, + "learning_rate": 0.0001482, + "loss": 2.2612, + "step": 503 + }, + { + "epoch": 0.8064, + "grad_norm": 2.705599069595337, + "learning_rate": 0.00014849999999999998, + "loss": 2.6145, + "step": 504 + }, + { + "epoch": 0.808, + "grad_norm": 1.0577749013900757, + "learning_rate": 0.00014879999999999998, + "loss": 2.305, + "step": 505 + }, + { + "epoch": 0.8096, + "grad_norm": 2.958197593688965, + "learning_rate": 0.0001491, + "loss": 2.2658, + "step": 506 + }, + { + "epoch": 0.8112, + "grad_norm": 1.8943955898284912, + "learning_rate": 0.0001494, + "loss": 2.3687, + "step": 507 + }, + { + "epoch": 0.8128, + "grad_norm": 1.8153882026672363, + "learning_rate": 0.00014969999999999998, + "loss": 1.9508, + "step": 508 + }, + { + "epoch": 0.8144, + "grad_norm": 1.5181827545166016, + "learning_rate": 0.00015, + "loss": 2.1255, + "step": 509 + }, + { + "epoch": 0.816, + "grad_norm": 1.3731355667114258, + "learning_rate": 0.0001503, + "loss": 2.1064, + "step": 510 + }, + { + "epoch": 0.8176, + "grad_norm": 3.0102975368499756, + "learning_rate": 0.00015059999999999997, + "loss": 2.3098, + "step": 511 + }, + { + "epoch": 0.8192, + "grad_norm": 3.4178948402404785, + "learning_rate": 0.00015089999999999998, + "loss": 1.9189, + "step": 512 + }, + { + "epoch": 0.8208, + "grad_norm": 2.359294891357422, + "learning_rate": 0.0001512, + "loss": 2.2278, + "step": 513 + }, + { + "epoch": 0.8224, + "grad_norm": 1.2816225290298462, + "learning_rate": 0.0001515, + "loss": 1.7545, + "step": 514 + }, + { + "epoch": 0.824, + "grad_norm": 1.676662564277649, + "learning_rate": 0.00015179999999999998, + "loss": 2.0563, + "step": 515 + }, + { + "epoch": 0.8256, + "grad_norm": 1.036680817604065, + "learning_rate": 0.00015209999999999998, + "loss": 1.9042, + "step": 516 + }, + { + "epoch": 0.8272, + "grad_norm": 0.8826329708099365, + "learning_rate": 0.0001524, + "loss": 1.9305, + "step": 517 + }, + { + "epoch": 0.8288, + "grad_norm": 6.336912631988525, + "learning_rate": 0.0001527, + "loss": 2.5653, + "step": 518 + }, + { + "epoch": 0.8304, + "grad_norm": 1.77652907371521, + "learning_rate": 0.00015299999999999998, + "loss": 1.8931, + "step": 519 + }, + { + "epoch": 0.832, + "grad_norm": 1.2775630950927734, + "learning_rate": 0.00015329999999999999, + "loss": 1.7564, + "step": 520 + }, + { + "epoch": 0.8336, + "grad_norm": 1.01466965675354, + "learning_rate": 0.0001536, + "loss": 1.7173, + "step": 521 + }, + { + "epoch": 0.8352, + "grad_norm": 1.1461132764816284, + "learning_rate": 0.0001539, + "loss": 1.9266, + "step": 522 + }, + { + "epoch": 0.8368, + "grad_norm": 1.6386278867721558, + "learning_rate": 0.00015419999999999998, + "loss": 1.6062, + "step": 523 + }, + { + "epoch": 0.8384, + "grad_norm": 1.1096084117889404, + "learning_rate": 0.0001545, + "loss": 1.6551, + "step": 524 + }, + { + "epoch": 0.84, + "grad_norm": 1.0475767850875854, + "learning_rate": 0.0001548, + "loss": 1.6422, + "step": 525 + }, + { + "epoch": 0.8416, + "grad_norm": 1.8000061511993408, + "learning_rate": 0.0001551, + "loss": 1.8283, + "step": 526 + }, + { + "epoch": 0.8432, + "grad_norm": 1.2773330211639404, + "learning_rate": 0.00015539999999999998, + "loss": 1.6057, + "step": 527 + }, + { + "epoch": 0.8448, + "grad_norm": 2.812037944793701, + "learning_rate": 0.0001557, + "loss": 1.9903, + "step": 528 + }, + { + "epoch": 0.8464, + "grad_norm": 1.1310831308364868, + "learning_rate": 0.000156, + "loss": 2.1515, + "step": 529 + }, + { + "epoch": 0.848, + "grad_norm": 1.6405916213989258, + "learning_rate": 0.0001563, + "loss": 1.6635, + "step": 530 + }, + { + "epoch": 0.8496, + "grad_norm": 1.143716812133789, + "learning_rate": 0.00015659999999999998, + "loss": 1.7994, + "step": 531 + }, + { + "epoch": 0.8512, + "grad_norm": 1.5526354312896729, + "learning_rate": 0.0001569, + "loss": 1.6934, + "step": 532 + }, + { + "epoch": 0.8528, + "grad_norm": 1.5130469799041748, + "learning_rate": 0.0001572, + "loss": 2.0692, + "step": 533 + }, + { + "epoch": 0.8544, + "grad_norm": 2.715682029724121, + "learning_rate": 0.00015749999999999998, + "loss": 2.178, + "step": 534 + }, + { + "epoch": 0.856, + "grad_norm": 1.2512975931167603, + "learning_rate": 0.0001578, + "loss": 2.0984, + "step": 535 + }, + { + "epoch": 0.8576, + "grad_norm": 1.528995394706726, + "learning_rate": 0.0001581, + "loss": 2.0069, + "step": 536 + }, + { + "epoch": 0.8592, + "grad_norm": 2.7194480895996094, + "learning_rate": 0.0001584, + "loss": 1.8528, + "step": 537 + }, + { + "epoch": 0.8608, + "grad_norm": 1.1557732820510864, + "learning_rate": 0.00015869999999999998, + "loss": 2.0246, + "step": 538 + }, + { + "epoch": 0.8624, + "grad_norm": 1.606906771659851, + "learning_rate": 0.000159, + "loss": 2.1268, + "step": 539 + }, + { + "epoch": 0.864, + "grad_norm": 1.176815152168274, + "learning_rate": 0.0001593, + "loss": 1.7476, + "step": 540 + }, + { + "epoch": 0.8656, + "grad_norm": 1.139367699623108, + "learning_rate": 0.0001596, + "loss": 2.1414, + "step": 541 + }, + { + "epoch": 0.8672, + "grad_norm": 1.1354289054870605, + "learning_rate": 0.00015989999999999998, + "loss": 2.1052, + "step": 542 + }, + { + "epoch": 0.8688, + "grad_norm": 1.374079942703247, + "learning_rate": 0.0001602, + "loss": 2.0711, + "step": 543 + }, + { + "epoch": 0.8704, + "grad_norm": 1.9494390487670898, + "learning_rate": 0.0001605, + "loss": 2.3558, + "step": 544 + }, + { + "epoch": 0.872, + "grad_norm": 1.1053131818771362, + "learning_rate": 0.0001608, + "loss": 2.0471, + "step": 545 + }, + { + "epoch": 0.8736, + "grad_norm": 1.0600978136062622, + "learning_rate": 0.00016109999999999999, + "loss": 1.6801, + "step": 546 + }, + { + "epoch": 0.8752, + "grad_norm": 1.5583279132843018, + "learning_rate": 0.0001614, + "loss": 2.222, + "step": 547 + }, + { + "epoch": 0.8768, + "grad_norm": 1.5233080387115479, + "learning_rate": 0.0001617, + "loss": 2.0222, + "step": 548 + }, + { + "epoch": 0.8784, + "grad_norm": 1.6784640550613403, + "learning_rate": 0.000162, + "loss": 2.3573, + "step": 549 + }, + { + "epoch": 0.88, + "grad_norm": NaN, + "learning_rate": 0.000162, + "loss": 1.9713, + "step": 550 + }, + { + "epoch": 0.8816, + "grad_norm": 22.64229393005371, + "learning_rate": 0.0001623, + "loss": 4.5456, + "step": 551 + }, + { + "epoch": 0.8832, + "grad_norm": 3.1680028438568115, + "learning_rate": 0.0001626, + "loss": 1.9073, + "step": 552 + }, + { + "epoch": 0.8848, + "grad_norm": 1.5461392402648926, + "learning_rate": 0.0001629, + "loss": 1.8575, + "step": 553 + }, + { + "epoch": 0.8864, + "grad_norm": 1.6578490734100342, + "learning_rate": 0.0001632, + "loss": 2.1306, + "step": 554 + }, + { + "epoch": 0.888, + "grad_norm": 2.54394793510437, + "learning_rate": 0.0001635, + "loss": 2.0791, + "step": 555 + }, + { + "epoch": 0.8896, + "grad_norm": 4.085062026977539, + "learning_rate": 0.0001638, + "loss": 2.0151, + "step": 556 + }, + { + "epoch": 0.8912, + "grad_norm": 1.3918102979660034, + "learning_rate": 0.0001641, + "loss": 1.9562, + "step": 557 + }, + { + "epoch": 0.8928, + "grad_norm": 1.7889790534973145, + "learning_rate": 0.0001644, + "loss": 2.3353, + "step": 558 + }, + { + "epoch": 0.8944, + "grad_norm": 1.9378011226654053, + "learning_rate": 0.0001647, + "loss": 1.5553, + "step": 559 + }, + { + "epoch": 0.896, + "grad_norm": 1.3917180299758911, + "learning_rate": 0.000165, + "loss": 1.6519, + "step": 560 + }, + { + "epoch": 0.8976, + "grad_norm": 5.907794952392578, + "learning_rate": 0.0001653, + "loss": 2.2321, + "step": 561 + }, + { + "epoch": 0.8992, + "grad_norm": 1.9292672872543335, + "learning_rate": 0.0001656, + "loss": 1.9859, + "step": 562 + }, + { + "epoch": 0.9008, + "grad_norm": 1.9389235973358154, + "learning_rate": 0.0001659, + "loss": 1.656, + "step": 563 + }, + { + "epoch": 0.9024, + "grad_norm": 1.0009876489639282, + "learning_rate": 0.0001662, + "loss": 1.7168, + "step": 564 + }, + { + "epoch": 0.904, + "grad_norm": 1.138958215713501, + "learning_rate": 0.0001665, + "loss": 1.3794, + "step": 565 + }, + { + "epoch": 0.9056, + "grad_norm": 1.236640453338623, + "learning_rate": 0.0001668, + "loss": 1.5395, + "step": 566 + }, + { + "epoch": 0.9072, + "grad_norm": 0.8538547158241272, + "learning_rate": 0.0001671, + "loss": 1.6903, + "step": 567 + }, + { + "epoch": 0.9088, + "grad_norm": 1.112789273262024, + "learning_rate": 0.0001674, + "loss": 1.6005, + "step": 568 + }, + { + "epoch": 0.9104, + "grad_norm": 0.9694105982780457, + "learning_rate": 0.0001677, + "loss": 1.6596, + "step": 569 + }, + { + "epoch": 0.912, + "grad_norm": 0.9604489803314209, + "learning_rate": 0.000168, + "loss": 1.3571, + "step": 570 + }, + { + "epoch": 0.9136, + "grad_norm": 1.1342942714691162, + "learning_rate": 0.0001683, + "loss": 1.9352, + "step": 571 + }, + { + "epoch": 0.9152, + "grad_norm": 1.059465765953064, + "learning_rate": 0.0001686, + "loss": 1.5642, + "step": 572 + }, + { + "epoch": 0.9168, + "grad_norm": 1.3122626543045044, + "learning_rate": 0.00016889999999999996, + "loss": 1.5281, + "step": 573 + }, + { + "epoch": 0.9184, + "grad_norm": 4.926758289337158, + "learning_rate": 0.00016919999999999997, + "loss": 2.0261, + "step": 574 + }, + { + "epoch": 0.92, + "grad_norm": 3.0546226501464844, + "learning_rate": 0.00016949999999999997, + "loss": 1.6621, + "step": 575 + }, + { + "epoch": 0.9216, + "grad_norm": 2.96925687789917, + "learning_rate": 0.00016979999999999998, + "loss": 2.0468, + "step": 576 + }, + { + "epoch": 0.9232, + "grad_norm": 1.883144736289978, + "learning_rate": 0.00017009999999999996, + "loss": 1.4683, + "step": 577 + }, + { + "epoch": 0.9248, + "grad_norm": 1.6988012790679932, + "learning_rate": 0.00017039999999999997, + "loss": 1.4475, + "step": 578 + }, + { + "epoch": 0.9264, + "grad_norm": 1.6453936100006104, + "learning_rate": 0.00017069999999999998, + "loss": 1.475, + "step": 579 + }, + { + "epoch": 0.928, + "grad_norm": 2.155768871307373, + "learning_rate": 0.00017099999999999998, + "loss": 1.5596, + "step": 580 + }, + { + "epoch": 0.9296, + "grad_norm": 1.20650053024292, + "learning_rate": 0.00017129999999999996, + "loss": 1.8129, + "step": 581 + }, + { + "epoch": 0.9312, + "grad_norm": 1.2697020769119263, + "learning_rate": 0.00017159999999999997, + "loss": 1.5443, + "step": 582 + }, + { + "epoch": 0.9328, + "grad_norm": 1.6629301309585571, + "learning_rate": 0.00017189999999999998, + "loss": 1.4867, + "step": 583 + }, + { + "epoch": 0.9344, + "grad_norm": 1.982050895690918, + "learning_rate": 0.00017219999999999998, + "loss": 1.7197, + "step": 584 + }, + { + "epoch": 0.936, + "grad_norm": 2.2001729011535645, + "learning_rate": 0.00017249999999999996, + "loss": 1.5931, + "step": 585 + }, + { + "epoch": 0.9376, + "grad_norm": 1.925163984298706, + "learning_rate": 0.00017279999999999997, + "loss": 1.6565, + "step": 586 + }, + { + "epoch": 0.9392, + "grad_norm": 1.163279414176941, + "learning_rate": 0.00017309999999999998, + "loss": 1.5769, + "step": 587 + }, + { + "epoch": 0.9408, + "grad_norm": 1.164640188217163, + "learning_rate": 0.00017339999999999996, + "loss": 1.4949, + "step": 588 + }, + { + "epoch": 0.9424, + "grad_norm": 1.5389615297317505, + "learning_rate": 0.00017369999999999997, + "loss": 1.6192, + "step": 589 + }, + { + "epoch": 0.944, + "grad_norm": 3.1581554412841797, + "learning_rate": 0.00017399999999999997, + "loss": 2.1662, + "step": 590 + }, + { + "epoch": 0.9456, + "grad_norm": 1.5341241359710693, + "learning_rate": 0.00017429999999999998, + "loss": 2.0104, + "step": 591 + }, + { + "epoch": 0.9472, + "grad_norm": 2.192890167236328, + "learning_rate": 0.00017459999999999996, + "loss": 1.6697, + "step": 592 + }, + { + "epoch": 0.9488, + "grad_norm": 1.1967136859893799, + "learning_rate": 0.00017489999999999997, + "loss": 2.1618, + "step": 593 + }, + { + "epoch": 0.9504, + "grad_norm": 1.4983010292053223, + "learning_rate": 0.00017519999999999998, + "loss": 1.6773, + "step": 594 + }, + { + "epoch": 0.952, + "grad_norm": 1.8660876750946045, + "learning_rate": 0.00017549999999999998, + "loss": 2.1347, + "step": 595 + }, + { + "epoch": 0.9536, + "grad_norm": 2.4708802700042725, + "learning_rate": 0.00017579999999999996, + "loss": 2.0521, + "step": 596 + }, + { + "epoch": 0.9552, + "grad_norm": 1.22525954246521, + "learning_rate": 0.00017609999999999997, + "loss": 2.3289, + "step": 597 + }, + { + "epoch": 0.9568, + "grad_norm": 2.8093461990356445, + "learning_rate": 0.00017639999999999998, + "loss": 2.3451, + "step": 598 + }, + { + "epoch": 0.9584, + "grad_norm": 2.3243844509124756, + "learning_rate": 0.00017669999999999999, + "loss": 1.8311, + "step": 599 + }, + { + "epoch": 0.96, + "grad_norm": NaN, + "learning_rate": 0.00017669999999999999, + "loss": 1.9821, + "step": 600 + }, + { + "epoch": 0.9616, + "grad_norm": 2.1118686199188232, + "learning_rate": 0.00017699999999999997, + "loss": 1.8856, + "step": 601 + }, + { + "epoch": 0.9632, + "grad_norm": 3.0222251415252686, + "learning_rate": 0.00017729999999999997, + "loss": 2.4598, + "step": 602 + }, + { + "epoch": 0.9648, + "grad_norm": 2.8710968494415283, + "learning_rate": 0.00017759999999999998, + "loss": 2.2719, + "step": 603 + }, + { + "epoch": 0.9664, + "grad_norm": 3.6922225952148438, + "learning_rate": 0.0001779, + "loss": 1.7081, + "step": 604 + }, + { + "epoch": 0.968, + "grad_norm": 4.826087951660156, + "learning_rate": 0.00017819999999999997, + "loss": 1.7774, + "step": 605 + }, + { + "epoch": 0.9696, + "grad_norm": 3.9629127979278564, + "learning_rate": 0.00017849999999999997, + "loss": 1.7018, + "step": 606 + }, + { + "epoch": 0.9712, + "grad_norm": 3.826023817062378, + "learning_rate": 0.00017879999999999998, + "loss": 1.5971, + "step": 607 + }, + { + "epoch": 0.9728, + "grad_norm": 1.6545122861862183, + "learning_rate": 0.0001791, + "loss": 1.9029, + "step": 608 + }, + { + "epoch": 0.9744, + "grad_norm": 1.552114725112915, + "learning_rate": 0.00017939999999999997, + "loss": 1.584, + "step": 609 + }, + { + "epoch": 0.976, + "grad_norm": 1.2096900939941406, + "learning_rate": 0.00017969999999999998, + "loss": 1.6172, + "step": 610 + }, + { + "epoch": 0.9776, + "grad_norm": 2.817974328994751, + "learning_rate": 0.00017999999999999998, + "loss": 1.2749, + "step": 611 + }, + { + "epoch": 0.9792, + "grad_norm": 3.8628475666046143, + "learning_rate": 0.00018029999999999996, + "loss": 2.0863, + "step": 612 + }, + { + "epoch": 0.9808, + "grad_norm": 4.558107852935791, + "learning_rate": 0.00018059999999999997, + "loss": 1.3319, + "step": 613 + }, + { + "epoch": 0.9824, + "grad_norm": 3.5585508346557617, + "learning_rate": 0.00018089999999999998, + "loss": 1.5316, + "step": 614 + }, + { + "epoch": 0.984, + "grad_norm": 3.8311831951141357, + "learning_rate": 0.00018119999999999999, + "loss": 1.7669, + "step": 615 + }, + { + "epoch": 0.9856, + "grad_norm": 3.750258207321167, + "learning_rate": 0.00018149999999999997, + "loss": 1.6969, + "step": 616 + }, + { + "epoch": 0.9872, + "grad_norm": 2.679804801940918, + "learning_rate": 0.00018179999999999997, + "loss": 1.6474, + "step": 617 + }, + { + "epoch": 0.9888, + "grad_norm": 1.5909662246704102, + "learning_rate": 0.00018209999999999998, + "loss": 1.5411, + "step": 618 + }, + { + "epoch": 0.9904, + "grad_norm": 1.123789668083191, + "learning_rate": 0.0001824, + "loss": 1.7017, + "step": 619 + }, + { + "epoch": 0.992, + "grad_norm": 1.4218626022338867, + "learning_rate": 0.00018269999999999997, + "loss": 1.3411, + "step": 620 + }, + { + "epoch": 0.9936, + "grad_norm": 1.5946886539459229, + "learning_rate": 0.00018299999999999998, + "loss": 2.159, + "step": 621 + }, + { + "epoch": 0.9952, + "grad_norm": 1.3549202680587769, + "learning_rate": 0.00018329999999999998, + "loss": 1.8949, + "step": 622 + }, + { + "epoch": 0.9968, + "grad_norm": 1.5221611261367798, + "learning_rate": 0.0001836, + "loss": 1.7881, + "step": 623 + }, + { + "epoch": 0.9984, + "grad_norm": 1.4437967538833618, + "learning_rate": 0.00018389999999999997, + "loss": 1.8924, + "step": 624 + }, + { + "epoch": 1.0, + "grad_norm": 2.5355899333953857, + "learning_rate": 0.00018419999999999998, + "loss": 2.3162, + "step": 625 + }, + { + "epoch": 1.0016, + "grad_norm": 17.70075798034668, + "learning_rate": 0.00018449999999999999, + "loss": 3.2662, + "step": 626 + }, + { + "epoch": 1.0032, + "grad_norm": 5.143967151641846, + "learning_rate": 0.0001848, + "loss": 2.1297, + "step": 627 + }, + { + "epoch": 1.0048, + "grad_norm": 7.86220645904541, + "learning_rate": 0.00018509999999999997, + "loss": 2.5585, + "step": 628 + }, + { + "epoch": 1.0064, + "grad_norm": 5.712104797363281, + "learning_rate": 0.00018539999999999998, + "loss": 2.479, + "step": 629 + }, + { + "epoch": 1.008, + "grad_norm": 8.05717658996582, + "learning_rate": 0.0001857, + "loss": 2.7763, + "step": 630 + }, + { + "epoch": 1.0096, + "grad_norm": 2.07607102394104, + "learning_rate": 0.000186, + "loss": 1.9077, + "step": 631 + }, + { + "epoch": 1.0112, + "grad_norm": 3.5380396842956543, + "learning_rate": 0.00018629999999999997, + "loss": 1.7525, + "step": 632 + }, + { + "epoch": 1.0128, + "grad_norm": 4.81820011138916, + "learning_rate": 0.00018659999999999998, + "loss": 1.8469, + "step": 633 + }, + { + "epoch": 1.0144, + "grad_norm": 4.835049629211426, + "learning_rate": 0.0001869, + "loss": 1.8129, + "step": 634 + }, + { + "epoch": 1.016, + "grad_norm": 5.322865962982178, + "learning_rate": 0.0001872, + "loss": 1.8239, + "step": 635 + }, + { + "epoch": 1.0176, + "grad_norm": 4.629301071166992, + "learning_rate": 0.00018749999999999998, + "loss": 1.5825, + "step": 636 + }, + { + "epoch": 1.0192, + "grad_norm": 1.5236226320266724, + "learning_rate": 0.00018779999999999998, + "loss": 1.8754, + "step": 637 + }, + { + "epoch": 1.0208, + "grad_norm": 1.857738733291626, + "learning_rate": 0.0001881, + "loss": 1.4363, + "step": 638 + }, + { + "epoch": 1.0224, + "grad_norm": 0.8692172765731812, + "learning_rate": 0.00018839999999999997, + "loss": 1.6648, + "step": 639 + }, + { + "epoch": 1.024, + "grad_norm": 3.064953088760376, + "learning_rate": 0.00018869999999999998, + "loss": 1.6678, + "step": 640 + }, + { + "epoch": 1.0256, + "grad_norm": 5.139404773712158, + "learning_rate": 0.00018899999999999999, + "loss": 1.9777, + "step": 641 + }, + { + "epoch": 1.0272, + "grad_norm": 2.796144723892212, + "learning_rate": 0.0001893, + "loss": 1.4951, + "step": 642 + }, + { + "epoch": 1.0288, + "grad_norm": 2.954390048980713, + "learning_rate": 0.00018959999999999997, + "loss": 1.5257, + "step": 643 + }, + { + "epoch": 1.0304, + "grad_norm": 3.8204727172851562, + "learning_rate": 0.00018989999999999998, + "loss": 1.8541, + "step": 644 + }, + { + "epoch": 1.032, + "grad_norm": 3.7258269786834717, + "learning_rate": 0.0001902, + "loss": 1.5582, + "step": 645 + }, + { + "epoch": 1.0336, + "grad_norm": 1.9401317834854126, + "learning_rate": 0.0001905, + "loss": 1.382, + "step": 646 + }, + { + "epoch": 1.0352, + "grad_norm": 2.4146273136138916, + "learning_rate": 0.00019079999999999998, + "loss": 1.3877, + "step": 647 + }, + { + "epoch": 1.0368, + "grad_norm": 1.2937108278274536, + "learning_rate": 0.00019109999999999998, + "loss": 1.476, + "step": 648 + }, + { + "epoch": 1.0384, + "grad_norm": 0.9674957394599915, + "learning_rate": 0.0001914, + "loss": 1.5663, + "step": 649 + }, + { + "epoch": 1.04, + "grad_norm": 1.0743024349212646, + "learning_rate": 0.0001917, + "loss": 1.3558, + "step": 650 + }, + { + "epoch": 1.0416, + "grad_norm": 1.2061973810195923, + "learning_rate": 0.00019199999999999998, + "loss": 1.2761, + "step": 651 + }, + { + "epoch": 1.0432, + "grad_norm": 1.1233325004577637, + "learning_rate": 0.00019229999999999999, + "loss": 1.5368, + "step": 652 + }, + { + "epoch": 1.0448, + "grad_norm": 0.894705593585968, + "learning_rate": 0.0001926, + "loss": 1.4238, + "step": 653 + }, + { + "epoch": 1.0464, + "grad_norm": 1.2992541790008545, + "learning_rate": 0.0001929, + "loss": 1.3999, + "step": 654 + }, + { + "epoch": 1.048, + "grad_norm": 1.919297695159912, + "learning_rate": 0.00019319999999999998, + "loss": 1.6165, + "step": 655 + }, + { + "epoch": 1.0496, + "grad_norm": 1.7403674125671387, + "learning_rate": 0.0001935, + "loss": 1.4777, + "step": 656 + }, + { + "epoch": 1.0512, + "grad_norm": 1.8304212093353271, + "learning_rate": 0.0001938, + "loss": 1.4263, + "step": 657 + }, + { + "epoch": 1.0528, + "grad_norm": 5.81217622756958, + "learning_rate": 0.0001941, + "loss": 1.5995, + "step": 658 + }, + { + "epoch": 1.0544, + "grad_norm": 2.001800775527954, + "learning_rate": 0.00019439999999999998, + "loss": 1.6648, + "step": 659 + }, + { + "epoch": 1.056, + "grad_norm": 2.0674045085906982, + "learning_rate": 0.0001947, + "loss": 1.9484, + "step": 660 + }, + { + "epoch": 1.0576, + "grad_norm": 1.6843963861465454, + "learning_rate": 0.000195, + "loss": 1.5397, + "step": 661 + }, + { + "epoch": 1.0592, + "grad_norm": 1.1019186973571777, + "learning_rate": 0.00019529999999999998, + "loss": 1.9416, + "step": 662 + }, + { + "epoch": 1.0608, + "grad_norm": 1.1988193988800049, + "learning_rate": 0.00019559999999999998, + "loss": 1.3641, + "step": 663 + }, + { + "epoch": 1.0624, + "grad_norm": 1.044538974761963, + "learning_rate": 0.0001959, + "loss": 1.5757, + "step": 664 + }, + { + "epoch": 1.064, + "grad_norm": 2.194228172302246, + "learning_rate": 0.0001962, + "loss": 1.6955, + "step": 665 + }, + { + "epoch": 1.0656, + "grad_norm": 1.3142859935760498, + "learning_rate": 0.00019649999999999998, + "loss": 1.7652, + "step": 666 + }, + { + "epoch": 1.0672, + "grad_norm": 2.0349411964416504, + "learning_rate": 0.00019679999999999999, + "loss": 1.5292, + "step": 667 + }, + { + "epoch": 1.0688, + "grad_norm": 2.3026349544525146, + "learning_rate": 0.0001971, + "loss": 2.1332, + "step": 668 + }, + { + "epoch": 1.0704, + "grad_norm": 1.8887361288070679, + "learning_rate": 0.0001974, + "loss": 1.6047, + "step": 669 + }, + { + "epoch": 1.072, + "grad_norm": 2.026123523712158, + "learning_rate": 0.00019769999999999998, + "loss": 1.4923, + "step": 670 + }, + { + "epoch": 1.0735999999999999, + "grad_norm": 1.346245527267456, + "learning_rate": 0.000198, + "loss": 2.2244, + "step": 671 + }, + { + "epoch": 1.0752, + "grad_norm": 1.6142189502716064, + "learning_rate": 0.0001983, + "loss": 2.0532, + "step": 672 + }, + { + "epoch": 1.0768, + "grad_norm": 1.4740869998931885, + "learning_rate": 0.0001986, + "loss": 1.4883, + "step": 673 + }, + { + "epoch": 1.0784, + "grad_norm": 2.6052584648132324, + "learning_rate": 0.00019889999999999998, + "loss": 2.5934, + "step": 674 + }, + { + "epoch": 1.08, + "grad_norm": 1.7282607555389404, + "learning_rate": 0.0001992, + "loss": 2.2343, + "step": 675 + }, + { + "epoch": 1.0816, + "grad_norm": 4.516241073608398, + "learning_rate": 0.0001995, + "loss": 2.2084, + "step": 676 + }, + { + "epoch": 1.0832, + "grad_norm": 4.2023162841796875, + "learning_rate": 0.0001998, + "loss": 2.0985, + "step": 677 + }, + { + "epoch": 1.0848, + "grad_norm": 7.1226325035095215, + "learning_rate": 0.00020009999999999998, + "loss": 2.6756, + "step": 678 + }, + { + "epoch": 1.0864, + "grad_norm": 2.7742109298706055, + "learning_rate": 0.0002004, + "loss": 2.074, + "step": 679 + }, + { + "epoch": 1.088, + "grad_norm": 1.4542862176895142, + "learning_rate": 0.0002007, + "loss": 2.0783, + "step": 680 + }, + { + "epoch": 1.0896, + "grad_norm": 3.168198585510254, + "learning_rate": 0.000201, + "loss": 1.7689, + "step": 681 + }, + { + "epoch": 1.0912, + "grad_norm": 3.7544302940368652, + "learning_rate": 0.0002013, + "loss": 1.9546, + "step": 682 + }, + { + "epoch": 1.0928, + "grad_norm": 3.380309820175171, + "learning_rate": 0.0002016, + "loss": 1.9613, + "step": 683 + }, + { + "epoch": 1.0944, + "grad_norm": 1.9851268529891968, + "learning_rate": 0.0002019, + "loss": 1.7351, + "step": 684 + }, + { + "epoch": 1.096, + "grad_norm": 1.1245828866958618, + "learning_rate": 0.0002022, + "loss": 1.6944, + "step": 685 + }, + { + "epoch": 1.0976, + "grad_norm": 1.3622814416885376, + "learning_rate": 0.0002025, + "loss": 1.176, + "step": 686 + }, + { + "epoch": 1.0992, + "grad_norm": 3.036677837371826, + "learning_rate": 0.0002028, + "loss": 1.4127, + "step": 687 + }, + { + "epoch": 1.1008, + "grad_norm": 5.2766337394714355, + "learning_rate": 0.0002031, + "loss": 1.9776, + "step": 688 + }, + { + "epoch": 1.1024, + "grad_norm": 3.7435624599456787, + "learning_rate": 0.00020339999999999998, + "loss": 1.6777, + "step": 689 + }, + { + "epoch": 1.104, + "grad_norm": 2.5562703609466553, + "learning_rate": 0.0002037, + "loss": 1.3577, + "step": 690 + }, + { + "epoch": 1.1056, + "grad_norm": 7.86569356918335, + "learning_rate": 0.000204, + "loss": 3.3387, + "step": 691 + }, + { + "epoch": 1.1072, + "grad_norm": 2.577002763748169, + "learning_rate": 0.0002043, + "loss": 1.5658, + "step": 692 + }, + { + "epoch": 1.1088, + "grad_norm": 1.7518302202224731, + "learning_rate": 0.00020459999999999999, + "loss": 1.7374, + "step": 693 + }, + { + "epoch": 1.1104, + "grad_norm": 1.0395740270614624, + "learning_rate": 0.0002049, + "loss": 1.3374, + "step": 694 + }, + { + "epoch": 1.112, + "grad_norm": 1.0535399913787842, + "learning_rate": 0.0002052, + "loss": 1.3842, + "step": 695 + }, + { + "epoch": 1.1136, + "grad_norm": 1.3788182735443115, + "learning_rate": 0.0002055, + "loss": 1.2359, + "step": 696 + }, + { + "epoch": 1.1152, + "grad_norm": 2.2207717895507812, + "learning_rate": 0.0002058, + "loss": 1.8388, + "step": 697 + }, + { + "epoch": 1.1168, + "grad_norm": 2.8598082065582275, + "learning_rate": 0.0002061, + "loss": 1.4118, + "step": 698 + }, + { + "epoch": 1.1184, + "grad_norm": 1.4320859909057617, + "learning_rate": 0.00020639999999999998, + "loss": 1.4376, + "step": 699 + }, + { + "epoch": 1.12, + "grad_norm": 0.9855818152427673, + "learning_rate": 0.00020669999999999996, + "loss": 1.3912, + "step": 700 + }, + { + "epoch": 1.1216, + "grad_norm": 1.0571755170822144, + "learning_rate": 0.00020699999999999996, + "loss": 1.3984, + "step": 701 + }, + { + "epoch": 1.1232, + "grad_norm": 1.7968964576721191, + "learning_rate": 0.00020729999999999997, + "loss": 1.5948, + "step": 702 + }, + { + "epoch": 1.1248, + "grad_norm": 2.7255570888519287, + "learning_rate": 0.00020759999999999998, + "loss": 1.8038, + "step": 703 + }, + { + "epoch": 1.1264, + "grad_norm": 3.1238272190093994, + "learning_rate": 0.00020789999999999996, + "loss": 1.6942, + "step": 704 + }, + { + "epoch": 1.1280000000000001, + "grad_norm": 2.947319984436035, + "learning_rate": 0.00020819999999999996, + "loss": 1.4243, + "step": 705 + }, + { + "epoch": 1.1296, + "grad_norm": 2.5516934394836426, + "learning_rate": 0.00020849999999999997, + "loss": 1.4778, + "step": 706 + }, + { + "epoch": 1.1312, + "grad_norm": 2.3497977256774902, + "learning_rate": 0.00020879999999999998, + "loss": 1.6869, + "step": 707 + }, + { + "epoch": 1.1328, + "grad_norm": 1.1373237371444702, + "learning_rate": 0.00020909999999999996, + "loss": 1.3769, + "step": 708 + }, + { + "epoch": 1.1344, + "grad_norm": 1.186307430267334, + "learning_rate": 0.00020939999999999997, + "loss": 1.5282, + "step": 709 + }, + { + "epoch": 1.1360000000000001, + "grad_norm": 1.152564525604248, + "learning_rate": 0.00020969999999999997, + "loss": 1.5939, + "step": 710 + }, + { + "epoch": 1.1376, + "grad_norm": 1.2839113473892212, + "learning_rate": 0.00020999999999999998, + "loss": 1.3393, + "step": 711 + }, + { + "epoch": 1.1392, + "grad_norm": 1.3416848182678223, + "learning_rate": 0.00021029999999999996, + "loss": 1.5591, + "step": 712 + }, + { + "epoch": 1.1408, + "grad_norm": 1.4146075248718262, + "learning_rate": 0.00021059999999999997, + "loss": 1.5227, + "step": 713 + }, + { + "epoch": 1.1424, + "grad_norm": 2.817870616912842, + "learning_rate": 0.00021089999999999998, + "loss": 2.0963, + "step": 714 + }, + { + "epoch": 1.144, + "grad_norm": 1.0793464183807373, + "learning_rate": 0.00021119999999999996, + "loss": 1.2714, + "step": 715 + }, + { + "epoch": 1.1456, + "grad_norm": 1.9559093713760376, + "learning_rate": 0.00021149999999999996, + "loss": 1.6881, + "step": 716 + }, + { + "epoch": 1.1472, + "grad_norm": 1.690061330795288, + "learning_rate": 0.00021179999999999997, + "loss": 1.7142, + "step": 717 + }, + { + "epoch": 1.1488, + "grad_norm": 1.6061111688613892, + "learning_rate": 0.00021209999999999998, + "loss": 1.5926, + "step": 718 + }, + { + "epoch": 1.1504, + "grad_norm": 1.7083910703659058, + "learning_rate": 0.00021239999999999996, + "loss": 1.9628, + "step": 719 + }, + { + "epoch": 1.152, + "grad_norm": 1.864904522895813, + "learning_rate": 0.00021269999999999997, + "loss": 1.6522, + "step": 720 + }, + { + "epoch": 1.1536, + "grad_norm": 1.7691465616226196, + "learning_rate": 0.00021299999999999997, + "loss": 1.7124, + "step": 721 + }, + { + "epoch": 1.1552, + "grad_norm": 1.6824963092803955, + "learning_rate": 0.00021329999999999998, + "loss": 1.6296, + "step": 722 + }, + { + "epoch": 1.1568, + "grad_norm": 1.7225028276443481, + "learning_rate": 0.00021359999999999996, + "loss": 1.9528, + "step": 723 + }, + { + "epoch": 1.1584, + "grad_norm": 2.096581220626831, + "learning_rate": 0.00021389999999999997, + "loss": 1.686, + "step": 724 + }, + { + "epoch": 1.16, + "grad_norm": NaN, + "learning_rate": 0.00021389999999999997, + "loss": 1.5142, + "step": 725 + }, + { + "epoch": 1.1616, + "grad_norm": 32.82332229614258, + "learning_rate": 0.00021419999999999998, + "loss": 7.8505, + "step": 726 + }, + { + "epoch": 1.1632, + "grad_norm": 6.196145534515381, + "learning_rate": 0.00021449999999999998, + "loss": 2.772, + "step": 727 + }, + { + "epoch": 1.1648, + "grad_norm": 1.9611283540725708, + "learning_rate": 0.00021479999999999996, + "loss": 1.8117, + "step": 728 + }, + { + "epoch": 1.1663999999999999, + "grad_norm": 2.0812253952026367, + "learning_rate": 0.00021509999999999997, + "loss": 2.1499, + "step": 729 + }, + { + "epoch": 1.168, + "grad_norm": 4.227649211883545, + "learning_rate": 0.00021539999999999998, + "loss": 1.8366, + "step": 730 + }, + { + "epoch": 1.1696, + "grad_norm": 3.2809646129608154, + "learning_rate": 0.00021569999999999998, + "loss": 1.9702, + "step": 731 + }, + { + "epoch": 1.1712, + "grad_norm": 3.782552480697632, + "learning_rate": 0.00021599999999999996, + "loss": 1.8251, + "step": 732 + }, + { + "epoch": 1.1728, + "grad_norm": 3.1371798515319824, + "learning_rate": 0.00021629999999999997, + "loss": 1.7244, + "step": 733 + }, + { + "epoch": 1.1743999999999999, + "grad_norm": 2.922322988510132, + "learning_rate": 0.00021659999999999998, + "loss": 1.58, + "step": 734 + }, + { + "epoch": 1.176, + "grad_norm": 1.2136024236679077, + "learning_rate": 0.0002169, + "loss": 1.7604, + "step": 735 + }, + { + "epoch": 1.1776, + "grad_norm": 1.53628408908844, + "learning_rate": 0.00021719999999999997, + "loss": 1.5032, + "step": 736 + }, + { + "epoch": 1.1792, + "grad_norm": 1.3501044511795044, + "learning_rate": 0.00021749999999999997, + "loss": 1.3641, + "step": 737 + }, + { + "epoch": 1.1808, + "grad_norm": 1.7561484575271606, + "learning_rate": 0.00021779999999999998, + "loss": 1.3152, + "step": 738 + }, + { + "epoch": 1.1824, + "grad_norm": 2.1395440101623535, + "learning_rate": 0.00021809999999999996, + "loss": 1.291, + "step": 739 + }, + { + "epoch": 1.184, + "grad_norm": 7.072542190551758, + "learning_rate": 0.00021839999999999997, + "loss": 2.4686, + "step": 740 + }, + { + "epoch": 1.1856, + "grad_norm": 2.9105377197265625, + "learning_rate": 0.00021869999999999998, + "loss": 1.4452, + "step": 741 + }, + { + "epoch": 1.1872, + "grad_norm": 2.0918514728546143, + "learning_rate": 0.00021899999999999998, + "loss": 1.6603, + "step": 742 + }, + { + "epoch": 1.1888, + "grad_norm": 6.477406024932861, + "learning_rate": 0.00021929999999999996, + "loss": 2.2748, + "step": 743 + }, + { + "epoch": 1.1904, + "grad_norm": 1.854599952697754, + "learning_rate": 0.00021959999999999997, + "loss": 1.6605, + "step": 744 + }, + { + "epoch": 1.192, + "grad_norm": 1.2816931009292603, + "learning_rate": 0.00021989999999999998, + "loss": 1.8005, + "step": 745 + }, + { + "epoch": 1.1936, + "grad_norm": 1.2794959545135498, + "learning_rate": 0.00022019999999999999, + "loss": 1.4313, + "step": 746 + }, + { + "epoch": 1.1952, + "grad_norm": 1.2156718969345093, + "learning_rate": 0.00022049999999999997, + "loss": 1.4062, + "step": 747 + }, + { + "epoch": 1.1968, + "grad_norm": 1.478219747543335, + "learning_rate": 0.00022079999999999997, + "loss": 1.4398, + "step": 748 + }, + { + "epoch": 1.1984, + "grad_norm": 1.091399908065796, + "learning_rate": 0.00022109999999999998, + "loss": 1.7798, + "step": 749 + }, + { + "epoch": 1.2, + "grad_norm": 2.1215689182281494, + "learning_rate": 0.0002214, + "loss": 1.4661, + "step": 750 + }, + { + "epoch": 1.2016, + "grad_norm": 1.341286301612854, + "learning_rate": 0.00022169999999999997, + "loss": 1.5076, + "step": 751 + }, + { + "epoch": 1.2032, + "grad_norm": 2.618873357772827, + "learning_rate": 0.00022199999999999998, + "loss": 1.8552, + "step": 752 + }, + { + "epoch": 1.2048, + "grad_norm": 1.0336085557937622, + "learning_rate": 0.00022229999999999998, + "loss": 1.3625, + "step": 753 + }, + { + "epoch": 1.2064, + "grad_norm": 1.1739717721939087, + "learning_rate": 0.0002226, + "loss": 1.4139, + "step": 754 + }, + { + "epoch": 1.208, + "grad_norm": 1.650221586227417, + "learning_rate": 0.00022289999999999997, + "loss": 1.2256, + "step": 755 + }, + { + "epoch": 1.2096, + "grad_norm": 1.2740941047668457, + "learning_rate": 0.00022319999999999998, + "loss": 1.7144, + "step": 756 + }, + { + "epoch": 1.2112, + "grad_norm": 1.0344876050949097, + "learning_rate": 0.00022349999999999998, + "loss": 1.7299, + "step": 757 + }, + { + "epoch": 1.2128, + "grad_norm": 1.5889973640441895, + "learning_rate": 0.0002238, + "loss": 1.4581, + "step": 758 + }, + { + "epoch": 1.2144, + "grad_norm": 1.274186372756958, + "learning_rate": 0.00022409999999999997, + "loss": 1.4694, + "step": 759 + }, + { + "epoch": 1.216, + "grad_norm": 1.1714391708374023, + "learning_rate": 0.00022439999999999998, + "loss": 1.5496, + "step": 760 + }, + { + "epoch": 1.2176, + "grad_norm": 1.1633304357528687, + "learning_rate": 0.0002247, + "loss": 1.6381, + "step": 761 + }, + { + "epoch": 1.2192, + "grad_norm": 1.6656583547592163, + "learning_rate": 0.000225, + "loss": 1.3814, + "step": 762 + }, + { + "epoch": 1.2208, + "grad_norm": 1.7100461721420288, + "learning_rate": 0.00022529999999999997, + "loss": 1.3805, + "step": 763 + }, + { + "epoch": 1.2224, + "grad_norm": 1.6684997081756592, + "learning_rate": 0.00022559999999999998, + "loss": 1.4335, + "step": 764 + }, + { + "epoch": 1.224, + "grad_norm": 1.9306715726852417, + "learning_rate": 0.0002259, + "loss": 1.6239, + "step": 765 + }, + { + "epoch": 1.2256, + "grad_norm": 1.488796353340149, + "learning_rate": 0.00022619999999999997, + "loss": 1.9, + "step": 766 + }, + { + "epoch": 1.2272, + "grad_norm": 1.65590500831604, + "learning_rate": 0.00022649999999999998, + "loss": 1.501, + "step": 767 + }, + { + "epoch": 1.2288000000000001, + "grad_norm": 1.105634331703186, + "learning_rate": 0.00022679999999999998, + "loss": 1.6754, + "step": 768 + }, + { + "epoch": 1.2304, + "grad_norm": 1.3590998649597168, + "learning_rate": 0.0002271, + "loss": 2.0468, + "step": 769 + }, + { + "epoch": 1.232, + "grad_norm": 1.2433547973632812, + "learning_rate": 0.00022739999999999997, + "loss": 1.4335, + "step": 770 + }, + { + "epoch": 1.2336, + "grad_norm": 1.7231038808822632, + "learning_rate": 0.00022769999999999998, + "loss": 1.465, + "step": 771 + }, + { + "epoch": 1.2352, + "grad_norm": 1.6311777830123901, + "learning_rate": 0.00022799999999999999, + "loss": 1.9477, + "step": 772 + }, + { + "epoch": 1.2368000000000001, + "grad_norm": 1.3222523927688599, + "learning_rate": 0.0002283, + "loss": 1.4479, + "step": 773 + }, + { + "epoch": 1.2384, + "grad_norm": 1.449326992034912, + "learning_rate": 0.00022859999999999997, + "loss": 1.7486, + "step": 774 + }, + { + "epoch": 1.24, + "grad_norm": NaN, + "learning_rate": 0.00022859999999999997, + "loss": 1.6244, + "step": 775 + }, + { + "epoch": 1.2416, + "grad_norm": 8.872183799743652, + "learning_rate": 0.00022889999999999998, + "loss": 3.0317, + "step": 776 + }, + { + "epoch": 1.2432, + "grad_norm": 3.2334752082824707, + "learning_rate": 0.0002292, + "loss": 1.7795, + "step": 777 + }, + { + "epoch": 1.2448, + "grad_norm": 2.064131259918213, + "learning_rate": 0.0002295, + "loss": 1.8597, + "step": 778 + }, + { + "epoch": 1.2464, + "grad_norm": 2.9841063022613525, + "learning_rate": 0.00022979999999999997, + "loss": 2.193, + "step": 779 + }, + { + "epoch": 1.248, + "grad_norm": 1.8428844213485718, + "learning_rate": 0.00023009999999999998, + "loss": 1.7662, + "step": 780 + }, + { + "epoch": 1.2496, + "grad_norm": 1.9402704238891602, + "learning_rate": 0.0002304, + "loss": 2.0471, + "step": 781 + }, + { + "epoch": 1.2511999999999999, + "grad_norm": 1.971617341041565, + "learning_rate": 0.0002307, + "loss": 1.8941, + "step": 782 + }, + { + "epoch": 1.2528000000000001, + "grad_norm": 3.7281742095947266, + "learning_rate": 0.00023099999999999998, + "loss": 1.8762, + "step": 783 + }, + { + "epoch": 1.2544, + "grad_norm": 1.6704293489456177, + "learning_rate": 0.00023129999999999998, + "loss": 1.7298, + "step": 784 + }, + { + "epoch": 1.256, + "grad_norm": 1.1760116815567017, + "learning_rate": 0.0002316, + "loss": 1.6819, + "step": 785 + }, + { + "epoch": 1.2576, + "grad_norm": 1.6610198020935059, + "learning_rate": 0.0002319, + "loss": 1.4823, + "step": 786 + }, + { + "epoch": 1.2591999999999999, + "grad_norm": 2.954132318496704, + "learning_rate": 0.00023219999999999998, + "loss": 1.7275, + "step": 787 + }, + { + "epoch": 1.2608, + "grad_norm": 4.86627197265625, + "learning_rate": 0.00023249999999999999, + "loss": 1.9728, + "step": 788 + }, + { + "epoch": 1.2624, + "grad_norm": 3.5794413089752197, + "learning_rate": 0.0002328, + "loss": 1.5752, + "step": 789 + }, + { + "epoch": 1.264, + "grad_norm": 3.866424322128296, + "learning_rate": 0.00023309999999999997, + "loss": 1.6367, + "step": 790 + }, + { + "epoch": 1.2656, + "grad_norm": 4.216460704803467, + "learning_rate": 0.00023339999999999998, + "loss": 1.8715, + "step": 791 + }, + { + "epoch": 1.2671999999999999, + "grad_norm": 2.919419288635254, + "learning_rate": 0.0002337, + "loss": 1.3454, + "step": 792 + }, + { + "epoch": 1.2688, + "grad_norm": 2.455012798309326, + "learning_rate": 0.000234, + "loss": 1.6547, + "step": 793 + }, + { + "epoch": 1.2704, + "grad_norm": 1.444284439086914, + "learning_rate": 0.00023429999999999998, + "loss": 1.4472, + "step": 794 + }, + { + "epoch": 1.272, + "grad_norm": 1.5226218700408936, + "learning_rate": 0.00023459999999999998, + "loss": 1.4115, + "step": 795 + }, + { + "epoch": 1.2736, + "grad_norm": 1.4698041677474976, + "learning_rate": 0.0002349, + "loss": 1.4998, + "step": 796 + }, + { + "epoch": 1.2752, + "grad_norm": 0.9360406994819641, + "learning_rate": 0.0002352, + "loss": 1.3246, + "step": 797 + }, + { + "epoch": 1.2768, + "grad_norm": 0.9899576902389526, + "learning_rate": 0.00023549999999999998, + "loss": 1.7864, + "step": 798 + }, + { + "epoch": 1.2784, + "grad_norm": 1.042768955230713, + "learning_rate": 0.00023579999999999999, + "loss": 1.7181, + "step": 799 + }, + { + "epoch": 1.28, + "grad_norm": 0.9345083236694336, + "learning_rate": 0.0002361, + "loss": 1.3929, + "step": 800 + }, + { + "epoch": 1.2816, + "grad_norm": 1.2019685506820679, + "learning_rate": 0.0002364, + "loss": 1.6232, + "step": 801 + }, + { + "epoch": 1.2832, + "grad_norm": 1.1262824535369873, + "learning_rate": 0.00023669999999999998, + "loss": 1.3208, + "step": 802 + }, + { + "epoch": 1.2848, + "grad_norm": 2.1480746269226074, + "learning_rate": 0.000237, + "loss": 1.484, + "step": 803 + }, + { + "epoch": 1.2864, + "grad_norm": 1.6011468172073364, + "learning_rate": 0.0002373, + "loss": 1.4428, + "step": 804 + }, + { + "epoch": 1.288, + "grad_norm": 1.681715965270996, + "learning_rate": 0.0002376, + "loss": 1.4815, + "step": 805 + }, + { + "epoch": 1.2896, + "grad_norm": 1.0105364322662354, + "learning_rate": 0.00023789999999999998, + "loss": 1.583, + "step": 806 + }, + { + "epoch": 1.2912, + "grad_norm": 2.4335265159606934, + "learning_rate": 0.0002382, + "loss": 1.6736, + "step": 807 + }, + { + "epoch": 1.2928, + "grad_norm": 1.1832919120788574, + "learning_rate": 0.0002385, + "loss": 1.7259, + "step": 808 + }, + { + "epoch": 1.2944, + "grad_norm": 1.4895908832550049, + "learning_rate": 0.0002388, + "loss": 1.5519, + "step": 809 + }, + { + "epoch": 1.296, + "grad_norm": 0.9306002855300903, + "learning_rate": 0.00023909999999999998, + "loss": 1.7638, + "step": 810 + }, + { + "epoch": 1.2976, + "grad_norm": 0.862756609916687, + "learning_rate": 0.0002394, + "loss": 1.3956, + "step": 811 + }, + { + "epoch": 1.2992, + "grad_norm": 0.9873697757720947, + "learning_rate": 0.0002397, + "loss": 1.741, + "step": 812 + }, + { + "epoch": 1.3008, + "grad_norm": 0.8924622535705566, + "learning_rate": 0.00023999999999999998, + "loss": 1.6026, + "step": 813 + }, + { + "epoch": 1.3024, + "grad_norm": 0.9441061615943909, + "learning_rate": 0.00024029999999999999, + "loss": 1.2882, + "step": 814 + }, + { + "epoch": 1.304, + "grad_norm": 1.0321191549301147, + "learning_rate": 0.0002406, + "loss": 1.4606, + "step": 815 + }, + { + "epoch": 1.3056, + "grad_norm": 1.8343979120254517, + "learning_rate": 0.0002409, + "loss": 1.6286, + "step": 816 + }, + { + "epoch": 1.3072, + "grad_norm": 1.3048229217529297, + "learning_rate": 0.00024119999999999998, + "loss": 1.8307, + "step": 817 + }, + { + "epoch": 1.3088, + "grad_norm": 1.3279473781585693, + "learning_rate": 0.0002415, + "loss": 1.8301, + "step": 818 + }, + { + "epoch": 1.3104, + "grad_norm": 1.0777429342269897, + "learning_rate": 0.0002418, + "loss": 1.5125, + "step": 819 + }, + { + "epoch": 1.312, + "grad_norm": 1.0291649103164673, + "learning_rate": 0.0002421, + "loss": 1.5399, + "step": 820 + }, + { + "epoch": 1.3136, + "grad_norm": 1.1538206338882446, + "learning_rate": 0.00024239999999999998, + "loss": 1.6562, + "step": 821 + }, + { + "epoch": 1.3152, + "grad_norm": 1.9606250524520874, + "learning_rate": 0.0002427, + "loss": 1.7324, + "step": 822 + }, + { + "epoch": 1.3168, + "grad_norm": 1.927132248878479, + "learning_rate": 0.000243, + "loss": 1.8767, + "step": 823 + }, + { + "epoch": 1.3184, + "grad_norm": 1.565555453300476, + "learning_rate": 0.0002433, + "loss": 2.3164, + "step": 824 + }, + { + "epoch": 1.32, + "grad_norm": NaN, + "learning_rate": 0.0002433, + "loss": 2.2162, + "step": 825 + }, + { + "epoch": 1.3216, + "grad_norm": 4.660001754760742, + "learning_rate": 0.00024359999999999999, + "loss": 1.9052, + "step": 826 + }, + { + "epoch": 1.3232, + "grad_norm": 4.901458740234375, + "learning_rate": 0.00024389999999999997, + "loss": 2.2262, + "step": 827 + }, + { + "epoch": 1.3248, + "grad_norm": 6.842830657958984, + "learning_rate": 0.00024419999999999997, + "loss": 2.609, + "step": 828 + }, + { + "epoch": 1.3264, + "grad_norm": 5.018871307373047, + "learning_rate": 0.0002445, + "loss": 1.9541, + "step": 829 + }, + { + "epoch": 1.328, + "grad_norm": 2.6428186893463135, + "learning_rate": 0.0002448, + "loss": 1.7591, + "step": 830 + }, + { + "epoch": 1.3296000000000001, + "grad_norm": 3.7194108963012695, + "learning_rate": 0.00024509999999999994, + "loss": 1.8893, + "step": 831 + }, + { + "epoch": 1.3312, + "grad_norm": 1.5559850931167603, + "learning_rate": 0.00024539999999999995, + "loss": 1.6199, + "step": 832 + }, + { + "epoch": 1.3328, + "grad_norm": 1.6611255407333374, + "learning_rate": 0.00024569999999999995, + "loss": 1.7948, + "step": 833 + }, + { + "epoch": 1.3344, + "grad_norm": 2.756481170654297, + "learning_rate": 0.00024599999999999996, + "loss": 1.3697, + "step": 834 + }, + { + "epoch": 1.336, + "grad_norm": 3.1236398220062256, + "learning_rate": 0.00024629999999999997, + "loss": 1.6154, + "step": 835 + }, + { + "epoch": 1.3376000000000001, + "grad_norm": 3.052428722381592, + "learning_rate": 0.0002466, + "loss": 1.7594, + "step": 836 + }, + { + "epoch": 1.3392, + "grad_norm": 2.0928995609283447, + "learning_rate": 0.0002469, + "loss": 2.4307, + "step": 837 + }, + { + "epoch": 1.3408, + "grad_norm": 1.914813756942749, + "learning_rate": 0.0002472, + "loss": 1.8485, + "step": 838 + }, + { + "epoch": 1.3424, + "grad_norm": 1.9151816368103027, + "learning_rate": 0.00024749999999999994, + "loss": 1.8105, + "step": 839 + }, + { + "epoch": 1.3439999999999999, + "grad_norm": 1.0234382152557373, + "learning_rate": 0.00024779999999999995, + "loss": 1.812, + "step": 840 + }, + { + "epoch": 1.3456000000000001, + "grad_norm": 2.383925199508667, + "learning_rate": 0.00024809999999999996, + "loss": 1.8388, + "step": 841 + }, + { + "epoch": 1.3472, + "grad_norm": 6.021166801452637, + "learning_rate": 0.00024839999999999997, + "loss": 2.2525, + "step": 842 + }, + { + "epoch": 1.3488, + "grad_norm": 5.078658580780029, + "learning_rate": 0.0002487, + "loss": 1.9297, + "step": 843 + }, + { + "epoch": 1.3504, + "grad_norm": 7.2157063484191895, + "learning_rate": 0.000249, + "loss": 2.1607, + "step": 844 + }, + { + "epoch": 1.3519999999999999, + "grad_norm": 7.366926193237305, + "learning_rate": 0.0002493, + "loss": 2.2679, + "step": 845 + }, + { + "epoch": 1.3536000000000001, + "grad_norm": 8.579809188842773, + "learning_rate": 0.00024959999999999994, + "loss": 2.5445, + "step": 846 + }, + { + "epoch": 1.3552, + "grad_norm": 9.298418998718262, + "learning_rate": 0.00024989999999999995, + "loss": 2.7447, + "step": 847 + }, + { + "epoch": 1.3568, + "grad_norm": 4.966423988342285, + "learning_rate": 0.00025019999999999996, + "loss": 2.3292, + "step": 848 + }, + { + "epoch": 1.3584, + "grad_norm": 3.2994112968444824, + "learning_rate": 0.00025049999999999996, + "loss": 2.1082, + "step": 849 + }, + { + "epoch": 1.3599999999999999, + "grad_norm": 2.276888370513916, + "learning_rate": 0.00025079999999999997, + "loss": 2.0868, + "step": 850 + }, + { + "epoch": 1.3616, + "grad_norm": 1.3332836627960205, + "learning_rate": 0.0002511, + "loss": 2.0474, + "step": 851 + }, + { + "epoch": 1.3632, + "grad_norm": 1.1775461435317993, + "learning_rate": 0.0002514, + "loss": 2.0204, + "step": 852 + }, + { + "epoch": 1.3648, + "grad_norm": 1.3132644891738892, + "learning_rate": 0.0002517, + "loss": 2.0342, + "step": 853 + }, + { + "epoch": 1.3664, + "grad_norm": 1.407052755355835, + "learning_rate": 0.00025199999999999995, + "loss": 2.0199, + "step": 854 + }, + { + "epoch": 1.3679999999999999, + "grad_norm": 0.9322783350944519, + "learning_rate": 0.00025229999999999995, + "loss": 1.9461, + "step": 855 + }, + { + "epoch": 1.3696, + "grad_norm": 1.6221606731414795, + "learning_rate": 0.00025259999999999996, + "loss": 1.9301, + "step": 856 + }, + { + "epoch": 1.3712, + "grad_norm": 1.9470421075820923, + "learning_rate": 0.00025289999999999997, + "loss": 2.1221, + "step": 857 + }, + { + "epoch": 1.3728, + "grad_norm": 2.600743532180786, + "learning_rate": 0.0002532, + "loss": 1.9257, + "step": 858 + }, + { + "epoch": 1.3744, + "grad_norm": 4.783127784729004, + "learning_rate": 0.0002535, + "loss": 2.6232, + "step": 859 + }, + { + "epoch": 1.376, + "grad_norm": 2.1699776649475098, + "learning_rate": 0.0002538, + "loss": 1.8527, + "step": 860 + }, + { + "epoch": 1.3776, + "grad_norm": 1.9697574377059937, + "learning_rate": 0.0002541, + "loss": 1.9053, + "step": 861 + }, + { + "epoch": 1.3792, + "grad_norm": 1.9653360843658447, + "learning_rate": 0.00025439999999999995, + "loss": 1.7472, + "step": 862 + }, + { + "epoch": 1.3808, + "grad_norm": 0.975303053855896, + "learning_rate": 0.00025469999999999996, + "loss": 1.5276, + "step": 863 + }, + { + "epoch": 1.3824, + "grad_norm": 0.8114136457443237, + "learning_rate": 0.00025499999999999996, + "loss": 1.9605, + "step": 864 + }, + { + "epoch": 1.384, + "grad_norm": 1.0743026733398438, + "learning_rate": 0.00025529999999999997, + "loss": 1.964, + "step": 865 + }, + { + "epoch": 1.3856, + "grad_norm": 0.572559118270874, + "learning_rate": 0.0002556, + "loss": 2.0769, + "step": 866 + }, + { + "epoch": 1.3872, + "grad_norm": 1.8745307922363281, + "learning_rate": 0.0002559, + "loss": 1.9341, + "step": 867 + }, + { + "epoch": 1.3888, + "grad_norm": 2.8381495475769043, + "learning_rate": 0.0002562, + "loss": 2.0994, + "step": 868 + }, + { + "epoch": 1.3904, + "grad_norm": 1.837712287902832, + "learning_rate": 0.00025649999999999995, + "loss": 1.6848, + "step": 869 + }, + { + "epoch": 1.392, + "grad_norm": 1.0406761169433594, + "learning_rate": 0.00025679999999999995, + "loss": 1.9728, + "step": 870 + }, + { + "epoch": 1.3936, + "grad_norm": 0.8133671879768372, + "learning_rate": 0.00025709999999999996, + "loss": 1.9183, + "step": 871 + }, + { + "epoch": 1.3952, + "grad_norm": 1.488303780555725, + "learning_rate": 0.00025739999999999997, + "loss": 1.7447, + "step": 872 + }, + { + "epoch": 1.3968, + "grad_norm": 0.8098030686378479, + "learning_rate": 0.0002577, + "loss": 2.1065, + "step": 873 + }, + { + "epoch": 1.3984, + "grad_norm": 1.3492127656936646, + "learning_rate": 0.000258, + "loss": 1.8567, + "step": 874 + }, + { + "epoch": 1.4, + "grad_norm": NaN, + "learning_rate": 0.000258, + "loss": 0.0, + "step": 875 + }, + { + "epoch": 1.4016, + "grad_norm": 3.2372701168060303, + "learning_rate": 0.0002583, + "loss": 2.1654, + "step": 876 + }, + { + "epoch": 1.4032, + "grad_norm": 1.9243760108947754, + "learning_rate": 0.0002586, + "loss": 1.8519, + "step": 877 + }, + { + "epoch": 1.4048, + "grad_norm": 1.633375883102417, + "learning_rate": 0.00025889999999999995, + "loss": 1.9246, + "step": 878 + }, + { + "epoch": 1.4064, + "grad_norm": 1.862678050994873, + "learning_rate": 0.00025919999999999996, + "loss": 1.796, + "step": 879 + }, + { + "epoch": 1.408, + "grad_norm": 1.930555820465088, + "learning_rate": 0.00025949999999999997, + "loss": 2.1167, + "step": 880 + }, + { + "epoch": 1.4096, + "grad_norm": 3.6139817237854004, + "learning_rate": 0.00025979999999999997, + "loss": 1.8125, + "step": 881 + }, + { + "epoch": 1.4112, + "grad_norm": 4.2841691970825195, + "learning_rate": 0.0002601, + "loss": 1.6891, + "step": 882 + }, + { + "epoch": 1.4128, + "grad_norm": 4.878454685211182, + "learning_rate": 0.0002604, + "loss": 1.9731, + "step": 883 + }, + { + "epoch": 1.4144, + "grad_norm": 2.765392541885376, + "learning_rate": 0.0002607, + "loss": 2.3051, + "step": 884 + }, + { + "epoch": 1.416, + "grad_norm": 5.951148986816406, + "learning_rate": 0.000261, + "loss": 2.044, + "step": 885 + }, + { + "epoch": 1.4176, + "grad_norm": 6.356194496154785, + "learning_rate": 0.00026129999999999995, + "loss": 2.267, + "step": 886 + }, + { + "epoch": 1.4192, + "grad_norm": 7.16682767868042, + "learning_rate": 0.00026159999999999996, + "loss": 2.3927, + "step": 887 + }, + { + "epoch": 1.4208, + "grad_norm": 6.482820987701416, + "learning_rate": 0.00026189999999999997, + "loss": 2.2649, + "step": 888 + }, + { + "epoch": 1.4224, + "grad_norm": 6.568836688995361, + "learning_rate": 0.0002622, + "loss": 2.4487, + "step": 889 + }, + { + "epoch": 1.424, + "grad_norm": 6.718606472015381, + "learning_rate": 0.0002625, + "loss": 2.3971, + "step": 890 + }, + { + "epoch": 1.4256, + "grad_norm": 5.949743747711182, + "learning_rate": 0.0002628, + "loss": 2.2921, + "step": 891 + }, + { + "epoch": 1.4272, + "grad_norm": 4.38700532913208, + "learning_rate": 0.0002631, + "loss": 2.0761, + "step": 892 + }, + { + "epoch": 1.4288, + "grad_norm": 4.077221393585205, + "learning_rate": 0.00026339999999999995, + "loss": 2.1059, + "step": 893 + }, + { + "epoch": 1.4304000000000001, + "grad_norm": 1.6001068353652954, + "learning_rate": 0.00026369999999999996, + "loss": 2.1777, + "step": 894 + }, + { + "epoch": 1.432, + "grad_norm": 1.2680809497833252, + "learning_rate": 0.00026399999999999997, + "loss": 2.216, + "step": 895 + }, + { + "epoch": 1.4336, + "grad_norm": 0.9712734222412109, + "learning_rate": 0.0002643, + "loss": 1.9073, + "step": 896 + }, + { + "epoch": 1.4352, + "grad_norm": 2.31636381149292, + "learning_rate": 0.0002646, + "loss": 1.8749, + "step": 897 + }, + { + "epoch": 1.4368, + "grad_norm": 3.8894119262695312, + "learning_rate": 0.0002649, + "loss": 2.1202, + "step": 898 + }, + { + "epoch": 1.4384000000000001, + "grad_norm": 5.789621353149414, + "learning_rate": 0.0002652, + "loss": 1.9118, + "step": 899 + }, + { + "epoch": 1.44, + "grad_norm": 7.5779619216918945, + "learning_rate": 0.0002655, + "loss": 2.2179, + "step": 900 + }, + { + "epoch": 1.4416, + "grad_norm": 10.712891578674316, + "learning_rate": 0.00026579999999999996, + "loss": 2.8742, + "step": 901 + }, + { + "epoch": 1.4432, + "grad_norm": 8.757124900817871, + "learning_rate": 0.00026609999999999996, + "loss": 2.4353, + "step": 902 + }, + { + "epoch": 1.4447999999999999, + "grad_norm": 9.56799602508545, + "learning_rate": 0.00026639999999999997, + "loss": 2.646, + "step": 903 + }, + { + "epoch": 1.4464000000000001, + "grad_norm": 8.285284996032715, + "learning_rate": 0.0002667, + "loss": 2.5576, + "step": 904 + }, + { + "epoch": 1.448, + "grad_norm": 7.641500473022461, + "learning_rate": 0.000267, + "loss": 2.4627, + "step": 905 + }, + { + "epoch": 1.4496, + "grad_norm": 9.177940368652344, + "learning_rate": 0.0002673, + "loss": 2.4342, + "step": 906 + }, + { + "epoch": 1.4512, + "grad_norm": 8.318049430847168, + "learning_rate": 0.0002676, + "loss": 2.5166, + "step": 907 + }, + { + "epoch": 1.4527999999999999, + "grad_norm": 6.994287967681885, + "learning_rate": 0.0002679, + "loss": 2.4509, + "step": 908 + }, + { + "epoch": 1.4544000000000001, + "grad_norm": 6.329995632171631, + "learning_rate": 0.00026819999999999996, + "loss": 2.4563, + "step": 909 + }, + { + "epoch": 1.456, + "grad_norm": 5.76509952545166, + "learning_rate": 0.00026849999999999997, + "loss": 2.2929, + "step": 910 + }, + { + "epoch": 1.4576, + "grad_norm": 3.3530473709106445, + "learning_rate": 0.0002688, + "loss": 2.0108, + "step": 911 + }, + { + "epoch": 1.4592, + "grad_norm": 2.6256699562072754, + "learning_rate": 0.0002691, + "loss": 2.0606, + "step": 912 + }, + { + "epoch": 1.4607999999999999, + "grad_norm": 1.4539761543273926, + "learning_rate": 0.0002694, + "loss": 1.9538, + "step": 913 + }, + { + "epoch": 1.4624, + "grad_norm": 1.3113408088684082, + "learning_rate": 0.0002697, + "loss": 2.1957, + "step": 914 + }, + { + "epoch": 1.464, + "grad_norm": 1.1808632612228394, + "learning_rate": 0.00027, + "loss": 2.0123, + "step": 915 + }, + { + "epoch": 1.4656, + "grad_norm": 0.9282050728797913, + "learning_rate": 0.00027029999999999996, + "loss": 2.0029, + "step": 916 + }, + { + "epoch": 1.4672, + "grad_norm": 1.2517030239105225, + "learning_rate": 0.00027059999999999996, + "loss": 1.9543, + "step": 917 + }, + { + "epoch": 1.4687999999999999, + "grad_norm": 1.2844178676605225, + "learning_rate": 0.00027089999999999997, + "loss": 1.8131, + "step": 918 + }, + { + "epoch": 1.4704, + "grad_norm": 0.7779868841171265, + "learning_rate": 0.0002712, + "loss": 1.9653, + "step": 919 + }, + { + "epoch": 1.472, + "grad_norm": 1.8325669765472412, + "learning_rate": 0.0002715, + "loss": 1.8809, + "step": 920 + }, + { + "epoch": 1.4736, + "grad_norm": 2.3614306449890137, + "learning_rate": 0.0002718, + "loss": 1.9198, + "step": 921 + }, + { + "epoch": 1.4752, + "grad_norm": 1.5548219680786133, + "learning_rate": 0.0002721, + "loss": 2.319, + "step": 922 + }, + { + "epoch": 1.4768, + "grad_norm": 1.010625958442688, + "learning_rate": 0.0002724, + "loss": 2.2114, + "step": 923 + }, + { + "epoch": 1.4784, + "grad_norm": 2.4790525436401367, + "learning_rate": 0.00027269999999999996, + "loss": 2.4394, + "step": 924 + }, + { + "epoch": 1.48, + "grad_norm": 3.139253616333008, + "learning_rate": 0.00027299999999999997, + "loss": 2.2607, + "step": 925 + }, + { + "epoch": 1.4816, + "grad_norm": 21.243789672851562, + "learning_rate": 0.0002733, + "loss": 4.0305, + "step": 926 + }, + { + "epoch": 1.4832, + "grad_norm": 6.695805549621582, + "learning_rate": 0.0002736, + "loss": 2.5432, + "step": 927 + }, + { + "epoch": 1.4848, + "grad_norm": 2.8859128952026367, + "learning_rate": 0.0002739, + "loss": 2.1313, + "step": 928 + }, + { + "epoch": 1.4864, + "grad_norm": 2.9990243911743164, + "learning_rate": 0.0002742, + "loss": 2.0824, + "step": 929 + }, + { + "epoch": 1.488, + "grad_norm": 0.9504282474517822, + "learning_rate": 0.0002745, + "loss": 1.9647, + "step": 930 + }, + { + "epoch": 1.4896, + "grad_norm": 1.416487693786621, + "learning_rate": 0.0002748, + "loss": 1.9534, + "step": 931 + }, + { + "epoch": 1.4912, + "grad_norm": 3.924936532974243, + "learning_rate": 0.00027509999999999996, + "loss": 2.1581, + "step": 932 + }, + { + "epoch": 1.4928, + "grad_norm": 4.775065898895264, + "learning_rate": 0.00027539999999999997, + "loss": 2.0693, + "step": 933 + }, + { + "epoch": 1.4944, + "grad_norm": 4.829035758972168, + "learning_rate": 0.0002757, + "loss": 2.0134, + "step": 934 + }, + { + "epoch": 1.496, + "grad_norm": 4.46183443069458, + "learning_rate": 0.000276, + "loss": 2.2973, + "step": 935 + }, + { + "epoch": 1.4976, + "grad_norm": 5.705532550811768, + "learning_rate": 0.0002763, + "loss": 2.2353, + "step": 936 + }, + { + "epoch": 1.4992, + "grad_norm": 5.170881748199463, + "learning_rate": 0.0002766, + "loss": 2.4058, + "step": 937 + }, + { + "epoch": 1.5008, + "grad_norm": 5.087926864624023, + "learning_rate": 0.0002769, + "loss": 2.006, + "step": 938 + }, + { + "epoch": 1.5024, + "grad_norm": 5.493293762207031, + "learning_rate": 0.0002772, + "loss": 2.2158, + "step": 939 + }, + { + "epoch": 1.504, + "grad_norm": 4.74140739440918, + "learning_rate": 0.00027749999999999997, + "loss": 2.2086, + "step": 940 + }, + { + "epoch": 1.5056, + "grad_norm": 3.2799718379974365, + "learning_rate": 0.0002778, + "loss": 2.1634, + "step": 941 + }, + { + "epoch": 1.5072, + "grad_norm": 2.608619451522827, + "learning_rate": 0.0002781, + "loss": 2.2917, + "step": 942 + }, + { + "epoch": 1.5088, + "grad_norm": 1.8219605684280396, + "learning_rate": 0.0002784, + "loss": 2.0337, + "step": 943 + }, + { + "epoch": 1.5104, + "grad_norm": 0.5471847653388977, + "learning_rate": 0.0002787, + "loss": 2.0357, + "step": 944 + }, + { + "epoch": 1.512, + "grad_norm": 1.5796411037445068, + "learning_rate": 0.000279, + "loss": 1.8157, + "step": 945 + }, + { + "epoch": 1.5135999999999998, + "grad_norm": 3.0464274883270264, + "learning_rate": 0.0002793, + "loss": 1.9289, + "step": 946 + }, + { + "epoch": 1.5152, + "grad_norm": 4.224644660949707, + "learning_rate": 0.00027959999999999997, + "loss": 2.1435, + "step": 947 + }, + { + "epoch": 1.5168, + "grad_norm": 5.403804302215576, + "learning_rate": 0.0002799, + "loss": 1.9621, + "step": 948 + }, + { + "epoch": 1.5184, + "grad_norm": 4.696282863616943, + "learning_rate": 0.0002802, + "loss": 2.2405, + "step": 949 + }, + { + "epoch": 1.52, + "grad_norm": 5.692913055419922, + "learning_rate": 0.0002805, + "loss": 2.2135, + "step": 950 + }, + { + "epoch": 1.5215999999999998, + "grad_norm": 6.66227912902832, + "learning_rate": 0.0002808, + "loss": 2.1851, + "step": 951 + }, + { + "epoch": 1.5232, + "grad_norm": 7.124788761138916, + "learning_rate": 0.0002811, + "loss": 2.1455, + "step": 952 + }, + { + "epoch": 1.5248, + "grad_norm": 6.634387016296387, + "learning_rate": 0.00028139999999999996, + "loss": 2.2038, + "step": 953 + }, + { + "epoch": 1.5264, + "grad_norm": 4.773592948913574, + "learning_rate": 0.00028169999999999996, + "loss": 1.9655, + "step": 954 + }, + { + "epoch": 1.528, + "grad_norm": 6.511046409606934, + "learning_rate": 0.00028199999999999997, + "loss": 2.1953, + "step": 955 + }, + { + "epoch": 1.5295999999999998, + "grad_norm": 5.800004005432129, + "learning_rate": 0.0002823, + "loss": 2.298, + "step": 956 + }, + { + "epoch": 1.5312000000000001, + "grad_norm": 8.844927787780762, + "learning_rate": 0.0002826, + "loss": 2.6741, + "step": 957 + }, + { + "epoch": 1.5328, + "grad_norm": 2.801509141921997, + "learning_rate": 0.00028289999999999994, + "loss": 2.0405, + "step": 958 + }, + { + "epoch": 1.5344, + "grad_norm": 1.441400170326233, + "learning_rate": 0.00028319999999999994, + "loss": 2.0097, + "step": 959 + }, + { + "epoch": 1.536, + "grad_norm": 0.6606685519218445, + "learning_rate": 0.00028349999999999995, + "loss": 1.9012, + "step": 960 + }, + { + "epoch": 1.5375999999999999, + "grad_norm": 1.0121657848358154, + "learning_rate": 0.00028379999999999996, + "loss": 2.1164, + "step": 961 + }, + { + "epoch": 1.5392000000000001, + "grad_norm": 2.11448073387146, + "learning_rate": 0.00028409999999999997, + "loss": 2.1646, + "step": 962 + }, + { + "epoch": 1.5408, + "grad_norm": 1.660666823387146, + "learning_rate": 0.0002844, + "loss": 2.0593, + "step": 963 + }, + { + "epoch": 1.5424, + "grad_norm": 0.7296435236930847, + "learning_rate": 0.0002847, + "loss": 1.9912, + "step": 964 + }, + { + "epoch": 1.544, + "grad_norm": 0.8172574639320374, + "learning_rate": 0.000285, + "loss": 2.1308, + "step": 965 + }, + { + "epoch": 1.5455999999999999, + "grad_norm": 3.529291868209839, + "learning_rate": 0.00028529999999999994, + "loss": 2.4858, + "step": 966 + }, + { + "epoch": 1.5472000000000001, + "grad_norm": 3.8260555267333984, + "learning_rate": 0.00028559999999999995, + "loss": 2.3045, + "step": 967 + }, + { + "epoch": 1.5488, + "grad_norm": 5.841991901397705, + "learning_rate": 0.00028589999999999996, + "loss": 2.5007, + "step": 968 + }, + { + "epoch": 1.5504, + "grad_norm": 5.58250093460083, + "learning_rate": 0.00028619999999999996, + "loss": 2.1947, + "step": 969 + }, + { + "epoch": 1.552, + "grad_norm": 6.479895114898682, + "learning_rate": 0.00028649999999999997, + "loss": 2.2422, + "step": 970 + }, + { + "epoch": 1.5535999999999999, + "grad_norm": 5.577583312988281, + "learning_rate": 0.0002868, + "loss": 2.3023, + "step": 971 + }, + { + "epoch": 1.5552000000000001, + "grad_norm": 4.665249824523926, + "learning_rate": 0.0002871, + "loss": 2.7231, + "step": 972 + }, + { + "epoch": 1.5568, + "grad_norm": 3.4380602836608887, + "learning_rate": 0.00028739999999999994, + "loss": 2.3567, + "step": 973 + }, + { + "epoch": 1.5584, + "grad_norm": 2.9674761295318604, + "learning_rate": 0.00028769999999999995, + "loss": 2.7469, + "step": 974 + }, + { + "epoch": 1.56, + "grad_norm": 2.04423189163208, + "learning_rate": 0.00028799999999999995, + "loss": 2.4228, + "step": 975 + }, + { + "epoch": 1.5615999999999999, + "grad_norm": 10.190044403076172, + "learning_rate": 0.00028829999999999996, + "loss": 3.0674, + "step": 976 + }, + { + "epoch": 1.5632000000000001, + "grad_norm": 2.3500773906707764, + "learning_rate": 0.00028859999999999997, + "loss": 2.4471, + "step": 977 + }, + { + "epoch": 1.5648, + "grad_norm": 1.5877519845962524, + "learning_rate": 0.0002889, + "loss": 2.3215, + "step": 978 + }, + { + "epoch": 1.5664, + "grad_norm": 2.7796225547790527, + "learning_rate": 0.0002892, + "loss": 2.5779, + "step": 979 + }, + { + "epoch": 1.568, + "grad_norm": 2.8195817470550537, + "learning_rate": 0.0002895, + "loss": 2.5889, + "step": 980 + }, + { + "epoch": 1.5695999999999999, + "grad_norm": 2.0445921421051025, + "learning_rate": 0.00028979999999999994, + "loss": 2.4993, + "step": 981 + }, + { + "epoch": 1.5712000000000002, + "grad_norm": 3.8158059120178223, + "learning_rate": 0.00029009999999999995, + "loss": 2.29, + "step": 982 + }, + { + "epoch": 1.5728, + "grad_norm": 5.886443614959717, + "learning_rate": 0.00029039999999999996, + "loss": 2.4405, + "step": 983 + }, + { + "epoch": 1.5744, + "grad_norm": 4.135772228240967, + "learning_rate": 0.00029069999999999996, + "loss": 2.4418, + "step": 984 + }, + { + "epoch": 1.576, + "grad_norm": 6.405320644378662, + "learning_rate": 0.00029099999999999997, + "loss": 2.4061, + "step": 985 + }, + { + "epoch": 1.5776, + "grad_norm": 5.528068542480469, + "learning_rate": 0.0002913, + "loss": 2.0997, + "step": 986 + }, + { + "epoch": 1.5792000000000002, + "grad_norm": 5.618749618530273, + "learning_rate": 0.0002916, + "loss": 2.4048, + "step": 987 + }, + { + "epoch": 1.5808, + "grad_norm": 5.455833911895752, + "learning_rate": 0.0002919, + "loss": 2.3788, + "step": 988 + }, + { + "epoch": 1.5824, + "grad_norm": 5.045185089111328, + "learning_rate": 0.00029219999999999995, + "loss": 2.1703, + "step": 989 + }, + { + "epoch": 1.584, + "grad_norm": 3.192173480987549, + "learning_rate": 0.00029249999999999995, + "loss": 2.4022, + "step": 990 + }, + { + "epoch": 1.5856, + "grad_norm": 3.1793525218963623, + "learning_rate": 0.00029279999999999996, + "loss": 2.1956, + "step": 991 + }, + { + "epoch": 1.5872000000000002, + "grad_norm": 1.323548674583435, + "learning_rate": 0.00029309999999999997, + "loss": 2.1388, + "step": 992 + }, + { + "epoch": 1.5888, + "grad_norm": 1.1508042812347412, + "learning_rate": 0.0002934, + "loss": 2.1403, + "step": 993 + }, + { + "epoch": 1.5904, + "grad_norm": 2.786349058151245, + "learning_rate": 0.0002937, + "loss": 2.0843, + "step": 994 + }, + { + "epoch": 1.592, + "grad_norm": 4.274806499481201, + "learning_rate": 0.000294, + "loss": 2.2627, + "step": 995 + }, + { + "epoch": 1.5936, + "grad_norm": 4.96270751953125, + "learning_rate": 0.00029429999999999994, + "loss": 2.2727, + "step": 996 + }, + { + "epoch": 1.5952, + "grad_norm": 5.281583786010742, + "learning_rate": 0.00029459999999999995, + "loss": 2.2626, + "step": 997 + }, + { + "epoch": 1.5968, + "grad_norm": 6.537198066711426, + "learning_rate": 0.00029489999999999996, + "loss": 2.258, + "step": 998 + }, + { + "epoch": 1.5984, + "grad_norm": 7.0249223709106445, + "learning_rate": 0.00029519999999999997, + "loss": 2.3156, + "step": 999 + }, + { + "epoch": 1.6, + "grad_norm": 5.909872055053711, + "learning_rate": 0.00029549999999999997, + "loss": 2.3959, + "step": 1000 + }, + { + "epoch": 1.6016, + "grad_norm": 7.637903213500977, + "learning_rate": 0.0002958, + "loss": 2.4885, + "step": 1001 + }, + { + "epoch": 1.6032, + "grad_norm": 6.988182067871094, + "learning_rate": 0.0002961, + "loss": 2.3817, + "step": 1002 + }, + { + "epoch": 1.6048, + "grad_norm": 6.992365837097168, + "learning_rate": 0.0002964, + "loss": 2.4426, + "step": 1003 + }, + { + "epoch": 1.6064, + "grad_norm": 5.621988296508789, + "learning_rate": 0.00029669999999999995, + "loss": 2.397, + "step": 1004 + }, + { + "epoch": 1.608, + "grad_norm": 3.53421688079834, + "learning_rate": 0.00029699999999999996, + "loss": 2.403, + "step": 1005 + }, + { + "epoch": 1.6096, + "grad_norm": 2.6446614265441895, + "learning_rate": 0.00029729999999999996, + "loss": 2.2512, + "step": 1006 + }, + { + "epoch": 1.6112, + "grad_norm": 2.235408306121826, + "learning_rate": 0.00029759999999999997, + "loss": 2.0068, + "step": 1007 + }, + { + "epoch": 1.6128, + "grad_norm": 1.4316028356552124, + "learning_rate": 0.0002979, + "loss": 2.1133, + "step": 1008 + }, + { + "epoch": 1.6143999999999998, + "grad_norm": 1.0931628942489624, + "learning_rate": 0.0002982, + "loss": 2.1428, + "step": 1009 + }, + { + "epoch": 1.616, + "grad_norm": 1.0988737344741821, + "learning_rate": 0.0002985, + "loss": 2.212, + "step": 1010 + }, + { + "epoch": 1.6176, + "grad_norm": 0.8665049076080322, + "learning_rate": 0.0002988, + "loss": 2.2145, + "step": 1011 + }, + { + "epoch": 1.6192, + "grad_norm": 0.6521317362785339, + "learning_rate": 0.00029909999999999995, + "loss": 2.2136, + "step": 1012 + }, + { + "epoch": 1.6208, + "grad_norm": 1.2333879470825195, + "learning_rate": 0.00029939999999999996, + "loss": 2.2025, + "step": 1013 + }, + { + "epoch": 1.6223999999999998, + "grad_norm": 1.2490397691726685, + "learning_rate": 0.00029969999999999997, + "loss": 2.3032, + "step": 1014 + }, + { + "epoch": 1.624, + "grad_norm": 1.9451225996017456, + "learning_rate": 0.0003, + "loss": 2.6045, + "step": 1015 + }, + { + "epoch": 1.6256, + "grad_norm": 3.870406150817871, + "learning_rate": 0.0003002999999999999, + "loss": 2.2674, + "step": 1016 + }, + { + "epoch": 1.6272, + "grad_norm": 3.619330644607544, + "learning_rate": 0.0003006, + "loss": 2.2744, + "step": 1017 + }, + { + "epoch": 1.6288, + "grad_norm": 4.24730920791626, + "learning_rate": 0.00030089999999999994, + "loss": 2.3574, + "step": 1018 + }, + { + "epoch": 1.6303999999999998, + "grad_norm": 4.763134002685547, + "learning_rate": 0.00030119999999999995, + "loss": 2.4282, + "step": 1019 + }, + { + "epoch": 1.6320000000000001, + "grad_norm": 4.33074426651001, + "learning_rate": 0.00030149999999999996, + "loss": 2.6336, + "step": 1020 + }, + { + "epoch": 1.6336, + "grad_norm": 2.6049580574035645, + "learning_rate": 0.00030179999999999996, + "loss": 2.0806, + "step": 1021 + }, + { + "epoch": 1.6352, + "grad_norm": 6.9189839363098145, + "learning_rate": 0.0003020999999999999, + "loss": 2.5478, + "step": 1022 + }, + { + "epoch": 1.6368, + "grad_norm": 1.222205638885498, + "learning_rate": 0.0003024, + "loss": 2.7747, + "step": 1023 + }, + { + "epoch": 1.6383999999999999, + "grad_norm": 1.9943503141403198, + "learning_rate": 0.00030269999999999993, + "loss": 2.5852, + "step": 1024 + }, + { + "epoch": 1.6400000000000001, + "grad_norm": 0.903025209903717, + "learning_rate": 0.000303, + "loss": 2.8456, + "step": 1025 + }, + { + "epoch": 1.6416, + "grad_norm": 17.016794204711914, + "learning_rate": 0.00030329999999999995, + "loss": 3.5966, + "step": 1026 + }, + { + "epoch": 1.6432, + "grad_norm": 3.820300340652466, + "learning_rate": 0.00030359999999999995, + "loss": 2.6173, + "step": 1027 + }, + { + "epoch": 1.6448, + "grad_norm": 6.402599811553955, + "learning_rate": 0.00030389999999999996, + "loss": 2.834, + "step": 1028 + }, + { + "epoch": 1.6463999999999999, + "grad_norm": 10.509306907653809, + "learning_rate": 0.00030419999999999997, + "loss": 3.2254, + "step": 1029 + }, + { + "epoch": 1.6480000000000001, + "grad_norm": 2.164959192276001, + "learning_rate": 0.0003044999999999999, + "loss": 2.3757, + "step": 1030 + }, + { + "epoch": 1.6496, + "grad_norm": 1.8674250841140747, + "learning_rate": 0.0003048, + "loss": 2.371, + "step": 1031 + }, + { + "epoch": 1.6512, + "grad_norm": 4.218185901641846, + "learning_rate": 0.00030509999999999994, + "loss": 2.5602, + "step": 1032 + }, + { + "epoch": 1.6528, + "grad_norm": 4.413148880004883, + "learning_rate": 0.0003054, + "loss": 2.4103, + "step": 1033 + }, + { + "epoch": 1.6543999999999999, + "grad_norm": 5.600552558898926, + "learning_rate": 0.00030569999999999995, + "loss": 2.2613, + "step": 1034 + }, + { + "epoch": 1.6560000000000001, + "grad_norm": 4.799594402313232, + "learning_rate": 0.00030599999999999996, + "loss": 2.3748, + "step": 1035 + }, + { + "epoch": 1.6576, + "grad_norm": 1.1949158906936646, + "learning_rate": 0.00030629999999999996, + "loss": 2.3916, + "step": 1036 + }, + { + "epoch": 1.6592, + "grad_norm": 4.209189414978027, + "learning_rate": 0.00030659999999999997, + "loss": 2.3753, + "step": 1037 + }, + { + "epoch": 1.6608, + "grad_norm": 4.924483299255371, + "learning_rate": 0.0003068999999999999, + "loss": 2.2942, + "step": 1038 + }, + { + "epoch": 1.6623999999999999, + "grad_norm": 5.148475170135498, + "learning_rate": 0.0003072, + "loss": 2.2493, + "step": 1039 + }, + { + "epoch": 1.6640000000000001, + "grad_norm": 4.989164352416992, + "learning_rate": 0.00030749999999999994, + "loss": 2.1704, + "step": 1040 + }, + { + "epoch": 1.6656, + "grad_norm": 4.440720558166504, + "learning_rate": 0.0003078, + "loss": 2.4185, + "step": 1041 + }, + { + "epoch": 1.6672, + "grad_norm": 3.280564308166504, + "learning_rate": 0.00030809999999999995, + "loss": 2.1599, + "step": 1042 + }, + { + "epoch": 1.6688, + "grad_norm": 1.5916945934295654, + "learning_rate": 0.00030839999999999996, + "loss": 2.1695, + "step": 1043 + }, + { + "epoch": 1.6703999999999999, + "grad_norm": 1.1902886629104614, + "learning_rate": 0.00030869999999999997, + "loss": 2.1445, + "step": 1044 + }, + { + "epoch": 1.6720000000000002, + "grad_norm": 1.2258431911468506, + "learning_rate": 0.000309, + "loss": 2.1879, + "step": 1045 + }, + { + "epoch": 1.6736, + "grad_norm": 2.241992473602295, + "learning_rate": 0.00030929999999999993, + "loss": 1.9961, + "step": 1046 + }, + { + "epoch": 1.6752, + "grad_norm": 3.9534943103790283, + "learning_rate": 0.0003096, + "loss": 2.074, + "step": 1047 + }, + { + "epoch": 1.6768, + "grad_norm": 5.947170257568359, + "learning_rate": 0.00030989999999999994, + "loss": 2.2004, + "step": 1048 + }, + { + "epoch": 1.6784, + "grad_norm": 7.540464878082275, + "learning_rate": 0.0003102, + "loss": 2.4244, + "step": 1049 + }, + { + "epoch": 1.6800000000000002, + "grad_norm": 4.6932172775268555, + "learning_rate": 0.00031049999999999996, + "loss": 2.1152, + "step": 1050 + }, + { + "epoch": 1.6816, + "grad_norm": 6.5101318359375, + "learning_rate": 0.00031079999999999997, + "loss": 2.3487, + "step": 1051 + }, + { + "epoch": 1.6832, + "grad_norm": 5.6864013671875, + "learning_rate": 0.00031109999999999997, + "loss": 2.4663, + "step": 1052 + }, + { + "epoch": 1.6848, + "grad_norm": 5.695480823516846, + "learning_rate": 0.0003114, + "loss": 2.2667, + "step": 1053 + }, + { + "epoch": 1.6864, + "grad_norm": 5.523344993591309, + "learning_rate": 0.00031169999999999993, + "loss": 2.199, + "step": 1054 + }, + { + "epoch": 1.688, + "grad_norm": 4.021401405334473, + "learning_rate": 0.000312, + "loss": 2.2399, + "step": 1055 + }, + { + "epoch": 1.6896, + "grad_norm": 3.751222848892212, + "learning_rate": 0.00031229999999999995, + "loss": 2.235, + "step": 1056 + }, + { + "epoch": 1.6912, + "grad_norm": 2.249674081802368, + "learning_rate": 0.0003126, + "loss": 2.1857, + "step": 1057 + }, + { + "epoch": 1.6928, + "grad_norm": 2.001965284347534, + "learning_rate": 0.00031289999999999996, + "loss": 2.1195, + "step": 1058 + }, + { + "epoch": 1.6944, + "grad_norm": 1.0671782493591309, + "learning_rate": 0.00031319999999999997, + "loss": 2.1389, + "step": 1059 + }, + { + "epoch": 1.696, + "grad_norm": 0.9789913296699524, + "learning_rate": 0.0003135, + "loss": 2.4445, + "step": 1060 + }, + { + "epoch": 1.6976, + "grad_norm": 1.2904689311981201, + "learning_rate": 0.0003138, + "loss": 2.1766, + "step": 1061 + }, + { + "epoch": 1.6992, + "grad_norm": 0.7292619943618774, + "learning_rate": 0.00031409999999999994, + "loss": 2.221, + "step": 1062 + }, + { + "epoch": 1.7008, + "grad_norm": 0.7377645373344421, + "learning_rate": 0.0003144, + "loss": 2.1054, + "step": 1063 + }, + { + "epoch": 1.7024, + "grad_norm": 1.674381136894226, + "learning_rate": 0.00031469999999999995, + "loss": 2.362, + "step": 1064 + }, + { + "epoch": 1.704, + "grad_norm": 6.733413219451904, + "learning_rate": 0.00031499999999999996, + "loss": 2.634, + "step": 1065 + }, + { + "epoch": 1.7056, + "grad_norm": 3.1459693908691406, + "learning_rate": 0.00031529999999999997, + "loss": 2.3274, + "step": 1066 + }, + { + "epoch": 1.7072, + "grad_norm": 3.5573384761810303, + "learning_rate": 0.0003156, + "loss": 2.7282, + "step": 1067 + }, + { + "epoch": 1.7088, + "grad_norm": 3.3276679515838623, + "learning_rate": 0.0003158999999999999, + "loss": 2.4335, + "step": 1068 + }, + { + "epoch": 1.7104, + "grad_norm": 3.3725104331970215, + "learning_rate": 0.0003162, + "loss": 2.6206, + "step": 1069 + }, + { + "epoch": 1.712, + "grad_norm": 1.2234220504760742, + "learning_rate": 0.00031649999999999994, + "loss": 2.4624, + "step": 1070 + }, + { + "epoch": 1.7136, + "grad_norm": 0.8122461438179016, + "learning_rate": 0.0003168, + "loss": 2.5879, + "step": 1071 + }, + { + "epoch": 1.7151999999999998, + "grad_norm": 0.7520931363105774, + "learning_rate": 0.00031709999999999996, + "loss": 2.4474, + "step": 1072 + }, + { + "epoch": 1.7168, + "grad_norm": 2.023834228515625, + "learning_rate": 0.00031739999999999996, + "loss": 2.873, + "step": 1073 + }, + { + "epoch": 1.7184, + "grad_norm": 1.4827693700790405, + "learning_rate": 0.00031769999999999997, + "loss": 2.5504, + "step": 1074 + }, + { + "epoch": 1.72, + "grad_norm": NaN, + "learning_rate": 0.00031769999999999997, + "loss": 2.2876, + "step": 1075 + }, + { + "epoch": 1.7216, + "grad_norm": 16.413217544555664, + "learning_rate": 0.000318, + "loss": 3.5108, + "step": 1076 + }, + { + "epoch": 1.7231999999999998, + "grad_norm": 8.193089485168457, + "learning_rate": 0.00031829999999999993, + "loss": 3.0809, + "step": 1077 + }, + { + "epoch": 1.7248, + "grad_norm": 4.528853416442871, + "learning_rate": 0.0003186, + "loss": 2.8905, + "step": 1078 + }, + { + "epoch": 1.7264, + "grad_norm": 2.9132778644561768, + "learning_rate": 0.00031889999999999995, + "loss": 2.8106, + "step": 1079 + }, + { + "epoch": 1.728, + "grad_norm": 6.130316257476807, + "learning_rate": 0.0003192, + "loss": 3.121, + "step": 1080 + }, + { + "epoch": 1.7296, + "grad_norm": 1.5412960052490234, + "learning_rate": 0.00031949999999999996, + "loss": 2.7423, + "step": 1081 + }, + { + "epoch": 1.7311999999999999, + "grad_norm": 2.8891286849975586, + "learning_rate": 0.00031979999999999997, + "loss": 2.6641, + "step": 1082 + }, + { + "epoch": 1.7328000000000001, + "grad_norm": 1.3349493741989136, + "learning_rate": 0.0003201, + "loss": 2.7299, + "step": 1083 + }, + { + "epoch": 1.7344, + "grad_norm": 2.5229907035827637, + "learning_rate": 0.0003204, + "loss": 2.7078, + "step": 1084 + }, + { + "epoch": 1.736, + "grad_norm": 3.7961184978485107, + "learning_rate": 0.00032069999999999993, + "loss": 2.4799, + "step": 1085 + }, + { + "epoch": 1.7376, + "grad_norm": 3.6919217109680176, + "learning_rate": 0.000321, + "loss": 2.4731, + "step": 1086 + }, + { + "epoch": 1.7391999999999999, + "grad_norm": 3.6608409881591797, + "learning_rate": 0.00032129999999999995, + "loss": 2.6901, + "step": 1087 + }, + { + "epoch": 1.7408000000000001, + "grad_norm": 2.5811784267425537, + "learning_rate": 0.0003216, + "loss": 2.4138, + "step": 1088 + }, + { + "epoch": 1.7424, + "grad_norm": 1.4929311275482178, + "learning_rate": 0.00032189999999999996, + "loss": 2.3692, + "step": 1089 + }, + { + "epoch": 1.744, + "grad_norm": 1.622512698173523, + "learning_rate": 0.00032219999999999997, + "loss": 2.4005, + "step": 1090 + }, + { + "epoch": 1.7456, + "grad_norm": 1.0638054609298706, + "learning_rate": 0.0003225, + "loss": 2.4661, + "step": 1091 + }, + { + "epoch": 1.7471999999999999, + "grad_norm": 1.3987395763397217, + "learning_rate": 0.0003228, + "loss": 2.4535, + "step": 1092 + }, + { + "epoch": 1.7488000000000001, + "grad_norm": 3.8728277683258057, + "learning_rate": 0.00032309999999999994, + "loss": 2.5941, + "step": 1093 + }, + { + "epoch": 1.7504, + "grad_norm": 2.725276470184326, + "learning_rate": 0.0003234, + "loss": 2.553, + "step": 1094 + }, + { + "epoch": 1.752, + "grad_norm": 2.4425482749938965, + "learning_rate": 0.00032369999999999995, + "loss": 2.2451, + "step": 1095 + }, + { + "epoch": 1.7536, + "grad_norm": 2.31581711769104, + "learning_rate": 0.000324, + "loss": 2.4025, + "step": 1096 + }, + { + "epoch": 1.7551999999999999, + "grad_norm": 6.327169418334961, + "learning_rate": 0.00032429999999999997, + "loss": 2.6523, + "step": 1097 + }, + { + "epoch": 1.7568000000000001, + "grad_norm": 6.215856552124023, + "learning_rate": 0.0003246, + "loss": 2.7057, + "step": 1098 + }, + { + "epoch": 1.7584, + "grad_norm": 0.9872040748596191, + "learning_rate": 0.0003249, + "loss": 2.4095, + "step": 1099 + }, + { + "epoch": 1.76, + "grad_norm": 1.473762035369873, + "learning_rate": 0.0003252, + "loss": 2.3173, + "step": 1100 + }, + { + "epoch": 1.7616, + "grad_norm": 1.889335036277771, + "learning_rate": 0.00032549999999999994, + "loss": 2.4215, + "step": 1101 + }, + { + "epoch": 1.7631999999999999, + "grad_norm": 2.3463101387023926, + "learning_rate": 0.0003258, + "loss": 2.2987, + "step": 1102 + }, + { + "epoch": 1.7648000000000001, + "grad_norm": 1.228926420211792, + "learning_rate": 0.00032609999999999996, + "loss": 2.2642, + "step": 1103 + }, + { + "epoch": 1.7664, + "grad_norm": 0.8833368420600891, + "learning_rate": 0.0003264, + "loss": 2.2548, + "step": 1104 + }, + { + "epoch": 1.768, + "grad_norm": 0.7233026027679443, + "learning_rate": 0.00032669999999999997, + "loss": 2.4638, + "step": 1105 + }, + { + "epoch": 1.7696, + "grad_norm": 0.7442961931228638, + "learning_rate": 0.000327, + "loss": 2.3391, + "step": 1106 + }, + { + "epoch": 1.7711999999999999, + "grad_norm": 0.688629150390625, + "learning_rate": 0.0003273, + "loss": 2.1627, + "step": 1107 + }, + { + "epoch": 1.7728000000000002, + "grad_norm": 2.3164734840393066, + "learning_rate": 0.0003276, + "loss": 2.6653, + "step": 1108 + }, + { + "epoch": 1.7744, + "grad_norm": 0.7344738245010376, + "learning_rate": 0.00032789999999999995, + "loss": 2.3615, + "step": 1109 + }, + { + "epoch": 1.776, + "grad_norm": 3.150681734085083, + "learning_rate": 0.0003282, + "loss": 2.2903, + "step": 1110 + }, + { + "epoch": 1.7776, + "grad_norm": 3.1520416736602783, + "learning_rate": 0.00032849999999999996, + "loss": 2.3161, + "step": 1111 + }, + { + "epoch": 1.7792, + "grad_norm": 1.8366395235061646, + "learning_rate": 0.0003288, + "loss": 2.3981, + "step": 1112 + }, + { + "epoch": 1.7808000000000002, + "grad_norm": 1.1598656177520752, + "learning_rate": 0.0003291, + "loss": 2.279, + "step": 1113 + }, + { + "epoch": 1.7824, + "grad_norm": 1.057368278503418, + "learning_rate": 0.0003294, + "loss": 2.4606, + "step": 1114 + }, + { + "epoch": 1.784, + "grad_norm": 0.8108821511268616, + "learning_rate": 0.0003297, + "loss": 2.3197, + "step": 1115 + }, + { + "epoch": 1.7856, + "grad_norm": 0.7972817420959473, + "learning_rate": 0.00033, + "loss": 2.3071, + "step": 1116 + }, + { + "epoch": 1.7872, + "grad_norm": 0.7329958081245422, + "learning_rate": 0.00033029999999999995, + "loss": 2.36, + "step": 1117 + }, + { + "epoch": 1.7888, + "grad_norm": 0.9914326667785645, + "learning_rate": 0.0003306, + "loss": 2.494, + "step": 1118 + }, + { + "epoch": 1.7904, + "grad_norm": 1.2104322910308838, + "learning_rate": 0.00033089999999999997, + "loss": 2.1681, + "step": 1119 + }, + { + "epoch": 1.792, + "grad_norm": 2.939289093017578, + "learning_rate": 0.0003312, + "loss": 2.707, + "step": 1120 + }, + { + "epoch": 1.7936, + "grad_norm": 1.4465129375457764, + "learning_rate": 0.0003315, + "loss": 2.6272, + "step": 1121 + }, + { + "epoch": 1.7952, + "grad_norm": 0.9196242094039917, + "learning_rate": 0.0003318, + "loss": 2.7562, + "step": 1122 + }, + { + "epoch": 1.7968, + "grad_norm": 3.128715991973877, + "learning_rate": 0.00033209999999999994, + "loss": 2.7036, + "step": 1123 + }, + { + "epoch": 1.7984, + "grad_norm": 1.085696816444397, + "learning_rate": 0.0003324, + "loss": 2.412, + "step": 1124 + }, + { + "epoch": 1.8, + "grad_norm": 2.5047450065612793, + "learning_rate": 0.00033269999999999996, + "loss": 2.7249, + "step": 1125 + }, + { + "epoch": 1.8016, + "grad_norm": 4.3478922843933105, + "learning_rate": 0.000333, + "loss": 2.7462, + "step": 1126 + }, + { + "epoch": 1.8032, + "grad_norm": 3.4352104663848877, + "learning_rate": 0.00033329999999999997, + "loss": 2.7354, + "step": 1127 + }, + { + "epoch": 1.8048, + "grad_norm": 8.34579849243164, + "learning_rate": 0.0003336, + "loss": 2.9036, + "step": 1128 + }, + { + "epoch": 1.8064, + "grad_norm": 1.7377263307571411, + "learning_rate": 0.0003339, + "loss": 2.6318, + "step": 1129 + }, + { + "epoch": 1.808, + "grad_norm": 0.8710009455680847, + "learning_rate": 0.0003342, + "loss": 2.4611, + "step": 1130 + }, + { + "epoch": 1.8096, + "grad_norm": 0.7961373329162598, + "learning_rate": 0.00033449999999999994, + "loss": 2.6149, + "step": 1131 + }, + { + "epoch": 1.8112, + "grad_norm": 1.2493680715560913, + "learning_rate": 0.0003348, + "loss": 2.4834, + "step": 1132 + }, + { + "epoch": 1.8128, + "grad_norm": 2.2496654987335205, + "learning_rate": 0.00033509999999999996, + "loss": 2.4858, + "step": 1133 + }, + { + "epoch": 1.8144, + "grad_norm": 1.7312769889831543, + "learning_rate": 0.0003354, + "loss": 2.5253, + "step": 1134 + }, + { + "epoch": 1.8159999999999998, + "grad_norm": 1.8472027778625488, + "learning_rate": 0.0003357, + "loss": 2.8019, + "step": 1135 + }, + { + "epoch": 1.8176, + "grad_norm": 3.3026609420776367, + "learning_rate": 0.000336, + "loss": 2.383, + "step": 1136 + }, + { + "epoch": 1.8192, + "grad_norm": 2.8907511234283447, + "learning_rate": 0.0003363, + "loss": 2.4224, + "step": 1137 + }, + { + "epoch": 1.8208, + "grad_norm": 2.530062437057495, + "learning_rate": 0.0003366, + "loss": 2.2473, + "step": 1138 + }, + { + "epoch": 1.8224, + "grad_norm": 1.1952489614486694, + "learning_rate": 0.00033689999999999995, + "loss": 2.3729, + "step": 1139 + }, + { + "epoch": 1.8239999999999998, + "grad_norm": 1.2493127584457397, + "learning_rate": 0.0003372, + "loss": 2.2322, + "step": 1140 + }, + { + "epoch": 1.8256000000000001, + "grad_norm": 1.3810757398605347, + "learning_rate": 0.00033749999999999996, + "loss": 2.1936, + "step": 1141 + }, + { + "epoch": 1.8272, + "grad_norm": 0.6716127395629883, + "learning_rate": 0.0003377999999999999, + "loss": 2.1006, + "step": 1142 + }, + { + "epoch": 1.8288, + "grad_norm": 4.641638278961182, + "learning_rate": 0.0003381, + "loss": 2.5418, + "step": 1143 + }, + { + "epoch": 1.8304, + "grad_norm": 0.7420514225959778, + "learning_rate": 0.00033839999999999993, + "loss": 2.1303, + "step": 1144 + }, + { + "epoch": 1.8319999999999999, + "grad_norm": 0.6666918992996216, + "learning_rate": 0.0003387, + "loss": 2.2456, + "step": 1145 + }, + { + "epoch": 1.8336000000000001, + "grad_norm": 0.5805254578590393, + "learning_rate": 0.00033899999999999995, + "loss": 2.2268, + "step": 1146 + }, + { + "epoch": 1.8352, + "grad_norm": 0.6321543455123901, + "learning_rate": 0.00033929999999999995, + "loss": 2.2048, + "step": 1147 + }, + { + "epoch": 1.8368, + "grad_norm": 3.640864849090576, + "learning_rate": 0.00033959999999999996, + "loss": 2.5881, + "step": 1148 + }, + { + "epoch": 1.8384, + "grad_norm": 2.1340978145599365, + "learning_rate": 0.00033989999999999997, + "loss": 2.1501, + "step": 1149 + }, + { + "epoch": 1.8399999999999999, + "grad_norm": 2.5168700218200684, + "learning_rate": 0.0003401999999999999, + "loss": 2.4225, + "step": 1150 + }, + { + "epoch": 1.8416000000000001, + "grad_norm": 1.6082960367202759, + "learning_rate": 0.0003405, + "loss": 2.172, + "step": 1151 + }, + { + "epoch": 1.8432, + "grad_norm": 2.326335906982422, + "learning_rate": 0.00034079999999999994, + "loss": 2.1434, + "step": 1152 + }, + { + "epoch": 1.8448, + "grad_norm": 0.920198380947113, + "learning_rate": 0.0003411, + "loss": 2.1625, + "step": 1153 + }, + { + "epoch": 1.8464, + "grad_norm": 1.2494739294052124, + "learning_rate": 0.00034139999999999995, + "loss": 2.1452, + "step": 1154 + }, + { + "epoch": 1.8479999999999999, + "grad_norm": 1.2549406290054321, + "learning_rate": 0.00034169999999999996, + "loss": 2.171, + "step": 1155 + }, + { + "epoch": 1.8496000000000001, + "grad_norm": 1.5478543043136597, + "learning_rate": 0.00034199999999999996, + "loss": 2.1439, + "step": 1156 + }, + { + "epoch": 1.8512, + "grad_norm": 1.768579125404358, + "learning_rate": 0.00034229999999999997, + "loss": 2.056, + "step": 1157 + }, + { + "epoch": 1.8528, + "grad_norm": 0.6443532109260559, + "learning_rate": 0.0003425999999999999, + "loss": 2.3397, + "step": 1158 + }, + { + "epoch": 1.8544, + "grad_norm": 0.5668714046478271, + "learning_rate": 0.0003429, + "loss": 2.2165, + "step": 1159 + }, + { + "epoch": 1.8559999999999999, + "grad_norm": 1.3049060106277466, + "learning_rate": 0.00034319999999999994, + "loss": 2.33, + "step": 1160 + }, + { + "epoch": 1.8576000000000001, + "grad_norm": 0.6785088777542114, + "learning_rate": 0.0003435, + "loss": 2.1462, + "step": 1161 + }, + { + "epoch": 1.8592, + "grad_norm": 1.5797127485275269, + "learning_rate": 0.00034379999999999995, + "loss": 2.5508, + "step": 1162 + }, + { + "epoch": 1.8608, + "grad_norm": 1.1962186098098755, + "learning_rate": 0.00034409999999999996, + "loss": 2.3856, + "step": 1163 + }, + { + "epoch": 1.8624, + "grad_norm": 0.6989690661430359, + "learning_rate": 0.00034439999999999997, + "loss": 2.0005, + "step": 1164 + }, + { + "epoch": 1.8639999999999999, + "grad_norm": 0.8127622604370117, + "learning_rate": 0.0003447, + "loss": 2.3053, + "step": 1165 + }, + { + "epoch": 1.8656000000000001, + "grad_norm": 0.8483240604400635, + "learning_rate": 0.00034499999999999993, + "loss": 2.1876, + "step": 1166 + }, + { + "epoch": 1.8672, + "grad_norm": 0.9150612950325012, + "learning_rate": 0.0003453, + "loss": 2.5235, + "step": 1167 + }, + { + "epoch": 1.8688, + "grad_norm": 0.6177669167518616, + "learning_rate": 0.00034559999999999994, + "loss": 2.2692, + "step": 1168 + }, + { + "epoch": 1.8704, + "grad_norm": 0.7125779986381531, + "learning_rate": 0.00034589999999999995, + "loss": 2.3864, + "step": 1169 + }, + { + "epoch": 1.8719999999999999, + "grad_norm": 0.5423337817192078, + "learning_rate": 0.00034619999999999996, + "loss": 2.3914, + "step": 1170 + }, + { + "epoch": 1.8736000000000002, + "grad_norm": 0.8027876019477844, + "learning_rate": 0.00034649999999999997, + "loss": 2.2606, + "step": 1171 + }, + { + "epoch": 1.8752, + "grad_norm": 3.676344156265259, + "learning_rate": 0.0003467999999999999, + "loss": 2.435, + "step": 1172 + }, + { + "epoch": 1.8768, + "grad_norm": 0.8745720386505127, + "learning_rate": 0.0003471, + "loss": 2.4396, + "step": 1173 + }, + { + "epoch": 1.8784, + "grad_norm": 2.438420534133911, + "learning_rate": 0.00034739999999999993, + "loss": 2.505, + "step": 1174 + }, + { + "epoch": 1.88, + "grad_norm": NaN, + "learning_rate": 0.00034739999999999993, + "loss": 3.2354, + "step": 1175 + }, + { + "epoch": 1.8816000000000002, + "grad_norm": 13.595325469970703, + "learning_rate": 0.0003477, + "loss": 3.0164, + "step": 1176 + }, + { + "epoch": 1.8832, + "grad_norm": 4.952884674072266, + "learning_rate": 0.00034799999999999995, + "loss": 2.7017, + "step": 1177 + }, + { + "epoch": 1.8848, + "grad_norm": 3.7639901638031006, + "learning_rate": 0.00034829999999999996, + "loss": 2.4942, + "step": 1178 + }, + { + "epoch": 1.8864, + "grad_norm": 5.62662935256958, + "learning_rate": 0.00034859999999999996, + "loss": 2.7815, + "step": 1179 + }, + { + "epoch": 1.888, + "grad_norm": 1.9343209266662598, + "learning_rate": 0.00034889999999999997, + "loss": 2.5347, + "step": 1180 + }, + { + "epoch": 1.8896, + "grad_norm": 0.9629626870155334, + "learning_rate": 0.0003491999999999999, + "loss": 2.4285, + "step": 1181 + }, + { + "epoch": 1.8912, + "grad_norm": 3.4893319606781006, + "learning_rate": 0.0003495, + "loss": 2.4941, + "step": 1182 + }, + { + "epoch": 1.8928, + "grad_norm": 5.90111780166626, + "learning_rate": 0.00034979999999999994, + "loss": 2.3566, + "step": 1183 + }, + { + "epoch": 1.8944, + "grad_norm": 6.313168525695801, + "learning_rate": 0.0003501, + "loss": 2.3439, + "step": 1184 + }, + { + "epoch": 1.896, + "grad_norm": 5.150593280792236, + "learning_rate": 0.00035039999999999995, + "loss": 2.5063, + "step": 1185 + }, + { + "epoch": 1.8976, + "grad_norm": 5.239500522613525, + "learning_rate": 0.00035069999999999996, + "loss": 2.1988, + "step": 1186 + }, + { + "epoch": 1.8992, + "grad_norm": 5.032492160797119, + "learning_rate": 0.00035099999999999997, + "loss": 2.47, + "step": 1187 + }, + { + "epoch": 1.9008, + "grad_norm": 4.160557270050049, + "learning_rate": 0.0003513, + "loss": 2.477, + "step": 1188 + }, + { + "epoch": 1.9024, + "grad_norm": 2.1352810859680176, + "learning_rate": 0.0003515999999999999, + "loss": 2.2105, + "step": 1189 + }, + { + "epoch": 1.904, + "grad_norm": 2.577263832092285, + "learning_rate": 0.0003519, + "loss": 2.1445, + "step": 1190 + }, + { + "epoch": 1.9056, + "grad_norm": 1.5833125114440918, + "learning_rate": 0.00035219999999999994, + "loss": 2.0875, + "step": 1191 + }, + { + "epoch": 1.9072, + "grad_norm": 0.9110488295555115, + "learning_rate": 0.0003525, + "loss": 2.357, + "step": 1192 + }, + { + "epoch": 1.9088, + "grad_norm": 2.1670477390289307, + "learning_rate": 0.00035279999999999996, + "loss": 2.2269, + "step": 1193 + }, + { + "epoch": 1.9104, + "grad_norm": 2.896484136581421, + "learning_rate": 0.00035309999999999996, + "loss": 2.277, + "step": 1194 + }, + { + "epoch": 1.912, + "grad_norm": 2.396368980407715, + "learning_rate": 0.00035339999999999997, + "loss": 2.2654, + "step": 1195 + }, + { + "epoch": 1.9136, + "grad_norm": 2.8805601596832275, + "learning_rate": 0.0003537, + "loss": 2.1887, + "step": 1196 + }, + { + "epoch": 1.9152, + "grad_norm": 1.5967847108840942, + "learning_rate": 0.00035399999999999993, + "loss": 2.3864, + "step": 1197 + }, + { + "epoch": 1.9167999999999998, + "grad_norm": 3.0907437801361084, + "learning_rate": 0.0003543, + "loss": 2.0803, + "step": 1198 + }, + { + "epoch": 1.9184, + "grad_norm": 3.411078691482544, + "learning_rate": 0.00035459999999999995, + "loss": 2.4314, + "step": 1199 + }, + { + "epoch": 1.92, + "grad_norm": 3.378351926803589, + "learning_rate": 0.0003549, + "loss": 2.2706, + "step": 1200 + }, + { + "epoch": 1.9216, + "grad_norm": 0.8546525835990906, + "learning_rate": 0.00035519999999999996, + "loss": 2.0895, + "step": 1201 + }, + { + "epoch": 1.9232, + "grad_norm": 1.635798454284668, + "learning_rate": 0.00035549999999999997, + "loss": 2.1679, + "step": 1202 + }, + { + "epoch": 1.9247999999999998, + "grad_norm": 1.9297326803207397, + "learning_rate": 0.0003558, + "loss": 2.2493, + "step": 1203 + }, + { + "epoch": 1.9264000000000001, + "grad_norm": 2.8344545364379883, + "learning_rate": 0.0003561, + "loss": 2.0785, + "step": 1204 + }, + { + "epoch": 1.928, + "grad_norm": 2.0780208110809326, + "learning_rate": 0.00035639999999999994, + "loss": 2.3996, + "step": 1205 + }, + { + "epoch": 1.9296, + "grad_norm": 1.769015908241272, + "learning_rate": 0.0003567, + "loss": 1.9828, + "step": 1206 + }, + { + "epoch": 1.9312, + "grad_norm": 1.2962875366210938, + "learning_rate": 0.00035699999999999995, + "loss": 2.0324, + "step": 1207 + }, + { + "epoch": 1.9327999999999999, + "grad_norm": 1.9676580429077148, + "learning_rate": 0.0003573, + "loss": 2.228, + "step": 1208 + }, + { + "epoch": 1.9344000000000001, + "grad_norm": 1.318317174911499, + "learning_rate": 0.00035759999999999996, + "loss": 2.0663, + "step": 1209 + }, + { + "epoch": 1.936, + "grad_norm": 1.430823564529419, + "learning_rate": 0.00035789999999999997, + "loss": 2.2146, + "step": 1210 + }, + { + "epoch": 1.9376, + "grad_norm": 1.1980717182159424, + "learning_rate": 0.0003582, + "loss": 2.2954, + "step": 1211 + }, + { + "epoch": 1.9392, + "grad_norm": 0.9021552801132202, + "learning_rate": 0.0003585, + "loss": 2.3383, + "step": 1212 + }, + { + "epoch": 1.9407999999999999, + "grad_norm": 0.9370974898338318, + "learning_rate": 0.00035879999999999994, + "loss": 2.3008, + "step": 1213 + }, + { + "epoch": 1.9424000000000001, + "grad_norm": 0.6908053755760193, + "learning_rate": 0.0003591, + "loss": 2.1052, + "step": 1214 + }, + { + "epoch": 1.944, + "grad_norm": 1.6524698734283447, + "learning_rate": 0.00035939999999999995, + "loss": 2.4055, + "step": 1215 + }, + { + "epoch": 1.9456, + "grad_norm": 2.5479085445404053, + "learning_rate": 0.00035969999999999996, + "loss": 2.4952, + "step": 1216 + }, + { + "epoch": 1.9472, + "grad_norm": 0.8158350586891174, + "learning_rate": 0.00035999999999999997, + "loss": 2.0211, + "step": 1217 + }, + { + "epoch": 1.9487999999999999, + "grad_norm": 1.3673202991485596, + "learning_rate": 0.0003603, + "loss": 2.3557, + "step": 1218 + }, + { + "epoch": 1.9504000000000001, + "grad_norm": 2.5113394260406494, + "learning_rate": 0.00036059999999999993, + "loss": 2.1276, + "step": 1219 + }, + { + "epoch": 1.952, + "grad_norm": 1.964048147201538, + "learning_rate": 0.0003609, + "loss": 2.3532, + "step": 1220 + }, + { + "epoch": 1.9536, + "grad_norm": 1.1110976934432983, + "learning_rate": 0.00036119999999999994, + "loss": 2.082, + "step": 1221 + }, + { + "epoch": 1.9552, + "grad_norm": 1.0433621406555176, + "learning_rate": 0.0003615, + "loss": 2.2084, + "step": 1222 + }, + { + "epoch": 1.9567999999999999, + "grad_norm": 0.9254918694496155, + "learning_rate": 0.00036179999999999996, + "loss": 2.5833, + "step": 1223 + }, + { + "epoch": 1.9584000000000001, + "grad_norm": 1.4809350967407227, + "learning_rate": 0.00036209999999999997, + "loss": 2.3376, + "step": 1224 + }, + { + "epoch": 1.96, + "grad_norm": 3.355405569076538, + "learning_rate": 0.00036239999999999997, + "loss": 3.0309, + "step": 1225 + }, + { + "epoch": 1.9616, + "grad_norm": 17.264118194580078, + "learning_rate": 0.0003627, + "loss": 3.2238, + "step": 1226 + }, + { + "epoch": 1.9632, + "grad_norm": 1.5916332006454468, + "learning_rate": 0.00036299999999999993, + "loss": 2.3729, + "step": 1227 + }, + { + "epoch": 1.9647999999999999, + "grad_norm": 2.5827670097351074, + "learning_rate": 0.0003633, + "loss": 2.4696, + "step": 1228 + }, + { + "epoch": 1.9664000000000001, + "grad_norm": 1.2073217630386353, + "learning_rate": 0.00036359999999999995, + "loss": 2.241, + "step": 1229 + }, + { + "epoch": 1.968, + "grad_norm": 1.464860200881958, + "learning_rate": 0.0003639, + "loss": 2.3808, + "step": 1230 + }, + { + "epoch": 1.9696, + "grad_norm": 3.0144131183624268, + "learning_rate": 0.00036419999999999996, + "loss": 2.1668, + "step": 1231 + }, + { + "epoch": 1.9712, + "grad_norm": 3.701151132583618, + "learning_rate": 0.00036449999999999997, + "loss": 2.1334, + "step": 1232 + }, + { + "epoch": 1.9727999999999999, + "grad_norm": 2.9317760467529297, + "learning_rate": 0.0003648, + "loss": 2.0689, + "step": 1233 + }, + { + "epoch": 1.9744000000000002, + "grad_norm": 1.4119415283203125, + "learning_rate": 0.0003651, + "loss": 1.8842, + "step": 1234 + }, + { + "epoch": 1.976, + "grad_norm": 1.5282409191131592, + "learning_rate": 0.00036539999999999994, + "loss": 2.0602, + "step": 1235 + }, + { + "epoch": 1.9776, + "grad_norm": 1.0516180992126465, + "learning_rate": 0.0003657, + "loss": 2.1239, + "step": 1236 + }, + { + "epoch": 1.9792, + "grad_norm": 0.7870113253593445, + "learning_rate": 0.00036599999999999995, + "loss": 2.1571, + "step": 1237 + }, + { + "epoch": 1.9808, + "grad_norm": 2.4591481685638428, + "learning_rate": 0.0003663, + "loss": 1.8778, + "step": 1238 + }, + { + "epoch": 1.9824000000000002, + "grad_norm": 1.552952527999878, + "learning_rate": 0.00036659999999999997, + "loss": 2.1993, + "step": 1239 + }, + { + "epoch": 1.984, + "grad_norm": 4.520159721374512, + "learning_rate": 0.0003669, + "loss": 2.1846, + "step": 1240 + }, + { + "epoch": 1.9856, + "grad_norm": 2.377891778945923, + "learning_rate": 0.0003672, + "loss": 1.9337, + "step": 1241 + }, + { + "epoch": 1.9872, + "grad_norm": 0.9092377424240112, + "learning_rate": 0.0003675, + "loss": 2.068, + "step": 1242 + }, + { + "epoch": 1.9888, + "grad_norm": 2.321300506591797, + "learning_rate": 0.00036779999999999994, + "loss": 2.0708, + "step": 1243 + }, + { + "epoch": 1.9904, + "grad_norm": 0.7131252884864807, + "learning_rate": 0.0003681, + "loss": 2.1614, + "step": 1244 + }, + { + "epoch": 1.992, + "grad_norm": 0.8142818212509155, + "learning_rate": 0.00036839999999999996, + "loss": 2.1221, + "step": 1245 + }, + { + "epoch": 1.9936, + "grad_norm": 1.1560027599334717, + "learning_rate": 0.0003687, + "loss": 2.434, + "step": 1246 + }, + { + "epoch": 1.9952, + "grad_norm": 2.076456069946289, + "learning_rate": 0.00036899999999999997, + "loss": 2.3639, + "step": 1247 + }, + { + "epoch": 1.9968, + "grad_norm": 3.1738593578338623, + "learning_rate": 0.0003693, + "loss": 2.463, + "step": 1248 + }, + { + "epoch": 1.9984, + "grad_norm": 1.0733180046081543, + "learning_rate": 0.0003696, + "loss": 2.7723, + "step": 1249 + }, + { + "epoch": 2.0, + "grad_norm": 4.47441291809082, + "learning_rate": 0.0003699, + "loss": 2.8313, + "step": 1250 + }, + { + "epoch": 2.0016, + "grad_norm": 16.142398834228516, + "learning_rate": 0.00037019999999999995, + "loss": 3.2352, + "step": 1251 + }, + { + "epoch": 2.0032, + "grad_norm": 1.3956218957901, + "learning_rate": 0.0003705, + "loss": 2.2324, + "step": 1252 + }, + { + "epoch": 2.0048, + "grad_norm": 3.2037904262542725, + "learning_rate": 0.00037079999999999996, + "loss": 2.4111, + "step": 1253 + }, + { + "epoch": 2.0064, + "grad_norm": 3.279060125350952, + "learning_rate": 0.0003711, + "loss": 2.4119, + "step": 1254 + }, + { + "epoch": 2.008, + "grad_norm": 2.2401888370513916, + "learning_rate": 0.0003714, + "loss": 2.1714, + "step": 1255 + }, + { + "epoch": 2.0096, + "grad_norm": 1.0563182830810547, + "learning_rate": 0.0003717, + "loss": 2.5566, + "step": 1256 + }, + { + "epoch": 2.0112, + "grad_norm": 2.3503923416137695, + "learning_rate": 0.000372, + "loss": 2.2229, + "step": 1257 + }, + { + "epoch": 2.0128, + "grad_norm": 2.3854825496673584, + "learning_rate": 0.0003723, + "loss": 2.1712, + "step": 1258 + }, + { + "epoch": 2.0144, + "grad_norm": 1.7842551469802856, + "learning_rate": 0.00037259999999999995, + "loss": 2.2321, + "step": 1259 + }, + { + "epoch": 2.016, + "grad_norm": 0.7852015495300293, + "learning_rate": 0.0003729, + "loss": 2.3389, + "step": 1260 + }, + { + "epoch": 2.0176, + "grad_norm": 1.1106822490692139, + "learning_rate": 0.00037319999999999996, + "loss": 1.9826, + "step": 1261 + }, + { + "epoch": 2.0192, + "grad_norm": 0.7619666457176208, + "learning_rate": 0.0003735, + "loss": 1.9273, + "step": 1262 + }, + { + "epoch": 2.0208, + "grad_norm": 0.7985825538635254, + "learning_rate": 0.0003738, + "loss": 1.978, + "step": 1263 + }, + { + "epoch": 2.0224, + "grad_norm": 0.5993479490280151, + "learning_rate": 0.0003741, + "loss": 2.0277, + "step": 1264 + }, + { + "epoch": 2.024, + "grad_norm": 5.699611186981201, + "learning_rate": 0.0003744, + "loss": 2.3771, + "step": 1265 + }, + { + "epoch": 2.0256, + "grad_norm": 1.6110634803771973, + "learning_rate": 0.0003747, + "loss": 2.2158, + "step": 1266 + }, + { + "epoch": 2.0272, + "grad_norm": 2.9092864990234375, + "learning_rate": 0.00037499999999999995, + "loss": 2.0358, + "step": 1267 + }, + { + "epoch": 2.0288, + "grad_norm": 1.1097922325134277, + "learning_rate": 0.00037529999999999996, + "loss": 2.1937, + "step": 1268 + }, + { + "epoch": 2.0304, + "grad_norm": 2.0541391372680664, + "learning_rate": 0.00037559999999999997, + "loss": 2.2668, + "step": 1269 + }, + { + "epoch": 2.032, + "grad_norm": 0.8676146268844604, + "learning_rate": 0.0003758999999999999, + "loss": 2.0615, + "step": 1270 + }, + { + "epoch": 2.0336, + "grad_norm": 1.396476149559021, + "learning_rate": 0.0003762, + "loss": 2.0727, + "step": 1271 + }, + { + "epoch": 2.0352, + "grad_norm": 0.6552128791809082, + "learning_rate": 0.00037649999999999994, + "loss": 1.957, + "step": 1272 + }, + { + "epoch": 2.0368, + "grad_norm": 0.5469757318496704, + "learning_rate": 0.00037679999999999994, + "loss": 1.9888, + "step": 1273 + }, + { + "epoch": 2.0384, + "grad_norm": 1.1260696649551392, + "learning_rate": 0.00037709999999999995, + "loss": 2.0644, + "step": 1274 + }, + { + "epoch": 2.04, + "grad_norm": 0.48484399914741516, + "learning_rate": 0.00037739999999999996, + "loss": 2.09, + "step": 1275 + }, + { + "epoch": 2.0416, + "grad_norm": 1.1584820747375488, + "learning_rate": 0.0003776999999999999, + "loss": 2.0482, + "step": 1276 + }, + { + "epoch": 2.0432, + "grad_norm": 1.8108872175216675, + "learning_rate": 0.00037799999999999997, + "loss": 2.0385, + "step": 1277 + }, + { + "epoch": 2.0448, + "grad_norm": 1.7837122678756714, + "learning_rate": 0.0003782999999999999, + "loss": 1.9413, + "step": 1278 + }, + { + "epoch": 2.0464, + "grad_norm": 1.3678404092788696, + "learning_rate": 0.0003786, + "loss": 1.968, + "step": 1279 + }, + { + "epoch": 2.048, + "grad_norm": 0.7513062953948975, + "learning_rate": 0.00037889999999999994, + "loss": 1.9819, + "step": 1280 + }, + { + "epoch": 2.0496, + "grad_norm": 0.7749692797660828, + "learning_rate": 0.00037919999999999995, + "loss": 1.9809, + "step": 1281 + }, + { + "epoch": 2.0512, + "grad_norm": 1.3065149784088135, + "learning_rate": 0.00037949999999999995, + "loss": 1.9025, + "step": 1282 + }, + { + "epoch": 2.0528, + "grad_norm": 0.6770661473274231, + "learning_rate": 0.00037979999999999996, + "loss": 2.1342, + "step": 1283 + }, + { + "epoch": 2.0544, + "grad_norm": 1.5681331157684326, + "learning_rate": 0.0003800999999999999, + "loss": 2.0686, + "step": 1284 + }, + { + "epoch": 2.056, + "grad_norm": 1.1625335216522217, + "learning_rate": 0.0003804, + "loss": 2.0118, + "step": 1285 + }, + { + "epoch": 2.0576, + "grad_norm": 0.9901454448699951, + "learning_rate": 0.00038069999999999993, + "loss": 2.0614, + "step": 1286 + }, + { + "epoch": 2.0592, + "grad_norm": 1.948578119277954, + "learning_rate": 0.000381, + "loss": 2.0299, + "step": 1287 + }, + { + "epoch": 2.0608, + "grad_norm": 1.5582636594772339, + "learning_rate": 0.00038129999999999994, + "loss": 2.0481, + "step": 1288 + }, + { + "epoch": 2.0624, + "grad_norm": 1.2568862438201904, + "learning_rate": 0.00038159999999999995, + "loss": 2.3433, + "step": 1289 + }, + { + "epoch": 2.064, + "grad_norm": 1.8581323623657227, + "learning_rate": 0.00038189999999999996, + "loss": 2.0881, + "step": 1290 + }, + { + "epoch": 2.0656, + "grad_norm": 1.954257607460022, + "learning_rate": 0.00038219999999999997, + "loss": 2.6312, + "step": 1291 + }, + { + "epoch": 2.0672, + "grad_norm": 2.38853120803833, + "learning_rate": 0.0003824999999999999, + "loss": 1.9754, + "step": 1292 + }, + { + "epoch": 2.0688, + "grad_norm": 0.8991557955741882, + "learning_rate": 0.0003828, + "loss": 2.2999, + "step": 1293 + }, + { + "epoch": 2.0704, + "grad_norm": 2.9160866737365723, + "learning_rate": 0.00038309999999999993, + "loss": 2.0209, + "step": 1294 + }, + { + "epoch": 2.072, + "grad_norm": 2.4676144123077393, + "learning_rate": 0.0003834, + "loss": 2.1019, + "step": 1295 + }, + { + "epoch": 2.0736, + "grad_norm": 1.0132110118865967, + "learning_rate": 0.00038369999999999995, + "loss": 2.355, + "step": 1296 + }, + { + "epoch": 2.0752, + "grad_norm": 1.3484665155410767, + "learning_rate": 0.00038399999999999996, + "loss": 2.1014, + "step": 1297 + }, + { + "epoch": 2.0768, + "grad_norm": 2.196345567703247, + "learning_rate": 0.00038429999999999996, + "loss": 2.1293, + "step": 1298 + }, + { + "epoch": 2.0784, + "grad_norm": 2.6476728916168213, + "learning_rate": 0.00038459999999999997, + "loss": 2.347, + "step": 1299 + }, + { + "epoch": 2.08, + "grad_norm": 1.3819998502731323, + "learning_rate": 0.0003848999999999999, + "loss": 2.6989, + "step": 1300 + }, + { + "epoch": 2.0816, + "grad_norm": 8.731172561645508, + "learning_rate": 0.0003852, + "loss": 2.7856, + "step": 1301 + }, + { + "epoch": 2.0832, + "grad_norm": 11.080965042114258, + "learning_rate": 0.00038549999999999994, + "loss": 2.9622, + "step": 1302 + }, + { + "epoch": 2.0848, + "grad_norm": 2.8074939250946045, + "learning_rate": 0.0003858, + "loss": 2.3932, + "step": 1303 + }, + { + "epoch": 2.0864, + "grad_norm": 2.6564383506774902, + "learning_rate": 0.00038609999999999995, + "loss": 2.2092, + "step": 1304 + }, + { + "epoch": 2.088, + "grad_norm": 4.33972692489624, + "learning_rate": 0.00038639999999999996, + "loss": 2.0047, + "step": 1305 + }, + { + "epoch": 2.0896, + "grad_norm": 1.5930198431015015, + "learning_rate": 0.00038669999999999997, + "loss": 2.2548, + "step": 1306 + }, + { + "epoch": 2.0912, + "grad_norm": 4.321974754333496, + "learning_rate": 0.000387, + "loss": 2.0923, + "step": 1307 + }, + { + "epoch": 2.0928, + "grad_norm": 3.0230000019073486, + "learning_rate": 0.00038729999999999993, + "loss": 2.2425, + "step": 1308 + }, + { + "epoch": 2.0944, + "grad_norm": 4.086137294769287, + "learning_rate": 0.0003876, + "loss": 1.9095, + "step": 1309 + }, + { + "epoch": 2.096, + "grad_norm": 3.24074387550354, + "learning_rate": 0.00038789999999999994, + "loss": 1.9992, + "step": 1310 + }, + { + "epoch": 2.0976, + "grad_norm": 2.870512008666992, + "learning_rate": 0.0003882, + "loss": 1.7777, + "step": 1311 + }, + { + "epoch": 2.0992, + "grad_norm": 1.1426403522491455, + "learning_rate": 0.00038849999999999996, + "loss": 2.0465, + "step": 1312 + }, + { + "epoch": 2.1008, + "grad_norm": 8.122888565063477, + "learning_rate": 0.00038879999999999996, + "loss": 2.5261, + "step": 1313 + }, + { + "epoch": 2.1024, + "grad_norm": 1.0429184436798096, + "learning_rate": 0.00038909999999999997, + "loss": 2.1316, + "step": 1314 + }, + { + "epoch": 2.104, + "grad_norm": 2.0756289958953857, + "learning_rate": 0.0003894, + "loss": 2.1585, + "step": 1315 + }, + { + "epoch": 2.1056, + "grad_norm": 2.4256415367126465, + "learning_rate": 0.00038969999999999993, + "loss": 2.2013, + "step": 1316 + }, + { + "epoch": 2.1072, + "grad_norm": 0.8627563118934631, + "learning_rate": 0.00039, + "loss": 1.8589, + "step": 1317 + }, + { + "epoch": 2.1088, + "grad_norm": 1.7571260929107666, + "learning_rate": 0.00039029999999999995, + "loss": 1.8394, + "step": 1318 + }, + { + "epoch": 2.1104, + "grad_norm": 1.1963249444961548, + "learning_rate": 0.00039059999999999995, + "loss": 2.2086, + "step": 1319 + }, + { + "epoch": 2.112, + "grad_norm": 0.7731038331985474, + "learning_rate": 0.00039089999999999996, + "loss": 1.8113, + "step": 1320 + }, + { + "epoch": 2.1136, + "grad_norm": 0.5587062835693359, + "learning_rate": 0.00039119999999999997, + "loss": 1.8982, + "step": 1321 + }, + { + "epoch": 2.1152, + "grad_norm": 0.5626364946365356, + "learning_rate": 0.0003914999999999999, + "loss": 1.8954, + "step": 1322 + }, + { + "epoch": 2.1168, + "grad_norm": 1.408499002456665, + "learning_rate": 0.0003918, + "loss": 1.7425, + "step": 1323 + }, + { + "epoch": 2.1184, + "grad_norm": 1.6526440382003784, + "learning_rate": 0.00039209999999999994, + "loss": 2.1664, + "step": 1324 + }, + { + "epoch": 2.12, + "grad_norm": 0.4821324050426483, + "learning_rate": 0.0003924, + "loss": 1.9262, + "step": 1325 + }, + { + "epoch": 2.1216, + "grad_norm": 0.7852998375892639, + "learning_rate": 0.00039269999999999995, + "loss": 1.6703, + "step": 1326 + }, + { + "epoch": 2.1232, + "grad_norm": 3.967268943786621, + "learning_rate": 0.00039299999999999996, + "loss": 1.8964, + "step": 1327 + }, + { + "epoch": 2.1248, + "grad_norm": 0.44754117727279663, + "learning_rate": 0.00039329999999999996, + "loss": 2.0235, + "step": 1328 + }, + { + "epoch": 2.1264, + "grad_norm": 0.6954207420349121, + "learning_rate": 0.00039359999999999997, + "loss": 1.9802, + "step": 1329 + }, + { + "epoch": 2.128, + "grad_norm": 1.3386781215667725, + "learning_rate": 0.0003938999999999999, + "loss": 1.9033, + "step": 1330 + }, + { + "epoch": 2.1296, + "grad_norm": 1.4398391246795654, + "learning_rate": 0.0003942, + "loss": 2.1433, + "step": 1331 + }, + { + "epoch": 2.1312, + "grad_norm": 3.3130927085876465, + "learning_rate": 0.00039449999999999994, + "loss": 2.2894, + "step": 1332 + }, + { + "epoch": 2.1328, + "grad_norm": 0.6074516177177429, + "learning_rate": 0.0003948, + "loss": 1.6613, + "step": 1333 + }, + { + "epoch": 2.1344, + "grad_norm": 0.8905891180038452, + "learning_rate": 0.00039509999999999995, + "loss": 1.9249, + "step": 1334 + }, + { + "epoch": 2.136, + "grad_norm": 0.693198025226593, + "learning_rate": 0.00039539999999999996, + "loss": 2.1552, + "step": 1335 + }, + { + "epoch": 2.1376, + "grad_norm": 1.595231056213379, + "learning_rate": 0.00039569999999999997, + "loss": 2.1719, + "step": 1336 + }, + { + "epoch": 2.1391999999999998, + "grad_norm": 0.9253241419792175, + "learning_rate": 0.000396, + "loss": 2.2147, + "step": 1337 + }, + { + "epoch": 2.1408, + "grad_norm": 0.9581746459007263, + "learning_rate": 0.00039629999999999993, + "loss": 1.7626, + "step": 1338 + }, + { + "epoch": 2.1424, + "grad_norm": 0.6418152451515198, + "learning_rate": 0.0003966, + "loss": 2.1862, + "step": 1339 + }, + { + "epoch": 2.144, + "grad_norm": 2.4077084064483643, + "learning_rate": 0.00039689999999999994, + "loss": 1.7163, + "step": 1340 + }, + { + "epoch": 2.1456, + "grad_norm": 0.8226627111434937, + "learning_rate": 0.0003972, + "loss": 2.1605, + "step": 1341 + }, + { + "epoch": 2.1471999999999998, + "grad_norm": 0.8403061628341675, + "learning_rate": 0.00039749999999999996, + "loss": 1.9728, + "step": 1342 + }, + { + "epoch": 2.1488, + "grad_norm": 0.5522119402885437, + "learning_rate": 0.00039779999999999997, + "loss": 1.7164, + "step": 1343 + }, + { + "epoch": 2.1504, + "grad_norm": 0.8088909387588501, + "learning_rate": 0.0003981, + "loss": 2.069, + "step": 1344 + }, + { + "epoch": 2.152, + "grad_norm": 0.649395763874054, + "learning_rate": 0.0003984, + "loss": 2.0172, + "step": 1345 + }, + { + "epoch": 2.1536, + "grad_norm": 1.245216965675354, + "learning_rate": 0.00039869999999999993, + "loss": 2.0648, + "step": 1346 + }, + { + "epoch": 2.1552, + "grad_norm": 2.028209686279297, + "learning_rate": 0.000399, + "loss": 2.2123, + "step": 1347 + }, + { + "epoch": 2.1568, + "grad_norm": 1.16299569606781, + "learning_rate": 0.00039929999999999995, + "loss": 2.1696, + "step": 1348 + }, + { + "epoch": 2.1584, + "grad_norm": 2.0417158603668213, + "learning_rate": 0.0003996, + "loss": 2.5852, + "step": 1349 + }, + { + "epoch": 2.16, + "grad_norm": 1.194480538368225, + "learning_rate": 0.00039989999999999996, + "loss": 2.6632, + "step": 1350 + }, + { + "epoch": 2.1616, + "grad_norm": 4.301017761230469, + "learning_rate": 0.00040019999999999997, + "loss": 2.2968, + "step": 1351 + }, + { + "epoch": 2.1632, + "grad_norm": 5.509984970092773, + "learning_rate": 0.0004005, + "loss": 2.2294, + "step": 1352 + }, + { + "epoch": 2.1648, + "grad_norm": 2.9739537239074707, + "learning_rate": 0.0004008, + "loss": 2.4747, + "step": 1353 + }, + { + "epoch": 2.1664, + "grad_norm": 5.930892467498779, + "learning_rate": 0.00040109999999999994, + "loss": 2.4353, + "step": 1354 + }, + { + "epoch": 2.168, + "grad_norm": 3.7599174976348877, + "learning_rate": 0.0004014, + "loss": 2.3522, + "step": 1355 + }, + { + "epoch": 2.1696, + "grad_norm": 1.8498942852020264, + "learning_rate": 0.00040169999999999995, + "loss": 2.6316, + "step": 1356 + }, + { + "epoch": 2.1712, + "grad_norm": 6.10044002532959, + "learning_rate": 0.000402, + "loss": 2.7012, + "step": 1357 + }, + { + "epoch": 2.1728, + "grad_norm": 4.401482582092285, + "learning_rate": 0.00040229999999999997, + "loss": 2.4794, + "step": 1358 + }, + { + "epoch": 2.1744, + "grad_norm": 5.523442268371582, + "learning_rate": 0.0004026, + "loss": 2.4327, + "step": 1359 + }, + { + "epoch": 2.176, + "grad_norm": 4.25037956237793, + "learning_rate": 0.0004029, + "loss": 2.2722, + "step": 1360 + }, + { + "epoch": 2.1776, + "grad_norm": 5.356450080871582, + "learning_rate": 0.0004032, + "loss": 2.9387, + "step": 1361 + }, + { + "epoch": 2.1792, + "grad_norm": 3.4604556560516357, + "learning_rate": 0.00040349999999999994, + "loss": 2.5264, + "step": 1362 + }, + { + "epoch": 2.1808, + "grad_norm": 4.26166296005249, + "learning_rate": 0.0004038, + "loss": 2.3071, + "step": 1363 + }, + { + "epoch": 2.1824, + "grad_norm": 4.863035202026367, + "learning_rate": 0.00040409999999999996, + "loss": 2.2555, + "step": 1364 + }, + { + "epoch": 2.184, + "grad_norm": 4.790585041046143, + "learning_rate": 0.0004044, + "loss": 2.383, + "step": 1365 + }, + { + "epoch": 2.1856, + "grad_norm": 4.248513698577881, + "learning_rate": 0.00040469999999999997, + "loss": 2.3042, + "step": 1366 + }, + { + "epoch": 2.1872, + "grad_norm": 3.236246109008789, + "learning_rate": 0.000405, + "loss": 2.3719, + "step": 1367 + }, + { + "epoch": 2.1888, + "grad_norm": 1.0238962173461914, + "learning_rate": 0.00040529999999999993, + "loss": 2.3266, + "step": 1368 + }, + { + "epoch": 2.1904, + "grad_norm": 0.859204888343811, + "learning_rate": 0.0004056, + "loss": 2.3052, + "step": 1369 + }, + { + "epoch": 2.192, + "grad_norm": 0.8815178275108337, + "learning_rate": 0.00040589999999999995, + "loss": 2.3872, + "step": 1370 + }, + { + "epoch": 2.1936, + "grad_norm": 1.2759977579116821, + "learning_rate": 0.0004062, + "loss": 2.1615, + "step": 1371 + }, + { + "epoch": 2.1952, + "grad_norm": 1.6016004085540771, + "learning_rate": 0.00040649999999999996, + "loss": 2.2023, + "step": 1372 + }, + { + "epoch": 2.1968, + "grad_norm": 1.93936026096344, + "learning_rate": 0.00040679999999999997, + "loss": 2.1016, + "step": 1373 + }, + { + "epoch": 2.1984, + "grad_norm": 0.9935839772224426, + "learning_rate": 0.0004071, + "loss": 2.3746, + "step": 1374 + }, + { + "epoch": 2.2, + "grad_norm": 2.0738706588745117, + "learning_rate": 0.0004074, + "loss": 2.1396, + "step": 1375 + }, + { + "epoch": 2.2016, + "grad_norm": 0.7379969954490662, + "learning_rate": 0.00040769999999999994, + "loss": 2.0896, + "step": 1376 + }, + { + "epoch": 2.2032, + "grad_norm": 1.2633837461471558, + "learning_rate": 0.000408, + "loss": 2.1598, + "step": 1377 + }, + { + "epoch": 2.2048, + "grad_norm": 0.7961741089820862, + "learning_rate": 0.00040829999999999995, + "loss": 2.1767, + "step": 1378 + }, + { + "epoch": 2.2064, + "grad_norm": 1.836298942565918, + "learning_rate": 0.0004086, + "loss": 2.3359, + "step": 1379 + }, + { + "epoch": 2.208, + "grad_norm": 2.328185796737671, + "learning_rate": 0.00040889999999999996, + "loss": 2.3435, + "step": 1380 + }, + { + "epoch": 2.2096, + "grad_norm": 1.825183629989624, + "learning_rate": 0.00040919999999999997, + "loss": 2.5522, + "step": 1381 + }, + { + "epoch": 2.2112, + "grad_norm": 0.8028872609138489, + "learning_rate": 0.0004095, + "loss": 2.3269, + "step": 1382 + }, + { + "epoch": 2.2128, + "grad_norm": 0.9807519316673279, + "learning_rate": 0.0004098, + "loss": 2.4156, + "step": 1383 + }, + { + "epoch": 2.2144, + "grad_norm": 2.0771546363830566, + "learning_rate": 0.00041009999999999994, + "loss": 2.0652, + "step": 1384 + }, + { + "epoch": 2.216, + "grad_norm": 2.1903600692749023, + "learning_rate": 0.0004104, + "loss": 2.2988, + "step": 1385 + }, + { + "epoch": 2.2176, + "grad_norm": 2.105729579925537, + "learning_rate": 0.00041069999999999995, + "loss": 2.3518, + "step": 1386 + }, + { + "epoch": 2.2192, + "grad_norm": 2.26141619682312, + "learning_rate": 0.000411, + "loss": 2.4327, + "step": 1387 + }, + { + "epoch": 2.2208, + "grad_norm": 2.7038395404815674, + "learning_rate": 0.00041129999999999997, + "loss": 2.2783, + "step": 1388 + }, + { + "epoch": 2.2224, + "grad_norm": 0.9944401383399963, + "learning_rate": 0.0004116, + "loss": 2.4459, + "step": 1389 + }, + { + "epoch": 2.224, + "grad_norm": 0.8764882683753967, + "learning_rate": 0.0004119, + "loss": 2.3234, + "step": 1390 + }, + { + "epoch": 2.2256, + "grad_norm": 3.0021326541900635, + "learning_rate": 0.0004122, + "loss": 2.5689, + "step": 1391 + }, + { + "epoch": 2.2272, + "grad_norm": 2.121433734893799, + "learning_rate": 0.00041249999999999994, + "loss": 2.2978, + "step": 1392 + }, + { + "epoch": 2.2288, + "grad_norm": 4.0318121910095215, + "learning_rate": 0.00041279999999999995, + "loss": 2.3756, + "step": 1393 + }, + { + "epoch": 2.2304, + "grad_norm": 4.119312763214111, + "learning_rate": 0.00041309999999999996, + "loss": 2.4183, + "step": 1394 + }, + { + "epoch": 2.232, + "grad_norm": 6.8856940269470215, + "learning_rate": 0.0004133999999999999, + "loss": 2.6106, + "step": 1395 + }, + { + "epoch": 2.2336, + "grad_norm": 5.660776138305664, + "learning_rate": 0.00041369999999999997, + "loss": 2.5119, + "step": 1396 + }, + { + "epoch": 2.2352, + "grad_norm": 4.50926399230957, + "learning_rate": 0.0004139999999999999, + "loss": 2.1634, + "step": 1397 + }, + { + "epoch": 2.2368, + "grad_norm": 6.12561559677124, + "learning_rate": 0.0004143, + "loss": 2.7285, + "step": 1398 + }, + { + "epoch": 2.2384, + "grad_norm": 2.332390546798706, + "learning_rate": 0.00041459999999999994, + "loss": 2.4774, + "step": 1399 + }, + { + "epoch": 2.24, + "grad_norm": 3.783649444580078, + "learning_rate": 0.00041489999999999995, + "loss": 2.7869, + "step": 1400 + }, + { + "epoch": 2.2416, + "grad_norm": 15.439468383789062, + "learning_rate": 0.00041519999999999995, + "loss": 3.6933, + "step": 1401 + }, + { + "epoch": 2.2432, + "grad_norm": 4.923689842224121, + "learning_rate": 0.00041549999999999996, + "loss": 2.9789, + "step": 1402 + }, + { + "epoch": 2.2448, + "grad_norm": 10.209508895874023, + "learning_rate": 0.0004157999999999999, + "loss": 2.9776, + "step": 1403 + }, + { + "epoch": 2.2464, + "grad_norm": 4.802323818206787, + "learning_rate": 0.0004161, + "loss": 2.8077, + "step": 1404 + }, + { + "epoch": 2.248, + "grad_norm": 10.365413665771484, + "learning_rate": 0.00041639999999999993, + "loss": 3.0581, + "step": 1405 + }, + { + "epoch": 2.2496, + "grad_norm": 3.0222530364990234, + "learning_rate": 0.0004167, + "loss": 2.5003, + "step": 1406 + }, + { + "epoch": 2.2512, + "grad_norm": 4.738245010375977, + "learning_rate": 0.00041699999999999994, + "loss": 2.664, + "step": 1407 + }, + { + "epoch": 2.2528, + "grad_norm": 0.694865882396698, + "learning_rate": 0.00041729999999999995, + "loss": 2.661, + "step": 1408 + }, + { + "epoch": 2.2544, + "grad_norm": 1.4679259061813354, + "learning_rate": 0.00041759999999999996, + "loss": 2.5017, + "step": 1409 + }, + { + "epoch": 2.2560000000000002, + "grad_norm": 3.7848410606384277, + "learning_rate": 0.00041789999999999997, + "loss": 2.6454, + "step": 1410 + }, + { + "epoch": 2.2576, + "grad_norm": 2.437657594680786, + "learning_rate": 0.0004181999999999999, + "loss": 2.3759, + "step": 1411 + }, + { + "epoch": 2.2592, + "grad_norm": 0.982265830039978, + "learning_rate": 0.0004185, + "loss": 2.6115, + "step": 1412 + }, + { + "epoch": 2.2608, + "grad_norm": 5.4903788566589355, + "learning_rate": 0.00041879999999999993, + "loss": 2.2263, + "step": 1413 + }, + { + "epoch": 2.2624, + "grad_norm": 6.109436511993408, + "learning_rate": 0.0004191, + "loss": 2.2023, + "step": 1414 + }, + { + "epoch": 2.2640000000000002, + "grad_norm": 8.365518569946289, + "learning_rate": 0.00041939999999999995, + "loss": 2.5806, + "step": 1415 + }, + { + "epoch": 2.2656, + "grad_norm": 8.561690330505371, + "learning_rate": 0.00041969999999999996, + "loss": 2.7747, + "step": 1416 + }, + { + "epoch": 2.2672, + "grad_norm": 8.934793472290039, + "learning_rate": 0.00041999999999999996, + "loss": 2.6602, + "step": 1417 + }, + { + "epoch": 2.2688, + "grad_norm": 4.877744197845459, + "learning_rate": 0.00042029999999999997, + "loss": 2.6439, + "step": 1418 + }, + { + "epoch": 2.2704, + "grad_norm": 7.499762535095215, + "learning_rate": 0.0004205999999999999, + "loss": 2.4732, + "step": 1419 + }, + { + "epoch": 2.2720000000000002, + "grad_norm": 6.877169609069824, + "learning_rate": 0.0004209, + "loss": 2.5265, + "step": 1420 + }, + { + "epoch": 2.2736, + "grad_norm": 5.535276889801025, + "learning_rate": 0.00042119999999999994, + "loss": 2.3668, + "step": 1421 + }, + { + "epoch": 2.2752, + "grad_norm": 4.566375255584717, + "learning_rate": 0.00042149999999999995, + "loss": 2.4487, + "step": 1422 + }, + { + "epoch": 2.2768, + "grad_norm": 2.4168081283569336, + "learning_rate": 0.00042179999999999995, + "loss": 2.426, + "step": 1423 + }, + { + "epoch": 2.2784, + "grad_norm": 3.585942506790161, + "learning_rate": 0.00042209999999999996, + "loss": 3.1391, + "step": 1424 + }, + { + "epoch": 2.2800000000000002, + "grad_norm": 1.2044767141342163, + "learning_rate": 0.0004223999999999999, + "loss": 2.4168, + "step": 1425 + }, + { + "epoch": 2.2816, + "grad_norm": 1.4629987478256226, + "learning_rate": 0.0004227, + "loss": 2.2756, + "step": 1426 + }, + { + "epoch": 2.2832, + "grad_norm": 1.5924323797225952, + "learning_rate": 0.00042299999999999993, + "loss": 2.2438, + "step": 1427 + }, + { + "epoch": 2.2848, + "grad_norm": 0.9790437817573547, + "learning_rate": 0.0004233, + "loss": 2.3787, + "step": 1428 + }, + { + "epoch": 2.2864, + "grad_norm": 2.070406675338745, + "learning_rate": 0.00042359999999999994, + "loss": 2.5208, + "step": 1429 + }, + { + "epoch": 2.288, + "grad_norm": 1.5404846668243408, + "learning_rate": 0.00042389999999999995, + "loss": 2.2397, + "step": 1430 + }, + { + "epoch": 2.2896, + "grad_norm": 1.684266448020935, + "learning_rate": 0.00042419999999999996, + "loss": 2.4266, + "step": 1431 + }, + { + "epoch": 2.2912, + "grad_norm": 2.2368626594543457, + "learning_rate": 0.00042449999999999996, + "loss": 2.2046, + "step": 1432 + }, + { + "epoch": 2.2928, + "grad_norm": 1.125388264656067, + "learning_rate": 0.0004247999999999999, + "loss": 2.2177, + "step": 1433 + }, + { + "epoch": 2.2944, + "grad_norm": 1.5939106941223145, + "learning_rate": 0.0004251, + "loss": 2.3616, + "step": 1434 + }, + { + "epoch": 2.296, + "grad_norm": 1.2981842756271362, + "learning_rate": 0.00042539999999999993, + "loss": 2.3863, + "step": 1435 + }, + { + "epoch": 2.2976, + "grad_norm": 0.7831336259841919, + "learning_rate": 0.0004257, + "loss": 2.6792, + "step": 1436 + }, + { + "epoch": 2.2992, + "grad_norm": 1.4698004722595215, + "learning_rate": 0.00042599999999999995, + "loss": 2.3886, + "step": 1437 + }, + { + "epoch": 2.3008, + "grad_norm": 0.9203685522079468, + "learning_rate": 0.00042629999999999995, + "loss": 2.3127, + "step": 1438 + }, + { + "epoch": 2.3024, + "grad_norm": 0.5544130206108093, + "learning_rate": 0.00042659999999999996, + "loss": 2.3799, + "step": 1439 + }, + { + "epoch": 2.304, + "grad_norm": 2.3812172412872314, + "learning_rate": 0.00042689999999999997, + "loss": 2.4408, + "step": 1440 + }, + { + "epoch": 2.3056, + "grad_norm": 2.6237330436706543, + "learning_rate": 0.0004271999999999999, + "loss": 2.3087, + "step": 1441 + }, + { + "epoch": 2.3072, + "grad_norm": 2.7588069438934326, + "learning_rate": 0.0004275, + "loss": 2.1371, + "step": 1442 + }, + { + "epoch": 2.3088, + "grad_norm": 0.6106075048446655, + "learning_rate": 0.00042779999999999994, + "loss": 2.2978, + "step": 1443 + }, + { + "epoch": 2.3104, + "grad_norm": 0.6980108022689819, + "learning_rate": 0.0004281, + "loss": 2.4839, + "step": 1444 + }, + { + "epoch": 2.312, + "grad_norm": 1.6969398260116577, + "learning_rate": 0.00042839999999999995, + "loss": 2.5316, + "step": 1445 + }, + { + "epoch": 2.3136, + "grad_norm": 1.0797035694122314, + "learning_rate": 0.00042869999999999996, + "loss": 2.3753, + "step": 1446 + }, + { + "epoch": 2.3152, + "grad_norm": 1.6315916776657104, + "learning_rate": 0.00042899999999999997, + "loss": 2.9357, + "step": 1447 + }, + { + "epoch": 2.3168, + "grad_norm": 2.026221990585327, + "learning_rate": 0.00042929999999999997, + "loss": 2.4578, + "step": 1448 + }, + { + "epoch": 2.3184, + "grad_norm": 1.5287202596664429, + "learning_rate": 0.0004295999999999999, + "loss": 2.5209, + "step": 1449 + }, + { + "epoch": 2.32, + "grad_norm": 3.8067626953125, + "learning_rate": 0.0004299, + "loss": 3.0858, + "step": 1450 + }, + { + "epoch": 2.3216, + "grad_norm": 16.87346076965332, + "learning_rate": 0.00043019999999999994, + "loss": 3.6091, + "step": 1451 + }, + { + "epoch": 2.3232, + "grad_norm": 2.06990385055542, + "learning_rate": 0.0004305, + "loss": 2.7344, + "step": 1452 + }, + { + "epoch": 2.3247999999999998, + "grad_norm": 2.008976459503174, + "learning_rate": 0.00043079999999999995, + "loss": 2.7445, + "step": 1453 + }, + { + "epoch": 2.3264, + "grad_norm": 1.7746312618255615, + "learning_rate": 0.00043109999999999996, + "loss": 2.8305, + "step": 1454 + }, + { + "epoch": 2.328, + "grad_norm": 3.5500175952911377, + "learning_rate": 0.00043139999999999997, + "loss": 2.884, + "step": 1455 + }, + { + "epoch": 2.3296, + "grad_norm": 5.436251163482666, + "learning_rate": 0.0004317, + "loss": 2.774, + "step": 1456 + }, + { + "epoch": 2.3312, + "grad_norm": 7.481227874755859, + "learning_rate": 0.00043199999999999993, + "loss": 2.6914, + "step": 1457 + }, + { + "epoch": 2.3327999999999998, + "grad_norm": 4.893389701843262, + "learning_rate": 0.0004323, + "loss": 2.5741, + "step": 1458 + }, + { + "epoch": 2.3344, + "grad_norm": 7.032483100891113, + "learning_rate": 0.00043259999999999994, + "loss": 2.5954, + "step": 1459 + }, + { + "epoch": 2.336, + "grad_norm": 6.124364852905273, + "learning_rate": 0.0004329, + "loss": 2.6022, + "step": 1460 + }, + { + "epoch": 2.3376, + "grad_norm": 7.075193881988525, + "learning_rate": 0.00043319999999999996, + "loss": 2.6718, + "step": 1461 + }, + { + "epoch": 2.3392, + "grad_norm": 4.350131988525391, + "learning_rate": 0.00043349999999999997, + "loss": 2.5274, + "step": 1462 + }, + { + "epoch": 2.3407999999999998, + "grad_norm": 1.6107730865478516, + "learning_rate": 0.0004338, + "loss": 2.3757, + "step": 1463 + }, + { + "epoch": 2.3424, + "grad_norm": 2.455488443374634, + "learning_rate": 0.0004341, + "loss": 2.6995, + "step": 1464 + }, + { + "epoch": 2.344, + "grad_norm": 1.2672028541564941, + "learning_rate": 0.00043439999999999993, + "loss": 2.5708, + "step": 1465 + }, + { + "epoch": 2.3456, + "grad_norm": 2.2092881202697754, + "learning_rate": 0.0004347, + "loss": 2.5716, + "step": 1466 + }, + { + "epoch": 2.3472, + "grad_norm": 1.2420680522918701, + "learning_rate": 0.00043499999999999995, + "loss": 2.294, + "step": 1467 + }, + { + "epoch": 2.3487999999999998, + "grad_norm": 2.2376654148101807, + "learning_rate": 0.00043529999999999996, + "loss": 2.2936, + "step": 1468 + }, + { + "epoch": 2.3504, + "grad_norm": 3.4342873096466064, + "learning_rate": 0.00043559999999999996, + "loss": 2.4192, + "step": 1469 + }, + { + "epoch": 2.352, + "grad_norm": 5.39082670211792, + "learning_rate": 0.00043589999999999997, + "loss": 2.6571, + "step": 1470 + }, + { + "epoch": 2.3536, + "grad_norm": 1.075791597366333, + "learning_rate": 0.0004361999999999999, + "loss": 2.2135, + "step": 1471 + }, + { + "epoch": 2.3552, + "grad_norm": 2.4903030395507812, + "learning_rate": 0.0004365, + "loss": 2.4113, + "step": 1472 + }, + { + "epoch": 2.3568, + "grad_norm": 2.7636349201202393, + "learning_rate": 0.00043679999999999994, + "loss": 2.2446, + "step": 1473 + }, + { + "epoch": 2.3584, + "grad_norm": 3.5851831436157227, + "learning_rate": 0.0004371, + "loss": 2.333, + "step": 1474 + }, + { + "epoch": 2.36, + "grad_norm": 3.596482515335083, + "learning_rate": 0.00043739999999999995, + "loss": 2.4006, + "step": 1475 + }, + { + "epoch": 2.3616, + "grad_norm": 3.66230845451355, + "learning_rate": 0.00043769999999999996, + "loss": 2.336, + "step": 1476 + }, + { + "epoch": 2.3632, + "grad_norm": 3.6934595108032227, + "learning_rate": 0.00043799999999999997, + "loss": 2.4821, + "step": 1477 + }, + { + "epoch": 2.3648, + "grad_norm": 1.861146330833435, + "learning_rate": 0.0004383, + "loss": 2.3221, + "step": 1478 + }, + { + "epoch": 2.3664, + "grad_norm": 3.078958511352539, + "learning_rate": 0.00043859999999999993, + "loss": 2.4531, + "step": 1479 + }, + { + "epoch": 2.368, + "grad_norm": 0.9115749597549438, + "learning_rate": 0.0004389, + "loss": 2.3416, + "step": 1480 + }, + { + "epoch": 2.3696, + "grad_norm": 3.4467458724975586, + "learning_rate": 0.00043919999999999994, + "loss": 2.2444, + "step": 1481 + }, + { + "epoch": 2.3712, + "grad_norm": 0.5655205845832825, + "learning_rate": 0.0004395, + "loss": 2.6449, + "step": 1482 + }, + { + "epoch": 2.3728, + "grad_norm": 3.295214891433716, + "learning_rate": 0.00043979999999999996, + "loss": 2.3346, + "step": 1483 + }, + { + "epoch": 2.3744, + "grad_norm": 4.051139831542969, + "learning_rate": 0.00044009999999999996, + "loss": 2.2891, + "step": 1484 + }, + { + "epoch": 2.376, + "grad_norm": 1.997973084449768, + "learning_rate": 0.00044039999999999997, + "loss": 2.2741, + "step": 1485 + }, + { + "epoch": 2.3776, + "grad_norm": 1.6853876113891602, + "learning_rate": 0.0004407, + "loss": 2.4009, + "step": 1486 + }, + { + "epoch": 2.3792, + "grad_norm": 1.9763381481170654, + "learning_rate": 0.00044099999999999993, + "loss": 2.3812, + "step": 1487 + }, + { + "epoch": 2.3808, + "grad_norm": 1.633191466331482, + "learning_rate": 0.0004413, + "loss": 2.3768, + "step": 1488 + }, + { + "epoch": 2.3824, + "grad_norm": 1.3585143089294434, + "learning_rate": 0.00044159999999999995, + "loss": 2.2952, + "step": 1489 + }, + { + "epoch": 2.384, + "grad_norm": 0.6596494913101196, + "learning_rate": 0.0004419, + "loss": 2.6077, + "step": 1490 + }, + { + "epoch": 2.3856, + "grad_norm": 1.646826982498169, + "learning_rate": 0.00044219999999999996, + "loss": 2.3303, + "step": 1491 + }, + { + "epoch": 2.3872, + "grad_norm": 1.6970372200012207, + "learning_rate": 0.00044249999999999997, + "loss": 2.3328, + "step": 1492 + }, + { + "epoch": 2.3888, + "grad_norm": 2.27344012260437, + "learning_rate": 0.0004428, + "loss": 2.6301, + "step": 1493 + }, + { + "epoch": 2.3904, + "grad_norm": 0.710404098033905, + "learning_rate": 0.0004431, + "loss": 2.4525, + "step": 1494 + }, + { + "epoch": 2.392, + "grad_norm": 0.9620774388313293, + "learning_rate": 0.00044339999999999994, + "loss": 2.7767, + "step": 1495 + }, + { + "epoch": 2.3936, + "grad_norm": 1.4343668222427368, + "learning_rate": 0.0004437, + "loss": 2.546, + "step": 1496 + }, + { + "epoch": 2.3952, + "grad_norm": 2.0133700370788574, + "learning_rate": 0.00044399999999999995, + "loss": 2.7277, + "step": 1497 + }, + { + "epoch": 2.3968, + "grad_norm": 1.0265851020812988, + "learning_rate": 0.0004443, + "loss": 2.4666, + "step": 1498 + }, + { + "epoch": 2.3984, + "grad_norm": 4.100027561187744, + "learning_rate": 0.00044459999999999996, + "loss": 3.0378, + "step": 1499 + }, + { + "epoch": 2.4, + "grad_norm": 5.678374767303467, + "learning_rate": 0.00044489999999999997, + "loss": 3.3142, + "step": 1500 + }, + { + "epoch": 2.4016, + "grad_norm": 4.942265510559082, + "learning_rate": 0.0004452, + "loss": 3.2102, + "step": 1501 + }, + { + "epoch": 2.4032, + "grad_norm": 1.7186299562454224, + "learning_rate": 0.0004455, + "loss": 2.9047, + "step": 1502 + }, + { + "epoch": 2.4048, + "grad_norm": 2.8937320709228516, + "learning_rate": 0.00044579999999999994, + "loss": 3.0197, + "step": 1503 + }, + { + "epoch": 2.4064, + "grad_norm": 3.7428622245788574, + "learning_rate": 0.0004461, + "loss": 2.9995, + "step": 1504 + }, + { + "epoch": 2.408, + "grad_norm": 4.141154766082764, + "learning_rate": 0.00044639999999999995, + "loss": 2.9252, + "step": 1505 + }, + { + "epoch": 2.4096, + "grad_norm": 4.670098304748535, + "learning_rate": 0.0004467, + "loss": 3.0031, + "step": 1506 + }, + { + "epoch": 2.4112, + "grad_norm": 5.129536151885986, + "learning_rate": 0.00044699999999999997, + "loss": 2.7867, + "step": 1507 + }, + { + "epoch": 2.4128, + "grad_norm": 5.81210470199585, + "learning_rate": 0.0004473, + "loss": 2.9749, + "step": 1508 + }, + { + "epoch": 2.4144, + "grad_norm": 2.4536852836608887, + "learning_rate": 0.0004476, + "loss": 2.6915, + "step": 1509 + }, + { + "epoch": 2.416, + "grad_norm": 2.283811569213867, + "learning_rate": 0.0004479, + "loss": 2.6715, + "step": 1510 + }, + { + "epoch": 2.4176, + "grad_norm": 1.1368249654769897, + "learning_rate": 0.00044819999999999994, + "loss": 2.5841, + "step": 1511 + }, + { + "epoch": 2.4192, + "grad_norm": 2.950226068496704, + "learning_rate": 0.0004485, + "loss": 2.7656, + "step": 1512 + }, + { + "epoch": 2.4208, + "grad_norm": 3.2112069129943848, + "learning_rate": 0.00044879999999999996, + "loss": 2.5429, + "step": 1513 + }, + { + "epoch": 2.4224, + "grad_norm": 3.645280361175537, + "learning_rate": 0.0004491, + "loss": 2.5638, + "step": 1514 + }, + { + "epoch": 2.424, + "grad_norm": 1.3241567611694336, + "learning_rate": 0.0004494, + "loss": 2.8091, + "step": 1515 + }, + { + "epoch": 2.4256, + "grad_norm": 2.1697170734405518, + "learning_rate": 0.0004497, + "loss": 2.5652, + "step": 1516 + }, + { + "epoch": 2.4272, + "grad_norm": 0.6105273962020874, + "learning_rate": 0.00045, + "loss": 2.5103, + "step": 1517 + }, + { + "epoch": 2.4288, + "grad_norm": 0.6202393174171448, + "learning_rate": 0.00045029999999999994, + "loss": 2.6103, + "step": 1518 + }, + { + "epoch": 2.4304, + "grad_norm": 1.3199430704116821, + "learning_rate": 0.00045059999999999995, + "loss": 2.6113, + "step": 1519 + }, + { + "epoch": 2.432, + "grad_norm": 1.5840743780136108, + "learning_rate": 0.0004508999999999999, + "loss": 2.6316, + "step": 1520 + }, + { + "epoch": 2.4336, + "grad_norm": 1.0554815530776978, + "learning_rate": 0.00045119999999999996, + "loss": 2.483, + "step": 1521 + }, + { + "epoch": 2.4352, + "grad_norm": 2.019838333129883, + "learning_rate": 0.0004514999999999999, + "loss": 2.623, + "step": 1522 + }, + { + "epoch": 2.4368, + "grad_norm": 1.290846586227417, + "learning_rate": 0.0004518, + "loss": 2.562, + "step": 1523 + }, + { + "epoch": 2.4384, + "grad_norm": 1.3342317342758179, + "learning_rate": 0.00045209999999999993, + "loss": 2.7231, + "step": 1524 + }, + { + "epoch": 2.44, + "grad_norm": 2.423088788986206, + "learning_rate": 0.00045239999999999994, + "loss": 2.6006, + "step": 1525 + }, + { + "epoch": 2.4416, + "grad_norm": 1.3409308195114136, + "learning_rate": 0.00045269999999999994, + "loss": 2.6124, + "step": 1526 + }, + { + "epoch": 2.4432, + "grad_norm": 1.8538460731506348, + "learning_rate": 0.00045299999999999995, + "loss": 2.3889, + "step": 1527 + }, + { + "epoch": 2.4448, + "grad_norm": 1.0645954608917236, + "learning_rate": 0.0004532999999999999, + "loss": 2.4413, + "step": 1528 + }, + { + "epoch": 2.4464, + "grad_norm": 1.1003177165985107, + "learning_rate": 0.00045359999999999997, + "loss": 2.7019, + "step": 1529 + }, + { + "epoch": 2.448, + "grad_norm": 1.1214135885238647, + "learning_rate": 0.0004538999999999999, + "loss": 2.6847, + "step": 1530 + }, + { + "epoch": 2.4496, + "grad_norm": 0.5774716734886169, + "learning_rate": 0.0004542, + "loss": 2.7017, + "step": 1531 + }, + { + "epoch": 2.4512, + "grad_norm": 4.226199150085449, + "learning_rate": 0.00045449999999999993, + "loss": 2.5284, + "step": 1532 + }, + { + "epoch": 2.4528, + "grad_norm": 1.6718403100967407, + "learning_rate": 0.00045479999999999994, + "loss": 2.5032, + "step": 1533 + }, + { + "epoch": 2.4544, + "grad_norm": 0.8635501861572266, + "learning_rate": 0.00045509999999999995, + "loss": 2.4638, + "step": 1534 + }, + { + "epoch": 2.456, + "grad_norm": 0.8180009722709656, + "learning_rate": 0.00045539999999999996, + "loss": 2.5861, + "step": 1535 + }, + { + "epoch": 2.4576000000000002, + "grad_norm": 1.2235063314437866, + "learning_rate": 0.0004556999999999999, + "loss": 2.6967, + "step": 1536 + }, + { + "epoch": 2.4592, + "grad_norm": 1.6099307537078857, + "learning_rate": 0.00045599999999999997, + "loss": 2.7573, + "step": 1537 + }, + { + "epoch": 2.4608, + "grad_norm": 0.6623131036758423, + "learning_rate": 0.0004562999999999999, + "loss": 2.3741, + "step": 1538 + }, + { + "epoch": 2.4624, + "grad_norm": 0.9379790425300598, + "learning_rate": 0.0004566, + "loss": 2.6461, + "step": 1539 + }, + { + "epoch": 2.464, + "grad_norm": 1.9863440990447998, + "learning_rate": 0.00045689999999999994, + "loss": 2.5278, + "step": 1540 + }, + { + "epoch": 2.4656000000000002, + "grad_norm": 1.1242375373840332, + "learning_rate": 0.00045719999999999995, + "loss": 2.6311, + "step": 1541 + }, + { + "epoch": 2.4672, + "grad_norm": 1.4005846977233887, + "learning_rate": 0.00045749999999999995, + "loss": 2.5263, + "step": 1542 + }, + { + "epoch": 2.4688, + "grad_norm": 3.9240052700042725, + "learning_rate": 0.00045779999999999996, + "loss": 2.7778, + "step": 1543 + }, + { + "epoch": 2.4704, + "grad_norm": 1.3024487495422363, + "learning_rate": 0.0004580999999999999, + "loss": 2.8233, + "step": 1544 + }, + { + "epoch": 2.472, + "grad_norm": 2.5846259593963623, + "learning_rate": 0.0004584, + "loss": 2.5476, + "step": 1545 + }, + { + "epoch": 2.4736000000000002, + "grad_norm": 0.9445663690567017, + "learning_rate": 0.00045869999999999993, + "loss": 2.7712, + "step": 1546 + }, + { + "epoch": 2.4752, + "grad_norm": 2.509721279144287, + "learning_rate": 0.000459, + "loss": 2.6738, + "step": 1547 + }, + { + "epoch": 2.4768, + "grad_norm": 2.4532852172851562, + "learning_rate": 0.00045929999999999994, + "loss": 2.7306, + "step": 1548 + }, + { + "epoch": 2.4784, + "grad_norm": 2.668494462966919, + "learning_rate": 0.00045959999999999995, + "loss": 2.7596, + "step": 1549 + }, + { + "epoch": 2.48, + "grad_norm": NaN, + "learning_rate": 0.00045959999999999995, + "loss": 2.7976, + "step": 1550 + }, + { + "epoch": 2.4816, + "grad_norm": 6.9989013671875, + "learning_rate": 0.00045989999999999996, + "loss": 3.2758, + "step": 1551 + }, + { + "epoch": 2.4832, + "grad_norm": 5.862189769744873, + "learning_rate": 0.00046019999999999996, + "loss": 3.071, + "step": 1552 + }, + { + "epoch": 2.4848, + "grad_norm": 8.821229934692383, + "learning_rate": 0.0004604999999999999, + "loss": 3.192, + "step": 1553 + }, + { + "epoch": 2.4864, + "grad_norm": 5.29599666595459, + "learning_rate": 0.0004608, + "loss": 3.1139, + "step": 1554 + }, + { + "epoch": 2.488, + "grad_norm": 3.378632068634033, + "learning_rate": 0.00046109999999999993, + "loss": 3.1756, + "step": 1555 + }, + { + "epoch": 2.4896, + "grad_norm": 7.168362140655518, + "learning_rate": 0.0004614, + "loss": 3.141, + "step": 1556 + }, + { + "epoch": 2.4912, + "grad_norm": 2.826796770095825, + "learning_rate": 0.00046169999999999995, + "loss": 2.991, + "step": 1557 + }, + { + "epoch": 2.4928, + "grad_norm": 1.08731210231781, + "learning_rate": 0.00046199999999999995, + "loss": 2.7976, + "step": 1558 + }, + { + "epoch": 2.4944, + "grad_norm": 1.3502941131591797, + "learning_rate": 0.00046229999999999996, + "loss": 2.8825, + "step": 1559 + }, + { + "epoch": 2.496, + "grad_norm": 0.9165865182876587, + "learning_rate": 0.00046259999999999997, + "loss": 2.7296, + "step": 1560 + }, + { + "epoch": 2.4976, + "grad_norm": 0.846776008605957, + "learning_rate": 0.0004628999999999999, + "loss": 2.9291, + "step": 1561 + }, + { + "epoch": 2.4992, + "grad_norm": 1.5572160482406616, + "learning_rate": 0.0004632, + "loss": 2.7399, + "step": 1562 + }, + { + "epoch": 2.5008, + "grad_norm": 1.5787986516952515, + "learning_rate": 0.00046349999999999994, + "loss": 2.839, + "step": 1563 + }, + { + "epoch": 2.5023999999999997, + "grad_norm": 2.3490312099456787, + "learning_rate": 0.0004638, + "loss": 3.2249, + "step": 1564 + }, + { + "epoch": 2.504, + "grad_norm": 5.96341609954834, + "learning_rate": 0.00046409999999999995, + "loss": 2.565, + "step": 1565 + }, + { + "epoch": 2.5056000000000003, + "grad_norm": 6.891876697540283, + "learning_rate": 0.00046439999999999996, + "loss": 2.7132, + "step": 1566 + }, + { + "epoch": 2.5072, + "grad_norm": 8.918244361877441, + "learning_rate": 0.00046469999999999997, + "loss": 2.7519, + "step": 1567 + }, + { + "epoch": 2.5088, + "grad_norm": 9.990086555480957, + "learning_rate": 0.00046499999999999997, + "loss": 2.7456, + "step": 1568 + }, + { + "epoch": 2.5103999999999997, + "grad_norm": 8.708830833435059, + "learning_rate": 0.0004652999999999999, + "loss": 2.6391, + "step": 1569 + }, + { + "epoch": 2.512, + "grad_norm": 6.798332214355469, + "learning_rate": 0.0004656, + "loss": 2.6381, + "step": 1570 + }, + { + "epoch": 2.5136, + "grad_norm": 7.431624412536621, + "learning_rate": 0.00046589999999999994, + "loss": 2.6923, + "step": 1571 + }, + { + "epoch": 2.5152, + "grad_norm": 6.813911437988281, + "learning_rate": 0.00046619999999999995, + "loss": 2.7874, + "step": 1572 + }, + { + "epoch": 2.5168, + "grad_norm": 3.9901123046875, + "learning_rate": 0.00046649999999999996, + "loss": 2.6047, + "step": 1573 + }, + { + "epoch": 2.5183999999999997, + "grad_norm": 2.224478006362915, + "learning_rate": 0.00046679999999999996, + "loss": 2.8424, + "step": 1574 + }, + { + "epoch": 2.52, + "grad_norm": 4.089131832122803, + "learning_rate": 0.0004670999999999999, + "loss": 2.8194, + "step": 1575 + }, + { + "epoch": 2.5216, + "grad_norm": 2.967945098876953, + "learning_rate": 0.0004674, + "loss": 2.6133, + "step": 1576 + }, + { + "epoch": 2.5232, + "grad_norm": 2.259937286376953, + "learning_rate": 0.00046769999999999993, + "loss": 2.6445, + "step": 1577 + }, + { + "epoch": 2.5248, + "grad_norm": 2.3867623805999756, + "learning_rate": 0.000468, + "loss": 2.6527, + "step": 1578 + }, + { + "epoch": 2.5263999999999998, + "grad_norm": 0.7020841836929321, + "learning_rate": 0.00046829999999999994, + "loss": 2.6492, + "step": 1579 + }, + { + "epoch": 2.528, + "grad_norm": 2.115121841430664, + "learning_rate": 0.00046859999999999995, + "loss": 2.5377, + "step": 1580 + }, + { + "epoch": 2.5296, + "grad_norm": 1.013954997062683, + "learning_rate": 0.00046889999999999996, + "loss": 2.4602, + "step": 1581 + }, + { + "epoch": 2.5312, + "grad_norm": 5.609471797943115, + "learning_rate": 0.00046919999999999997, + "loss": 2.7264, + "step": 1582 + }, + { + "epoch": 2.5328, + "grad_norm": 1.7620587348937988, + "learning_rate": 0.0004694999999999999, + "loss": 2.7404, + "step": 1583 + }, + { + "epoch": 2.5343999999999998, + "grad_norm": 2.578216075897217, + "learning_rate": 0.0004698, + "loss": 2.6936, + "step": 1584 + }, + { + "epoch": 2.536, + "grad_norm": 2.9530296325683594, + "learning_rate": 0.00047009999999999993, + "loss": 2.7589, + "step": 1585 + }, + { + "epoch": 2.5376, + "grad_norm": 2.248521566390991, + "learning_rate": 0.0004704, + "loss": 2.6856, + "step": 1586 + }, + { + "epoch": 2.5392, + "grad_norm": 0.7185359001159668, + "learning_rate": 0.00047069999999999995, + "loss": 2.5873, + "step": 1587 + }, + { + "epoch": 2.5408, + "grad_norm": 0.950563371181488, + "learning_rate": 0.00047099999999999996, + "loss": 2.702, + "step": 1588 + }, + { + "epoch": 2.5423999999999998, + "grad_norm": 4.2564191818237305, + "learning_rate": 0.00047129999999999996, + "loss": 2.5006, + "step": 1589 + }, + { + "epoch": 2.544, + "grad_norm": 0.7628245949745178, + "learning_rate": 0.00047159999999999997, + "loss": 2.6586, + "step": 1590 + }, + { + "epoch": 2.5456, + "grad_norm": 3.4643051624298096, + "learning_rate": 0.0004718999999999999, + "loss": 2.5925, + "step": 1591 + }, + { + "epoch": 2.5472, + "grad_norm": 2.054396152496338, + "learning_rate": 0.0004722, + "loss": 2.6902, + "step": 1592 + }, + { + "epoch": 2.5488, + "grad_norm": 1.0989121198654175, + "learning_rate": 0.00047249999999999994, + "loss": 2.4568, + "step": 1593 + }, + { + "epoch": 2.5504, + "grad_norm": 2.788877248764038, + "learning_rate": 0.0004728, + "loss": 2.4971, + "step": 1594 + }, + { + "epoch": 2.552, + "grad_norm": 2.4059903621673584, + "learning_rate": 0.00047309999999999995, + "loss": 2.6616, + "step": 1595 + }, + { + "epoch": 2.5536, + "grad_norm": 2.420016288757324, + "learning_rate": 0.00047339999999999996, + "loss": 2.6371, + "step": 1596 + }, + { + "epoch": 2.5552, + "grad_norm": 3.306741952896118, + "learning_rate": 0.00047369999999999997, + "loss": 2.7256, + "step": 1597 + }, + { + "epoch": 2.5568, + "grad_norm": 1.3698875904083252, + "learning_rate": 0.000474, + "loss": 2.6461, + "step": 1598 + }, + { + "epoch": 2.5584, + "grad_norm": 3.0524322986602783, + "learning_rate": 0.00047429999999999993, + "loss": 2.8663, + "step": 1599 + }, + { + "epoch": 2.56, + "grad_norm": 4.398044586181641, + "learning_rate": 0.0004746, + "loss": 3.3413, + "step": 1600 + }, + { + "epoch": 2.5616, + "grad_norm": 11.20898151397705, + "learning_rate": 0.00047489999999999994, + "loss": 3.4181, + "step": 1601 + }, + { + "epoch": 2.5632, + "grad_norm": 2.485570192337036, + "learning_rate": 0.0004752, + "loss": 2.9986, + "step": 1602 + }, + { + "epoch": 2.5648, + "grad_norm": 9.538570404052734, + "learning_rate": 0.00047549999999999996, + "loss": 3.2206, + "step": 1603 + }, + { + "epoch": 2.5664, + "grad_norm": 4.369180679321289, + "learning_rate": 0.00047579999999999996, + "loss": 2.9533, + "step": 1604 + }, + { + "epoch": 2.568, + "grad_norm": 9.760275840759277, + "learning_rate": 0.00047609999999999997, + "loss": 3.2309, + "step": 1605 + }, + { + "epoch": 2.5696, + "grad_norm": 1.91250741481781, + "learning_rate": 0.0004764, + "loss": 2.7691, + "step": 1606 + }, + { + "epoch": 2.5712, + "grad_norm": 0.9459549188613892, + "learning_rate": 0.00047669999999999993, + "loss": 2.8651, + "step": 1607 + }, + { + "epoch": 2.5728, + "grad_norm": 0.77849942445755, + "learning_rate": 0.000477, + "loss": 2.9374, + "step": 1608 + }, + { + "epoch": 2.5744, + "grad_norm": 3.6616194248199463, + "learning_rate": 0.00047729999999999995, + "loss": 2.8133, + "step": 1609 + }, + { + "epoch": 2.576, + "grad_norm": 3.393961191177368, + "learning_rate": 0.0004776, + "loss": 2.7227, + "step": 1610 + }, + { + "epoch": 2.5776, + "grad_norm": 1.0827534198760986, + "learning_rate": 0.00047789999999999996, + "loss": 2.9257, + "step": 1611 + }, + { + "epoch": 2.5792, + "grad_norm": 1.3316205739974976, + "learning_rate": 0.00047819999999999997, + "loss": 2.7961, + "step": 1612 + }, + { + "epoch": 2.5808, + "grad_norm": 2.686330795288086, + "learning_rate": 0.0004785, + "loss": 2.6238, + "step": 1613 + }, + { + "epoch": 2.5824, + "grad_norm": 0.7391074895858765, + "learning_rate": 0.0004788, + "loss": 2.5468, + "step": 1614 + }, + { + "epoch": 2.584, + "grad_norm": 3.1132357120513916, + "learning_rate": 0.00047909999999999994, + "loss": 2.5595, + "step": 1615 + }, + { + "epoch": 2.5856, + "grad_norm": 1.5804221630096436, + "learning_rate": 0.0004794, + "loss": 2.6619, + "step": 1616 + }, + { + "epoch": 2.5872, + "grad_norm": 0.633274257183075, + "learning_rate": 0.00047969999999999995, + "loss": 2.6271, + "step": 1617 + }, + { + "epoch": 2.5888, + "grad_norm": 0.8176840543746948, + "learning_rate": 0.00047999999999999996, + "loss": 2.6379, + "step": 1618 + }, + { + "epoch": 2.5904, + "grad_norm": 0.6381813287734985, + "learning_rate": 0.00048029999999999997, + "loss": 2.6628, + "step": 1619 + }, + { + "epoch": 2.592, + "grad_norm": 0.7543144226074219, + "learning_rate": 0.00048059999999999997, + "loss": 2.6104, + "step": 1620 + }, + { + "epoch": 2.5936, + "grad_norm": 1.0972641706466675, + "learning_rate": 0.0004808999999999999, + "loss": 2.6753, + "step": 1621 + }, + { + "epoch": 2.5952, + "grad_norm": 1.9060057401657104, + "learning_rate": 0.0004812, + "loss": 2.5652, + "step": 1622 + }, + { + "epoch": 2.5968, + "grad_norm": 1.4949616193771362, + "learning_rate": 0.00048149999999999994, + "loss": 2.6658, + "step": 1623 + }, + { + "epoch": 2.5984, + "grad_norm": 1.245529055595398, + "learning_rate": 0.0004818, + "loss": 2.6667, + "step": 1624 + }, + { + "epoch": 2.6, + "grad_norm": 0.5955816507339478, + "learning_rate": 0.00048209999999999995, + "loss": 2.6539, + "step": 1625 + }, + { + "epoch": 2.6016, + "grad_norm": 4.172550201416016, + "learning_rate": 0.00048239999999999996, + "loss": 2.6025, + "step": 1626 + }, + { + "epoch": 2.6032, + "grad_norm": 1.587763786315918, + "learning_rate": 0.00048269999999999997, + "loss": 2.5869, + "step": 1627 + }, + { + "epoch": 2.6048, + "grad_norm": 1.7045336961746216, + "learning_rate": 0.000483, + "loss": 2.5114, + "step": 1628 + }, + { + "epoch": 2.6064, + "grad_norm": 1.0802538394927979, + "learning_rate": 0.00048329999999999993, + "loss": 2.6275, + "step": 1629 + }, + { + "epoch": 2.608, + "grad_norm": 1.2180992364883423, + "learning_rate": 0.0004836, + "loss": 2.6264, + "step": 1630 + }, + { + "epoch": 2.6096, + "grad_norm": 0.6582339406013489, + "learning_rate": 0.00048389999999999994, + "loss": 2.6329, + "step": 1631 + }, + { + "epoch": 2.6112, + "grad_norm": 0.9185123443603516, + "learning_rate": 0.0004842, + "loss": 2.5624, + "step": 1632 + }, + { + "epoch": 2.6128, + "grad_norm": 0.6815674304962158, + "learning_rate": 0.00048449999999999996, + "loss": 2.4889, + "step": 1633 + }, + { + "epoch": 2.6144, + "grad_norm": 1.176547646522522, + "learning_rate": 0.00048479999999999997, + "loss": 2.5491, + "step": 1634 + }, + { + "epoch": 2.616, + "grad_norm": 7.645500659942627, + "learning_rate": 0.0004851, + "loss": 2.5999, + "step": 1635 + }, + { + "epoch": 2.6176, + "grad_norm": 0.7159596681594849, + "learning_rate": 0.0004854, + "loss": 2.6153, + "step": 1636 + }, + { + "epoch": 2.6192, + "grad_norm": 0.8762513995170593, + "learning_rate": 0.00048569999999999993, + "loss": 2.4501, + "step": 1637 + }, + { + "epoch": 2.6208, + "grad_norm": 2.3339498043060303, + "learning_rate": 0.000486, + "loss": 2.7459, + "step": 1638 + }, + { + "epoch": 2.6224, + "grad_norm": 3.838904619216919, + "learning_rate": 0.00048629999999999995, + "loss": 2.7411, + "step": 1639 + }, + { + "epoch": 2.624, + "grad_norm": 2.539436101913452, + "learning_rate": 0.0004866, + "loss": 2.4974, + "step": 1640 + }, + { + "epoch": 2.6256, + "grad_norm": 1.0629886388778687, + "learning_rate": 0.00048689999999999996, + "loss": 2.6559, + "step": 1641 + }, + { + "epoch": 2.6272, + "grad_norm": 0.8675243258476257, + "learning_rate": 0.00048719999999999997, + "loss": 2.5446, + "step": 1642 + }, + { + "epoch": 2.6288, + "grad_norm": 1.6415584087371826, + "learning_rate": 0.0004875, + "loss": 2.3971, + "step": 1643 + }, + { + "epoch": 2.6304, + "grad_norm": 0.995968222618103, + "learning_rate": 0.00048779999999999993, + "loss": 2.5962, + "step": 1644 + }, + { + "epoch": 2.632, + "grad_norm": 0.7832150459289551, + "learning_rate": 0.00048809999999999994, + "loss": 2.6166, + "step": 1645 + }, + { + "epoch": 2.6336, + "grad_norm": 2.025477170944214, + "learning_rate": 0.0004883999999999999, + "loss": 2.7621, + "step": 1646 + }, + { + "epoch": 2.6352, + "grad_norm": 1.517505407333374, + "learning_rate": 0.0004887, + "loss": 2.8696, + "step": 1647 + }, + { + "epoch": 2.6368, + "grad_norm": 1.3974066972732544, + "learning_rate": 0.000489, + "loss": 2.6362, + "step": 1648 + }, + { + "epoch": 2.6384, + "grad_norm": 1.3067556619644165, + "learning_rate": 0.0004892999999999999, + "loss": 2.6683, + "step": 1649 + }, + { + "epoch": 2.64, + "grad_norm": NaN, + "learning_rate": 0.0004892999999999999, + "loss": 2.3979, + "step": 1650 + }, + { + "epoch": 2.6416, + "grad_norm": 7.526185035705566, + "learning_rate": 0.0004896, + "loss": 3.2359, + "step": 1651 + }, + { + "epoch": 2.6432, + "grad_norm": 1.8751859664916992, + "learning_rate": 0.0004898999999999999, + "loss": 3.0194, + "step": 1652 + }, + { + "epoch": 2.6448, + "grad_norm": 1.6169847249984741, + "learning_rate": 0.0004901999999999999, + "loss": 2.9794, + "step": 1653 + }, + { + "epoch": 2.6464, + "grad_norm": 3.470064878463745, + "learning_rate": 0.0004904999999999999, + "loss": 3.23, + "step": 1654 + }, + { + "epoch": 2.648, + "grad_norm": 1.2958786487579346, + "learning_rate": 0.0004907999999999999, + "loss": 3.0643, + "step": 1655 + }, + { + "epoch": 2.6496, + "grad_norm": 3.670785903930664, + "learning_rate": 0.0004911, + "loss": 2.6688, + "step": 1656 + }, + { + "epoch": 2.6512000000000002, + "grad_norm": 5.36950159072876, + "learning_rate": 0.0004913999999999999, + "loss": 2.6845, + "step": 1657 + }, + { + "epoch": 2.6528, + "grad_norm": 5.3715338706970215, + "learning_rate": 0.0004917, + "loss": 2.807, + "step": 1658 + }, + { + "epoch": 2.6544, + "grad_norm": 1.8697935342788696, + "learning_rate": 0.0004919999999999999, + "loss": 2.984, + "step": 1659 + }, + { + "epoch": 2.656, + "grad_norm": 2.8437161445617676, + "learning_rate": 0.0004923, + "loss": 2.7468, + "step": 1660 + }, + { + "epoch": 2.6576, + "grad_norm": 1.6597505807876587, + "learning_rate": 0.0004925999999999999, + "loss": 2.6693, + "step": 1661 + }, + { + "epoch": 2.6592000000000002, + "grad_norm": 1.2704110145568848, + "learning_rate": 0.0004929, + "loss": 2.7273, + "step": 1662 + }, + { + "epoch": 2.6608, + "grad_norm": 1.4806967973709106, + "learning_rate": 0.0004932, + "loss": 2.6624, + "step": 1663 + }, + { + "epoch": 2.6624, + "grad_norm": 3.6257212162017822, + "learning_rate": 0.0004935, + "loss": 2.7391, + "step": 1664 + }, + { + "epoch": 2.664, + "grad_norm": 3.1111960411071777, + "learning_rate": 0.0004938, + "loss": 2.5829, + "step": 1665 + }, + { + "epoch": 2.6656, + "grad_norm": 3.902526617050171, + "learning_rate": 0.0004940999999999999, + "loss": 2.5868, + "step": 1666 + }, + { + "epoch": 2.6672000000000002, + "grad_norm": 5.665694236755371, + "learning_rate": 0.0004944, + "loss": 2.7117, + "step": 1667 + }, + { + "epoch": 2.6688, + "grad_norm": 2.3080103397369385, + "learning_rate": 0.0004946999999999999, + "loss": 2.512, + "step": 1668 + }, + { + "epoch": 2.6704, + "grad_norm": 1.7370351552963257, + "learning_rate": 0.0004949999999999999, + "loss": 2.5534, + "step": 1669 + }, + { + "epoch": 2.672, + "grad_norm": 1.2155604362487793, + "learning_rate": 0.0004953, + "loss": 2.5723, + "step": 1670 + }, + { + "epoch": 2.6736, + "grad_norm": 1.4938063621520996, + "learning_rate": 0.0004955999999999999, + "loss": 2.4879, + "step": 1671 + }, + { + "epoch": 2.6752000000000002, + "grad_norm": 4.0337419509887695, + "learning_rate": 0.0004959, + "loss": 2.6292, + "step": 1672 + }, + { + "epoch": 2.6768, + "grad_norm": 2.6993560791015625, + "learning_rate": 0.0004961999999999999, + "loss": 2.7822, + "step": 1673 + }, + { + "epoch": 2.6784, + "grad_norm": 1.474229097366333, + "learning_rate": 0.0004965, + "loss": 2.5086, + "step": 1674 + }, + { + "epoch": 2.68, + "grad_norm": 2.241969108581543, + "learning_rate": 0.0004967999999999999, + "loss": 2.408, + "step": 1675 + }, + { + "epoch": 2.6816, + "grad_norm": 1.253790259361267, + "learning_rate": 0.0004971, + "loss": 2.5079, + "step": 1676 + }, + { + "epoch": 2.6832000000000003, + "grad_norm": 2.400538206100464, + "learning_rate": 0.0004974, + "loss": 2.5499, + "step": 1677 + }, + { + "epoch": 2.6848, + "grad_norm": 2.866384267807007, + "learning_rate": 0.0004977, + "loss": 2.4777, + "step": 1678 + }, + { + "epoch": 2.6864, + "grad_norm": 3.607046604156494, + "learning_rate": 0.000498, + "loss": 2.4503, + "step": 1679 + }, + { + "epoch": 2.6879999999999997, + "grad_norm": 4.765000343322754, + "learning_rate": 0.0004982999999999999, + "loss": 2.5933, + "step": 1680 + }, + { + "epoch": 2.6896, + "grad_norm": 2.517519950866699, + "learning_rate": 0.0004986, + "loss": 2.5683, + "step": 1681 + }, + { + "epoch": 2.6912000000000003, + "grad_norm": 0.8712690472602844, + "learning_rate": 0.0004988999999999999, + "loss": 2.7788, + "step": 1682 + }, + { + "epoch": 2.6928, + "grad_norm": 1.0803961753845215, + "learning_rate": 0.0004991999999999999, + "loss": 2.4722, + "step": 1683 + }, + { + "epoch": 2.6944, + "grad_norm": 0.6999735832214355, + "learning_rate": 0.0004994999999999999, + "loss": 2.486, + "step": 1684 + }, + { + "epoch": 2.6959999999999997, + "grad_norm": 1.2827504873275757, + "learning_rate": 0.0004997999999999999, + "loss": 2.48, + "step": 1685 + }, + { + "epoch": 2.6976, + "grad_norm": 0.8345691561698914, + "learning_rate": 0.0005001, + "loss": 2.3659, + "step": 1686 + }, + { + "epoch": 2.6992000000000003, + "grad_norm": 0.798511266708374, + "learning_rate": 0.0005003999999999999, + "loss": 2.4629, + "step": 1687 + }, + { + "epoch": 2.7008, + "grad_norm": 1.9487131834030151, + "learning_rate": 0.0005007, + "loss": 2.523, + "step": 1688 + }, + { + "epoch": 2.7024, + "grad_norm": 0.9477632641792297, + "learning_rate": 0.0005009999999999999, + "loss": 2.4581, + "step": 1689 + }, + { + "epoch": 2.7039999999999997, + "grad_norm": 1.6193513870239258, + "learning_rate": 0.0005013, + "loss": 2.5933, + "step": 1690 + }, + { + "epoch": 2.7056, + "grad_norm": 1.8013644218444824, + "learning_rate": 0.0005015999999999999, + "loss": 2.8349, + "step": 1691 + }, + { + "epoch": 2.7072000000000003, + "grad_norm": 2.3376691341400146, + "learning_rate": 0.0005019, + "loss": 2.8021, + "step": 1692 + }, + { + "epoch": 2.7088, + "grad_norm": 3.678694009780884, + "learning_rate": 0.0005022, + "loss": 2.6318, + "step": 1693 + }, + { + "epoch": 2.7104, + "grad_norm": 1.4150782823562622, + "learning_rate": 0.0005025, + "loss": 2.8046, + "step": 1694 + }, + { + "epoch": 2.7119999999999997, + "grad_norm": 3.379930019378662, + "learning_rate": 0.0005028, + "loss": 2.7794, + "step": 1695 + }, + { + "epoch": 2.7136, + "grad_norm": 1.2813504934310913, + "learning_rate": 0.0005030999999999999, + "loss": 2.6174, + "step": 1696 + }, + { + "epoch": 2.7152, + "grad_norm": 0.9052807688713074, + "learning_rate": 0.0005034, + "loss": 2.9687, + "step": 1697 + }, + { + "epoch": 2.7168, + "grad_norm": 2.1118052005767822, + "learning_rate": 0.0005036999999999999, + "loss": 2.773, + "step": 1698 + }, + { + "epoch": 2.7184, + "grad_norm": 1.609099268913269, + "learning_rate": 0.0005039999999999999, + "loss": 2.7635, + "step": 1699 + }, + { + "epoch": 2.7199999999999998, + "grad_norm": NaN, + "learning_rate": 0.0005039999999999999, + "loss": 3.1193, + "step": 1700 + }, + { + "epoch": 2.7216, + "grad_norm": 13.232641220092773, + "learning_rate": 0.0005043, + "loss": 3.4776, + "step": 1701 + }, + { + "epoch": 2.7232, + "grad_norm": 7.2539873123168945, + "learning_rate": 0.0005045999999999999, + "loss": 3.0714, + "step": 1702 + }, + { + "epoch": 2.7248, + "grad_norm": 5.081768035888672, + "learning_rate": 0.0005049, + "loss": 3.1389, + "step": 1703 + }, + { + "epoch": 2.7264, + "grad_norm": 2.3188507556915283, + "learning_rate": 0.0005051999999999999, + "loss": 3.0545, + "step": 1704 + }, + { + "epoch": 2.7279999999999998, + "grad_norm": 3.5873944759368896, + "learning_rate": 0.0005055, + "loss": 3.0174, + "step": 1705 + }, + { + "epoch": 2.7296, + "grad_norm": 4.156705379486084, + "learning_rate": 0.0005057999999999999, + "loss": 2.8348, + "step": 1706 + }, + { + "epoch": 2.7312, + "grad_norm": 2.0585386753082275, + "learning_rate": 0.0005061, + "loss": 2.9599, + "step": 1707 + }, + { + "epoch": 2.7328, + "grad_norm": 4.402909278869629, + "learning_rate": 0.0005064, + "loss": 2.9435, + "step": 1708 + }, + { + "epoch": 2.7344, + "grad_norm": 1.3773367404937744, + "learning_rate": 0.0005067, + "loss": 2.7322, + "step": 1709 + }, + { + "epoch": 2.7359999999999998, + "grad_norm": 1.3306688070297241, + "learning_rate": 0.000507, + "loss": 2.6582, + "step": 1710 + }, + { + "epoch": 2.7376, + "grad_norm": 6.887178897857666, + "learning_rate": 0.0005073, + "loss": 2.7469, + "step": 1711 + }, + { + "epoch": 2.7392, + "grad_norm": 5.434063911437988, + "learning_rate": 0.0005076, + "loss": 2.7386, + "step": 1712 + }, + { + "epoch": 2.7408, + "grad_norm": 2.059262275695801, + "learning_rate": 0.0005078999999999999, + "loss": 2.7037, + "step": 1713 + }, + { + "epoch": 2.7424, + "grad_norm": 3.2821879386901855, + "learning_rate": 0.0005082, + "loss": 2.6632, + "step": 1714 + }, + { + "epoch": 2.7439999999999998, + "grad_norm": 1.6987167596817017, + "learning_rate": 0.0005085, + "loss": 2.8453, + "step": 1715 + }, + { + "epoch": 2.7456, + "grad_norm": 1.0021125078201294, + "learning_rate": 0.0005087999999999999, + "loss": 2.7244, + "step": 1716 + }, + { + "epoch": 2.7472, + "grad_norm": 1.5165038108825684, + "learning_rate": 0.0005091, + "loss": 2.6291, + "step": 1717 + }, + { + "epoch": 2.7488, + "grad_norm": 5.8641767501831055, + "learning_rate": 0.0005093999999999999, + "loss": 2.8931, + "step": 1718 + }, + { + "epoch": 2.7504, + "grad_norm": 1.7871204614639282, + "learning_rate": 0.0005097, + "loss": 2.586, + "step": 1719 + }, + { + "epoch": 2.752, + "grad_norm": 1.9642881155014038, + "learning_rate": 0.0005099999999999999, + "loss": 2.593, + "step": 1720 + }, + { + "epoch": 2.7536, + "grad_norm": 1.1813987493515015, + "learning_rate": 0.0005103, + "loss": 2.6729, + "step": 1721 + }, + { + "epoch": 2.7552, + "grad_norm": 1.2921805381774902, + "learning_rate": 0.0005105999999999999, + "loss": 2.4355, + "step": 1722 + }, + { + "epoch": 2.7568, + "grad_norm": 0.7591195702552795, + "learning_rate": 0.0005109, + "loss": 2.5492, + "step": 1723 + }, + { + "epoch": 2.7584, + "grad_norm": 0.9900460243225098, + "learning_rate": 0.0005112, + "loss": 2.5241, + "step": 1724 + }, + { + "epoch": 2.76, + "grad_norm": 0.8340640068054199, + "learning_rate": 0.0005115, + "loss": 2.5323, + "step": 1725 + }, + { + "epoch": 2.7616, + "grad_norm": 0.8422091007232666, + "learning_rate": 0.0005118, + "loss": 2.6332, + "step": 1726 + }, + { + "epoch": 2.7632, + "grad_norm": 1.318156123161316, + "learning_rate": 0.0005120999999999999, + "loss": 2.5152, + "step": 1727 + }, + { + "epoch": 2.7648, + "grad_norm": 0.7750735282897949, + "learning_rate": 0.0005124, + "loss": 2.4252, + "step": 1728 + }, + { + "epoch": 2.7664, + "grad_norm": 1.509163737297058, + "learning_rate": 0.0005126999999999999, + "loss": 2.5499, + "step": 1729 + }, + { + "epoch": 2.768, + "grad_norm": 0.9951829314231873, + "learning_rate": 0.0005129999999999999, + "loss": 2.555, + "step": 1730 + }, + { + "epoch": 2.7696, + "grad_norm": 0.9104136228561401, + "learning_rate": 0.0005133, + "loss": 2.4509, + "step": 1731 + }, + { + "epoch": 2.7712, + "grad_norm": 1.038901925086975, + "learning_rate": 0.0005135999999999999, + "loss": 2.5816, + "step": 1732 + }, + { + "epoch": 2.7728, + "grad_norm": 2.0173180103302, + "learning_rate": 0.0005139, + "loss": 2.4119, + "step": 1733 + }, + { + "epoch": 2.7744, + "grad_norm": 2.1461172103881836, + "learning_rate": 0.0005141999999999999, + "loss": 2.4564, + "step": 1734 + }, + { + "epoch": 2.776, + "grad_norm": 0.8707074522972107, + "learning_rate": 0.0005145, + "loss": 2.5007, + "step": 1735 + }, + { + "epoch": 2.7776, + "grad_norm": 1.6499031782150269, + "learning_rate": 0.0005147999999999999, + "loss": 2.6064, + "step": 1736 + }, + { + "epoch": 2.7792, + "grad_norm": 0.86881422996521, + "learning_rate": 0.0005151, + "loss": 2.4206, + "step": 1737 + }, + { + "epoch": 2.7808, + "grad_norm": 0.8201207518577576, + "learning_rate": 0.0005154, + "loss": 2.3982, + "step": 1738 + }, + { + "epoch": 2.7824, + "grad_norm": 1.2034120559692383, + "learning_rate": 0.0005157, + "loss": 2.578, + "step": 1739 + }, + { + "epoch": 2.784, + "grad_norm": 0.7777190804481506, + "learning_rate": 0.000516, + "loss": 2.3112, + "step": 1740 + }, + { + "epoch": 2.7856, + "grad_norm": 0.82183837890625, + "learning_rate": 0.0005163, + "loss": 2.5456, + "step": 1741 + }, + { + "epoch": 2.7872, + "grad_norm": 1.6097009181976318, + "learning_rate": 0.0005166, + "loss": 2.4209, + "step": 1742 + }, + { + "epoch": 2.7888, + "grad_norm": 0.803339421749115, + "learning_rate": 0.0005168999999999999, + "loss": 2.6127, + "step": 1743 + }, + { + "epoch": 2.7904, + "grad_norm": 1.109894871711731, + "learning_rate": 0.0005172, + "loss": 2.7093, + "step": 1744 + }, + { + "epoch": 2.792, + "grad_norm": 3.9808509349823, + "learning_rate": 0.0005175, + "loss": 2.394, + "step": 1745 + }, + { + "epoch": 2.7936, + "grad_norm": 1.2084410190582275, + "learning_rate": 0.0005177999999999999, + "loss": 2.8151, + "step": 1746 + }, + { + "epoch": 2.7952, + "grad_norm": 1.0662058591842651, + "learning_rate": 0.0005181, + "loss": 2.5713, + "step": 1747 + }, + { + "epoch": 2.7968, + "grad_norm": 0.6551415324211121, + "learning_rate": 0.0005183999999999999, + "loss": 2.5872, + "step": 1748 + }, + { + "epoch": 2.7984, + "grad_norm": 1.0831410884857178, + "learning_rate": 0.0005187, + "loss": 2.4416, + "step": 1749 + }, + { + "epoch": 2.8, + "grad_norm": 1.6357882022857666, + "learning_rate": 0.0005189999999999999, + "loss": 3.438, + "step": 1750 + }, + { + "epoch": 2.8016, + "grad_norm": 15.630077362060547, + "learning_rate": 0.0005193, + "loss": 3.6123, + "step": 1751 + }, + { + "epoch": 2.8032, + "grad_norm": 4.824563980102539, + "learning_rate": 0.0005195999999999999, + "loss": 3.0755, + "step": 1752 + }, + { + "epoch": 2.8048, + "grad_norm": 3.7035696506500244, + "learning_rate": 0.0005199, + "loss": 3.1672, + "step": 1753 + }, + { + "epoch": 2.8064, + "grad_norm": 1.7311122417449951, + "learning_rate": 0.0005202, + "loss": 2.9867, + "step": 1754 + }, + { + "epoch": 2.808, + "grad_norm": 1.3840744495391846, + "learning_rate": 0.0005205, + "loss": 2.8987, + "step": 1755 + }, + { + "epoch": 2.8096, + "grad_norm": 5.936819076538086, + "learning_rate": 0.0005208, + "loss": 2.8573, + "step": 1756 + }, + { + "epoch": 2.8112, + "grad_norm": 4.638219356536865, + "learning_rate": 0.0005211, + "loss": 2.6146, + "step": 1757 + }, + { + "epoch": 2.8128, + "grad_norm": 5.056271553039551, + "learning_rate": 0.0005214, + "loss": 2.6315, + "step": 1758 + }, + { + "epoch": 2.8144, + "grad_norm": 6.352563381195068, + "learning_rate": 0.0005216999999999999, + "loss": 2.7745, + "step": 1759 + }, + { + "epoch": 2.816, + "grad_norm": 2.4548048973083496, + "learning_rate": 0.000522, + "loss": 2.7158, + "step": 1760 + }, + { + "epoch": 2.8176, + "grad_norm": 1.3754136562347412, + "learning_rate": 0.0005223, + "loss": 2.7262, + "step": 1761 + }, + { + "epoch": 2.8192, + "grad_norm": 1.041412353515625, + "learning_rate": 0.0005225999999999999, + "loss": 2.6656, + "step": 1762 + }, + { + "epoch": 2.8208, + "grad_norm": 2.493197441101074, + "learning_rate": 0.0005229, + "loss": 2.4963, + "step": 1763 + }, + { + "epoch": 2.8224, + "grad_norm": 0.7722348570823669, + "learning_rate": 0.0005231999999999999, + "loss": 2.5916, + "step": 1764 + }, + { + "epoch": 2.824, + "grad_norm": 2.9039366245269775, + "learning_rate": 0.0005235, + "loss": 2.5123, + "step": 1765 + }, + { + "epoch": 2.8256, + "grad_norm": 2.1121346950531006, + "learning_rate": 0.0005237999999999999, + "loss": 2.6058, + "step": 1766 + }, + { + "epoch": 2.8272, + "grad_norm": 0.9812942743301392, + "learning_rate": 0.0005241, + "loss": 2.4736, + "step": 1767 + }, + { + "epoch": 2.8288, + "grad_norm": 1.2289108037948608, + "learning_rate": 0.0005244, + "loss": 2.5537, + "step": 1768 + }, + { + "epoch": 2.8304, + "grad_norm": 3.8981456756591797, + "learning_rate": 0.0005247, + "loss": 2.6518, + "step": 1769 + }, + { + "epoch": 2.832, + "grad_norm": 2.036348342895508, + "learning_rate": 0.000525, + "loss": 2.5605, + "step": 1770 + }, + { + "epoch": 2.8336, + "grad_norm": 1.4323660135269165, + "learning_rate": 0.0005252999999999999, + "loss": 2.4834, + "step": 1771 + }, + { + "epoch": 2.8352, + "grad_norm": 1.9249393939971924, + "learning_rate": 0.0005256, + "loss": 2.5572, + "step": 1772 + }, + { + "epoch": 2.8368, + "grad_norm": 1.6773697137832642, + "learning_rate": 0.0005258999999999999, + "loss": 2.3995, + "step": 1773 + }, + { + "epoch": 2.8384, + "grad_norm": 1.1707098484039307, + "learning_rate": 0.0005262, + "loss": 2.5901, + "step": 1774 + }, + { + "epoch": 2.84, + "grad_norm": 0.7177858948707581, + "learning_rate": 0.0005265, + "loss": 2.5273, + "step": 1775 + }, + { + "epoch": 2.8416, + "grad_norm": 0.754425048828125, + "learning_rate": 0.0005267999999999999, + "loss": 2.4347, + "step": 1776 + }, + { + "epoch": 2.8432, + "grad_norm": 2.790383815765381, + "learning_rate": 0.0005271, + "loss": 2.5219, + "step": 1777 + }, + { + "epoch": 2.8448, + "grad_norm": 4.15537166595459, + "learning_rate": 0.0005273999999999999, + "loss": 2.5522, + "step": 1778 + }, + { + "epoch": 2.8464, + "grad_norm": 3.3372390270233154, + "learning_rate": 0.0005276999999999999, + "loss": 2.5126, + "step": 1779 + }, + { + "epoch": 2.848, + "grad_norm": 1.7653952836990356, + "learning_rate": 0.0005279999999999999, + "loss": 2.4318, + "step": 1780 + }, + { + "epoch": 2.8496, + "grad_norm": 1.198250412940979, + "learning_rate": 0.0005282999999999999, + "loss": 2.5744, + "step": 1781 + }, + { + "epoch": 2.8512, + "grad_norm": 2.5732247829437256, + "learning_rate": 0.0005286, + "loss": 2.4265, + "step": 1782 + }, + { + "epoch": 2.8528000000000002, + "grad_norm": 1.0462865829467773, + "learning_rate": 0.0005288999999999999, + "loss": 2.4897, + "step": 1783 + }, + { + "epoch": 2.8544, + "grad_norm": 0.891711950302124, + "learning_rate": 0.0005292, + "loss": 2.2267, + "step": 1784 + }, + { + "epoch": 2.856, + "grad_norm": 1.0404731035232544, + "learning_rate": 0.0005294999999999999, + "loss": 2.467, + "step": 1785 + }, + { + "epoch": 2.8576, + "grad_norm": 0.6127783060073853, + "learning_rate": 0.0005298, + "loss": 2.4414, + "step": 1786 + }, + { + "epoch": 2.8592, + "grad_norm": 2.071681022644043, + "learning_rate": 0.0005300999999999999, + "loss": 2.5055, + "step": 1787 + }, + { + "epoch": 2.8608000000000002, + "grad_norm": 1.3703131675720215, + "learning_rate": 0.0005304, + "loss": 2.4044, + "step": 1788 + }, + { + "epoch": 2.8624, + "grad_norm": 1.315367341041565, + "learning_rate": 0.0005306999999999999, + "loss": 2.7501, + "step": 1789 + }, + { + "epoch": 2.864, + "grad_norm": 1.6619805097579956, + "learning_rate": 0.000531, + "loss": 2.4622, + "step": 1790 + }, + { + "epoch": 2.8656, + "grad_norm": 0.6953228116035461, + "learning_rate": 0.0005313, + "loss": 2.5714, + "step": 1791 + }, + { + "epoch": 2.8672, + "grad_norm": 1.0954629182815552, + "learning_rate": 0.0005315999999999999, + "loss": 2.4817, + "step": 1792 + }, + { + "epoch": 2.8688000000000002, + "grad_norm": 2.2827250957489014, + "learning_rate": 0.0005319, + "loss": 2.5893, + "step": 1793 + }, + { + "epoch": 2.8704, + "grad_norm": 0.9044419527053833, + "learning_rate": 0.0005321999999999999, + "loss": 2.5491, + "step": 1794 + }, + { + "epoch": 2.872, + "grad_norm": 1.5412472486495972, + "learning_rate": 0.0005324999999999999, + "loss": 2.8427, + "step": 1795 + }, + { + "epoch": 2.8736, + "grad_norm": 1.7915852069854736, + "learning_rate": 0.0005327999999999999, + "loss": 2.7391, + "step": 1796 + }, + { + "epoch": 2.8752, + "grad_norm": 0.6722069382667542, + "learning_rate": 0.0005330999999999999, + "loss": 2.6614, + "step": 1797 + }, + { + "epoch": 2.8768000000000002, + "grad_norm": 3.3082215785980225, + "learning_rate": 0.0005334, + "loss": 2.8621, + "step": 1798 + }, + { + "epoch": 2.8784, + "grad_norm": 1.5828150510787964, + "learning_rate": 0.0005336999999999999, + "loss": 2.7225, + "step": 1799 + }, + { + "epoch": 2.88, + "grad_norm": NaN, + "learning_rate": 0.0005336999999999999, + "loss": 2.5517, + "step": 1800 + }, + { + "epoch": 2.8816, + "grad_norm": 11.708396911621094, + "learning_rate": 0.000534, + "loss": 3.5065, + "step": 1801 + }, + { + "epoch": 2.8832, + "grad_norm": 4.113176345825195, + "learning_rate": 0.0005342999999999999, + "loss": 3.1792, + "step": 1802 + }, + { + "epoch": 2.8848000000000003, + "grad_norm": 2.708540916442871, + "learning_rate": 0.0005346, + "loss": 3.097, + "step": 1803 + }, + { + "epoch": 2.8864, + "grad_norm": 1.2130831480026245, + "learning_rate": 0.0005348999999999999, + "loss": 2.9355, + "step": 1804 + }, + { + "epoch": 2.888, + "grad_norm": 3.173293113708496, + "learning_rate": 0.0005352, + "loss": 2.4884, + "step": 1805 + }, + { + "epoch": 2.8895999999999997, + "grad_norm": 1.250343918800354, + "learning_rate": 0.0005355, + "loss": 2.7102, + "step": 1806 + }, + { + "epoch": 2.8912, + "grad_norm": 1.3402652740478516, + "learning_rate": 0.0005358, + "loss": 2.6662, + "step": 1807 + }, + { + "epoch": 2.8928000000000003, + "grad_norm": 0.8727527856826782, + "learning_rate": 0.0005361, + "loss": 2.696, + "step": 1808 + }, + { + "epoch": 2.8944, + "grad_norm": 2.339231014251709, + "learning_rate": 0.0005363999999999999, + "loss": 2.4603, + "step": 1809 + }, + { + "epoch": 2.896, + "grad_norm": 13.718860626220703, + "learning_rate": 0.0005367, + "loss": 3.0912, + "step": 1810 + }, + { + "epoch": 2.8975999999999997, + "grad_norm": 3.7470011711120605, + "learning_rate": 0.0005369999999999999, + "loss": 2.4037, + "step": 1811 + }, + { + "epoch": 2.8992, + "grad_norm": 7.758830547332764, + "learning_rate": 0.0005372999999999999, + "loss": 2.7615, + "step": 1812 + }, + { + "epoch": 2.9008000000000003, + "grad_norm": 1.4061205387115479, + "learning_rate": 0.0005376, + "loss": 2.6428, + "step": 1813 + }, + { + "epoch": 2.9024, + "grad_norm": 1.3114807605743408, + "learning_rate": 0.0005378999999999999, + "loss": 2.5062, + "step": 1814 + }, + { + "epoch": 2.904, + "grad_norm": 1.0546765327453613, + "learning_rate": 0.0005382, + "loss": 2.5246, + "step": 1815 + }, + { + "epoch": 2.9055999999999997, + "grad_norm": 1.1753357648849487, + "learning_rate": 0.0005384999999999999, + "loss": 2.7387, + "step": 1816 + }, + { + "epoch": 2.9072, + "grad_norm": 0.7613487839698792, + "learning_rate": 0.0005388, + "loss": 2.4745, + "step": 1817 + }, + { + "epoch": 2.9088000000000003, + "grad_norm": 0.9472857713699341, + "learning_rate": 0.0005390999999999999, + "loss": 2.5202, + "step": 1818 + }, + { + "epoch": 2.9104, + "grad_norm": 0.7422348856925964, + "learning_rate": 0.0005394, + "loss": 2.5207, + "step": 1819 + }, + { + "epoch": 2.912, + "grad_norm": 4.3653883934021, + "learning_rate": 0.0005396999999999999, + "loss": 2.4471, + "step": 1820 + }, + { + "epoch": 2.9135999999999997, + "grad_norm": 2.8848469257354736, + "learning_rate": 0.00054, + "loss": 2.2509, + "step": 1821 + }, + { + "epoch": 2.9152, + "grad_norm": 1.0910775661468506, + "learning_rate": 0.0005403, + "loss": 2.3507, + "step": 1822 + }, + { + "epoch": 2.9168, + "grad_norm": 1.115868330001831, + "learning_rate": 0.0005405999999999999, + "loss": 2.4938, + "step": 1823 + }, + { + "epoch": 2.9184, + "grad_norm": 4.200908184051514, + "learning_rate": 0.0005409, + "loss": 2.8903, + "step": 1824 + }, + { + "epoch": 2.92, + "grad_norm": 0.9364967346191406, + "learning_rate": 0.0005411999999999999, + "loss": 2.3932, + "step": 1825 + }, + { + "epoch": 2.9215999999999998, + "grad_norm": 1.9644137620925903, + "learning_rate": 0.0005414999999999999, + "loss": 2.3253, + "step": 1826 + }, + { + "epoch": 2.9232, + "grad_norm": 0.847099244594574, + "learning_rate": 0.0005417999999999999, + "loss": 2.475, + "step": 1827 + }, + { + "epoch": 2.9248, + "grad_norm": 4.294593811035156, + "learning_rate": 0.0005420999999999999, + "loss": 2.6673, + "step": 1828 + }, + { + "epoch": 2.9264, + "grad_norm": 1.6129556894302368, + "learning_rate": 0.0005424, + "loss": 2.3189, + "step": 1829 + }, + { + "epoch": 2.928, + "grad_norm": 1.0951628684997559, + "learning_rate": 0.0005426999999999999, + "loss": 2.4761, + "step": 1830 + }, + { + "epoch": 2.9295999999999998, + "grad_norm": 0.8047678470611572, + "learning_rate": 0.000543, + "loss": 2.5185, + "step": 1831 + }, + { + "epoch": 2.9312, + "grad_norm": 0.8008071184158325, + "learning_rate": 0.0005432999999999999, + "loss": 2.3002, + "step": 1832 + }, + { + "epoch": 2.9328, + "grad_norm": 2.075387477874756, + "learning_rate": 0.0005436, + "loss": 2.2973, + "step": 1833 + }, + { + "epoch": 2.9344, + "grad_norm": 0.7697364091873169, + "learning_rate": 0.0005438999999999999, + "loss": 2.4137, + "step": 1834 + }, + { + "epoch": 2.936, + "grad_norm": 1.1987799406051636, + "learning_rate": 0.0005442, + "loss": 2.5328, + "step": 1835 + }, + { + "epoch": 2.9375999999999998, + "grad_norm": 1.0034129619598389, + "learning_rate": 0.0005445, + "loss": 2.3681, + "step": 1836 + }, + { + "epoch": 2.9392, + "grad_norm": 0.910660982131958, + "learning_rate": 0.0005448, + "loss": 2.5604, + "step": 1837 + }, + { + "epoch": 2.9408, + "grad_norm": 4.149611949920654, + "learning_rate": 0.0005451, + "loss": 2.4479, + "step": 1838 + }, + { + "epoch": 2.9424, + "grad_norm": 1.1839433908462524, + "learning_rate": 0.0005453999999999999, + "loss": 2.4518, + "step": 1839 + }, + { + "epoch": 2.944, + "grad_norm": 3.1271791458129883, + "learning_rate": 0.0005457, + "loss": 2.5315, + "step": 1840 + }, + { + "epoch": 2.9455999999999998, + "grad_norm": 1.6314308643341064, + "learning_rate": 0.0005459999999999999, + "loss": 2.3553, + "step": 1841 + }, + { + "epoch": 2.9472, + "grad_norm": 1.4227911233901978, + "learning_rate": 0.0005462999999999999, + "loss": 2.4249, + "step": 1842 + }, + { + "epoch": 2.9488, + "grad_norm": 1.3367630243301392, + "learning_rate": 0.0005466, + "loss": 2.3452, + "step": 1843 + }, + { + "epoch": 2.9504, + "grad_norm": 1.0855189561843872, + "learning_rate": 0.0005468999999999999, + "loss": 2.243, + "step": 1844 + }, + { + "epoch": 2.952, + "grad_norm": 2.145113945007324, + "learning_rate": 0.0005472, + "loss": 2.4281, + "step": 1845 + }, + { + "epoch": 2.9536, + "grad_norm": 2.266225576400757, + "learning_rate": 0.0005474999999999999, + "loss": 2.5112, + "step": 1846 + }, + { + "epoch": 2.9552, + "grad_norm": 1.2084014415740967, + "learning_rate": 0.0005478, + "loss": 2.6551, + "step": 1847 + }, + { + "epoch": 2.9568, + "grad_norm": 2.3133509159088135, + "learning_rate": 0.0005480999999999999, + "loss": 2.8256, + "step": 1848 + }, + { + "epoch": 2.9584, + "grad_norm": 2.073927879333496, + "learning_rate": 0.0005484, + "loss": 2.549, + "step": 1849 + }, + { + "epoch": 2.96, + "grad_norm": NaN, + "learning_rate": 0.0005484, + "loss": 3.4224, + "step": 1850 + }, + { + "epoch": 2.9616, + "grad_norm": 16.866920471191406, + "learning_rate": 0.0005487, + "loss": 3.5822, + "step": 1851 + }, + { + "epoch": 2.9632, + "grad_norm": 6.754800319671631, + "learning_rate": 0.000549, + "loss": 3.0191, + "step": 1852 + }, + { + "epoch": 2.9648, + "grad_norm": 1.5004948377609253, + "learning_rate": 0.0005493, + "loss": 2.8276, + "step": 1853 + }, + { + "epoch": 2.9664, + "grad_norm": 14.171708106994629, + "learning_rate": 0.0005496, + "loss": 3.3777, + "step": 1854 + }, + { + "epoch": 2.968, + "grad_norm": 1.8601189851760864, + "learning_rate": 0.0005499, + "loss": 2.5501, + "step": 1855 + }, + { + "epoch": 2.9696, + "grad_norm": 2.106257677078247, + "learning_rate": 0.0005501999999999999, + "loss": 2.7187, + "step": 1856 + }, + { + "epoch": 2.9712, + "grad_norm": 1.215472936630249, + "learning_rate": 0.0005505, + "loss": 2.6169, + "step": 1857 + }, + { + "epoch": 2.9728, + "grad_norm": 1.326351523399353, + "learning_rate": 0.0005507999999999999, + "loss": 2.5181, + "step": 1858 + }, + { + "epoch": 2.9744, + "grad_norm": 1.5959700345993042, + "learning_rate": 0.0005510999999999999, + "loss": 2.5987, + "step": 1859 + }, + { + "epoch": 2.976, + "grad_norm": 1.368794322013855, + "learning_rate": 0.0005514, + "loss": 2.3588, + "step": 1860 + }, + { + "epoch": 2.9776, + "grad_norm": 0.9404537677764893, + "learning_rate": 0.0005516999999999999, + "loss": 2.398, + "step": 1861 + }, + { + "epoch": 2.9792, + "grad_norm": 3.2774293422698975, + "learning_rate": 0.000552, + "loss": 2.3012, + "step": 1862 + }, + { + "epoch": 2.9808, + "grad_norm": 1.7845696210861206, + "learning_rate": 0.0005522999999999999, + "loss": 2.6416, + "step": 1863 + }, + { + "epoch": 2.9824, + "grad_norm": 2.4613161087036133, + "learning_rate": 0.0005526, + "loss": 2.708, + "step": 1864 + }, + { + "epoch": 2.984, + "grad_norm": 3.9170267581939697, + "learning_rate": 0.0005528999999999999, + "loss": 2.4808, + "step": 1865 + }, + { + "epoch": 2.9856, + "grad_norm": 3.9187498092651367, + "learning_rate": 0.0005532, + "loss": 2.1382, + "step": 1866 + }, + { + "epoch": 2.9872, + "grad_norm": 3.5691933631896973, + "learning_rate": 0.0005535, + "loss": 2.6534, + "step": 1867 + }, + { + "epoch": 2.9888, + "grad_norm": 1.9193207025527954, + "learning_rate": 0.0005538, + "loss": 2.4064, + "step": 1868 + }, + { + "epoch": 2.9904, + "grad_norm": 0.9679790735244751, + "learning_rate": 0.0005541, + "loss": 2.2612, + "step": 1869 + }, + { + "epoch": 2.992, + "grad_norm": 0.8689792156219482, + "learning_rate": 0.0005544, + "loss": 2.3644, + "step": 1870 + }, + { + "epoch": 2.9936, + "grad_norm": 1.476846694946289, + "learning_rate": 0.0005547, + "loss": 2.6756, + "step": 1871 + }, + { + "epoch": 2.9952, + "grad_norm": 1.7874386310577393, + "learning_rate": 0.0005549999999999999, + "loss": 2.5747, + "step": 1872 + }, + { + "epoch": 2.9968, + "grad_norm": 1.3470101356506348, + "learning_rate": 0.0005552999999999999, + "loss": 2.5029, + "step": 1873 + }, + { + "epoch": 2.9984, + "grad_norm": 1.982383131980896, + "learning_rate": 0.0005556, + "loss": 2.9071, + "step": 1874 + }, + { + "epoch": 3.0, + "grad_norm": NaN, + "learning_rate": 0.0005556, + "loss": 3.1407, + "step": 1875 + }, + { + "epoch": 3.0016, + "grad_norm": 25.29146385192871, + "learning_rate": 0.0005558999999999999, + "loss": 3.837, + "step": 1876 + }, + { + "epoch": 3.0032, + "grad_norm": 9.291911125183105, + "learning_rate": 0.0005562, + "loss": 2.9761, + "step": 1877 + }, + { + "epoch": 3.0048, + "grad_norm": 4.966324329376221, + "learning_rate": 0.0005564999999999999, + "loss": 3.0185, + "step": 1878 + }, + { + "epoch": 3.0064, + "grad_norm": 2.462636709213257, + "learning_rate": 0.0005568, + "loss": 2.7161, + "step": 1879 + }, + { + "epoch": 3.008, + "grad_norm": 1.5384327173233032, + "learning_rate": 0.0005570999999999999, + "loss": 2.669, + "step": 1880 + }, + { + "epoch": 3.0096, + "grad_norm": 2.980968713760376, + "learning_rate": 0.0005574, + "loss": 2.8147, + "step": 1881 + }, + { + "epoch": 3.0112, + "grad_norm": 5.0100908279418945, + "learning_rate": 0.0005577, + "loss": 2.6561, + "step": 1882 + }, + { + "epoch": 3.0128, + "grad_norm": 2.8315351009368896, + "learning_rate": 0.000558, + "loss": 2.8438, + "step": 1883 + }, + { + "epoch": 3.0144, + "grad_norm": 2.2698962688446045, + "learning_rate": 0.0005583, + "loss": 2.7352, + "step": 1884 + }, + { + "epoch": 3.016, + "grad_norm": 1.6499841213226318, + "learning_rate": 0.0005586, + "loss": 2.3665, + "step": 1885 + }, + { + "epoch": 3.0176, + "grad_norm": 7.6851091384887695, + "learning_rate": 0.0005589, + "loss": 2.9253, + "step": 1886 + }, + { + "epoch": 3.0192, + "grad_norm": 9.859707832336426, + "learning_rate": 0.0005591999999999999, + "loss": 2.848, + "step": 1887 + }, + { + "epoch": 3.0208, + "grad_norm": 2.2398126125335693, + "learning_rate": 0.0005595, + "loss": 2.5253, + "step": 1888 + }, + { + "epoch": 3.0224, + "grad_norm": 3.5129432678222656, + "learning_rate": 0.0005598, + "loss": 2.4375, + "step": 1889 + }, + { + "epoch": 3.024, + "grad_norm": 2.765233278274536, + "learning_rate": 0.0005600999999999999, + "loss": 2.3901, + "step": 1890 + }, + { + "epoch": 3.0256, + "grad_norm": 2.178379774093628, + "learning_rate": 0.0005604, + "loss": 2.4001, + "step": 1891 + }, + { + "epoch": 3.0272, + "grad_norm": 5.110830307006836, + "learning_rate": 0.0005606999999999999, + "loss": 2.7783, + "step": 1892 + }, + { + "epoch": 3.0288, + "grad_norm": 1.4218653440475464, + "learning_rate": 0.000561, + "loss": 2.4922, + "step": 1893 + }, + { + "epoch": 3.0304, + "grad_norm": 2.5468997955322266, + "learning_rate": 0.0005612999999999999, + "loss": 2.4771, + "step": 1894 + }, + { + "epoch": 3.032, + "grad_norm": 1.7513118982315063, + "learning_rate": 0.0005616, + "loss": 2.4955, + "step": 1895 + }, + { + "epoch": 3.0336, + "grad_norm": 1.6879457235336304, + "learning_rate": 0.0005618999999999999, + "loss": 2.3897, + "step": 1896 + }, + { + "epoch": 3.0352, + "grad_norm": 1.5849945545196533, + "learning_rate": 0.0005622, + "loss": 2.2501, + "step": 1897 + }, + { + "epoch": 3.0368, + "grad_norm": 1.4570404291152954, + "learning_rate": 0.0005625, + "loss": 2.3634, + "step": 1898 + }, + { + "epoch": 3.0384, + "grad_norm": 1.2244665622711182, + "learning_rate": 0.0005627999999999999, + "loss": 2.5295, + "step": 1899 + }, + { + "epoch": 3.04, + "grad_norm": 1.8930087089538574, + "learning_rate": 0.0005631, + "loss": 2.2068, + "step": 1900 + }, + { + "epoch": 3.0416, + "grad_norm": 1.3034899234771729, + "learning_rate": 0.0005633999999999999, + "loss": 2.3666, + "step": 1901 + }, + { + "epoch": 3.0432, + "grad_norm": 2.7985405921936035, + "learning_rate": 0.0005637, + "loss": 2.2791, + "step": 1902 + }, + { + "epoch": 3.0448, + "grad_norm": 1.738775610923767, + "learning_rate": 0.0005639999999999999, + "loss": 2.3868, + "step": 1903 + }, + { + "epoch": 3.0464, + "grad_norm": 4.025388240814209, + "learning_rate": 0.0005643, + "loss": 2.4722, + "step": 1904 + }, + { + "epoch": 3.048, + "grad_norm": 2.4662373065948486, + "learning_rate": 0.0005646, + "loss": 2.3586, + "step": 1905 + }, + { + "epoch": 3.0496, + "grad_norm": 2.3425092697143555, + "learning_rate": 0.0005648999999999999, + "loss": 2.5808, + "step": 1906 + }, + { + "epoch": 3.0512, + "grad_norm": 3.2466976642608643, + "learning_rate": 0.0005652, + "loss": 2.2679, + "step": 1907 + }, + { + "epoch": 3.0528, + "grad_norm": 2.766129493713379, + "learning_rate": 0.0005654999999999999, + "loss": 2.5386, + "step": 1908 + }, + { + "epoch": 3.0544, + "grad_norm": 1.7741740942001343, + "learning_rate": 0.0005657999999999999, + "loss": 2.7793, + "step": 1909 + }, + { + "epoch": 3.056, + "grad_norm": 1.4344654083251953, + "learning_rate": 0.0005660999999999999, + "loss": 2.3047, + "step": 1910 + }, + { + "epoch": 3.0576, + "grad_norm": 2.42335844039917, + "learning_rate": 0.0005663999999999999, + "loss": 2.3451, + "step": 1911 + }, + { + "epoch": 3.0592, + "grad_norm": 1.0476360321044922, + "learning_rate": 0.0005667, + "loss": 2.1713, + "step": 1912 + }, + { + "epoch": 3.0608, + "grad_norm": 1.0051932334899902, + "learning_rate": 0.0005669999999999999, + "loss": 2.2454, + "step": 1913 + }, + { + "epoch": 3.0624, + "grad_norm": 3.0773088932037354, + "learning_rate": 0.0005673, + "loss": 2.6067, + "step": 1914 + }, + { + "epoch": 3.064, + "grad_norm": 4.21511173248291, + "learning_rate": 0.0005675999999999999, + "loss": 2.3492, + "step": 1915 + }, + { + "epoch": 3.0656, + "grad_norm": 2.3126461505889893, + "learning_rate": 0.0005679, + "loss": 2.5079, + "step": 1916 + }, + { + "epoch": 3.0672, + "grad_norm": 2.4149043560028076, + "learning_rate": 0.0005681999999999999, + "loss": 2.7256, + "step": 1917 + }, + { + "epoch": 3.0688, + "grad_norm": 2.686859607696533, + "learning_rate": 0.0005685, + "loss": 2.7068, + "step": 1918 + }, + { + "epoch": 3.0704, + "grad_norm": 3.6169612407684326, + "learning_rate": 0.0005688, + "loss": 2.6735, + "step": 1919 + }, + { + "epoch": 3.072, + "grad_norm": 1.6046686172485352, + "learning_rate": 0.0005691, + "loss": 2.3301, + "step": 1920 + }, + { + "epoch": 3.0736, + "grad_norm": 2.102391242980957, + "learning_rate": 0.0005694, + "loss": 2.6546, + "step": 1921 + }, + { + "epoch": 3.0752, + "grad_norm": 2.0405120849609375, + "learning_rate": 0.0005696999999999999, + "loss": 2.6443, + "step": 1922 + }, + { + "epoch": 3.0768, + "grad_norm": 2.9234347343444824, + "learning_rate": 0.00057, + "loss": 2.3251, + "step": 1923 + }, + { + "epoch": 3.0784, + "grad_norm": 1.348129153251648, + "learning_rate": 0.0005702999999999999, + "loss": 2.4312, + "step": 1924 + }, + { + "epoch": 3.08, + "grad_norm": NaN, + "learning_rate": 0.0005702999999999999, + "loss": 2.1661, + "step": 1925 + }, + { + "epoch": 3.0816, + "grad_norm": 2.0966081619262695, + "learning_rate": 0.0005705999999999999, + "loss": 2.9023, + "step": 1926 + }, + { + "epoch": 3.0832, + "grad_norm": 1.7861617803573608, + "learning_rate": 0.0005708999999999999, + "loss": 2.7179, + "step": 1927 + }, + { + "epoch": 3.0848, + "grad_norm": 10.112266540527344, + "learning_rate": 0.0005711999999999999, + "loss": 3.2361, + "step": 1928 + }, + { + "epoch": 3.0864, + "grad_norm": 11.517034530639648, + "learning_rate": 0.0005715, + "loss": 3.095, + "step": 1929 + }, + { + "epoch": 3.088, + "grad_norm": 11.801773071289062, + "learning_rate": 0.0005717999999999999, + "loss": 2.9547, + "step": 1930 + }, + { + "epoch": 3.0896, + "grad_norm": 5.743291854858398, + "learning_rate": 0.0005721, + "loss": 2.4528, + "step": 1931 + }, + { + "epoch": 3.0912, + "grad_norm": 2.2978971004486084, + "learning_rate": 0.0005723999999999999, + "loss": 2.7744, + "step": 1932 + }, + { + "epoch": 3.0928, + "grad_norm": 3.952575206756592, + "learning_rate": 0.0005727, + "loss": 2.6278, + "step": 1933 + }, + { + "epoch": 3.0944, + "grad_norm": 1.861054539680481, + "learning_rate": 0.0005729999999999999, + "loss": 2.51, + "step": 1934 + }, + { + "epoch": 3.096, + "grad_norm": 4.444493293762207, + "learning_rate": 0.0005733, + "loss": 2.7883, + "step": 1935 + }, + { + "epoch": 3.0976, + "grad_norm": 2.688218593597412, + "learning_rate": 0.0005736, + "loss": 2.4545, + "step": 1936 + }, + { + "epoch": 3.0992, + "grad_norm": 3.14324951171875, + "learning_rate": 0.0005738999999999999, + "loss": 2.5803, + "step": 1937 + }, + { + "epoch": 3.1008, + "grad_norm": 2.0566539764404297, + "learning_rate": 0.0005742, + "loss": 2.3113, + "step": 1938 + }, + { + "epoch": 3.1024, + "grad_norm": 1.8503978252410889, + "learning_rate": 0.0005744999999999999, + "loss": 2.4176, + "step": 1939 + }, + { + "epoch": 3.104, + "grad_norm": 2.6355438232421875, + "learning_rate": 0.0005747999999999999, + "loss": 2.4964, + "step": 1940 + }, + { + "epoch": 3.1056, + "grad_norm": 1.8768773078918457, + "learning_rate": 0.0005750999999999999, + "loss": 2.4573, + "step": 1941 + }, + { + "epoch": 3.1072, + "grad_norm": 3.4046988487243652, + "learning_rate": 0.0005753999999999999, + "loss": 2.6418, + "step": 1942 + }, + { + "epoch": 3.1088, + "grad_norm": 1.7340015172958374, + "learning_rate": 0.0005757, + "loss": 2.4697, + "step": 1943 + }, + { + "epoch": 3.1104, + "grad_norm": 2.1641948223114014, + "learning_rate": 0.0005759999999999999, + "loss": 2.5012, + "step": 1944 + }, + { + "epoch": 3.112, + "grad_norm": 3.7307941913604736, + "learning_rate": 0.0005763, + "loss": 2.4757, + "step": 1945 + }, + { + "epoch": 3.1136, + "grad_norm": 2.3845276832580566, + "learning_rate": 0.0005765999999999999, + "loss": 2.4441, + "step": 1946 + }, + { + "epoch": 3.1152, + "grad_norm": 2.855138063430786, + "learning_rate": 0.0005769, + "loss": 2.2891, + "step": 1947 + }, + { + "epoch": 3.1168, + "grad_norm": 2.2033305168151855, + "learning_rate": 0.0005771999999999999, + "loss": 2.1741, + "step": 1948 + }, + { + "epoch": 3.1184, + "grad_norm": 3.501863479614258, + "learning_rate": 0.0005775, + "loss": 2.849, + "step": 1949 + }, + { + "epoch": 3.12, + "grad_norm": 2.6321678161621094, + "learning_rate": 0.0005778, + "loss": 2.3158, + "step": 1950 + }, + { + "epoch": 3.1216, + "grad_norm": 1.6110666990280151, + "learning_rate": 0.0005781, + "loss": 2.3013, + "step": 1951 + }, + { + "epoch": 3.1232, + "grad_norm": 1.646638035774231, + "learning_rate": 0.0005784, + "loss": 2.3759, + "step": 1952 + }, + { + "epoch": 3.1248, + "grad_norm": 3.533754587173462, + "learning_rate": 0.0005786999999999999, + "loss": 2.3256, + "step": 1953 + }, + { + "epoch": 3.1264, + "grad_norm": 2.3070497512817383, + "learning_rate": 0.000579, + "loss": 2.4358, + "step": 1954 + }, + { + "epoch": 3.128, + "grad_norm": 2.581441879272461, + "learning_rate": 0.0005792999999999999, + "loss": 2.4647, + "step": 1955 + }, + { + "epoch": 3.1296, + "grad_norm": 2.834535598754883, + "learning_rate": 0.0005795999999999999, + "loss": 2.3173, + "step": 1956 + }, + { + "epoch": 3.1312, + "grad_norm": 1.163590908050537, + "learning_rate": 0.0005799, + "loss": 2.7147, + "step": 1957 + }, + { + "epoch": 3.1328, + "grad_norm": 5.1326375007629395, + "learning_rate": 0.0005801999999999999, + "loss": 2.7112, + "step": 1958 + }, + { + "epoch": 3.1344, + "grad_norm": 2.296224594116211, + "learning_rate": 0.0005805, + "loss": 2.4191, + "step": 1959 + }, + { + "epoch": 3.136, + "grad_norm": 3.148940086364746, + "learning_rate": 0.0005807999999999999, + "loss": 2.3378, + "step": 1960 + }, + { + "epoch": 3.1376, + "grad_norm": 1.3591207265853882, + "learning_rate": 0.0005811, + "loss": 2.5719, + "step": 1961 + }, + { + "epoch": 3.1391999999999998, + "grad_norm": 2.0001440048217773, + "learning_rate": 0.0005813999999999999, + "loss": 2.1871, + "step": 1962 + }, + { + "epoch": 3.1408, + "grad_norm": 1.4580248594284058, + "learning_rate": 0.0005817, + "loss": 2.4727, + "step": 1963 + }, + { + "epoch": 3.1424, + "grad_norm": 1.5752590894699097, + "learning_rate": 0.0005819999999999999, + "loss": 2.4377, + "step": 1964 + }, + { + "epoch": 3.144, + "grad_norm": 2.6746673583984375, + "learning_rate": 0.0005823, + "loss": 2.2526, + "step": 1965 + }, + { + "epoch": 3.1456, + "grad_norm": 1.5736886262893677, + "learning_rate": 0.0005826, + "loss": 2.5686, + "step": 1966 + }, + { + "epoch": 3.1471999999999998, + "grad_norm": 1.3866643905639648, + "learning_rate": 0.0005829, + "loss": 2.5909, + "step": 1967 + }, + { + "epoch": 3.1488, + "grad_norm": 1.2823047637939453, + "learning_rate": 0.0005832, + "loss": 2.7079, + "step": 1968 + }, + { + "epoch": 3.1504, + "grad_norm": 1.206445336341858, + "learning_rate": 0.0005834999999999999, + "loss": 2.6945, + "step": 1969 + }, + { + "epoch": 3.152, + "grad_norm": 1.8165123462677002, + "learning_rate": 0.0005838, + "loss": 2.4911, + "step": 1970 + }, + { + "epoch": 3.1536, + "grad_norm": 4.777324199676514, + "learning_rate": 0.0005840999999999999, + "loss": 2.4485, + "step": 1971 + }, + { + "epoch": 3.1552, + "grad_norm": 2.511350393295288, + "learning_rate": 0.0005843999999999999, + "loss": 2.8384, + "step": 1972 + }, + { + "epoch": 3.1568, + "grad_norm": 2.9431605339050293, + "learning_rate": 0.0005847, + "loss": 2.7788, + "step": 1973 + }, + { + "epoch": 3.1584, + "grad_norm": 2.0299463272094727, + "learning_rate": 0.0005849999999999999, + "loss": 2.6043, + "step": 1974 + }, + { + "epoch": 3.16, + "grad_norm": 1.5065081119537354, + "learning_rate": 0.0005853, + "loss": 3.039, + "step": 1975 + }, + { + "epoch": 3.1616, + "grad_norm": 15.988265991210938, + "learning_rate": 0.0005855999999999999, + "loss": 3.272, + "step": 1976 + }, + { + "epoch": 3.1632, + "grad_norm": 3.07070255279541, + "learning_rate": 0.0005859, + "loss": 3.2078, + "step": 1977 + }, + { + "epoch": 3.1648, + "grad_norm": 8.927387237548828, + "learning_rate": 0.0005861999999999999, + "loss": 3.2354, + "step": 1978 + }, + { + "epoch": 3.1664, + "grad_norm": 14.451622009277344, + "learning_rate": 0.0005865, + "loss": 3.3638, + "step": 1979 + }, + { + "epoch": 3.168, + "grad_norm": 2.183037757873535, + "learning_rate": 0.0005868, + "loss": 2.9126, + "step": 1980 + }, + { + "epoch": 3.1696, + "grad_norm": 3.3495078086853027, + "learning_rate": 0.0005871, + "loss": 3.0289, + "step": 1981 + }, + { + "epoch": 3.1712, + "grad_norm": 2.572366714477539, + "learning_rate": 0.0005874, + "loss": 2.6892, + "step": 1982 + }, + { + "epoch": 3.1728, + "grad_norm": 1.9703586101531982, + "learning_rate": 0.0005876999999999999, + "loss": 3.0979, + "step": 1983 + }, + { + "epoch": 3.1744, + "grad_norm": 3.3965091705322266, + "learning_rate": 0.000588, + "loss": 2.8213, + "step": 1984 + }, + { + "epoch": 3.176, + "grad_norm": 2.5969173908233643, + "learning_rate": 0.0005882999999999999, + "loss": 2.7087, + "step": 1985 + }, + { + "epoch": 3.1776, + "grad_norm": 3.7670791149139404, + "learning_rate": 0.0005885999999999999, + "loss": 2.3701, + "step": 1986 + }, + { + "epoch": 3.1792, + "grad_norm": 2.7813947200775146, + "learning_rate": 0.0005889, + "loss": 2.6189, + "step": 1987 + }, + { + "epoch": 3.1808, + "grad_norm": 2.1592915058135986, + "learning_rate": 0.0005891999999999999, + "loss": 2.4803, + "step": 1988 + }, + { + "epoch": 3.1824, + "grad_norm": 1.573058009147644, + "learning_rate": 0.0005895, + "loss": 2.504, + "step": 1989 + }, + { + "epoch": 3.184, + "grad_norm": 2.9862351417541504, + "learning_rate": 0.0005897999999999999, + "loss": 2.4572, + "step": 1990 + }, + { + "epoch": 3.1856, + "grad_norm": 2.136756420135498, + "learning_rate": 0.0005901, + "loss": 2.4623, + "step": 1991 + }, + { + "epoch": 3.1872, + "grad_norm": 2.619926929473877, + "learning_rate": 0.0005903999999999999, + "loss": 2.3356, + "step": 1992 + }, + { + "epoch": 3.1888, + "grad_norm": 12.829959869384766, + "learning_rate": 0.0005907, + "loss": 3.5301, + "step": 1993 + }, + { + "epoch": 3.1904, + "grad_norm": 2.9119503498077393, + "learning_rate": 0.0005909999999999999, + "loss": 2.3499, + "step": 1994 + }, + { + "epoch": 3.192, + "grad_norm": 6.323517799377441, + "learning_rate": 0.0005913, + "loss": 2.7329, + "step": 1995 + }, + { + "epoch": 3.1936, + "grad_norm": 3.644813060760498, + "learning_rate": 0.0005916, + "loss": 2.5666, + "step": 1996 + }, + { + "epoch": 3.1952, + "grad_norm": 5.822663307189941, + "learning_rate": 0.0005919, + "loss": 2.7859, + "step": 1997 + }, + { + "epoch": 3.1968, + "grad_norm": 2.3552889823913574, + "learning_rate": 0.0005922, + "loss": 2.1818, + "step": 1998 + }, + { + "epoch": 3.1984, + "grad_norm": 3.3351669311523438, + "learning_rate": 0.0005924999999999999, + "loss": 2.6502, + "step": 1999 + }, + { + "epoch": 3.2, + "grad_norm": 3.1912646293640137, + "learning_rate": 0.0005928, + "loss": 2.4085, + "step": 2000 + }, + { + "epoch": 3.2016, + "grad_norm": 2.0900375843048096, + "learning_rate": 0.0005930999999999999, + "loss": 2.3544, + "step": 2001 + }, + { + "epoch": 3.2032, + "grad_norm": 1.7966182231903076, + "learning_rate": 0.0005933999999999999, + "loss": 2.4058, + "step": 2002 + }, + { + "epoch": 3.2048, + "grad_norm": 4.029713153839111, + "learning_rate": 0.0005937, + "loss": 2.2235, + "step": 2003 + }, + { + "epoch": 3.2064, + "grad_norm": 5.860076427459717, + "learning_rate": 0.0005939999999999999, + "loss": 2.6818, + "step": 2004 + }, + { + "epoch": 3.208, + "grad_norm": 9.5676851272583, + "learning_rate": 0.0005943, + "loss": 2.3783, + "step": 2005 + }, + { + "epoch": 3.2096, + "grad_norm": 3.6261396408081055, + "learning_rate": 0.0005945999999999999, + "loss": 2.2907, + "step": 2006 + }, + { + "epoch": 3.2112, + "grad_norm": 1.9345386028289795, + "learning_rate": 0.0005949, + "loss": 2.249, + "step": 2007 + }, + { + "epoch": 3.2128, + "grad_norm": 1.471825122833252, + "learning_rate": 0.0005951999999999999, + "loss": 2.3189, + "step": 2008 + }, + { + "epoch": 3.2144, + "grad_norm": 1.4700121879577637, + "learning_rate": 0.0005955, + "loss": 2.4003, + "step": 2009 + }, + { + "epoch": 3.216, + "grad_norm": 1.4592626094818115, + "learning_rate": 0.0005958, + "loss": 2.3744, + "step": 2010 + }, + { + "epoch": 3.2176, + "grad_norm": 1.8481483459472656, + "learning_rate": 0.0005961, + "loss": 2.4525, + "step": 2011 + }, + { + "epoch": 3.2192, + "grad_norm": 2.111783266067505, + "learning_rate": 0.0005964, + "loss": 2.6025, + "step": 2012 + }, + { + "epoch": 3.2208, + "grad_norm": 1.3971112966537476, + "learning_rate": 0.0005967, + "loss": 2.4959, + "step": 2013 + }, + { + "epoch": 3.2224, + "grad_norm": 2.5589585304260254, + "learning_rate": 0.000597, + "loss": 2.6663, + "step": 2014 + }, + { + "epoch": 3.224, + "grad_norm": 4.524598598480225, + "learning_rate": 0.0005972999999999999, + "loss": 2.5798, + "step": 2015 + }, + { + "epoch": 3.2256, + "grad_norm": 6.212462902069092, + "learning_rate": 0.0005976, + "loss": 2.3817, + "step": 2016 + }, + { + "epoch": 3.2272, + "grad_norm": 3.290090560913086, + "learning_rate": 0.0005979, + "loss": 2.7171, + "step": 2017 + }, + { + "epoch": 3.2288, + "grad_norm": 4.479714393615723, + "learning_rate": 0.0005981999999999999, + "loss": 2.2308, + "step": 2018 + }, + { + "epoch": 3.2304, + "grad_norm": 1.692572832107544, + "learning_rate": 0.0005985, + "loss": 2.7637, + "step": 2019 + }, + { + "epoch": 3.232, + "grad_norm": 1.6728739738464355, + "learning_rate": 0.0005987999999999999, + "loss": 2.615, + "step": 2020 + }, + { + "epoch": 3.2336, + "grad_norm": 1.3036847114562988, + "learning_rate": 0.0005991, + "loss": 2.6417, + "step": 2021 + }, + { + "epoch": 3.2352, + "grad_norm": 1.48746919631958, + "learning_rate": 0.0005993999999999999, + "loss": 2.8424, + "step": 2022 + }, + { + "epoch": 3.2368, + "grad_norm": 1.4318227767944336, + "learning_rate": 0.0005997, + "loss": 2.469, + "step": 2023 + }, + { + "epoch": 3.2384, + "grad_norm": 5.054473876953125, + "learning_rate": 0.0006, + "loss": 3.4372, + "step": 2024 + }, + { + "epoch": 3.24, + "grad_norm": NaN, + "learning_rate": 0.0006, + "loss": 2.7082, + "step": 2025 + }, + { + "epoch": 3.2416, + "grad_norm": 34.61737823486328, + "learning_rate": 0.0005997999999999999, + "loss": 4.3039, + "step": 2026 + }, + { + "epoch": 3.2432, + "grad_norm": 3.473752021789551, + "learning_rate": 0.0005995999999999999, + "loss": 2.9205, + "step": 2027 + }, + { + "epoch": 3.2448, + "grad_norm": 3.742560625076294, + "learning_rate": 0.0005993999999999999, + "loss": 3.0236, + "step": 2028 + }, + { + "epoch": 3.2464, + "grad_norm": 2.6675426959991455, + "learning_rate": 0.0005991999999999999, + "loss": 2.862, + "step": 2029 + }, + { + "epoch": 3.248, + "grad_norm": 3.681030750274658, + "learning_rate": 0.0005989999999999999, + "loss": 3.2253, + "step": 2030 + }, + { + "epoch": 3.2496, + "grad_norm": 3.5721378326416016, + "learning_rate": 0.0005987999999999999, + "loss": 3.0141, + "step": 2031 + }, + { + "epoch": 3.2512, + "grad_norm": 2.2624754905700684, + "learning_rate": 0.0005986, + "loss": 2.6375, + "step": 2032 + }, + { + "epoch": 3.2528, + "grad_norm": 2.0107228755950928, + "learning_rate": 0.0005983999999999999, + "loss": 2.6202, + "step": 2033 + }, + { + "epoch": 3.2544, + "grad_norm": 2.3134191036224365, + "learning_rate": 0.0005981999999999999, + "loss": 2.6874, + "step": 2034 + }, + { + "epoch": 3.2560000000000002, + "grad_norm": 27.417612075805664, + "learning_rate": 0.000598, + "loss": 3.7857, + "step": 2035 + }, + { + "epoch": 3.2576, + "grad_norm": 10.154114723205566, + "learning_rate": 0.0005977999999999999, + "loss": 2.9672, + "step": 2036 + }, + { + "epoch": 3.2592, + "grad_norm": 5.68153715133667, + "learning_rate": 0.0005976, + "loss": 2.6946, + "step": 2037 + }, + { + "epoch": 3.2608, + "grad_norm": 3.4215259552001953, + "learning_rate": 0.0005974, + "loss": 2.5735, + "step": 2038 + }, + { + "epoch": 3.2624, + "grad_norm": 9.281172752380371, + "learning_rate": 0.0005971999999999999, + "loss": 2.7988, + "step": 2039 + }, + { + "epoch": 3.2640000000000002, + "grad_norm": 4.832975387573242, + "learning_rate": 0.000597, + "loss": 2.6848, + "step": 2040 + }, + { + "epoch": 3.2656, + "grad_norm": 4.111706256866455, + "learning_rate": 0.0005968, + "loss": 2.951, + "step": 2041 + }, + { + "epoch": 3.2672, + "grad_norm": 2.7047650814056396, + "learning_rate": 0.0005966, + "loss": 2.9283, + "step": 2042 + }, + { + "epoch": 3.2688, + "grad_norm": 3.965059995651245, + "learning_rate": 0.0005964, + "loss": 2.6947, + "step": 2043 + }, + { + "epoch": 3.2704, + "grad_norm": 2.8490593433380127, + "learning_rate": 0.0005962, + "loss": 2.4638, + "step": 2044 + }, + { + "epoch": 3.2720000000000002, + "grad_norm": 3.065896511077881, + "learning_rate": 0.000596, + "loss": 2.3995, + "step": 2045 + }, + { + "epoch": 3.2736, + "grad_norm": 4.135379314422607, + "learning_rate": 0.0005958, + "loss": 2.5321, + "step": 2046 + }, + { + "epoch": 3.2752, + "grad_norm": 3.0422980785369873, + "learning_rate": 0.0005956, + "loss": 2.3867, + "step": 2047 + }, + { + "epoch": 3.2768, + "grad_norm": 3.1054189205169678, + "learning_rate": 0.0005953999999999999, + "loss": 2.3821, + "step": 2048 + }, + { + "epoch": 3.2784, + "grad_norm": 1.8611743450164795, + "learning_rate": 0.0005951999999999999, + "loss": 2.4836, + "step": 2049 + }, + { + "epoch": 3.2800000000000002, + "grad_norm": 3.28997540473938, + "learning_rate": 0.0005949999999999999, + "loss": 2.7678, + "step": 2050 + }, + { + "epoch": 3.2816, + "grad_norm": 3.161947011947632, + "learning_rate": 0.0005947999999999999, + "loss": 2.4457, + "step": 2051 + }, + { + "epoch": 3.2832, + "grad_norm": 3.553466558456421, + "learning_rate": 0.0005945999999999999, + "loss": 2.2187, + "step": 2052 + }, + { + "epoch": 3.2848, + "grad_norm": 2.138315439224243, + "learning_rate": 0.0005943999999999999, + "loss": 2.4322, + "step": 2053 + }, + { + "epoch": 3.2864, + "grad_norm": 2.237607717514038, + "learning_rate": 0.0005941999999999999, + "loss": 2.5734, + "step": 2054 + }, + { + "epoch": 3.288, + "grad_norm": 2.6869661808013916, + "learning_rate": 0.0005939999999999999, + "loss": 2.4021, + "step": 2055 + }, + { + "epoch": 3.2896, + "grad_norm": 1.8337856531143188, + "learning_rate": 0.0005938, + "loss": 2.533, + "step": 2056 + }, + { + "epoch": 3.2912, + "grad_norm": 4.238888263702393, + "learning_rate": 0.0005935999999999999, + "loss": 2.5023, + "step": 2057 + }, + { + "epoch": 3.2928, + "grad_norm": 5.389216899871826, + "learning_rate": 0.0005933999999999999, + "loss": 2.6831, + "step": 2058 + }, + { + "epoch": 3.2944, + "grad_norm": 7.200102806091309, + "learning_rate": 0.0005932, + "loss": 2.7044, + "step": 2059 + }, + { + "epoch": 3.296, + "grad_norm": 6.210408687591553, + "learning_rate": 0.0005929999999999999, + "loss": 2.7682, + "step": 2060 + }, + { + "epoch": 3.2976, + "grad_norm": 2.685565233230591, + "learning_rate": 0.0005928, + "loss": 2.504, + "step": 2061 + }, + { + "epoch": 3.2992, + "grad_norm": 3.0856990814208984, + "learning_rate": 0.0005926, + "loss": 2.4779, + "step": 2062 + }, + { + "epoch": 3.3008, + "grad_norm": 2.0148448944091797, + "learning_rate": 0.0005923999999999999, + "loss": 2.3513, + "step": 2063 + }, + { + "epoch": 3.3024, + "grad_norm": 2.684429883956909, + "learning_rate": 0.0005922, + "loss": 2.612, + "step": 2064 + }, + { + "epoch": 3.304, + "grad_norm": 1.8254823684692383, + "learning_rate": 0.000592, + "loss": 2.2172, + "step": 2065 + }, + { + "epoch": 3.3056, + "grad_norm": 2.231571674346924, + "learning_rate": 0.0005918, + "loss": 2.5103, + "step": 2066 + }, + { + "epoch": 3.3072, + "grad_norm": 1.586329698562622, + "learning_rate": 0.0005916, + "loss": 2.5327, + "step": 2067 + }, + { + "epoch": 3.3088, + "grad_norm": 2.507460832595825, + "learning_rate": 0.0005914, + "loss": 3.0766, + "step": 2068 + }, + { + "epoch": 3.3104, + "grad_norm": 2.443678855895996, + "learning_rate": 0.0005912, + "loss": 2.6095, + "step": 2069 + }, + { + "epoch": 3.312, + "grad_norm": 5.117314338684082, + "learning_rate": 0.0005909999999999999, + "loss": 2.3947, + "step": 2070 + }, + { + "epoch": 3.3136, + "grad_norm": 8.64423656463623, + "learning_rate": 0.0005907999999999999, + "loss": 3.0027, + "step": 2071 + }, + { + "epoch": 3.3152, + "grad_norm": 3.596308946609497, + "learning_rate": 0.0005905999999999999, + "loss": 2.4804, + "step": 2072 + }, + { + "epoch": 3.3168, + "grad_norm": 2.0442867279052734, + "learning_rate": 0.0005903999999999999, + "loss": 2.5104, + "step": 2073 + }, + { + "epoch": 3.3184, + "grad_norm": 1.5353082418441772, + "learning_rate": 0.0005901999999999999, + "loss": 2.6682, + "step": 2074 + }, + { + "epoch": 3.32, + "grad_norm": NaN, + "learning_rate": 0.0005901999999999999, + "loss": 3.3712, + "step": 2075 + }, + { + "epoch": 3.3216, + "grad_norm": 5.499607563018799, + "learning_rate": 0.0005899999999999999, + "loss": 3.4196, + "step": 2076 + }, + { + "epoch": 3.3232, + "grad_norm": 2.8983004093170166, + "learning_rate": 0.0005897999999999999, + "loss": 2.9806, + "step": 2077 + }, + { + "epoch": 3.3247999999999998, + "grad_norm": 3.1704025268554688, + "learning_rate": 0.0005895999999999999, + "loss": 2.9869, + "step": 2078 + }, + { + "epoch": 3.3264, + "grad_norm": 14.21891975402832, + "learning_rate": 0.0005893999999999999, + "loss": 3.5013, + "step": 2079 + }, + { + "epoch": 3.328, + "grad_norm": 11.844083786010742, + "learning_rate": 0.0005891999999999999, + "loss": 3.1626, + "step": 2080 + }, + { + "epoch": 3.3296, + "grad_norm": 18.769367218017578, + "learning_rate": 0.000589, + "loss": 3.3018, + "step": 2081 + }, + { + "epoch": 3.3312, + "grad_norm": 5.585927963256836, + "learning_rate": 0.0005887999999999999, + "loss": 3.0258, + "step": 2082 + }, + { + "epoch": 3.3327999999999998, + "grad_norm": 6.997848987579346, + "learning_rate": 0.0005885999999999999, + "loss": 2.7815, + "step": 2083 + }, + { + "epoch": 3.3344, + "grad_norm": 7.808254718780518, + "learning_rate": 0.0005884, + "loss": 2.7257, + "step": 2084 + }, + { + "epoch": 3.336, + "grad_norm": 3.34100604057312, + "learning_rate": 0.0005881999999999999, + "loss": 2.631, + "step": 2085 + }, + { + "epoch": 3.3376, + "grad_norm": 2.386023998260498, + "learning_rate": 0.000588, + "loss": 2.6013, + "step": 2086 + }, + { + "epoch": 3.3392, + "grad_norm": 2.3267204761505127, + "learning_rate": 0.0005878, + "loss": 2.4455, + "step": 2087 + }, + { + "epoch": 3.3407999999999998, + "grad_norm": 3.6707804203033447, + "learning_rate": 0.0005875999999999999, + "loss": 2.5229, + "step": 2088 + }, + { + "epoch": 3.3424, + "grad_norm": 2.1236207485198975, + "learning_rate": 0.0005874, + "loss": 2.5236, + "step": 2089 + }, + { + "epoch": 3.344, + "grad_norm": 3.2923009395599365, + "learning_rate": 0.0005872, + "loss": 2.642, + "step": 2090 + }, + { + "epoch": 3.3456, + "grad_norm": 5.383463382720947, + "learning_rate": 0.000587, + "loss": 2.5033, + "step": 2091 + }, + { + "epoch": 3.3472, + "grad_norm": 18.03877067565918, + "learning_rate": 0.0005868, + "loss": 3.8571, + "step": 2092 + }, + { + "epoch": 3.3487999999999998, + "grad_norm": 3.8737637996673584, + "learning_rate": 0.0005866, + "loss": 2.5824, + "step": 2093 + }, + { + "epoch": 3.3504, + "grad_norm": 3.6687588691711426, + "learning_rate": 0.0005863999999999999, + "loss": 2.6839, + "step": 2094 + }, + { + "epoch": 3.352, + "grad_norm": 2.334989547729492, + "learning_rate": 0.0005861999999999999, + "loss": 2.9145, + "step": 2095 + }, + { + "epoch": 3.3536, + "grad_norm": 2.323333501815796, + "learning_rate": 0.0005859999999999999, + "loss": 2.7732, + "step": 2096 + }, + { + "epoch": 3.3552, + "grad_norm": 2.2676596641540527, + "learning_rate": 0.0005857999999999999, + "loss": 2.5097, + "step": 2097 + }, + { + "epoch": 3.3568, + "grad_norm": 6.144402503967285, + "learning_rate": 0.0005855999999999999, + "loss": 2.5556, + "step": 2098 + }, + { + "epoch": 3.3584, + "grad_norm": 3.086674690246582, + "learning_rate": 0.0005853999999999999, + "loss": 2.8186, + "step": 2099 + }, + { + "epoch": 3.36, + "grad_norm": 2.0040550231933594, + "learning_rate": 0.0005852, + "loss": 2.7517, + "step": 2100 + }, + { + "epoch": 3.3616, + "grad_norm": 2.953817367553711, + "learning_rate": 0.0005849999999999999, + "loss": 2.2774, + "step": 2101 + }, + { + "epoch": 3.3632, + "grad_norm": 3.084397554397583, + "learning_rate": 0.0005848, + "loss": 2.5014, + "step": 2102 + }, + { + "epoch": 3.3648, + "grad_norm": 4.549737453460693, + "learning_rate": 0.0005846, + "loss": 2.5644, + "step": 2103 + }, + { + "epoch": 3.3664, + "grad_norm": 4.3595356941223145, + "learning_rate": 0.0005843999999999999, + "loss": 2.4014, + "step": 2104 + }, + { + "epoch": 3.368, + "grad_norm": 7.38173770904541, + "learning_rate": 0.0005842, + "loss": 2.6751, + "step": 2105 + }, + { + "epoch": 3.3696, + "grad_norm": 2.3727738857269287, + "learning_rate": 0.000584, + "loss": 2.5866, + "step": 2106 + }, + { + "epoch": 3.3712, + "grad_norm": 2.507899045944214, + "learning_rate": 0.0005838, + "loss": 2.735, + "step": 2107 + }, + { + "epoch": 3.3728, + "grad_norm": 1.9190720319747925, + "learning_rate": 0.0005836, + "loss": 2.9674, + "step": 2108 + }, + { + "epoch": 3.3744, + "grad_norm": 3.665943145751953, + "learning_rate": 0.0005834, + "loss": 2.5277, + "step": 2109 + }, + { + "epoch": 3.376, + "grad_norm": 6.320660591125488, + "learning_rate": 0.0005832, + "loss": 2.5458, + "step": 2110 + }, + { + "epoch": 3.3776, + "grad_norm": 4.5971503257751465, + "learning_rate": 0.000583, + "loss": 2.1244, + "step": 2111 + }, + { + "epoch": 3.3792, + "grad_norm": 4.167581558227539, + "learning_rate": 0.0005828, + "loss": 2.5984, + "step": 2112 + }, + { + "epoch": 3.3808, + "grad_norm": 3.1671297550201416, + "learning_rate": 0.0005826, + "loss": 2.6531, + "step": 2113 + }, + { + "epoch": 3.3824, + "grad_norm": 1.7537808418273926, + "learning_rate": 0.0005824, + "loss": 2.2473, + "step": 2114 + }, + { + "epoch": 3.384, + "grad_norm": 2.429500102996826, + "learning_rate": 0.0005822, + "loss": 2.6241, + "step": 2115 + }, + { + "epoch": 3.3856, + "grad_norm": 7.364534378051758, + "learning_rate": 0.0005819999999999999, + "loss": 2.5622, + "step": 2116 + }, + { + "epoch": 3.3872, + "grad_norm": 2.009014844894409, + "learning_rate": 0.0005817999999999999, + "loss": 2.7381, + "step": 2117 + }, + { + "epoch": 3.3888, + "grad_norm": 4.46831750869751, + "learning_rate": 0.0005815999999999999, + "loss": 2.6733, + "step": 2118 + }, + { + "epoch": 3.3904, + "grad_norm": 1.760084629058838, + "learning_rate": 0.0005813999999999999, + "loss": 2.7984, + "step": 2119 + }, + { + "epoch": 3.392, + "grad_norm": 1.8434828519821167, + "learning_rate": 0.0005811999999999999, + "loss": 3.2522, + "step": 2120 + }, + { + "epoch": 3.3936, + "grad_norm": 5.597031593322754, + "learning_rate": 0.000581, + "loss": 2.8036, + "step": 2121 + }, + { + "epoch": 3.3952, + "grad_norm": 5.435385227203369, + "learning_rate": 0.0005807999999999999, + "loss": 2.7302, + "step": 2122 + }, + { + "epoch": 3.3968, + "grad_norm": 3.0995161533355713, + "learning_rate": 0.0005805999999999999, + "loss": 2.7353, + "step": 2123 + }, + { + "epoch": 3.3984, + "grad_norm": 3.709087610244751, + "learning_rate": 0.0005804, + "loss": 2.9114, + "step": 2124 + }, + { + "epoch": 3.4, + "grad_norm": NaN, + "learning_rate": 0.0005804, + "loss": 2.6715, + "step": 2125 + }, + { + "epoch": 3.4016, + "grad_norm": 36.76521682739258, + "learning_rate": 0.0005801999999999999, + "loss": 4.7221, + "step": 2126 + }, + { + "epoch": 3.4032, + "grad_norm": 2.8888540267944336, + "learning_rate": 0.00058, + "loss": 3.1805, + "step": 2127 + }, + { + "epoch": 3.4048, + "grad_norm": 3.92763614654541, + "learning_rate": 0.0005798, + "loss": 3.751, + "step": 2128 + }, + { + "epoch": 3.4064, + "grad_norm": 3.051140785217285, + "learning_rate": 0.0005795999999999999, + "loss": 3.4025, + "step": 2129 + }, + { + "epoch": 3.408, + "grad_norm": 4.9858574867248535, + "learning_rate": 0.0005794, + "loss": 3.3216, + "step": 2130 + }, + { + "epoch": 3.4096, + "grad_norm": 1.9907106161117554, + "learning_rate": 0.0005792, + "loss": 3.0417, + "step": 2131 + }, + { + "epoch": 3.4112, + "grad_norm": 5.9723920822143555, + "learning_rate": 0.000579, + "loss": 2.9545, + "step": 2132 + }, + { + "epoch": 3.4128, + "grad_norm": 4.6594929695129395, + "learning_rate": 0.0005788, + "loss": 3.3907, + "step": 2133 + }, + { + "epoch": 3.4144, + "grad_norm": 3.687283754348755, + "learning_rate": 0.0005786, + "loss": 2.7718, + "step": 2134 + }, + { + "epoch": 3.416, + "grad_norm": 2.3645780086517334, + "learning_rate": 0.0005784, + "loss": 2.6572, + "step": 2135 + }, + { + "epoch": 3.4176, + "grad_norm": 4.219537734985352, + "learning_rate": 0.0005782, + "loss": 3.1178, + "step": 2136 + }, + { + "epoch": 3.4192, + "grad_norm": 8.279850006103516, + "learning_rate": 0.000578, + "loss": 2.7765, + "step": 2137 + }, + { + "epoch": 3.4208, + "grad_norm": 8.785918235778809, + "learning_rate": 0.0005778, + "loss": 2.9478, + "step": 2138 + }, + { + "epoch": 3.4224, + "grad_norm": 4.895820617675781, + "learning_rate": 0.0005775999999999999, + "loss": 2.7274, + "step": 2139 + }, + { + "epoch": 3.424, + "grad_norm": 4.281044006347656, + "learning_rate": 0.0005773999999999999, + "loss": 2.761, + "step": 2140 + }, + { + "epoch": 3.4256, + "grad_norm": 6.915059566497803, + "learning_rate": 0.0005771999999999999, + "loss": 2.6285, + "step": 2141 + }, + { + "epoch": 3.4272, + "grad_norm": 4.3538737297058105, + "learning_rate": 0.0005769999999999999, + "loss": 2.5224, + "step": 2142 + }, + { + "epoch": 3.4288, + "grad_norm": 3.245569944381714, + "learning_rate": 0.0005767999999999999, + "loss": 3.0929, + "step": 2143 + }, + { + "epoch": 3.4304, + "grad_norm": 3.608419179916382, + "learning_rate": 0.0005765999999999999, + "loss": 2.7562, + "step": 2144 + }, + { + "epoch": 3.432, + "grad_norm": 3.6872401237487793, + "learning_rate": 0.0005763999999999999, + "loss": 3.4305, + "step": 2145 + }, + { + "epoch": 3.4336, + "grad_norm": 6.966886043548584, + "learning_rate": 0.0005762, + "loss": 3.7485, + "step": 2146 + }, + { + "epoch": 3.4352, + "grad_norm": 3.857985496520996, + "learning_rate": 0.0005759999999999999, + "loss": 2.6345, + "step": 2147 + }, + { + "epoch": 3.4368, + "grad_norm": 2.1783053874969482, + "learning_rate": 0.0005757999999999999, + "loss": 2.4628, + "step": 2148 + }, + { + "epoch": 3.4384, + "grad_norm": 3.5218007564544678, + "learning_rate": 0.0005756, + "loss": 2.4832, + "step": 2149 + }, + { + "epoch": 3.44, + "grad_norm": 3.616251230239868, + "learning_rate": 0.0005753999999999999, + "loss": 2.6505, + "step": 2150 + }, + { + "epoch": 3.4416, + "grad_norm": 9.68862533569336, + "learning_rate": 0.0005752, + "loss": 2.672, + "step": 2151 + }, + { + "epoch": 3.4432, + "grad_norm": 8.697070121765137, + "learning_rate": 0.000575, + "loss": 2.7261, + "step": 2152 + }, + { + "epoch": 3.4448, + "grad_norm": 4.014763832092285, + "learning_rate": 0.0005747999999999999, + "loss": 2.955, + "step": 2153 + }, + { + "epoch": 3.4464, + "grad_norm": 6.026673793792725, + "learning_rate": 0.0005746, + "loss": 2.541, + "step": 2154 + }, + { + "epoch": 3.448, + "grad_norm": 2.990135908126831, + "learning_rate": 0.0005744, + "loss": 2.7884, + "step": 2155 + }, + { + "epoch": 3.4496, + "grad_norm": 2.3250844478607178, + "learning_rate": 0.0005742, + "loss": 2.6858, + "step": 2156 + }, + { + "epoch": 3.4512, + "grad_norm": 2.917367696762085, + "learning_rate": 0.000574, + "loss": 2.6458, + "step": 2157 + }, + { + "epoch": 3.4528, + "grad_norm": 3.353424310684204, + "learning_rate": 0.0005738, + "loss": 2.7368, + "step": 2158 + }, + { + "epoch": 3.4544, + "grad_norm": 4.0551323890686035, + "learning_rate": 0.0005736, + "loss": 2.8494, + "step": 2159 + }, + { + "epoch": 3.456, + "grad_norm": 2.2357308864593506, + "learning_rate": 0.0005734, + "loss": 2.4518, + "step": 2160 + }, + { + "epoch": 3.4576000000000002, + "grad_norm": 2.1687746047973633, + "learning_rate": 0.0005732, + "loss": 2.4907, + "step": 2161 + }, + { + "epoch": 3.4592, + "grad_norm": 2.188624143600464, + "learning_rate": 0.0005729999999999999, + "loss": 3.0182, + "step": 2162 + }, + { + "epoch": 3.4608, + "grad_norm": 3.905832052230835, + "learning_rate": 0.0005727999999999999, + "loss": 2.6255, + "step": 2163 + }, + { + "epoch": 3.4624, + "grad_norm": 2.897252321243286, + "learning_rate": 0.0005725999999999999, + "loss": 2.6728, + "step": 2164 + }, + { + "epoch": 3.464, + "grad_norm": 4.757837772369385, + "learning_rate": 0.0005723999999999999, + "loss": 2.8212, + "step": 2165 + }, + { + "epoch": 3.4656000000000002, + "grad_norm": 5.804827690124512, + "learning_rate": 0.0005721999999999999, + "loss": 3.5648, + "step": 2166 + }, + { + "epoch": 3.4672, + "grad_norm": 4.273700714111328, + "learning_rate": 0.0005719999999999999, + "loss": 2.8125, + "step": 2167 + }, + { + "epoch": 3.4688, + "grad_norm": 3.1538400650024414, + "learning_rate": 0.0005717999999999999, + "loss": 2.6748, + "step": 2168 + }, + { + "epoch": 3.4704, + "grad_norm": 2.156485080718994, + "learning_rate": 0.0005715999999999999, + "loss": 2.8008, + "step": 2169 + }, + { + "epoch": 3.472, + "grad_norm": 2.4508657455444336, + "learning_rate": 0.0005714, + "loss": 3.0445, + "step": 2170 + }, + { + "epoch": 3.4736000000000002, + "grad_norm": 2.667672634124756, + "learning_rate": 0.0005711999999999999, + "loss": 3.0013, + "step": 2171 + }, + { + "epoch": 3.4752, + "grad_norm": 2.244243860244751, + "learning_rate": 0.0005709999999999999, + "loss": 2.8604, + "step": 2172 + }, + { + "epoch": 3.4768, + "grad_norm": 2.0999715328216553, + "learning_rate": 0.0005708, + "loss": 3.1734, + "step": 2173 + }, + { + "epoch": 3.4784, + "grad_norm": 3.0818936824798584, + "learning_rate": 0.0005705999999999999, + "loss": 3.456, + "step": 2174 + }, + { + "epoch": 3.48, + "grad_norm": 2.844378709793091, + "learning_rate": 0.0005704, + "loss": 3.211, + "step": 2175 + }, + { + "epoch": 3.4816, + "grad_norm": 19.489864349365234, + "learning_rate": 0.0005702, + "loss": 4.3387, + "step": 2176 + }, + { + "epoch": 3.4832, + "grad_norm": 20.620067596435547, + "learning_rate": 0.00057, + "loss": 3.8751, + "step": 2177 + }, + { + "epoch": 3.4848, + "grad_norm": 20.510961532592773, + "learning_rate": 0.0005698, + "loss": 3.9736, + "step": 2178 + }, + { + "epoch": 3.4864, + "grad_norm": 13.105198860168457, + "learning_rate": 0.0005696, + "loss": 3.1943, + "step": 2179 + }, + { + "epoch": 3.488, + "grad_norm": 16.707609176635742, + "learning_rate": 0.0005694, + "loss": 3.9101, + "step": 2180 + }, + { + "epoch": 3.4896, + "grad_norm": 11.06164836883545, + "learning_rate": 0.0005692, + "loss": 3.3959, + "step": 2181 + }, + { + "epoch": 3.4912, + "grad_norm": 10.18997573852539, + "learning_rate": 0.000569, + "loss": 4.1144, + "step": 2182 + }, + { + "epoch": 3.4928, + "grad_norm": 3.5979034900665283, + "learning_rate": 0.0005688, + "loss": 3.6712, + "step": 2183 + }, + { + "epoch": 3.4944, + "grad_norm": 5.594130992889404, + "learning_rate": 0.0005685999999999999, + "loss": 3.3385, + "step": 2184 + }, + { + "epoch": 3.496, + "grad_norm": 4.669001579284668, + "learning_rate": 0.0005683999999999999, + "loss": 2.8457, + "step": 2185 + }, + { + "epoch": 3.4976, + "grad_norm": 3.709428310394287, + "learning_rate": 0.0005681999999999999, + "loss": 3.1379, + "step": 2186 + }, + { + "epoch": 3.4992, + "grad_norm": 2.764406204223633, + "learning_rate": 0.0005679999999999999, + "loss": 2.9813, + "step": 2187 + }, + { + "epoch": 3.5008, + "grad_norm": 5.064258575439453, + "learning_rate": 0.0005677999999999999, + "loss": 3.1149, + "step": 2188 + }, + { + "epoch": 3.5023999999999997, + "grad_norm": 7.851861953735352, + "learning_rate": 0.0005675999999999999, + "loss": 2.8176, + "step": 2189 + }, + { + "epoch": 3.504, + "grad_norm": 11.211861610412598, + "learning_rate": 0.0005673999999999999, + "loss": 3.3833, + "step": 2190 + }, + { + "epoch": 3.5056000000000003, + "grad_norm": 11.95122241973877, + "learning_rate": 0.0005672, + "loss": 3.6712, + "step": 2191 + }, + { + "epoch": 3.5072, + "grad_norm": 10.484594345092773, + "learning_rate": 0.0005669999999999999, + "loss": 3.233, + "step": 2192 + }, + { + "epoch": 3.5088, + "grad_norm": 34.02505874633789, + "learning_rate": 0.0005667999999999999, + "loss": 4.0065, + "step": 2193 + }, + { + "epoch": 3.5103999999999997, + "grad_norm": 7.815525531768799, + "learning_rate": 0.0005666, + "loss": 2.5542, + "step": 2194 + }, + { + "epoch": 3.512, + "grad_norm": 2.106872320175171, + "learning_rate": 0.0005663999999999999, + "loss": 2.8195, + "step": 2195 + }, + { + "epoch": 3.5136, + "grad_norm": 3.031090497970581, + "learning_rate": 0.0005662, + "loss": 3.145, + "step": 2196 + }, + { + "epoch": 3.5152, + "grad_norm": 2.442329168319702, + "learning_rate": 0.000566, + "loss": 2.8421, + "step": 2197 + }, + { + "epoch": 3.5168, + "grad_norm": 2.5574185848236084, + "learning_rate": 0.0005657999999999999, + "loss": 3.4521, + "step": 2198 + }, + { + "epoch": 3.5183999999999997, + "grad_norm": 4.203116416931152, + "learning_rate": 0.0005656, + "loss": 2.9609, + "step": 2199 + }, + { + "epoch": 3.52, + "grad_norm": 6.534364223480225, + "learning_rate": 0.0005654, + "loss": 3.5474, + "step": 2200 + }, + { + "epoch": 3.5216, + "grad_norm": 8.213438034057617, + "learning_rate": 0.0005652, + "loss": 3.1475, + "step": 2201 + }, + { + "epoch": 3.5232, + "grad_norm": 8.855327606201172, + "learning_rate": 0.000565, + "loss": 3.3337, + "step": 2202 + }, + { + "epoch": 3.5248, + "grad_norm": 5.681925296783447, + "learning_rate": 0.0005648, + "loss": 2.9456, + "step": 2203 + }, + { + "epoch": 3.5263999999999998, + "grad_norm": 10.223812103271484, + "learning_rate": 0.0005646, + "loss": 3.1749, + "step": 2204 + }, + { + "epoch": 3.528, + "grad_norm": 4.700658321380615, + "learning_rate": 0.0005644, + "loss": 3.2825, + "step": 2205 + }, + { + "epoch": 3.5296, + "grad_norm": 3.4023351669311523, + "learning_rate": 0.0005641999999999999, + "loss": 2.734, + "step": 2206 + }, + { + "epoch": 3.5312, + "grad_norm": 4.814945220947266, + "learning_rate": 0.0005639999999999999, + "loss": 2.5598, + "step": 2207 + }, + { + "epoch": 3.5328, + "grad_norm": 4.566649436950684, + "learning_rate": 0.0005637999999999999, + "loss": 2.9104, + "step": 2208 + }, + { + "epoch": 3.5343999999999998, + "grad_norm": 2.31830096244812, + "learning_rate": 0.0005635999999999999, + "loss": 2.575, + "step": 2209 + }, + { + "epoch": 3.536, + "grad_norm": 4.133569717407227, + "learning_rate": 0.0005633999999999999, + "loss": 3.195, + "step": 2210 + }, + { + "epoch": 3.5376, + "grad_norm": 2.573678731918335, + "learning_rate": 0.0005631999999999999, + "loss": 3.7206, + "step": 2211 + }, + { + "epoch": 3.5392, + "grad_norm": 7.069974422454834, + "learning_rate": 0.0005629999999999999, + "loss": 3.3338, + "step": 2212 + }, + { + "epoch": 3.5408, + "grad_norm": 3.81504225730896, + "learning_rate": 0.0005627999999999999, + "loss": 2.5689, + "step": 2213 + }, + { + "epoch": 3.5423999999999998, + "grad_norm": 8.163509368896484, + "learning_rate": 0.0005625999999999999, + "loss": 3.3907, + "step": 2214 + }, + { + "epoch": 3.544, + "grad_norm": 5.199379920959473, + "learning_rate": 0.0005624, + "loss": 2.9082, + "step": 2215 + }, + { + "epoch": 3.5456, + "grad_norm": 3.8271634578704834, + "learning_rate": 0.0005622, + "loss": 2.9105, + "step": 2216 + }, + { + "epoch": 3.5472, + "grad_norm": 3.3696279525756836, + "learning_rate": 0.0005619999999999999, + "loss": 2.8964, + "step": 2217 + }, + { + "epoch": 3.5488, + "grad_norm": 2.7918050289154053, + "learning_rate": 0.0005618, + "loss": 2.9344, + "step": 2218 + }, + { + "epoch": 3.5504, + "grad_norm": 2.721285343170166, + "learning_rate": 0.0005616, + "loss": 3.0515, + "step": 2219 + }, + { + "epoch": 3.552, + "grad_norm": 2.7775137424468994, + "learning_rate": 0.0005614, + "loss": 2.8442, + "step": 2220 + }, + { + "epoch": 3.5536, + "grad_norm": 2.965029001235962, + "learning_rate": 0.0005612, + "loss": 3.2262, + "step": 2221 + }, + { + "epoch": 3.5552, + "grad_norm": 2.9676098823547363, + "learning_rate": 0.000561, + "loss": 3.0243, + "step": 2222 + }, + { + "epoch": 3.5568, + "grad_norm": 2.839040756225586, + "learning_rate": 0.0005608, + "loss": 3.2154, + "step": 2223 + }, + { + "epoch": 3.5584, + "grad_norm": 5.983321666717529, + "learning_rate": 0.0005606, + "loss": 2.9206, + "step": 2224 + }, + { + "epoch": 3.56, + "grad_norm": NaN, + "learning_rate": 0.0005606, + "loss": 3.0367, + "step": 2225 + }, + { + "epoch": 3.5616, + "grad_norm": 55.5407600402832, + "learning_rate": 0.0005604, + "loss": 6.4386, + "step": 2226 + }, + { + "epoch": 3.5632, + "grad_norm": 9.289352416992188, + "learning_rate": 0.0005602, + "loss": 4.0924, + "step": 2227 + }, + { + "epoch": 3.5648, + "grad_norm": 14.431886672973633, + "learning_rate": 0.00056, + "loss": 4.1028, + "step": 2228 + }, + { + "epoch": 3.5664, + "grad_norm": 24.23824119567871, + "learning_rate": 0.0005598, + "loss": 5.3427, + "step": 2229 + }, + { + "epoch": 3.568, + "grad_norm": 8.971092224121094, + "learning_rate": 0.0005595999999999999, + "loss": 4.0412, + "step": 2230 + }, + { + "epoch": 3.5696, + "grad_norm": 17.683528900146484, + "learning_rate": 0.0005593999999999999, + "loss": 3.9749, + "step": 2231 + }, + { + "epoch": 3.5712, + "grad_norm": 3.934190511703491, + "learning_rate": 0.0005591999999999999, + "loss": 4.0728, + "step": 2232 + }, + { + "epoch": 3.5728, + "grad_norm": 4.2022705078125, + "learning_rate": 0.0005589999999999999, + "loss": 3.2269, + "step": 2233 + }, + { + "epoch": 3.5744, + "grad_norm": 6.388844013214111, + "learning_rate": 0.0005587999999999999, + "loss": 3.4502, + "step": 2234 + }, + { + "epoch": 3.576, + "grad_norm": 2.6255085468292236, + "learning_rate": 0.0005586, + "loss": 3.6688, + "step": 2235 + }, + { + "epoch": 3.5776, + "grad_norm": 3.0032479763031006, + "learning_rate": 0.0005583999999999999, + "loss": 3.9305, + "step": 2236 + }, + { + "epoch": 3.5792, + "grad_norm": 2.0182929039001465, + "learning_rate": 0.0005581999999999999, + "loss": 3.3079, + "step": 2237 + }, + { + "epoch": 3.5808, + "grad_norm": 6.957843780517578, + "learning_rate": 0.000558, + "loss": 3.6745, + "step": 2238 + }, + { + "epoch": 3.5824, + "grad_norm": 4.60112190246582, + "learning_rate": 0.0005577999999999999, + "loss": 2.9602, + "step": 2239 + }, + { + "epoch": 3.584, + "grad_norm": 6.672953128814697, + "learning_rate": 0.0005576, + "loss": 3.9476, + "step": 2240 + }, + { + "epoch": 3.5856, + "grad_norm": 16.332260131835938, + "learning_rate": 0.0005574, + "loss": 3.1406, + "step": 2241 + }, + { + "epoch": 3.5872, + "grad_norm": 7.7880401611328125, + "learning_rate": 0.0005571999999999999, + "loss": 3.3363, + "step": 2242 + }, + { + "epoch": 3.5888, + "grad_norm": 9.519548416137695, + "learning_rate": 0.000557, + "loss": 3.4756, + "step": 2243 + }, + { + "epoch": 3.5904, + "grad_norm": 7.101512908935547, + "learning_rate": 0.0005568, + "loss": 3.4255, + "step": 2244 + }, + { + "epoch": 3.592, + "grad_norm": 5.652818202972412, + "learning_rate": 0.0005566, + "loss": 2.7506, + "step": 2245 + }, + { + "epoch": 3.5936, + "grad_norm": 5.916102886199951, + "learning_rate": 0.0005564, + "loss": 2.9571, + "step": 2246 + }, + { + "epoch": 3.5952, + "grad_norm": 2.533904790878296, + "learning_rate": 0.0005562, + "loss": 2.5523, + "step": 2247 + }, + { + "epoch": 3.5968, + "grad_norm": 2.5718328952789307, + "learning_rate": 0.000556, + "loss": 3.6372, + "step": 2248 + }, + { + "epoch": 3.5984, + "grad_norm": 2.4668445587158203, + "learning_rate": 0.0005558, + "loss": 4.7817, + "step": 2249 + }, + { + "epoch": 3.6, + "grad_norm": 3.1797170639038086, + "learning_rate": 0.0005556, + "loss": 3.0265, + "step": 2250 + }, + { + "epoch": 3.6016, + "grad_norm": 2.3671298027038574, + "learning_rate": 0.0005554, + "loss": 2.495, + "step": 2251 + }, + { + "epoch": 3.6032, + "grad_norm": 7.401203155517578, + "learning_rate": 0.0005551999999999999, + "loss": 3.3338, + "step": 2252 + }, + { + "epoch": 3.6048, + "grad_norm": 7.648886680603027, + "learning_rate": 0.0005549999999999999, + "loss": 3.638, + "step": 2253 + }, + { + "epoch": 3.6064, + "grad_norm": 9.498307228088379, + "learning_rate": 0.0005547999999999999, + "loss": 3.1476, + "step": 2254 + }, + { + "epoch": 3.608, + "grad_norm": 10.403427124023438, + "learning_rate": 0.0005545999999999999, + "loss": 2.9936, + "step": 2255 + }, + { + "epoch": 3.6096, + "grad_norm": 10.709396362304688, + "learning_rate": 0.0005544, + "loss": 3.9072, + "step": 2256 + }, + { + "epoch": 3.6112, + "grad_norm": 5.007658958435059, + "learning_rate": 0.0005541999999999999, + "loss": 3.3232, + "step": 2257 + }, + { + "epoch": 3.6128, + "grad_norm": 5.138033390045166, + "learning_rate": 0.0005539999999999999, + "loss": 3.2679, + "step": 2258 + }, + { + "epoch": 3.6144, + "grad_norm": 4.138444423675537, + "learning_rate": 0.0005538, + "loss": 3.2151, + "step": 2259 + }, + { + "epoch": 3.616, + "grad_norm": 2.749028205871582, + "learning_rate": 0.0005535999999999999, + "loss": 3.6335, + "step": 2260 + }, + { + "epoch": 3.6176, + "grad_norm": 2.672708511352539, + "learning_rate": 0.0005534, + "loss": 2.5717, + "step": 2261 + }, + { + "epoch": 3.6192, + "grad_norm": 2.188662528991699, + "learning_rate": 0.0005532, + "loss": 3.0575, + "step": 2262 + }, + { + "epoch": 3.6208, + "grad_norm": 4.486994743347168, + "learning_rate": 0.0005529999999999999, + "loss": 3.5481, + "step": 2263 + }, + { + "epoch": 3.6224, + "grad_norm": 2.5265097618103027, + "learning_rate": 0.0005528, + "loss": 3.0689, + "step": 2264 + }, + { + "epoch": 3.624, + "grad_norm": 3.3407726287841797, + "learning_rate": 0.0005526, + "loss": 3.017, + "step": 2265 + }, + { + "epoch": 3.6256, + "grad_norm": 6.142095565795898, + "learning_rate": 0.0005524, + "loss": 2.7306, + "step": 2266 + }, + { + "epoch": 3.6272, + "grad_norm": 7.358902931213379, + "learning_rate": 0.0005522, + "loss": 5.4735, + "step": 2267 + }, + { + "epoch": 3.6288, + "grad_norm": 9.835092544555664, + "learning_rate": 0.000552, + "loss": 5.4242, + "step": 2268 + }, + { + "epoch": 3.6304, + "grad_norm": 6.334985733032227, + "learning_rate": 0.0005518, + "loss": 2.8247, + "step": 2269 + }, + { + "epoch": 3.632, + "grad_norm": 3.05710506439209, + "learning_rate": 0.0005516, + "loss": 3.1278, + "step": 2270 + }, + { + "epoch": 3.6336, + "grad_norm": 3.3504600524902344, + "learning_rate": 0.0005514, + "loss": 2.4559, + "step": 2271 + }, + { + "epoch": 3.6352, + "grad_norm": 4.229508876800537, + "learning_rate": 0.0005512, + "loss": 3.9111, + "step": 2272 + }, + { + "epoch": 3.6368, + "grad_norm": 2.7114548683166504, + "learning_rate": 0.000551, + "loss": 3.8723, + "step": 2273 + }, + { + "epoch": 3.6384, + "grad_norm": 6.074462413787842, + "learning_rate": 0.0005507999999999999, + "loss": 3.4165, + "step": 2274 + }, + { + "epoch": 3.64, + "grad_norm": NaN, + "learning_rate": 0.0005507999999999999, + "loss": 3.5407, + "step": 2275 + }, + { + "epoch": 3.6416, + "grad_norm": 7.515296936035156, + "learning_rate": 0.0005505999999999999, + "loss": 4.6945, + "step": 2276 + }, + { + "epoch": 3.6432, + "grad_norm": 0.9642918109893799, + "learning_rate": 0.0005503999999999999, + "loss": 4.4873, + "step": 2277 + }, + { + "epoch": 3.6448, + "grad_norm": 3.470863103866577, + "learning_rate": 0.0005501999999999999, + "loss": 4.744, + "step": 2278 + }, + { + "epoch": 3.6464, + "grad_norm": 1.6295921802520752, + "learning_rate": 0.0005499999999999999, + "loss": 3.9066, + "step": 2279 + }, + { + "epoch": 3.648, + "grad_norm": 0.9528029561042786, + "learning_rate": 0.0005497999999999999, + "loss": 3.3472, + "step": 2280 + }, + { + "epoch": 3.6496, + "grad_norm": 0.9496387839317322, + "learning_rate": 0.0005496, + "loss": 3.4575, + "step": 2281 + }, + { + "epoch": 3.6512000000000002, + "grad_norm": 1.0395222902297974, + "learning_rate": 0.0005493999999999999, + "loss": 3.2515, + "step": 2282 + }, + { + "epoch": 3.6528, + "grad_norm": 1.5674383640289307, + "learning_rate": 0.0005491999999999999, + "loss": 3.6947, + "step": 2283 + }, + { + "epoch": 3.6544, + "grad_norm": 1.2627230882644653, + "learning_rate": 0.000549, + "loss": 3.0947, + "step": 2284 + }, + { + "epoch": 3.656, + "grad_norm": 1.4390543699264526, + "learning_rate": 0.0005487999999999999, + "loss": 4.4352, + "step": 2285 + }, + { + "epoch": 3.6576, + "grad_norm": 1.4020869731903076, + "learning_rate": 0.0005486, + "loss": 3.9516, + "step": 2286 + }, + { + "epoch": 3.6592000000000002, + "grad_norm": 3.6883347034454346, + "learning_rate": 0.0005484, + "loss": 4.0214, + "step": 2287 + }, + { + "epoch": 3.6608, + "grad_norm": 4.250986576080322, + "learning_rate": 0.0005481999999999999, + "loss": 3.2447, + "step": 2288 + }, + { + "epoch": 3.6624, + "grad_norm": 2.007300615310669, + "learning_rate": 0.000548, + "loss": 4.1801, + "step": 2289 + }, + { + "epoch": 3.664, + "grad_norm": 3.120436191558838, + "learning_rate": 0.0005478, + "loss": 4.7246, + "step": 2290 + }, + { + "epoch": 3.6656, + "grad_norm": 3.354717254638672, + "learning_rate": 0.0005476, + "loss": 3.4444, + "step": 2291 + }, + { + "epoch": 3.6672000000000002, + "grad_norm": 2.1865100860595703, + "learning_rate": 0.0005474, + "loss": 3.7239, + "step": 2292 + }, + { + "epoch": 3.6688, + "grad_norm": 2.3343372344970703, + "learning_rate": 0.0005472, + "loss": 4.016, + "step": 2293 + }, + { + "epoch": 3.6704, + "grad_norm": 3.9303016662597656, + "learning_rate": 0.000547, + "loss": 4.0466, + "step": 2294 + }, + { + "epoch": 3.672, + "grad_norm": 4.819889068603516, + "learning_rate": 0.0005468, + "loss": 3.3989, + "step": 2295 + }, + { + "epoch": 3.6736, + "grad_norm": 3.13912296295166, + "learning_rate": 0.0005466, + "loss": 4.0242, + "step": 2296 + }, + { + "epoch": 3.6752000000000002, + "grad_norm": 2.8301594257354736, + "learning_rate": 0.0005463999999999999, + "loss": 3.1699, + "step": 2297 + }, + { + "epoch": 3.6768, + "grad_norm": 1.9209338426589966, + "learning_rate": 0.0005461999999999999, + "loss": 3.1893, + "step": 2298 + }, + { + "epoch": 3.6784, + "grad_norm": 1.8804454803466797, + "learning_rate": 0.0005459999999999999, + "loss": 3.2806, + "step": 2299 + }, + { + "epoch": 3.68, + "grad_norm": 2.186152458190918, + "learning_rate": 0.0005457999999999999, + "loss": 2.8408, + "step": 2300 + }, + { + "epoch": 3.6816, + "grad_norm": 4.055525779724121, + "learning_rate": 0.0005455999999999999, + "loss": 3.7474, + "step": 2301 + }, + { + "epoch": 3.6832000000000003, + "grad_norm": 4.328959941864014, + "learning_rate": 0.0005453999999999999, + "loss": 3.3388, + "step": 2302 + }, + { + "epoch": 3.6848, + "grad_norm": 12.9420804977417, + "learning_rate": 0.0005451999999999999, + "loss": 4.7525, + "step": 2303 + }, + { + "epoch": 3.6864, + "grad_norm": 3.576298236846924, + "learning_rate": 0.0005449999999999999, + "loss": 3.9676, + "step": 2304 + }, + { + "epoch": 3.6879999999999997, + "grad_norm": 2.730264186859131, + "learning_rate": 0.0005448, + "loss": 3.3703, + "step": 2305 + }, + { + "epoch": 3.6896, + "grad_norm": 1.9057469367980957, + "learning_rate": 0.0005445999999999999, + "loss": 3.0003, + "step": 2306 + }, + { + "epoch": 3.6912000000000003, + "grad_norm": 2.068631172180176, + "learning_rate": 0.0005443999999999999, + "loss": 3.6165, + "step": 2307 + }, + { + "epoch": 3.6928, + "grad_norm": 3.0546584129333496, + "learning_rate": 0.0005442, + "loss": 4.0022, + "step": 2308 + }, + { + "epoch": 3.6944, + "grad_norm": 5.953769207000732, + "learning_rate": 0.0005439999999999999, + "loss": 3.903, + "step": 2309 + }, + { + "epoch": 3.6959999999999997, + "grad_norm": 3.818378448486328, + "learning_rate": 0.0005438, + "loss": 2.8817, + "step": 2310 + }, + { + "epoch": 3.6976, + "grad_norm": 6.505511283874512, + "learning_rate": 0.0005436, + "loss": 4.2538, + "step": 2311 + }, + { + "epoch": 3.6992000000000003, + "grad_norm": 3.884477138519287, + "learning_rate": 0.0005433999999999999, + "loss": 2.9049, + "step": 2312 + }, + { + "epoch": 3.7008, + "grad_norm": 5.818808555603027, + "learning_rate": 0.0005432, + "loss": 3.8679, + "step": 2313 + }, + { + "epoch": 3.7024, + "grad_norm": 2.5341482162475586, + "learning_rate": 0.000543, + "loss": 3.1927, + "step": 2314 + }, + { + "epoch": 3.7039999999999997, + "grad_norm": 2.6208629608154297, + "learning_rate": 0.0005428, + "loss": 4.2163, + "step": 2315 + }, + { + "epoch": 3.7056, + "grad_norm": 2.189680337905884, + "learning_rate": 0.0005426, + "loss": 2.5342, + "step": 2316 + }, + { + "epoch": 3.7072000000000003, + "grad_norm": 2.4418387413024902, + "learning_rate": 0.0005424, + "loss": 3.01, + "step": 2317 + }, + { + "epoch": 3.7088, + "grad_norm": 2.613811731338501, + "learning_rate": 0.0005422, + "loss": 2.6786, + "step": 2318 + }, + { + "epoch": 3.7104, + "grad_norm": 8.510128021240234, + "learning_rate": 0.000542, + "loss": 3.3229, + "step": 2319 + }, + { + "epoch": 3.7119999999999997, + "grad_norm": 3.6487812995910645, + "learning_rate": 0.0005417999999999999, + "loss": 2.6784, + "step": 2320 + }, + { + "epoch": 3.7136, + "grad_norm": 3.9220001697540283, + "learning_rate": 0.0005415999999999999, + "loss": 3.1996, + "step": 2321 + }, + { + "epoch": 3.7152, + "grad_norm": 2.069011688232422, + "learning_rate": 0.0005413999999999999, + "loss": 3.2684, + "step": 2322 + }, + { + "epoch": 3.7168, + "grad_norm": 3.0379621982574463, + "learning_rate": 0.0005411999999999999, + "loss": 3.0774, + "step": 2323 + }, + { + "epoch": 3.7184, + "grad_norm": 2.2943036556243896, + "learning_rate": 0.0005409999999999999, + "loss": 3.2742, + "step": 2324 + }, + { + "epoch": 3.7199999999999998, + "grad_norm": 6.881977081298828, + "learning_rate": 0.0005407999999999999, + "loss": 4.7077, + "step": 2325 + }, + { + "epoch": 3.7216, + "grad_norm": 81.76461791992188, + "learning_rate": 0.0005405999999999999, + "loss": 7.5684, + "step": 2326 + }, + { + "epoch": 3.7232, + "grad_norm": 10.492643356323242, + "learning_rate": 0.0005403999999999999, + "loss": 4.9986, + "step": 2327 + }, + { + "epoch": 3.7248, + "grad_norm": 4.756387710571289, + "learning_rate": 0.0005401999999999999, + "loss": 3.9451, + "step": 2328 + }, + { + "epoch": 3.7264, + "grad_norm": 2.5230603218078613, + "learning_rate": 0.00054, + "loss": 3.8227, + "step": 2329 + }, + { + "epoch": 3.7279999999999998, + "grad_norm": 2.209505796432495, + "learning_rate": 0.0005397999999999999, + "loss": 3.8123, + "step": 2330 + }, + { + "epoch": 3.7296, + "grad_norm": 2.965146064758301, + "learning_rate": 0.0005396, + "loss": 4.5036, + "step": 2331 + }, + { + "epoch": 3.7312, + "grad_norm": 2.212378978729248, + "learning_rate": 0.0005394, + "loss": 4.0939, + "step": 2332 + }, + { + "epoch": 3.7328, + "grad_norm": 1.8609493970870972, + "learning_rate": 0.0005391999999999999, + "loss": 3.5066, + "step": 2333 + }, + { + "epoch": 3.7344, + "grad_norm": 2.2023518085479736, + "learning_rate": 0.000539, + "loss": 3.911, + "step": 2334 + }, + { + "epoch": 3.7359999999999998, + "grad_norm": 2.8470547199249268, + "learning_rate": 0.0005388, + "loss": 3.2395, + "step": 2335 + }, + { + "epoch": 3.7376, + "grad_norm": 2.1184237003326416, + "learning_rate": 0.0005386, + "loss": 3.7863, + "step": 2336 + }, + { + "epoch": 3.7392, + "grad_norm": 2.1715586185455322, + "learning_rate": 0.0005384, + "loss": 4.5799, + "step": 2337 + }, + { + "epoch": 3.7408, + "grad_norm": 2.3556015491485596, + "learning_rate": 0.0005382, + "loss": 3.8685, + "step": 2338 + }, + { + "epoch": 3.7424, + "grad_norm": 1.3282142877578735, + "learning_rate": 0.000538, + "loss": 3.316, + "step": 2339 + }, + { + "epoch": 3.7439999999999998, + "grad_norm": 4.5761542320251465, + "learning_rate": 0.0005378, + "loss": 4.0411, + "step": 2340 + }, + { + "epoch": 3.7456, + "grad_norm": 4.5847554206848145, + "learning_rate": 0.0005376, + "loss": 4.7927, + "step": 2341 + }, + { + "epoch": 3.7472, + "grad_norm": 12.918116569519043, + "learning_rate": 0.0005373999999999999, + "loss": 3.7438, + "step": 2342 + }, + { + "epoch": 3.7488, + "grad_norm": 4.8914361000061035, + "learning_rate": 0.0005371999999999999, + "loss": 4.3877, + "step": 2343 + }, + { + "epoch": 3.7504, + "grad_norm": 5.574761867523193, + "learning_rate": 0.0005369999999999999, + "loss": 3.7056, + "step": 2344 + }, + { + "epoch": 3.752, + "grad_norm": 6.177387237548828, + "learning_rate": 0.0005368, + "loss": 3.6515, + "step": 2345 + }, + { + "epoch": 3.7536, + "grad_norm": 4.47343635559082, + "learning_rate": 0.0005365999999999999, + "loss": 3.5961, + "step": 2346 + }, + { + "epoch": 3.7552, + "grad_norm": 6.870029449462891, + "learning_rate": 0.0005363999999999999, + "loss": 4.1159, + "step": 2347 + }, + { + "epoch": 3.7568, + "grad_norm": 3.4203975200653076, + "learning_rate": 0.0005362, + "loss": 3.3731, + "step": 2348 + }, + { + "epoch": 3.7584, + "grad_norm": 5.828915119171143, + "learning_rate": 0.0005359999999999999, + "loss": 3.5227, + "step": 2349 + }, + { + "epoch": 3.76, + "grad_norm": 2.20630145072937, + "learning_rate": 0.0005358, + "loss": 3.7634, + "step": 2350 + }, + { + "epoch": 3.7616, + "grad_norm": 4.080089569091797, + "learning_rate": 0.0005356, + "loss": 5.0694, + "step": 2351 + }, + { + "epoch": 3.7632, + "grad_norm": 3.561363697052002, + "learning_rate": 0.0005353999999999999, + "loss": 3.7534, + "step": 2352 + }, + { + "epoch": 3.7648, + "grad_norm": 3.118837833404541, + "learning_rate": 0.0005352, + "loss": 4.8316, + "step": 2353 + }, + { + "epoch": 3.7664, + "grad_norm": 3.6192729473114014, + "learning_rate": 0.000535, + "loss": 3.5515, + "step": 2354 + }, + { + "epoch": 3.768, + "grad_norm": 3.632134437561035, + "learning_rate": 0.0005348, + "loss": 3.4193, + "step": 2355 + }, + { + "epoch": 3.7696, + "grad_norm": 2.139364719390869, + "learning_rate": 0.0005346, + "loss": 4.3545, + "step": 2356 + }, + { + "epoch": 3.7712, + "grad_norm": 4.8385419845581055, + "learning_rate": 0.0005344, + "loss": 4.3522, + "step": 2357 + }, + { + "epoch": 3.7728, + "grad_norm": 5.820652484893799, + "learning_rate": 0.0005342, + "loss": 3.4902, + "step": 2358 + }, + { + "epoch": 3.7744, + "grad_norm": 7.365394592285156, + "learning_rate": 0.000534, + "loss": 3.2174, + "step": 2359 + }, + { + "epoch": 3.776, + "grad_norm": 8.943775177001953, + "learning_rate": 0.0005338, + "loss": 3.7161, + "step": 2360 + }, + { + "epoch": 3.7776, + "grad_norm": 12.740777015686035, + "learning_rate": 0.0005336, + "loss": 4.0428, + "step": 2361 + }, + { + "epoch": 3.7792, + "grad_norm": 12.166577339172363, + "learning_rate": 0.0005334, + "loss": 3.3414, + "step": 2362 + }, + { + "epoch": 3.7808, + "grad_norm": 8.097125053405762, + "learning_rate": 0.0005332, + "loss": 3.7573, + "step": 2363 + }, + { + "epoch": 3.7824, + "grad_norm": 8.305039405822754, + "learning_rate": 0.0005329999999999999, + "loss": 4.6625, + "step": 2364 + }, + { + "epoch": 3.784, + "grad_norm": 4.6849260330200195, + "learning_rate": 0.0005327999999999999, + "loss": 3.0211, + "step": 2365 + }, + { + "epoch": 3.7856, + "grad_norm": 2.537353992462158, + "learning_rate": 0.0005325999999999999, + "loss": 4.5432, + "step": 2366 + }, + { + "epoch": 3.7872, + "grad_norm": 2.6421570777893066, + "learning_rate": 0.0005323999999999999, + "loss": 2.6823, + "step": 2367 + }, + { + "epoch": 3.7888, + "grad_norm": 2.8577213287353516, + "learning_rate": 0.0005321999999999999, + "loss": 3.7853, + "step": 2368 + }, + { + "epoch": 3.7904, + "grad_norm": 2.6776928901672363, + "learning_rate": 0.000532, + "loss": 3.8282, + "step": 2369 + }, + { + "epoch": 3.792, + "grad_norm": 2.697481632232666, + "learning_rate": 0.0005317999999999999, + "loss": 3.5799, + "step": 2370 + }, + { + "epoch": 3.7936, + "grad_norm": 4.719255447387695, + "learning_rate": 0.0005315999999999999, + "loss": 3.7143, + "step": 2371 + }, + { + "epoch": 3.7952, + "grad_norm": 5.107699871063232, + "learning_rate": 0.0005314, + "loss": 4.2279, + "step": 2372 + }, + { + "epoch": 3.7968, + "grad_norm": 9.929701805114746, + "learning_rate": 0.0005311999999999999, + "loss": 3.9941, + "step": 2373 + }, + { + "epoch": 3.7984, + "grad_norm": 11.735438346862793, + "learning_rate": 0.000531, + "loss": 3.7405, + "step": 2374 + }, + { + "epoch": 3.8, + "grad_norm": 5.764272689819336, + "learning_rate": 0.0005308, + "loss": 4.5021, + "step": 2375 + }, + { + "epoch": 3.8016, + "grad_norm": 18.376399993896484, + "learning_rate": 0.0005305999999999999, + "loss": 6.781, + "step": 2376 + }, + { + "epoch": 3.8032, + "grad_norm": 14.563941955566406, + "learning_rate": 0.0005304, + "loss": 7.7015, + "step": 2377 + }, + { + "epoch": 3.8048, + "grad_norm": 1.9453444480895996, + "learning_rate": 0.0005302, + "loss": 4.7244, + "step": 2378 + }, + { + "epoch": 3.8064, + "grad_norm": 4.174312114715576, + "learning_rate": 0.00053, + "loss": 4.4179, + "step": 2379 + }, + { + "epoch": 3.808, + "grad_norm": 0.9180065393447876, + "learning_rate": 0.0005298, + "loss": 3.8616, + "step": 2380 + }, + { + "epoch": 3.8096, + "grad_norm": 2.4266366958618164, + "learning_rate": 0.0005296, + "loss": 4.7, + "step": 2381 + }, + { + "epoch": 3.8112, + "grad_norm": 2.9785518646240234, + "learning_rate": 0.0005294, + "loss": 5.1105, + "step": 2382 + }, + { + "epoch": 3.8128, + "grad_norm": 3.8676366806030273, + "learning_rate": 0.0005292, + "loss": 4.7435, + "step": 2383 + }, + { + "epoch": 3.8144, + "grad_norm": 2.722191572189331, + "learning_rate": 0.000529, + "loss": 4.3771, + "step": 2384 + }, + { + "epoch": 3.816, + "grad_norm": 1.83686101436615, + "learning_rate": 0.0005288, + "loss": 3.7945, + "step": 2385 + }, + { + "epoch": 3.8176, + "grad_norm": 1.6038964986801147, + "learning_rate": 0.0005286, + "loss": 3.8832, + "step": 2386 + }, + { + "epoch": 3.8192, + "grad_norm": 1.0230307579040527, + "learning_rate": 0.0005283999999999999, + "loss": 4.6346, + "step": 2387 + }, + { + "epoch": 3.8208, + "grad_norm": 1.8154736757278442, + "learning_rate": 0.0005281999999999999, + "loss": 3.3476, + "step": 2388 + }, + { + "epoch": 3.8224, + "grad_norm": 9.780559539794922, + "learning_rate": 0.0005279999999999999, + "loss": 5.038, + "step": 2389 + }, + { + "epoch": 3.824, + "grad_norm": 6.866722106933594, + "learning_rate": 0.0005277999999999999, + "loss": 5.6834, + "step": 2390 + }, + { + "epoch": 3.8256, + "grad_norm": 5.508349418640137, + "learning_rate": 0.0005275999999999999, + "loss": 3.5608, + "step": 2391 + }, + { + "epoch": 3.8272, + "grad_norm": 16.235729217529297, + "learning_rate": 0.0005273999999999999, + "loss": 5.1608, + "step": 2392 + }, + { + "epoch": 3.8288, + "grad_norm": 8.720699310302734, + "learning_rate": 0.0005272, + "loss": 5.3787, + "step": 2393 + }, + { + "epoch": 3.8304, + "grad_norm": 16.350589752197266, + "learning_rate": 0.0005269999999999999, + "loss": 4.7983, + "step": 2394 + }, + { + "epoch": 3.832, + "grad_norm": 6.838560104370117, + "learning_rate": 0.0005267999999999999, + "loss": 3.8524, + "step": 2395 + }, + { + "epoch": 3.8336, + "grad_norm": 11.337925910949707, + "learning_rate": 0.0005266, + "loss": 4.7016, + "step": 2396 + }, + { + "epoch": 3.8352, + "grad_norm": 7.908731937408447, + "learning_rate": 0.0005263999999999999, + "loss": 4.3663, + "step": 2397 + }, + { + "epoch": 3.8368, + "grad_norm": 4.7114787101745605, + "learning_rate": 0.0005262, + "loss": 3.8907, + "step": 2398 + }, + { + "epoch": 3.8384, + "grad_norm": 4.332289695739746, + "learning_rate": 0.000526, + "loss": 3.3793, + "step": 2399 + }, + { + "epoch": 3.84, + "grad_norm": 2.8754754066467285, + "learning_rate": 0.0005257999999999999, + "loss": 4.017, + "step": 2400 + }, + { + "epoch": 3.8416, + "grad_norm": 3.1306464672088623, + "learning_rate": 0.0005256, + "loss": 4.1431, + "step": 2401 + }, + { + "epoch": 3.8432, + "grad_norm": 3.269331455230713, + "learning_rate": 0.0005254, + "loss": 3.5724, + "step": 2402 + }, + { + "epoch": 3.8448, + "grad_norm": 2.9769835472106934, + "learning_rate": 0.0005252, + "loss": 3.1223, + "step": 2403 + }, + { + "epoch": 3.8464, + "grad_norm": 2.5394082069396973, + "learning_rate": 0.000525, + "loss": 4.9138, + "step": 2404 + }, + { + "epoch": 3.848, + "grad_norm": 2.2424628734588623, + "learning_rate": 0.0005248, + "loss": 3.1959, + "step": 2405 + }, + { + "epoch": 3.8496, + "grad_norm": 7.9830803871154785, + "learning_rate": 0.0005246, + "loss": 3.0957, + "step": 2406 + }, + { + "epoch": 3.8512, + "grad_norm": 6.655513763427734, + "learning_rate": 0.0005244, + "loss": 4.3865, + "step": 2407 + }, + { + "epoch": 3.8528000000000002, + "grad_norm": 6.7835211753845215, + "learning_rate": 0.0005242, + "loss": 3.7315, + "step": 2408 + }, + { + "epoch": 3.8544, + "grad_norm": 9.676862716674805, + "learning_rate": 0.0005239999999999999, + "loss": 6.1095, + "step": 2409 + }, + { + "epoch": 3.856, + "grad_norm": 6.528567314147949, + "learning_rate": 0.0005237999999999999, + "loss": 4.2052, + "step": 2410 + }, + { + "epoch": 3.8576, + "grad_norm": 6.409698963165283, + "learning_rate": 0.0005235999999999999, + "loss": 3.2719, + "step": 2411 + }, + { + "epoch": 3.8592, + "grad_norm": 5.9662628173828125, + "learning_rate": 0.0005233999999999999, + "loss": 3.7916, + "step": 2412 + }, + { + "epoch": 3.8608000000000002, + "grad_norm": 11.17818832397461, + "learning_rate": 0.0005231999999999999, + "loss": 7.6018, + "step": 2413 + }, + { + "epoch": 3.8624, + "grad_norm": 2.0362625122070312, + "learning_rate": 0.000523, + "loss": 3.7233, + "step": 2414 + }, + { + "epoch": 3.864, + "grad_norm": 2.409855365753174, + "learning_rate": 0.0005227999999999999, + "loss": 3.1131, + "step": 2415 + }, + { + "epoch": 3.8656, + "grad_norm": 2.5807554721832275, + "learning_rate": 0.0005225999999999999, + "loss": 4.7464, + "step": 2416 + }, + { + "epoch": 3.8672, + "grad_norm": 2.853008985519409, + "learning_rate": 0.0005224, + "loss": 3.9214, + "step": 2417 + }, + { + "epoch": 3.8688000000000002, + "grad_norm": 2.9966490268707275, + "learning_rate": 0.0005221999999999999, + "loss": 3.1906, + "step": 2418 + }, + { + "epoch": 3.8704, + "grad_norm": 2.0658886432647705, + "learning_rate": 0.000522, + "loss": 3.5001, + "step": 2419 + }, + { + "epoch": 3.872, + "grad_norm": 8.184765815734863, + "learning_rate": 0.0005218, + "loss": 3.4756, + "step": 2420 + }, + { + "epoch": 3.8736, + "grad_norm": 4.851075649261475, + "learning_rate": 0.0005215999999999999, + "loss": 3.7197, + "step": 2421 + }, + { + "epoch": 3.8752, + "grad_norm": 11.42324447631836, + "learning_rate": 0.0005214, + "loss": 3.2422, + "step": 2422 + }, + { + "epoch": 3.8768000000000002, + "grad_norm": 8.599984169006348, + "learning_rate": 0.0005212, + "loss": 3.459, + "step": 2423 + }, + { + "epoch": 3.8784, + "grad_norm": 6.367029666900635, + "learning_rate": 0.000521, + "loss": 5.2247, + "step": 2424 + }, + { + "epoch": 3.88, + "grad_norm": 10.062922477722168, + "learning_rate": 0.0005208, + "loss": 3.6883, + "step": 2425 + }, + { + "epoch": 3.8816, + "grad_norm": 7.4639668464660645, + "learning_rate": 0.0005206, + "loss": 6.8429, + "step": 2426 + }, + { + "epoch": 3.8832, + "grad_norm": 7.481766700744629, + "learning_rate": 0.0005204, + "loss": 7.3029, + "step": 2427 + }, + { + "epoch": 3.8848000000000003, + "grad_norm": 9.236445426940918, + "learning_rate": 0.0005202, + "loss": 5.3437, + "step": 2428 + }, + { + "epoch": 3.8864, + "grad_norm": 0.8947409391403198, + "learning_rate": 0.00052, + "loss": 5.7565, + "step": 2429 + }, + { + "epoch": 3.888, + "grad_norm": 1.8359402418136597, + "learning_rate": 0.0005198, + "loss": 6.744, + "step": 2430 + }, + { + "epoch": 3.8895999999999997, + "grad_norm": 1.6362475156784058, + "learning_rate": 0.0005195999999999999, + "loss": 4.6432, + "step": 2431 + }, + { + "epoch": 3.8912, + "grad_norm": 3.711618423461914, + "learning_rate": 0.0005193999999999999, + "loss": 3.4999, + "step": 2432 + }, + { + "epoch": 3.8928000000000003, + "grad_norm": 3.213042974472046, + "learning_rate": 0.0005191999999999999, + "loss": 4.144, + "step": 2433 + }, + { + "epoch": 3.8944, + "grad_norm": 3.1469593048095703, + "learning_rate": 0.0005189999999999999, + "loss": 3.7807, + "step": 2434 + }, + { + "epoch": 3.896, + "grad_norm": 1.5552211999893188, + "learning_rate": 0.0005187999999999999, + "loss": 5.4997, + "step": 2435 + }, + { + "epoch": 3.8975999999999997, + "grad_norm": 12.214163780212402, + "learning_rate": 0.0005185999999999999, + "loss": 4.3924, + "step": 2436 + }, + { + "epoch": 3.8992, + "grad_norm": 2.1951797008514404, + "learning_rate": 0.0005183999999999999, + "loss": 5.7433, + "step": 2437 + }, + { + "epoch": 3.9008000000000003, + "grad_norm": 2.7041990756988525, + "learning_rate": 0.0005182, + "loss": 4.0962, + "step": 2438 + }, + { + "epoch": 3.9024, + "grad_norm": 2.592421293258667, + "learning_rate": 0.0005179999999999999, + "loss": 4.6299, + "step": 2439 + }, + { + "epoch": 3.904, + "grad_norm": 2.230567693710327, + "learning_rate": 0.0005177999999999999, + "loss": 3.002, + "step": 2440 + }, + { + "epoch": 3.9055999999999997, + "grad_norm": 3.1844351291656494, + "learning_rate": 0.0005176, + "loss": 3.6995, + "step": 2441 + }, + { + "epoch": 3.9072, + "grad_norm": 4.0518107414245605, + "learning_rate": 0.0005173999999999999, + "loss": 3.7061, + "step": 2442 + }, + { + "epoch": 3.9088000000000003, + "grad_norm": 4.409000873565674, + "learning_rate": 0.0005172, + "loss": 3.8965, + "step": 2443 + }, + { + "epoch": 3.9104, + "grad_norm": 2.557565689086914, + "learning_rate": 0.000517, + "loss": 3.649, + "step": 2444 + }, + { + "epoch": 3.912, + "grad_norm": 4.122837066650391, + "learning_rate": 0.0005167999999999999, + "loss": 4.4164, + "step": 2445 + }, + { + "epoch": 3.9135999999999997, + "grad_norm": 3.307701587677002, + "learning_rate": 0.0005166, + "loss": 3.8117, + "step": 2446 + }, + { + "epoch": 3.9152, + "grad_norm": 2.113417387008667, + "learning_rate": 0.0005164, + "loss": 3.5849, + "step": 2447 + }, + { + "epoch": 3.9168, + "grad_norm": 5.073932647705078, + "learning_rate": 0.0005162, + "loss": 4.0564, + "step": 2448 + }, + { + "epoch": 3.9184, + "grad_norm": 2.0293564796447754, + "learning_rate": 0.000516, + "loss": 4.6822, + "step": 2449 + }, + { + "epoch": 3.92, + "grad_norm": 2.3307976722717285, + "learning_rate": 0.0005158, + "loss": 5.6256, + "step": 2450 + }, + { + "epoch": 3.9215999999999998, + "grad_norm": 2.8671512603759766, + "learning_rate": 0.0005156, + "loss": 3.3186, + "step": 2451 + }, + { + "epoch": 3.9232, + "grad_norm": 2.8677480220794678, + "learning_rate": 0.0005154, + "loss": 3.3332, + "step": 2452 + }, + { + "epoch": 3.9248, + "grad_norm": 5.9910478591918945, + "learning_rate": 0.0005152, + "loss": 5.1829, + "step": 2453 + }, + { + "epoch": 3.9264, + "grad_norm": 3.645705461502075, + "learning_rate": 0.0005149999999999999, + "loss": 3.3698, + "step": 2454 + }, + { + "epoch": 3.928, + "grad_norm": 5.09229040145874, + "learning_rate": 0.0005147999999999999, + "loss": 3.4024, + "step": 2455 + }, + { + "epoch": 3.9295999999999998, + "grad_norm": 3.315643310546875, + "learning_rate": 0.0005145999999999999, + "loss": 2.8343, + "step": 2456 + }, + { + "epoch": 3.9312, + "grad_norm": 6.215513229370117, + "learning_rate": 0.0005143999999999999, + "loss": 4.926, + "step": 2457 + }, + { + "epoch": 3.9328, + "grad_norm": 7.02705717086792, + "learning_rate": 0.0005141999999999999, + "loss": 5.8658, + "step": 2458 + }, + { + "epoch": 3.9344, + "grad_norm": 3.302187442779541, + "learning_rate": 0.0005139999999999999, + "loss": 3.6607, + "step": 2459 + }, + { + "epoch": 3.936, + "grad_norm": 2.8426706790924072, + "learning_rate": 0.0005137999999999999, + "loss": 3.3349, + "step": 2460 + }, + { + "epoch": 3.9375999999999998, + "grad_norm": 2.594999074935913, + "learning_rate": 0.0005135999999999999, + "loss": 3.7503, + "step": 2461 + }, + { + "epoch": 3.9392, + "grad_norm": 2.7870514392852783, + "learning_rate": 0.0005134, + "loss": 5.2539, + "step": 2462 + }, + { + "epoch": 3.9408, + "grad_norm": 2.494035482406616, + "learning_rate": 0.0005131999999999999, + "loss": 3.7372, + "step": 2463 + }, + { + "epoch": 3.9424, + "grad_norm": 5.120970249176025, + "learning_rate": 0.0005129999999999999, + "loss": 4.3101, + "step": 2464 + }, + { + "epoch": 3.944, + "grad_norm": 5.125430583953857, + "learning_rate": 0.0005128, + "loss": 3.7904, + "step": 2465 + }, + { + "epoch": 3.9455999999999998, + "grad_norm": 8.505355834960938, + "learning_rate": 0.0005125999999999999, + "loss": 4.2762, + "step": 2466 + }, + { + "epoch": 3.9472, + "grad_norm": 8.884271621704102, + "learning_rate": 0.0005124, + "loss": 3.7425, + "step": 2467 + }, + { + "epoch": 3.9488, + "grad_norm": 5.896363735198975, + "learning_rate": 0.0005122, + "loss": 3.9677, + "step": 2468 + }, + { + "epoch": 3.9504, + "grad_norm": 5.016576766967773, + "learning_rate": 0.000512, + "loss": 3.5789, + "step": 2469 + }, + { + "epoch": 3.952, + "grad_norm": 2.708357572555542, + "learning_rate": 0.0005118, + "loss": 3.8632, + "step": 2470 + }, + { + "epoch": 3.9536, + "grad_norm": 2.9615776538848877, + "learning_rate": 0.0005116, + "loss": 2.7705, + "step": 2471 + }, + { + "epoch": 3.9552, + "grad_norm": 3.5062623023986816, + "learning_rate": 0.0005114, + "loss": 3.9674, + "step": 2472 + }, + { + "epoch": 3.9568, + "grad_norm": 2.6961495876312256, + "learning_rate": 0.0005112, + "loss": 3.4771, + "step": 2473 + }, + { + "epoch": 3.9584, + "grad_norm": 6.347679615020752, + "learning_rate": 0.000511, + "loss": 5.0042, + "step": 2474 + }, + { + "epoch": 3.96, + "grad_norm": NaN, + "learning_rate": 0.000511, + "loss": 2.9314, + "step": 2475 + }, + { + "epoch": 3.9616, + "grad_norm": 5.443944454193115, + "learning_rate": 0.0005108, + "loss": 5.7649, + "step": 2476 + }, + { + "epoch": 3.9632, + "grad_norm": 7.1263508796691895, + "learning_rate": 0.0005105999999999999, + "loss": 6.3727, + "step": 2477 + }, + { + "epoch": 3.9648, + "grad_norm": 1.0259019136428833, + "learning_rate": 0.0005103999999999999, + "loss": 4.1111, + "step": 2478 + }, + { + "epoch": 3.9664, + "grad_norm": 0.1592855453491211, + "learning_rate": 0.0005101999999999999, + "loss": 3.6376, + "step": 2479 + }, + { + "epoch": 3.968, + "grad_norm": 1.0972356796264648, + "learning_rate": 0.0005099999999999999, + "loss": 5.4418, + "step": 2480 + }, + { + "epoch": 3.9696, + "grad_norm": 0.4594857692718506, + "learning_rate": 0.0005097999999999999, + "loss": 4.4428, + "step": 2481 + }, + { + "epoch": 3.9712, + "grad_norm": 0.2423648089170456, + "learning_rate": 0.0005096, + "loss": 3.5892, + "step": 2482 + }, + { + "epoch": 3.9728, + "grad_norm": 1.5900893211364746, + "learning_rate": 0.0005093999999999999, + "loss": 4.7644, + "step": 2483 + }, + { + "epoch": 3.9744, + "grad_norm": 0.36498916149139404, + "learning_rate": 0.0005092, + "loss": 3.3058, + "step": 2484 + }, + { + "epoch": 3.976, + "grad_norm": 0.3362158536911011, + "learning_rate": 0.000509, + "loss": 3.8689, + "step": 2485 + }, + { + "epoch": 3.9776, + "grad_norm": 0.5637915134429932, + "learning_rate": 0.0005087999999999999, + "loss": 4.7144, + "step": 2486 + }, + { + "epoch": 3.9792, + "grad_norm": 0.690257728099823, + "learning_rate": 0.0005086, + "loss": 3.2482, + "step": 2487 + }, + { + "epoch": 3.9808, + "grad_norm": 1.2970272302627563, + "learning_rate": 0.0005084, + "loss": 3.5763, + "step": 2488 + }, + { + "epoch": 3.9824, + "grad_norm": 1.0402405261993408, + "learning_rate": 0.0005082, + "loss": 3.6031, + "step": 2489 + }, + { + "epoch": 3.984, + "grad_norm": 1.123017430305481, + "learning_rate": 0.000508, + "loss": 4.6232, + "step": 2490 + }, + { + "epoch": 3.9856, + "grad_norm": 1.0631844997406006, + "learning_rate": 0.0005078, + "loss": 3.987, + "step": 2491 + }, + { + "epoch": 3.9872, + "grad_norm": 1.2788798809051514, + "learning_rate": 0.0005076, + "loss": 4.0148, + "step": 2492 + }, + { + "epoch": 3.9888, + "grad_norm": 3.9951908588409424, + "learning_rate": 0.0005074, + "loss": 4.0768, + "step": 2493 + }, + { + "epoch": 3.9904, + "grad_norm": 1.8701856136322021, + "learning_rate": 0.0005072, + "loss": 2.5434, + "step": 2494 + }, + { + "epoch": 3.992, + "grad_norm": 11.19664478302002, + "learning_rate": 0.000507, + "loss": 8.6562, + "step": 2495 + }, + { + "epoch": 3.9936, + "grad_norm": 7.005781650543213, + "learning_rate": 0.0005068, + "loss": 3.8161, + "step": 2496 + }, + { + "epoch": 3.9952, + "grad_norm": 6.712696552276611, + "learning_rate": 0.0005066, + "loss": 5.4329, + "step": 2497 + }, + { + "epoch": 3.9968, + "grad_norm": 6.685918807983398, + "learning_rate": 0.0005064, + "loss": 5.2872, + "step": 2498 + }, + { + "epoch": 3.9984, + "grad_norm": 9.401721000671387, + "learning_rate": 0.0005061999999999999, + "loss": 4.5241, + "step": 2499 + }, + { + "epoch": 4.0, + "grad_norm": 6.013485908508301, + "learning_rate": 0.0005059999999999999, + "loss": 7.2804, + "step": 2500 + }, + { + "epoch": 4.0, + "eval_cer": 0.5062452549582437, + "eval_loss": 4.04433012008667, + "eval_runtime": 161.6965, + "eval_samples_per_second": 19.394, + "eval_steps_per_second": 1.212, + "eval_wer": 0.85816470337969, + "step": 2500 + }, + { + "epoch": 4.0016, + "grad_norm": 23.144800186157227, + "learning_rate": 0.0005057999999999999, + "loss": 6.8372, + "step": 2501 + }, + { + "epoch": 4.0032, + "grad_norm": 4.225520610809326, + "learning_rate": 0.0005055999999999999, + "loss": 3.9508, + "step": 2502 + }, + { + "epoch": 4.0048, + "grad_norm": 9.67379093170166, + "learning_rate": 0.0005054, + "loss": 6.3174, + "step": 2503 + }, + { + "epoch": 4.0064, + "grad_norm": 0.17756377160549164, + "learning_rate": 0.0005051999999999999, + "loss": 4.1162, + "step": 2504 + }, + { + "epoch": 4.008, + "grad_norm": 0.32008469104766846, + "learning_rate": 0.0005049999999999999, + "loss": 3.9213, + "step": 2505 + }, + { + "epoch": 4.0096, + "grad_norm": 0.21909049153327942, + "learning_rate": 0.0005048, + "loss": 5.2693, + "step": 2506 + }, + { + "epoch": 4.0112, + "grad_norm": 0.45780324935913086, + "learning_rate": 0.0005045999999999999, + "loss": 5.0734, + "step": 2507 + }, + { + "epoch": 4.0128, + "grad_norm": 0.569255530834198, + "learning_rate": 0.0005044, + "loss": 6.3472, + "step": 2508 + }, + { + "epoch": 4.0144, + "grad_norm": 0.4611165225505829, + "learning_rate": 0.0005042, + "loss": 4.9004, + "step": 2509 + }, + { + "epoch": 4.016, + "grad_norm": 0.1880490928888321, + "learning_rate": 0.0005039999999999999, + "loss": 4.5229, + "step": 2510 + }, + { + "epoch": 4.0176, + "grad_norm": 2.2325141429901123, + "learning_rate": 0.0005038, + "loss": 4.5049, + "step": 2511 + }, + { + "epoch": 4.0192, + "grad_norm": 1.0180236101150513, + "learning_rate": 0.0005036, + "loss": 5.1325, + "step": 2512 + }, + { + "epoch": 4.0208, + "grad_norm": 0.42767855525016785, + "learning_rate": 0.0005034, + "loss": 4.2691, + "step": 2513 + }, + { + "epoch": 4.0224, + "grad_norm": 0.16366739571094513, + "learning_rate": 0.0005032, + "loss": 4.6155, + "step": 2514 + }, + { + "epoch": 4.024, + "grad_norm": 0.6206148862838745, + "learning_rate": 0.000503, + "loss": 4.8346, + "step": 2515 + }, + { + "epoch": 4.0256, + "grad_norm": 0.4776545763015747, + "learning_rate": 0.0005028, + "loss": 4.7195, + "step": 2516 + }, + { + "epoch": 4.0272, + "grad_norm": 9.23447036743164, + "learning_rate": 0.0005026, + "loss": 7.7137, + "step": 2517 + }, + { + "epoch": 4.0288, + "grad_norm": 0.9635998010635376, + "learning_rate": 0.0005024, + "loss": 4.3765, + "step": 2518 + }, + { + "epoch": 4.0304, + "grad_norm": 0.444718599319458, + "learning_rate": 0.0005022, + "loss": 4.302, + "step": 2519 + }, + { + "epoch": 4.032, + "grad_norm": 0.35454219579696655, + "learning_rate": 0.000502, + "loss": 3.5834, + "step": 2520 + }, + { + "epoch": 4.0336, + "grad_norm": 0.2884537875652313, + "learning_rate": 0.0005017999999999999, + "loss": 3.9605, + "step": 2521 + }, + { + "epoch": 4.0352, + "grad_norm": 0.36135971546173096, + "learning_rate": 0.0005015999999999999, + "loss": 4.4906, + "step": 2522 + }, + { + "epoch": 4.0368, + "grad_norm": 0.40249496698379517, + "learning_rate": 0.0005013999999999999, + "loss": 2.7345, + "step": 2523 + }, + { + "epoch": 4.0384, + "grad_norm": 0.5062964558601379, + "learning_rate": 0.0005011999999999999, + "loss": 3.894, + "step": 2524 + }, + { + "epoch": 4.04, + "grad_norm": 0.5407964587211609, + "learning_rate": 0.0005009999999999999, + "loss": 3.2813, + "step": 2525 + }, + { + "epoch": 4.0416, + "grad_norm": 1.2305890321731567, + "learning_rate": 0.0005007999999999999, + "loss": 5.1225, + "step": 2526 + }, + { + "epoch": 4.0432, + "grad_norm": 1.5490796566009521, + "learning_rate": 0.0005006, + "loss": 4.7507, + "step": 2527 + }, + { + "epoch": 4.0448, + "grad_norm": 3.157155752182007, + "learning_rate": 0.0005003999999999999, + "loss": 4.7362, + "step": 2528 + }, + { + "epoch": 4.0464, + "grad_norm": 0.4881833791732788, + "learning_rate": 0.0005001999999999999, + "loss": 5.1957, + "step": 2529 + }, + { + "epoch": 4.048, + "grad_norm": 0.8387967944145203, + "learning_rate": 0.0005, + "loss": 4.8116, + "step": 2530 + }, + { + "epoch": 4.0496, + "grad_norm": 0.8843464255332947, + "learning_rate": 0.0004997999999999999, + "loss": 3.8705, + "step": 2531 + }, + { + "epoch": 4.0512, + "grad_norm": 1.8773194551467896, + "learning_rate": 0.0004996, + "loss": 6.0234, + "step": 2532 + }, + { + "epoch": 4.0528, + "grad_norm": 2.2637429237365723, + "learning_rate": 0.0004994, + "loss": 3.2086, + "step": 2533 + }, + { + "epoch": 4.0544, + "grad_norm": 1.3157998323440552, + "learning_rate": 0.0004991999999999999, + "loss": 3.8496, + "step": 2534 + }, + { + "epoch": 4.056, + "grad_norm": 2.5601677894592285, + "learning_rate": 0.000499, + "loss": 5.1782, + "step": 2535 + }, + { + "epoch": 4.0576, + "grad_norm": 2.2101492881774902, + "learning_rate": 0.0004988, + "loss": 4.1483, + "step": 2536 + }, + { + "epoch": 4.0592, + "grad_norm": 3.1673080921173096, + "learning_rate": 0.0004986, + "loss": 5.8835, + "step": 2537 + }, + { + "epoch": 4.0608, + "grad_norm": 2.5407838821411133, + "learning_rate": 0.0004984, + "loss": 6.0127, + "step": 2538 + }, + { + "epoch": 4.0624, + "grad_norm": 1.7373592853546143, + "learning_rate": 0.0004982, + "loss": 3.743, + "step": 2539 + }, + { + "epoch": 4.064, + "grad_norm": 1.6780894994735718, + "learning_rate": 0.000498, + "loss": 3.7424, + "step": 2540 + }, + { + "epoch": 4.0656, + "grad_norm": 1.9963005781173706, + "learning_rate": 0.0004978, + "loss": 3.5275, + "step": 2541 + }, + { + "epoch": 4.0672, + "grad_norm": 3.755784273147583, + "learning_rate": 0.0004976, + "loss": 3.3858, + "step": 2542 + }, + { + "epoch": 4.0688, + "grad_norm": 2.3292593955993652, + "learning_rate": 0.0004974, + "loss": 4.0922, + "step": 2543 + }, + { + "epoch": 4.0704, + "grad_norm": 1.708128809928894, + "learning_rate": 0.0004971999999999999, + "loss": 3.9157, + "step": 2544 + }, + { + "epoch": 4.072, + "grad_norm": 6.881852149963379, + "learning_rate": 0.0004969999999999999, + "loss": 3.9577, + "step": 2545 + }, + { + "epoch": 4.0736, + "grad_norm": 4.577492713928223, + "learning_rate": 0.0004967999999999999, + "loss": 4.825, + "step": 2546 + }, + { + "epoch": 4.0752, + "grad_norm": 10.244248390197754, + "learning_rate": 0.0004965999999999999, + "loss": 9.1545, + "step": 2547 + }, + { + "epoch": 4.0768, + "grad_norm": 5.325230121612549, + "learning_rate": 0.0004963999999999999, + "loss": 3.3126, + "step": 2548 + }, + { + "epoch": 4.0784, + "grad_norm": 5.925568580627441, + "learning_rate": 0.0004961999999999999, + "loss": 3.5161, + "step": 2549 + }, + { + "epoch": 4.08, + "grad_norm": NaN, + "learning_rate": 0.0004961999999999999, + "loss": 2.5839, + "step": 2550 + }, + { + "epoch": 4.0816, + "grad_norm": 5.299213886260986, + "learning_rate": 0.0004959999999999999, + "loss": 5.6916, + "step": 2551 + }, + { + "epoch": 4.0832, + "grad_norm": 0.0, + "learning_rate": 0.0004958, + "loss": 6.4078, + "step": 2552 + }, + { + "epoch": 4.0848, + "grad_norm": 8.862677574157715, + "learning_rate": 0.0004955999999999999, + "loss": 5.5826, + "step": 2553 + }, + { + "epoch": 4.0864, + "grad_norm": 1.0761029720306396, + "learning_rate": 0.0004953999999999999, + "loss": 5.3076, + "step": 2554 + }, + { + "epoch": 4.088, + "grad_norm": 0.0, + "learning_rate": 0.0004952, + "loss": 6.6558, + "step": 2555 + }, + { + "epoch": 4.0896, + "grad_norm": 0.29841262102127075, + "learning_rate": 0.0004949999999999999, + "loss": 6.0496, + "step": 2556 + }, + { + "epoch": 4.0912, + "grad_norm": 0.6810678243637085, + "learning_rate": 0.0004948, + "loss": 4.658, + "step": 2557 + }, + { + "epoch": 4.0928, + "grad_norm": 0.17521485686302185, + "learning_rate": 0.0004946, + "loss": 5.2457, + "step": 2558 + }, + { + "epoch": 4.0944, + "grad_norm": 0.0, + "learning_rate": 0.0004944, + "loss": 6.2678, + "step": 2559 + }, + { + "epoch": 4.096, + "grad_norm": 0.0, + "learning_rate": 0.0004942, + "loss": 4.4643, + "step": 2560 + }, + { + "epoch": 4.0976, + "grad_norm": 0.2048645168542862, + "learning_rate": 0.000494, + "loss": 5.4229, + "step": 2561 + }, + { + "epoch": 4.0992, + "grad_norm": 0.0, + "learning_rate": 0.0004938, + "loss": 7.4454, + "step": 2562 + }, + { + "epoch": 4.1008, + "grad_norm": 0.0, + "learning_rate": 0.0004936, + "loss": 6.2727, + "step": 2563 + }, + { + "epoch": 4.1024, + "grad_norm": 2.374197483062744, + "learning_rate": 0.0004934, + "loss": 7.3081, + "step": 2564 + }, + { + "epoch": 4.104, + "grad_norm": 0.0, + "learning_rate": 0.0004932, + "loss": 5.4504, + "step": 2565 + }, + { + "epoch": 4.1056, + "grad_norm": 0.19911202788352966, + "learning_rate": 0.000493, + "loss": 5.56, + "step": 2566 + }, + { + "epoch": 4.1072, + "grad_norm": 0.3077949285507202, + "learning_rate": 0.0004927999999999999, + "loss": 6.1582, + "step": 2567 + }, + { + "epoch": 4.1088, + "grad_norm": 0.0, + "learning_rate": 0.0004925999999999999, + "loss": 5.4011, + "step": 2568 + }, + { + "epoch": 4.1104, + "grad_norm": 0.0, + "learning_rate": 0.0004923999999999999, + "loss": 6.5703, + "step": 2569 + }, + { + "epoch": 4.112, + "grad_norm": 0.6070860624313354, + "learning_rate": 0.0004921999999999999, + "loss": 7.1556, + "step": 2570 + }, + { + "epoch": 4.1136, + "grad_norm": 0.0, + "learning_rate": 0.0004919999999999999, + "loss": 5.0671, + "step": 2571 + }, + { + "epoch": 4.1152, + "grad_norm": 0.35825300216674805, + "learning_rate": 0.0004917999999999999, + "loss": 4.6262, + "step": 2572 + }, + { + "epoch": 4.1168, + "grad_norm": 0.0, + "learning_rate": 0.0004916, + "loss": 5.4843, + "step": 2573 + }, + { + "epoch": 4.1184, + "grad_norm": 0.44535768032073975, + "learning_rate": 0.0004913999999999999, + "loss": 6.6016, + "step": 2574 + }, + { + "epoch": 4.12, + "grad_norm": 0.200550377368927, + "learning_rate": 0.0004911999999999999, + "loss": 6.1914, + "step": 2575 + }, + { + "epoch": 4.1216, + "grad_norm": 0.8900039196014404, + "learning_rate": 0.000491, + "loss": 6.965, + "step": 2576 + }, + { + "epoch": 4.1232, + "grad_norm": 0.5002645254135132, + "learning_rate": 0.0004907999999999999, + "loss": 6.0534, + "step": 2577 + }, + { + "epoch": 4.1248, + "grad_norm": 0.15191906690597534, + "learning_rate": 0.0004906, + "loss": 5.4136, + "step": 2578 + }, + { + "epoch": 4.1264, + "grad_norm": 0.5502559542655945, + "learning_rate": 0.0004904, + "loss": 4.4739, + "step": 2579 + }, + { + "epoch": 4.128, + "grad_norm": 0.8410767912864685, + "learning_rate": 0.0004901999999999999, + "loss": 4.0054, + "step": 2580 + }, + { + "epoch": 4.1296, + "grad_norm": 0.480327844619751, + "learning_rate": 0.00049, + "loss": 4.8526, + "step": 2581 + }, + { + "epoch": 4.1312, + "grad_norm": 0.0, + "learning_rate": 0.0004898, + "loss": 5.2028, + "step": 2582 + }, + { + "epoch": 4.1328, + "grad_norm": 0.5567176938056946, + "learning_rate": 0.0004896, + "loss": 5.6976, + "step": 2583 + }, + { + "epoch": 4.1344, + "grad_norm": 0.39206984639167786, + "learning_rate": 0.0004894, + "loss": 3.6201, + "step": 2584 + }, + { + "epoch": 4.136, + "grad_norm": 0.47318196296691895, + "learning_rate": 0.0004892, + "loss": 5.6331, + "step": 2585 + }, + { + "epoch": 4.1376, + "grad_norm": 0.2152274250984192, + "learning_rate": 0.000489, + "loss": 5.7551, + "step": 2586 + }, + { + "epoch": 4.1392, + "grad_norm": 0.1511097401380539, + "learning_rate": 0.0004888, + "loss": 3.9637, + "step": 2587 + }, + { + "epoch": 4.1408, + "grad_norm": 6.308553695678711, + "learning_rate": 0.0004886, + "loss": 8.2982, + "step": 2588 + }, + { + "epoch": 4.1424, + "grad_norm": 0.3845791816711426, + "learning_rate": 0.0004883999999999999, + "loss": 4.9296, + "step": 2589 + }, + { + "epoch": 4.144, + "grad_norm": 0.2978602945804596, + "learning_rate": 0.00048819999999999994, + "loss": 4.3235, + "step": 2590 + }, + { + "epoch": 4.1456, + "grad_norm": 0.5199742317199707, + "learning_rate": 0.000488, + "loss": 5.8864, + "step": 2591 + }, + { + "epoch": 4.1472, + "grad_norm": 1.0750148296356201, + "learning_rate": 0.00048779999999999993, + "loss": 5.0529, + "step": 2592 + }, + { + "epoch": 4.1488, + "grad_norm": 0.8279758095741272, + "learning_rate": 0.0004875999999999999, + "loss": 4.36, + "step": 2593 + }, + { + "epoch": 4.1504, + "grad_norm": 0.8125143647193909, + "learning_rate": 0.0004874, + "loss": 3.9805, + "step": 2594 + }, + { + "epoch": 4.152, + "grad_norm": 1.2143406867980957, + "learning_rate": 0.00048719999999999997, + "loss": 3.8903, + "step": 2595 + }, + { + "epoch": 4.1536, + "grad_norm": 0.8204197287559509, + "learning_rate": 0.00048699999999999997, + "loss": 3.9114, + "step": 2596 + }, + { + "epoch": 4.1552, + "grad_norm": 1.063886046409607, + "learning_rate": 0.00048679999999999996, + "loss": 5.1182, + "step": 2597 + }, + { + "epoch": 4.1568, + "grad_norm": 3.12077260017395, + "learning_rate": 0.0004866, + "loss": 4.672, + "step": 2598 + }, + { + "epoch": 4.1584, + "grad_norm": 1.1409622430801392, + "learning_rate": 0.00048639999999999995, + "loss": 3.5499, + "step": 2599 + }, + { + "epoch": 4.16, + "grad_norm": NaN, + "learning_rate": 0.00048639999999999995, + "loss": 5.9472, + "step": 2600 + }, + { + "epoch": 4.1616, + "grad_norm": 0.0, + "learning_rate": 0.00048619999999999995, + "loss": 7.79, + "step": 2601 + }, + { + "epoch": 4.1632, + "grad_norm": 0.0, + "learning_rate": 0.000486, + "loss": 7.946, + "step": 2602 + }, + { + "epoch": 4.1648, + "grad_norm": 0.0, + "learning_rate": 0.00048579999999999994, + "loss": 5.8051, + "step": 2603 + }, + { + "epoch": 4.1664, + "grad_norm": 0.0, + "learning_rate": 0.0004856, + "loss": 5.9305, + "step": 2604 + }, + { + "epoch": 4.168, + "grad_norm": 0.0, + "learning_rate": 0.0004854, + "loss": 4.823, + "step": 2605 + }, + { + "epoch": 4.1696, + "grad_norm": 0.0, + "learning_rate": 0.0004851999999999999, + "loss": 5.3295, + "step": 2606 + }, + { + "epoch": 4.1712, + "grad_norm": 0.0, + "learning_rate": 0.00048499999999999997, + "loss": 4.9052, + "step": 2607 + }, + { + "epoch": 4.1728, + "grad_norm": 0.0, + "learning_rate": 0.00048479999999999997, + "loss": 5.7401, + "step": 2608 + }, + { + "epoch": 4.1744, + "grad_norm": 0.0, + "learning_rate": 0.00048459999999999996, + "loss": 4.8954, + "step": 2609 + }, + { + "epoch": 4.176, + "grad_norm": 24.389549255371094, + "learning_rate": 0.00048439999999999996, + "loss": 6.1263, + "step": 2610 + }, + { + "epoch": 4.1776, + "grad_norm": 0.0, + "learning_rate": 0.0004842, + "loss": 5.6154, + "step": 2611 + }, + { + "epoch": 4.1792, + "grad_norm": 0.0, + "learning_rate": 0.00048399999999999995, + "loss": 4.0222, + "step": 2612 + }, + { + "epoch": 4.1808, + "grad_norm": 0.0, + "learning_rate": 0.00048379999999999994, + "loss": 5.6816, + "step": 2613 + }, + { + "epoch": 4.1824, + "grad_norm": 0.0, + "learning_rate": 0.0004836, + "loss": 5.705, + "step": 2614 + }, + { + "epoch": 4.184, + "grad_norm": 0.0, + "learning_rate": 0.00048339999999999993, + "loss": 4.51, + "step": 2615 + }, + { + "epoch": 4.1856, + "grad_norm": 0.0, + "learning_rate": 0.0004832, + "loss": 4.7296, + "step": 2616 + }, + { + "epoch": 4.1872, + "grad_norm": 0.0, + "learning_rate": 0.000483, + "loss": 5.9244, + "step": 2617 + }, + { + "epoch": 4.1888, + "grad_norm": 0.0, + "learning_rate": 0.0004827999999999999, + "loss": 5.6749, + "step": 2618 + }, + { + "epoch": 4.1904, + "grad_norm": 0.0, + "learning_rate": 0.00048259999999999997, + "loss": 4.6188, + "step": 2619 + }, + { + "epoch": 4.192, + "grad_norm": 0.0, + "learning_rate": 0.00048239999999999996, + "loss": 5.8702, + "step": 2620 + }, + { + "epoch": 4.1936, + "grad_norm": 0.17885775864124298, + "learning_rate": 0.00048219999999999996, + "loss": 5.862, + "step": 2621 + }, + { + "epoch": 4.1952, + "grad_norm": 0.0, + "learning_rate": 0.00048199999999999995, + "loss": 4.847, + "step": 2622 + }, + { + "epoch": 4.1968, + "grad_norm": 0.0, + "learning_rate": 0.0004818, + "loss": 5.8012, + "step": 2623 + }, + { + "epoch": 4.1984, + "grad_norm": 0.0, + "learning_rate": 0.00048159999999999994, + "loss": 4.8323, + "step": 2624 + }, + { + "epoch": 4.2, + "grad_norm": 3.3034310340881348, + "learning_rate": 0.00048139999999999994, + "loss": 5.2113, + "step": 2625 + }, + { + "epoch": 4.2016, + "grad_norm": 0.0, + "learning_rate": 0.0004812, + "loss": 4.9682, + "step": 2626 + }, + { + "epoch": 4.2032, + "grad_norm": 0.0, + "learning_rate": 0.00048099999999999993, + "loss": 4.3872, + "step": 2627 + }, + { + "epoch": 4.2048, + "grad_norm": 0.0, + "learning_rate": 0.0004808, + "loss": 4.8967, + "step": 2628 + }, + { + "epoch": 4.2064, + "grad_norm": 0.0, + "learning_rate": 0.00048059999999999997, + "loss": 7.0991, + "step": 2629 + }, + { + "epoch": 4.208, + "grad_norm": 0.0, + "learning_rate": 0.0004803999999999999, + "loss": 4.9539, + "step": 2630 + }, + { + "epoch": 4.2096, + "grad_norm": 0.0, + "learning_rate": 0.00048019999999999996, + "loss": 4.1062, + "step": 2631 + }, + { + "epoch": 4.2112, + "grad_norm": 0.0, + "learning_rate": 0.00047999999999999996, + "loss": 5.8075, + "step": 2632 + }, + { + "epoch": 4.2128, + "grad_norm": 0.0, + "learning_rate": 0.00047979999999999995, + "loss": 6.6011, + "step": 2633 + }, + { + "epoch": 4.2144, + "grad_norm": 0.0, + "learning_rate": 0.00047959999999999995, + "loss": 4.7376, + "step": 2634 + }, + { + "epoch": 4.216, + "grad_norm": 0.0, + "learning_rate": 0.0004794, + "loss": 4.6453, + "step": 2635 + }, + { + "epoch": 4.2176, + "grad_norm": 0.0, + "learning_rate": 0.00047919999999999994, + "loss": 5.1572, + "step": 2636 + }, + { + "epoch": 4.2192, + "grad_norm": 0.0, + "learning_rate": 0.000479, + "loss": 5.2229, + "step": 2637 + }, + { + "epoch": 4.2208, + "grad_norm": 0.0, + "learning_rate": 0.0004788, + "loss": 6.2779, + "step": 2638 + }, + { + "epoch": 4.2224, + "grad_norm": 0.0, + "learning_rate": 0.0004785999999999999, + "loss": 4.2566, + "step": 2639 + }, + { + "epoch": 4.224, + "grad_norm": 0.0, + "learning_rate": 0.0004784, + "loss": 5.1199, + "step": 2640 + }, + { + "epoch": 4.2256, + "grad_norm": 0.0, + "learning_rate": 0.00047819999999999997, + "loss": 4.7579, + "step": 2641 + }, + { + "epoch": 4.2272, + "grad_norm": 0.3036254048347473, + "learning_rate": 0.00047799999999999996, + "loss": 3.1542, + "step": 2642 + }, + { + "epoch": 4.2288, + "grad_norm": 0.0, + "learning_rate": 0.00047779999999999996, + "loss": 4.169, + "step": 2643 + }, + { + "epoch": 4.2304, + "grad_norm": 1.608218789100647, + "learning_rate": 0.0004776, + "loss": 5.7199, + "step": 2644 + }, + { + "epoch": 4.232, + "grad_norm": 0.5215432047843933, + "learning_rate": 0.00047739999999999995, + "loss": 3.9992, + "step": 2645 + }, + { + "epoch": 4.2336, + "grad_norm": 0.17919880151748657, + "learning_rate": 0.00047719999999999994, + "loss": 4.6033, + "step": 2646 + }, + { + "epoch": 4.2352, + "grad_norm": 0.4602564573287964, + "learning_rate": 0.000477, + "loss": 3.9578, + "step": 2647 + }, + { + "epoch": 4.2368, + "grad_norm": 1.4237133264541626, + "learning_rate": 0.00047679999999999993, + "loss": 3.9533, + "step": 2648 + }, + { + "epoch": 4.2384, + "grad_norm": 0.8177955746650696, + "learning_rate": 0.0004766, + "loss": 4.8144, + "step": 2649 + }, + { + "epoch": 4.24, + "grad_norm": 1.2843536138534546, + "learning_rate": 0.0004764, + "loss": 5.2531, + "step": 2650 + }, + { + "epoch": 4.2416, + "grad_norm": 2.9875361919403076, + "learning_rate": 0.0004761999999999999, + "loss": 7.2527, + "step": 2651 + }, + { + "epoch": 4.2432, + "grad_norm": 0.0, + "learning_rate": 0.00047599999999999997, + "loss": 7.248, + "step": 2652 + }, + { + "epoch": 4.2448, + "grad_norm": 0.0, + "learning_rate": 0.00047579999999999996, + "loss": 6.575, + "step": 2653 + }, + { + "epoch": 4.2464, + "grad_norm": 0.0, + "learning_rate": 0.00047559999999999996, + "loss": 5.8347, + "step": 2654 + }, + { + "epoch": 4.248, + "grad_norm": 0.0, + "learning_rate": 0.00047539999999999995, + "loss": 6.9052, + "step": 2655 + }, + { + "epoch": 4.2496, + "grad_norm": 0.0, + "learning_rate": 0.0004752, + "loss": 4.7331, + "step": 2656 + }, + { + "epoch": 4.2512, + "grad_norm": 0.0, + "learning_rate": 0.00047499999999999994, + "loss": 7.093, + "step": 2657 + }, + { + "epoch": 4.2528, + "grad_norm": 0.0, + "learning_rate": 0.00047479999999999994, + "loss": 6.3205, + "step": 2658 + }, + { + "epoch": 4.2544, + "grad_norm": 0.0, + "learning_rate": 0.0004746, + "loss": 5.4774, + "step": 2659 + }, + { + "epoch": 4.256, + "grad_norm": 0.0, + "learning_rate": 0.00047439999999999993, + "loss": 5.5052, + "step": 2660 + }, + { + "epoch": 4.2576, + "grad_norm": 2.182249069213867, + "learning_rate": 0.0004742, + "loss": 5.3668, + "step": 2661 + }, + { + "epoch": 4.2592, + "grad_norm": 0.0, + "learning_rate": 0.000474, + "loss": 5.3307, + "step": 2662 + }, + { + "epoch": 4.2608, + "grad_norm": 0.0, + "learning_rate": 0.0004737999999999999, + "loss": 5.2796, + "step": 2663 + }, + { + "epoch": 4.2624, + "grad_norm": 0.0, + "learning_rate": 0.00047359999999999997, + "loss": 5.1192, + "step": 2664 + }, + { + "epoch": 4.264, + "grad_norm": 0.0, + "learning_rate": 0.00047339999999999996, + "loss": 6.0993, + "step": 2665 + }, + { + "epoch": 4.2656, + "grad_norm": 0.0, + "learning_rate": 0.00047319999999999996, + "loss": 5.8218, + "step": 2666 + }, + { + "epoch": 4.2672, + "grad_norm": 0.0, + "learning_rate": 0.00047299999999999995, + "loss": 5.8958, + "step": 2667 + }, + { + "epoch": 4.2688, + "grad_norm": 0.0, + "learning_rate": 0.0004728, + "loss": 5.4774, + "step": 2668 + }, + { + "epoch": 4.2704, + "grad_norm": 2.3783512115478516, + "learning_rate": 0.00047259999999999994, + "loss": 6.3583, + "step": 2669 + }, + { + "epoch": 4.272, + "grad_norm": 0.0, + "learning_rate": 0.00047239999999999994, + "loss": 5.7838, + "step": 2670 + }, + { + "epoch": 4.2736, + "grad_norm": 0.0, + "learning_rate": 0.0004722, + "loss": 5.6294, + "step": 2671 + }, + { + "epoch": 4.2752, + "grad_norm": 0.0, + "learning_rate": 0.0004719999999999999, + "loss": 5.6274, + "step": 2672 + }, + { + "epoch": 4.2768, + "grad_norm": 0.0, + "learning_rate": 0.0004718, + "loss": 4.7142, + "step": 2673 + }, + { + "epoch": 4.2783999999999995, + "grad_norm": 0.0, + "learning_rate": 0.00047159999999999997, + "loss": 6.9477, + "step": 2674 + }, + { + "epoch": 4.28, + "grad_norm": 0.0, + "learning_rate": 0.0004713999999999999, + "loss": 5.3256, + "step": 2675 + }, + { + "epoch": 4.2816, + "grad_norm": 0.0, + "learning_rate": 0.00047119999999999996, + "loss": 5.4785, + "step": 2676 + }, + { + "epoch": 4.2832, + "grad_norm": 0.0, + "learning_rate": 0.00047099999999999996, + "loss": 6.1855, + "step": 2677 + }, + { + "epoch": 4.2848, + "grad_norm": 0.0, + "learning_rate": 0.00047079999999999995, + "loss": 4.6882, + "step": 2678 + }, + { + "epoch": 4.2864, + "grad_norm": 0.0, + "learning_rate": 0.00047059999999999995, + "loss": 5.5514, + "step": 2679 + }, + { + "epoch": 4.288, + "grad_norm": 0.0, + "learning_rate": 0.0004704, + "loss": 6.5728, + "step": 2680 + }, + { + "epoch": 4.2896, + "grad_norm": 0.0, + "learning_rate": 0.00047019999999999994, + "loss": 5.5523, + "step": 2681 + }, + { + "epoch": 4.2912, + "grad_norm": 0.0, + "learning_rate": 0.00046999999999999993, + "loss": 5.4308, + "step": 2682 + }, + { + "epoch": 4.2928, + "grad_norm": 0.0, + "learning_rate": 0.0004698, + "loss": 5.6038, + "step": 2683 + }, + { + "epoch": 4.2943999999999996, + "grad_norm": 0.0, + "learning_rate": 0.0004695999999999999, + "loss": 5.8146, + "step": 2684 + }, + { + "epoch": 4.296, + "grad_norm": 0.0, + "learning_rate": 0.00046939999999999997, + "loss": 5.6524, + "step": 2685 + }, + { + "epoch": 4.2976, + "grad_norm": 0.0, + "learning_rate": 0.00046919999999999997, + "loss": 4.9563, + "step": 2686 + }, + { + "epoch": 4.2992, + "grad_norm": 0.0, + "learning_rate": 0.0004689999999999999, + "loss": 5.0189, + "step": 2687 + }, + { + "epoch": 4.3008, + "grad_norm": 0.0, + "learning_rate": 0.00046879999999999996, + "loss": 6.2963, + "step": 2688 + }, + { + "epoch": 4.3024000000000004, + "grad_norm": 0.0, + "learning_rate": 0.00046859999999999995, + "loss": 7.2364, + "step": 2689 + }, + { + "epoch": 4.304, + "grad_norm": 0.5065405964851379, + "learning_rate": 0.00046839999999999995, + "loss": 6.5879, + "step": 2690 + }, + { + "epoch": 4.3056, + "grad_norm": 0.6606298685073853, + "learning_rate": 0.00046819999999999994, + "loss": 7.6232, + "step": 2691 + }, + { + "epoch": 4.3072, + "grad_norm": 0.28764608502388, + "learning_rate": 0.000468, + "loss": 6.2058, + "step": 2692 + }, + { + "epoch": 4.3088, + "grad_norm": 0.5861437916755676, + "learning_rate": 0.00046779999999999993, + "loss": 6.5446, + "step": 2693 + }, + { + "epoch": 4.3104, + "grad_norm": 0.0, + "learning_rate": 0.00046759999999999993, + "loss": 4.6033, + "step": 2694 + }, + { + "epoch": 4.312, + "grad_norm": 0.39768925309181213, + "learning_rate": 0.0004674, + "loss": 7.0912, + "step": 2695 + }, + { + "epoch": 4.3136, + "grad_norm": 0.4369770288467407, + "learning_rate": 0.0004671999999999999, + "loss": 4.8682, + "step": 2696 + }, + { + "epoch": 4.3152, + "grad_norm": 0.17879962921142578, + "learning_rate": 0.00046699999999999997, + "loss": 5.6477, + "step": 2697 + }, + { + "epoch": 4.3168, + "grad_norm": 2.0778021812438965, + "learning_rate": 0.00046679999999999996, + "loss": 4.4139, + "step": 2698 + }, + { + "epoch": 4.3184000000000005, + "grad_norm": 0.6356657147407532, + "learning_rate": 0.0004665999999999999, + "loss": 4.6318, + "step": 2699 + }, + { + "epoch": 4.32, + "grad_norm": 3.4940898418426514, + "learning_rate": 0.00046639999999999995, + "loss": 8.6823, + "step": 2700 + }, + { + "epoch": 4.3216, + "grad_norm": 3.1868233680725098, + "learning_rate": 0.00046619999999999995, + "loss": 8.6328, + "step": 2701 + }, + { + "epoch": 4.3232, + "grad_norm": 0.0, + "learning_rate": 0.00046599999999999994, + "loss": 9.1806, + "step": 2702 + }, + { + "epoch": 4.3248, + "grad_norm": 0.0, + "learning_rate": 0.00046579999999999994, + "loss": 7.04, + "step": 2703 + }, + { + "epoch": 4.3264, + "grad_norm": 0.0, + "learning_rate": 0.0004656, + "loss": 6.8505, + "step": 2704 + }, + { + "epoch": 4.328, + "grad_norm": 0.911113977432251, + "learning_rate": 0.00046539999999999993, + "loss": 5.8174, + "step": 2705 + }, + { + "epoch": 4.3296, + "grad_norm": 0.0, + "learning_rate": 0.0004651999999999999, + "loss": 4.6813, + "step": 2706 + }, + { + "epoch": 4.3312, + "grad_norm": 0.0, + "learning_rate": 0.00046499999999999997, + "loss": 4.9136, + "step": 2707 + }, + { + "epoch": 4.3328, + "grad_norm": 0.0, + "learning_rate": 0.0004647999999999999, + "loss": 5.1828, + "step": 2708 + }, + { + "epoch": 4.3344, + "grad_norm": 0.0, + "learning_rate": 0.00046459999999999996, + "loss": 5.4842, + "step": 2709 + }, + { + "epoch": 4.336, + "grad_norm": 0.0, + "learning_rate": 0.00046439999999999996, + "loss": 5.8971, + "step": 2710 + }, + { + "epoch": 4.3376, + "grad_norm": 0.0, + "learning_rate": 0.00046419999999999995, + "loss": 4.7415, + "step": 2711 + }, + { + "epoch": 4.3392, + "grad_norm": 0.0, + "learning_rate": 0.00046399999999999995, + "loss": 5.4208, + "step": 2712 + }, + { + "epoch": 4.3408, + "grad_norm": 0.0, + "learning_rate": 0.0004638, + "loss": 5.7258, + "step": 2713 + }, + { + "epoch": 4.3424, + "grad_norm": 0.0, + "learning_rate": 0.00046359999999999994, + "loss": 4.4678, + "step": 2714 + }, + { + "epoch": 4.344, + "grad_norm": 0.0, + "learning_rate": 0.00046339999999999993, + "loss": 6.3103, + "step": 2715 + }, + { + "epoch": 4.3456, + "grad_norm": 0.0, + "learning_rate": 0.0004632, + "loss": 5.1996, + "step": 2716 + }, + { + "epoch": 4.3472, + "grad_norm": 0.0, + "learning_rate": 0.0004629999999999999, + "loss": 4.8172, + "step": 2717 + }, + { + "epoch": 4.3488, + "grad_norm": 0.0, + "learning_rate": 0.0004628, + "loss": 6.0634, + "step": 2718 + }, + { + "epoch": 4.3504, + "grad_norm": 0.0, + "learning_rate": 0.00046259999999999997, + "loss": 5.0384, + "step": 2719 + }, + { + "epoch": 4.352, + "grad_norm": 0.0, + "learning_rate": 0.0004624, + "loss": 4.5593, + "step": 2720 + }, + { + "epoch": 4.3536, + "grad_norm": 0.0, + "learning_rate": 0.00046219999999999996, + "loss": 6.2736, + "step": 2721 + }, + { + "epoch": 4.3552, + "grad_norm": 0.0, + "learning_rate": 0.00046199999999999995, + "loss": 5.7264, + "step": 2722 + }, + { + "epoch": 4.3568, + "grad_norm": 0.0, + "learning_rate": 0.0004618, + "loss": 5.6145, + "step": 2723 + }, + { + "epoch": 4.3584, + "grad_norm": 0.17345993220806122, + "learning_rate": 0.00046159999999999994, + "loss": 7.1915, + "step": 2724 + }, + { + "epoch": 4.36, + "grad_norm": 0.0, + "learning_rate": 0.0004614, + "loss": 4.9104, + "step": 2725 + }, + { + "epoch": 4.3616, + "grad_norm": 0.0, + "learning_rate": 0.0004612, + "loss": 4.6579, + "step": 2726 + }, + { + "epoch": 4.3632, + "grad_norm": 0.0, + "learning_rate": 0.00046099999999999993, + "loss": 5.2145, + "step": 2727 + }, + { + "epoch": 4.3648, + "grad_norm": 0.0, + "learning_rate": 0.0004608, + "loss": 4.6131, + "step": 2728 + }, + { + "epoch": 4.3664, + "grad_norm": 0.0, + "learning_rate": 0.0004606, + "loss": 3.9284, + "step": 2729 + }, + { + "epoch": 4.368, + "grad_norm": 0.0, + "learning_rate": 0.00046039999999999997, + "loss": 4.8025, + "step": 2730 + }, + { + "epoch": 4.3696, + "grad_norm": 0.0, + "learning_rate": 0.00046019999999999996, + "loss": 5.654, + "step": 2731 + }, + { + "epoch": 4.3712, + "grad_norm": 0.0, + "learning_rate": 0.00046, + "loss": 4.4884, + "step": 2732 + }, + { + "epoch": 4.3728, + "grad_norm": 0.0, + "learning_rate": 0.00045979999999999995, + "loss": 6.9474, + "step": 2733 + }, + { + "epoch": 4.3744, + "grad_norm": 0.0, + "learning_rate": 0.00045959999999999995, + "loss": 4.3246, + "step": 2734 + }, + { + "epoch": 4.376, + "grad_norm": 0.0, + "learning_rate": 0.0004594, + "loss": 4.2979, + "step": 2735 + }, + { + "epoch": 4.3776, + "grad_norm": 0.0, + "learning_rate": 0.00045919999999999994, + "loss": 6.0484, + "step": 2736 + }, + { + "epoch": 4.3792, + "grad_norm": 0.0, + "learning_rate": 0.000459, + "loss": 6.9776, + "step": 2737 + }, + { + "epoch": 4.3808, + "grad_norm": 0.0, + "learning_rate": 0.0004588, + "loss": 4.6165, + "step": 2738 + }, + { + "epoch": 4.3824, + "grad_norm": 0.0, + "learning_rate": 0.0004585999999999999, + "loss": 5.9903, + "step": 2739 + }, + { + "epoch": 4.384, + "grad_norm": 0.0, + "learning_rate": 0.0004584, + "loss": 3.9935, + "step": 2740 + }, + { + "epoch": 4.3856, + "grad_norm": 0.0, + "learning_rate": 0.00045819999999999997, + "loss": 7.7079, + "step": 2741 + }, + { + "epoch": 4.3872, + "grad_norm": 0.0, + "learning_rate": 0.00045799999999999997, + "loss": 4.1452, + "step": 2742 + }, + { + "epoch": 4.3888, + "grad_norm": 0.0, + "learning_rate": 0.00045779999999999996, + "loss": 5.6116, + "step": 2743 + }, + { + "epoch": 4.3904, + "grad_norm": 0.0, + "learning_rate": 0.0004576, + "loss": 4.892, + "step": 2744 + }, + { + "epoch": 4.392, + "grad_norm": 0.0, + "learning_rate": 0.00045739999999999995, + "loss": 4.6825, + "step": 2745 + }, + { + "epoch": 4.3936, + "grad_norm": 0.0, + "learning_rate": 0.00045719999999999995, + "loss": 5.306, + "step": 2746 + }, + { + "epoch": 4.3952, + "grad_norm": 1.4116289615631104, + "learning_rate": 0.000457, + "loss": 5.0769, + "step": 2747 + }, + { + "epoch": 4.3968, + "grad_norm": 0.31743568181991577, + "learning_rate": 0.00045679999999999994, + "loss": 3.7618, + "step": 2748 + }, + { + "epoch": 4.3984, + "grad_norm": 0.9777640700340271, + "learning_rate": 0.0004566, + "loss": 4.3887, + "step": 2749 + }, + { + "epoch": 4.4, + "grad_norm": 2.4460084438323975, + "learning_rate": 0.0004564, + "loss": 6.2199, + "step": 2750 + }, + { + "epoch": 4.4016, + "grad_norm": 1.542219877243042, + "learning_rate": 0.0004561999999999999, + "loss": 6.9442, + "step": 2751 + }, + { + "epoch": 4.4032, + "grad_norm": 0.0, + "learning_rate": 0.00045599999999999997, + "loss": 7.7903, + "step": 2752 + }, + { + "epoch": 4.4048, + "grad_norm": 0.0, + "learning_rate": 0.00045579999999999997, + "loss": 5.6368, + "step": 2753 + }, + { + "epoch": 4.4064, + "grad_norm": 0.0, + "learning_rate": 0.00045559999999999996, + "loss": 5.115, + "step": 2754 + }, + { + "epoch": 4.408, + "grad_norm": 0.0, + "learning_rate": 0.00045539999999999996, + "loss": 5.4766, + "step": 2755 + }, + { + "epoch": 4.4096, + "grad_norm": 0.0, + "learning_rate": 0.0004552, + "loss": 4.7461, + "step": 2756 + }, + { + "epoch": 4.4112, + "grad_norm": 0.0, + "learning_rate": 0.00045499999999999995, + "loss": 6.3163, + "step": 2757 + }, + { + "epoch": 4.4128, + "grad_norm": 0.0, + "learning_rate": 0.00045479999999999994, + "loss": 5.9893, + "step": 2758 + }, + { + "epoch": 4.4144, + "grad_norm": 0.0, + "learning_rate": 0.0004546, + "loss": 6.0288, + "step": 2759 + }, + { + "epoch": 4.416, + "grad_norm": 0.0, + "learning_rate": 0.00045439999999999993, + "loss": 6.9431, + "step": 2760 + }, + { + "epoch": 4.4176, + "grad_norm": 0.0, + "learning_rate": 0.0004542, + "loss": 6.0607, + "step": 2761 + }, + { + "epoch": 4.4192, + "grad_norm": 0.0, + "learning_rate": 0.000454, + "loss": 4.9378, + "step": 2762 + }, + { + "epoch": 4.4208, + "grad_norm": 0.0, + "learning_rate": 0.0004537999999999999, + "loss": 6.1937, + "step": 2763 + }, + { + "epoch": 4.4224, + "grad_norm": 0.0, + "learning_rate": 0.00045359999999999997, + "loss": 4.2373, + "step": 2764 + }, + { + "epoch": 4.424, + "grad_norm": 0.0, + "learning_rate": 0.00045339999999999996, + "loss": 5.8447, + "step": 2765 + }, + { + "epoch": 4.4256, + "grad_norm": 0.0, + "learning_rate": 0.00045319999999999996, + "loss": 4.4457, + "step": 2766 + }, + { + "epoch": 4.4272, + "grad_norm": 0.0, + "learning_rate": 0.00045299999999999995, + "loss": 4.2722, + "step": 2767 + }, + { + "epoch": 4.4288, + "grad_norm": 0.0, + "learning_rate": 0.0004528, + "loss": 6.8676, + "step": 2768 + }, + { + "epoch": 4.4304, + "grad_norm": 0.0, + "learning_rate": 0.00045259999999999994, + "loss": 4.8872, + "step": 2769 + }, + { + "epoch": 4.432, + "grad_norm": 0.0, + "learning_rate": 0.00045239999999999994, + "loss": 5.9599, + "step": 2770 + }, + { + "epoch": 4.4336, + "grad_norm": 0.0, + "learning_rate": 0.0004522, + "loss": 4.9395, + "step": 2771 + }, + { + "epoch": 4.4352, + "grad_norm": 0.0, + "learning_rate": 0.00045199999999999993, + "loss": 5.1839, + "step": 2772 + }, + { + "epoch": 4.4368, + "grad_norm": 0.0, + "learning_rate": 0.0004518, + "loss": 4.0279, + "step": 2773 + }, + { + "epoch": 4.4384, + "grad_norm": 0.0, + "learning_rate": 0.00045159999999999997, + "loss": 5.7606, + "step": 2774 + }, + { + "epoch": 4.44, + "grad_norm": 0.0, + "learning_rate": 0.0004513999999999999, + "loss": 5.0386, + "step": 2775 + }, + { + "epoch": 4.4416, + "grad_norm": 0.0, + "learning_rate": 0.00045119999999999996, + "loss": 4.8579, + "step": 2776 + }, + { + "epoch": 4.4432, + "grad_norm": 0.0, + "learning_rate": 0.00045099999999999996, + "loss": 4.8091, + "step": 2777 + }, + { + "epoch": 4.4448, + "grad_norm": 0.0, + "learning_rate": 0.00045079999999999995, + "loss": 4.4286, + "step": 2778 + }, + { + "epoch": 4.4464, + "grad_norm": 0.0, + "learning_rate": 0.00045059999999999995, + "loss": 5.0576, + "step": 2779 + }, + { + "epoch": 4.448, + "grad_norm": 0.0, + "learning_rate": 0.0004504, + "loss": 4.2884, + "step": 2780 + }, + { + "epoch": 4.4496, + "grad_norm": 0.0, + "learning_rate": 0.00045019999999999994, + "loss": 6.9496, + "step": 2781 + }, + { + "epoch": 4.4512, + "grad_norm": 0.0, + "learning_rate": 0.00045, + "loss": 6.1365, + "step": 2782 + }, + { + "epoch": 4.4528, + "grad_norm": 0.0, + "learning_rate": 0.0004498, + "loss": 4.9852, + "step": 2783 + }, + { + "epoch": 4.4544, + "grad_norm": 0.17997673153877258, + "learning_rate": 0.0004495999999999999, + "loss": 7.4237, + "step": 2784 + }, + { + "epoch": 4.456, + "grad_norm": 0.0, + "learning_rate": 0.0004494, + "loss": 3.8001, + "step": 2785 + }, + { + "epoch": 4.4576, + "grad_norm": 0.1768631637096405, + "learning_rate": 0.00044919999999999997, + "loss": 4.2716, + "step": 2786 + }, + { + "epoch": 4.4592, + "grad_norm": 0.0, + "learning_rate": 0.00044899999999999996, + "loss": 3.8919, + "step": 2787 + }, + { + "epoch": 4.4608, + "grad_norm": 0.0, + "learning_rate": 0.00044879999999999996, + "loss": 4.659, + "step": 2788 + }, + { + "epoch": 4.4624, + "grad_norm": 0.0, + "learning_rate": 0.0004486, + "loss": 6.3004, + "step": 2789 + }, + { + "epoch": 4.464, + "grad_norm": 0.0, + "learning_rate": 0.00044839999999999995, + "loss": 4.3226, + "step": 2790 + }, + { + "epoch": 4.4656, + "grad_norm": 0.17757441103458405, + "learning_rate": 0.00044819999999999994, + "loss": 5.2982, + "step": 2791 + }, + { + "epoch": 4.4672, + "grad_norm": 0.1834985762834549, + "learning_rate": 0.000448, + "loss": 3.8272, + "step": 2792 + }, + { + "epoch": 4.4688, + "grad_norm": 0.0, + "learning_rate": 0.00044779999999999993, + "loss": 4.5373, + "step": 2793 + }, + { + "epoch": 4.4704, + "grad_norm": 0.0, + "learning_rate": 0.0004476, + "loss": 5.0192, + "step": 2794 + }, + { + "epoch": 4.4719999999999995, + "grad_norm": 0.0, + "learning_rate": 0.0004474, + "loss": 5.412, + "step": 2795 + }, + { + "epoch": 4.4736, + "grad_norm": 0.0, + "learning_rate": 0.0004471999999999999, + "loss": 4.5218, + "step": 2796 + }, + { + "epoch": 4.4752, + "grad_norm": 0.23554396629333496, + "learning_rate": 0.00044699999999999997, + "loss": 4.6174, + "step": 2797 + }, + { + "epoch": 4.4768, + "grad_norm": 0.40512022376060486, + "learning_rate": 0.00044679999999999996, + "loss": 5.9445, + "step": 2798 + }, + { + "epoch": 4.4784, + "grad_norm": 0.2543751299381256, + "learning_rate": 0.00044659999999999996, + "loss": 5.8068, + "step": 2799 + }, + { + "epoch": 4.48, + "grad_norm": NaN, + "learning_rate": 0.00044659999999999996, + "loss": 5.964, + "step": 2800 + }, + { + "epoch": 4.4816, + "grad_norm": 0.0, + "learning_rate": 0.00044639999999999995, + "loss": 8.724, + "step": 2801 + }, + { + "epoch": 4.4832, + "grad_norm": 0.0, + "learning_rate": 0.0004462, + "loss": 6.484, + "step": 2802 + }, + { + "epoch": 4.4848, + "grad_norm": 0.0, + "learning_rate": 0.00044599999999999994, + "loss": 6.6023, + "step": 2803 + }, + { + "epoch": 4.4864, + "grad_norm": 0.0, + "learning_rate": 0.00044579999999999994, + "loss": 6.6948, + "step": 2804 + }, + { + "epoch": 4.4879999999999995, + "grad_norm": 0.0, + "learning_rate": 0.0004456, + "loss": 5.974, + "step": 2805 + }, + { + "epoch": 4.4896, + "grad_norm": 0.0, + "learning_rate": 0.00044539999999999993, + "loss": 6.118, + "step": 2806 + }, + { + "epoch": 4.4912, + "grad_norm": 0.0, + "learning_rate": 0.0004452, + "loss": 4.0869, + "step": 2807 + }, + { + "epoch": 4.4928, + "grad_norm": 0.0, + "learning_rate": 0.000445, + "loss": 5.2811, + "step": 2808 + }, + { + "epoch": 4.4944, + "grad_norm": 0.0, + "learning_rate": 0.0004447999999999999, + "loss": 8.1623, + "step": 2809 + }, + { + "epoch": 4.496, + "grad_norm": 0.0, + "learning_rate": 0.00044459999999999996, + "loss": 5.452, + "step": 2810 + }, + { + "epoch": 4.4976, + "grad_norm": 0.0, + "learning_rate": 0.00044439999999999996, + "loss": 5.8144, + "step": 2811 + }, + { + "epoch": 4.4992, + "grad_norm": 0.0, + "learning_rate": 0.00044419999999999996, + "loss": 5.1539, + "step": 2812 + }, + { + "epoch": 4.5008, + "grad_norm": 0.0, + "learning_rate": 0.00044399999999999995, + "loss": 4.1331, + "step": 2813 + }, + { + "epoch": 4.5024, + "grad_norm": 0.0, + "learning_rate": 0.0004438, + "loss": 5.1579, + "step": 2814 + }, + { + "epoch": 4.504, + "grad_norm": 0.0, + "learning_rate": 0.00044359999999999994, + "loss": 5.5597, + "step": 2815 + }, + { + "epoch": 4.5056, + "grad_norm": 0.0, + "learning_rate": 0.00044339999999999994, + "loss": 5.3935, + "step": 2816 + }, + { + "epoch": 4.5072, + "grad_norm": 0.0, + "learning_rate": 0.0004432, + "loss": 5.3337, + "step": 2817 + }, + { + "epoch": 4.5088, + "grad_norm": 1.7325307130813599, + "learning_rate": 0.0004429999999999999, + "loss": 3.4717, + "step": 2818 + }, + { + "epoch": 4.5104, + "grad_norm": 0.0, + "learning_rate": 0.0004428, + "loss": 4.7052, + "step": 2819 + }, + { + "epoch": 4.5120000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00044259999999999997, + "loss": 5.0218, + "step": 2820 + }, + { + "epoch": 4.5136, + "grad_norm": 0.0, + "learning_rate": 0.0004423999999999999, + "loss": 5.7618, + "step": 2821 + }, + { + "epoch": 4.5152, + "grad_norm": 0.0, + "learning_rate": 0.00044219999999999996, + "loss": 6.152, + "step": 2822 + }, + { + "epoch": 4.5168, + "grad_norm": 0.0, + "learning_rate": 0.00044199999999999996, + "loss": 4.7174, + "step": 2823 + }, + { + "epoch": 4.5184, + "grad_norm": 0.0, + "learning_rate": 0.00044179999999999995, + "loss": 5.1476, + "step": 2824 + }, + { + "epoch": 4.52, + "grad_norm": 0.0, + "learning_rate": 0.00044159999999999995, + "loss": 5.0282, + "step": 2825 + }, + { + "epoch": 4.5216, + "grad_norm": 0.0, + "learning_rate": 0.0004414, + "loss": 4.2326, + "step": 2826 + }, + { + "epoch": 4.5232, + "grad_norm": 0.0, + "learning_rate": 0.00044119999999999994, + "loss": 4.6174, + "step": 2827 + }, + { + "epoch": 4.5248, + "grad_norm": 0.0, + "learning_rate": 0.00044099999999999993, + "loss": 6.1353, + "step": 2828 + }, + { + "epoch": 4.5264, + "grad_norm": 0.0, + "learning_rate": 0.0004408, + "loss": 4.0639, + "step": 2829 + }, + { + "epoch": 4.5280000000000005, + "grad_norm": 0.0, + "learning_rate": 0.0004405999999999999, + "loss": 4.8102, + "step": 2830 + }, + { + "epoch": 4.5296, + "grad_norm": 0.0, + "learning_rate": 0.00044039999999999997, + "loss": 4.3702, + "step": 2831 + }, + { + "epoch": 4.5312, + "grad_norm": 0.0, + "learning_rate": 0.00044019999999999997, + "loss": 5.6737, + "step": 2832 + }, + { + "epoch": 4.5328, + "grad_norm": 0.0, + "learning_rate": 0.0004399999999999999, + "loss": 5.2063, + "step": 2833 + }, + { + "epoch": 4.5344, + "grad_norm": 0.0, + "learning_rate": 0.00043979999999999996, + "loss": 4.1089, + "step": 2834 + }, + { + "epoch": 4.536, + "grad_norm": 0.0, + "learning_rate": 0.00043959999999999995, + "loss": 4.6398, + "step": 2835 + }, + { + "epoch": 4.5376, + "grad_norm": 0.0, + "learning_rate": 0.00043939999999999995, + "loss": 4.4789, + "step": 2836 + }, + { + "epoch": 4.5392, + "grad_norm": 0.0, + "learning_rate": 0.00043919999999999994, + "loss": 4.5969, + "step": 2837 + }, + { + "epoch": 4.5408, + "grad_norm": 0.0, + "learning_rate": 0.000439, + "loss": 5.2949, + "step": 2838 + }, + { + "epoch": 4.5424, + "grad_norm": 0.0, + "learning_rate": 0.00043879999999999993, + "loss": 5.4931, + "step": 2839 + }, + { + "epoch": 4.5440000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00043859999999999993, + "loss": 7.0166, + "step": 2840 + }, + { + "epoch": 4.5456, + "grad_norm": 0.0, + "learning_rate": 0.0004384, + "loss": 4.1068, + "step": 2841 + }, + { + "epoch": 4.5472, + "grad_norm": 0.0, + "learning_rate": 0.0004381999999999999, + "loss": 4.6902, + "step": 2842 + }, + { + "epoch": 4.5488, + "grad_norm": 0.0, + "learning_rate": 0.00043799999999999997, + "loss": 4.0462, + "step": 2843 + }, + { + "epoch": 4.5504, + "grad_norm": 0.0, + "learning_rate": 0.00043779999999999996, + "loss": 4.897, + "step": 2844 + }, + { + "epoch": 4.552, + "grad_norm": 0.0, + "learning_rate": 0.0004375999999999999, + "loss": 5.958, + "step": 2845 + }, + { + "epoch": 4.5536, + "grad_norm": 0.0, + "learning_rate": 0.00043739999999999995, + "loss": 4.2306, + "step": 2846 + }, + { + "epoch": 4.5552, + "grad_norm": 0.0, + "learning_rate": 0.00043719999999999995, + "loss": 5.5561, + "step": 2847 + }, + { + "epoch": 4.5568, + "grad_norm": 0.0, + "learning_rate": 0.000437, + "loss": 4.5923, + "step": 2848 + }, + { + "epoch": 4.5584, + "grad_norm": 0.0, + "learning_rate": 0.00043679999999999994, + "loss": 4.4687, + "step": 2849 + }, + { + "epoch": 4.5600000000000005, + "grad_norm": 0.0, + "learning_rate": 0.0004366, + "loss": 6.6604, + "step": 2850 + }, + { + "epoch": 4.5616, + "grad_norm": 0.0, + "learning_rate": 0.0004364, + "loss": 6.7391, + "step": 2851 + }, + { + "epoch": 4.5632, + "grad_norm": 0.0, + "learning_rate": 0.0004361999999999999, + "loss": 9.4595, + "step": 2852 + }, + { + "epoch": 4.5648, + "grad_norm": 0.0, + "learning_rate": 0.00043599999999999997, + "loss": 6.9115, + "step": 2853 + }, + { + "epoch": 4.5664, + "grad_norm": 0.0, + "learning_rate": 0.00043579999999999997, + "loss": 6.8698, + "step": 2854 + }, + { + "epoch": 4.568, + "grad_norm": 0.0, + "learning_rate": 0.00043559999999999996, + "loss": 5.8743, + "step": 2855 + }, + { + "epoch": 4.5696, + "grad_norm": 0.0, + "learning_rate": 0.00043539999999999996, + "loss": 6.1022, + "step": 2856 + }, + { + "epoch": 4.5712, + "grad_norm": 0.0, + "learning_rate": 0.0004352, + "loss": 6.6078, + "step": 2857 + }, + { + "epoch": 4.5728, + "grad_norm": 0.0, + "learning_rate": 0.00043499999999999995, + "loss": 5.2506, + "step": 2858 + }, + { + "epoch": 4.5744, + "grad_norm": 0.0, + "learning_rate": 0.0004348, + "loss": 5.4908, + "step": 2859 + }, + { + "epoch": 4.576, + "grad_norm": 0.0, + "learning_rate": 0.0004346, + "loss": 5.4408, + "step": 2860 + }, + { + "epoch": 4.5776, + "grad_norm": 0.0, + "learning_rate": 0.00043439999999999993, + "loss": 4.4981, + "step": 2861 + }, + { + "epoch": 4.5792, + "grad_norm": 0.0, + "learning_rate": 0.0004342, + "loss": 5.0869, + "step": 2862 + }, + { + "epoch": 4.5808, + "grad_norm": 0.0, + "learning_rate": 0.000434, + "loss": 6.4992, + "step": 2863 + }, + { + "epoch": 4.5824, + "grad_norm": 0.0, + "learning_rate": 0.0004338, + "loss": 5.6383, + "step": 2864 + }, + { + "epoch": 4.584, + "grad_norm": 0.0, + "learning_rate": 0.00043359999999999997, + "loss": 5.1236, + "step": 2865 + }, + { + "epoch": 4.5856, + "grad_norm": 0.0, + "learning_rate": 0.0004334, + "loss": 4.9954, + "step": 2866 + }, + { + "epoch": 4.5872, + "grad_norm": 0.0, + "learning_rate": 0.00043319999999999996, + "loss": 4.672, + "step": 2867 + }, + { + "epoch": 4.5888, + "grad_norm": 0.0, + "learning_rate": 0.00043299999999999995, + "loss": 4.0659, + "step": 2868 + }, + { + "epoch": 4.5904, + "grad_norm": 0.0, + "learning_rate": 0.0004328, + "loss": 5.3073, + "step": 2869 + }, + { + "epoch": 4.592, + "grad_norm": 0.0, + "learning_rate": 0.00043259999999999994, + "loss": 5.6311, + "step": 2870 + }, + { + "epoch": 4.5936, + "grad_norm": 0.0, + "learning_rate": 0.0004324, + "loss": 4.7312, + "step": 2871 + }, + { + "epoch": 4.5952, + "grad_norm": 0.0, + "learning_rate": 0.0004322, + "loss": 5.8437, + "step": 2872 + }, + { + "epoch": 4.5968, + "grad_norm": 0.0, + "learning_rate": 0.00043199999999999993, + "loss": 4.181, + "step": 2873 + }, + { + "epoch": 4.5984, + "grad_norm": 0.0, + "learning_rate": 0.0004318, + "loss": 5.6732, + "step": 2874 + }, + { + "epoch": 4.6, + "grad_norm": 0.0, + "learning_rate": 0.0004316, + "loss": 5.0165, + "step": 2875 + }, + { + "epoch": 4.6016, + "grad_norm": 0.0, + "learning_rate": 0.00043139999999999997, + "loss": 5.9094, + "step": 2876 + }, + { + "epoch": 4.6032, + "grad_norm": 0.0, + "learning_rate": 0.00043119999999999996, + "loss": 3.7399, + "step": 2877 + }, + { + "epoch": 4.6048, + "grad_norm": 0.0, + "learning_rate": 0.000431, + "loss": 4.8993, + "step": 2878 + }, + { + "epoch": 4.6064, + "grad_norm": 0.0, + "learning_rate": 0.00043079999999999995, + "loss": 4.6256, + "step": 2879 + }, + { + "epoch": 4.608, + "grad_norm": 0.0, + "learning_rate": 0.00043059999999999995, + "loss": 4.7072, + "step": 2880 + }, + { + "epoch": 4.6096, + "grad_norm": 0.0, + "learning_rate": 0.0004304, + "loss": 4.5723, + "step": 2881 + }, + { + "epoch": 4.6112, + "grad_norm": 0.0, + "learning_rate": 0.00043019999999999994, + "loss": 4.2221, + "step": 2882 + }, + { + "epoch": 4.6128, + "grad_norm": 0.0, + "learning_rate": 0.00043, + "loss": 4.2942, + "step": 2883 + }, + { + "epoch": 4.6144, + "grad_norm": 0.0, + "learning_rate": 0.0004298, + "loss": 6.2476, + "step": 2884 + }, + { + "epoch": 4.616, + "grad_norm": 0.0, + "learning_rate": 0.0004295999999999999, + "loss": 4.6567, + "step": 2885 + }, + { + "epoch": 4.6176, + "grad_norm": 0.0, + "learning_rate": 0.0004294, + "loss": 5.1709, + "step": 2886 + }, + { + "epoch": 4.6192, + "grad_norm": 0.0, + "learning_rate": 0.00042919999999999997, + "loss": 5.8477, + "step": 2887 + }, + { + "epoch": 4.6208, + "grad_norm": 0.0, + "learning_rate": 0.00042899999999999997, + "loss": 4.9164, + "step": 2888 + }, + { + "epoch": 4.6224, + "grad_norm": 0.0, + "learning_rate": 0.00042879999999999996, + "loss": 3.8585, + "step": 2889 + }, + { + "epoch": 4.624, + "grad_norm": 0.0, + "learning_rate": 0.0004286, + "loss": 4.5949, + "step": 2890 + }, + { + "epoch": 4.6256, + "grad_norm": 0.0, + "learning_rate": 0.00042839999999999995, + "loss": 3.6914, + "step": 2891 + }, + { + "epoch": 4.6272, + "grad_norm": 0.0, + "learning_rate": 0.00042819999999999995, + "loss": 5.2296, + "step": 2892 + }, + { + "epoch": 4.6288, + "grad_norm": 0.0, + "learning_rate": 0.000428, + "loss": 6.8092, + "step": 2893 + }, + { + "epoch": 4.6304, + "grad_norm": 0.0, + "learning_rate": 0.00042779999999999994, + "loss": 4.4948, + "step": 2894 + }, + { + "epoch": 4.632, + "grad_norm": 0.0, + "learning_rate": 0.0004276, + "loss": 4.3081, + "step": 2895 + }, + { + "epoch": 4.6336, + "grad_norm": 0.0, + "learning_rate": 0.0004274, + "loss": 4.1203, + "step": 2896 + }, + { + "epoch": 4.6352, + "grad_norm": 0.0, + "learning_rate": 0.0004271999999999999, + "loss": 6.3119, + "step": 2897 + }, + { + "epoch": 4.6368, + "grad_norm": 0.0, + "learning_rate": 0.00042699999999999997, + "loss": 6.0715, + "step": 2898 + }, + { + "epoch": 4.6384, + "grad_norm": 0.0, + "learning_rate": 0.00042679999999999997, + "loss": 5.0456, + "step": 2899 + }, + { + "epoch": 4.64, + "grad_norm": 0.5054150819778442, + "learning_rate": 0.00042659999999999996, + "loss": 7.2105, + "step": 2900 + }, + { + "epoch": 4.6416, + "grad_norm": 0.0, + "learning_rate": 0.00042639999999999996, + "loss": 7.5307, + "step": 2901 + }, + { + "epoch": 4.6432, + "grad_norm": 0.0, + "learning_rate": 0.0004262, + "loss": 5.5226, + "step": 2902 + }, + { + "epoch": 4.6448, + "grad_norm": 0.0, + "learning_rate": 0.00042599999999999995, + "loss": 7.2145, + "step": 2903 + }, + { + "epoch": 4.6464, + "grad_norm": 0.0, + "learning_rate": 0.00042579999999999994, + "loss": 7.0281, + "step": 2904 + }, + { + "epoch": 4.648, + "grad_norm": 0.0, + "learning_rate": 0.0004256, + "loss": 6.662, + "step": 2905 + }, + { + "epoch": 4.6495999999999995, + "grad_norm": 0.0, + "learning_rate": 0.00042539999999999993, + "loss": 5.4537, + "step": 2906 + }, + { + "epoch": 4.6512, + "grad_norm": 0.0, + "learning_rate": 0.0004252, + "loss": 5.9836, + "step": 2907 + }, + { + "epoch": 4.6528, + "grad_norm": 0.0, + "learning_rate": 0.000425, + "loss": 7.3074, + "step": 2908 + }, + { + "epoch": 4.6544, + "grad_norm": 0.0, + "learning_rate": 0.0004247999999999999, + "loss": 6.1179, + "step": 2909 + }, + { + "epoch": 4.656, + "grad_norm": 0.0, + "learning_rate": 0.00042459999999999997, + "loss": 6.4216, + "step": 2910 + }, + { + "epoch": 4.6576, + "grad_norm": 0.0, + "learning_rate": 0.00042439999999999996, + "loss": 4.8458, + "step": 2911 + }, + { + "epoch": 4.6592, + "grad_norm": 0.0, + "learning_rate": 0.00042419999999999996, + "loss": 6.493, + "step": 2912 + }, + { + "epoch": 4.6608, + "grad_norm": 0.0, + "learning_rate": 0.00042399999999999995, + "loss": 5.6785, + "step": 2913 + }, + { + "epoch": 4.6624, + "grad_norm": 0.0, + "learning_rate": 0.0004238, + "loss": 5.1236, + "step": 2914 + }, + { + "epoch": 4.664, + "grad_norm": 0.0, + "learning_rate": 0.00042359999999999994, + "loss": 5.6009, + "step": 2915 + }, + { + "epoch": 4.6655999999999995, + "grad_norm": 0.0, + "learning_rate": 0.00042339999999999994, + "loss": 4.3321, + "step": 2916 + }, + { + "epoch": 4.6672, + "grad_norm": 0.0, + "learning_rate": 0.0004232, + "loss": 5.8398, + "step": 2917 + }, + { + "epoch": 4.6688, + "grad_norm": 0.0, + "learning_rate": 0.00042299999999999993, + "loss": 5.6342, + "step": 2918 + }, + { + "epoch": 4.6704, + "grad_norm": 0.0, + "learning_rate": 0.0004228, + "loss": 5.2944, + "step": 2919 + }, + { + "epoch": 4.672, + "grad_norm": 0.0, + "learning_rate": 0.00042259999999999997, + "loss": 4.9023, + "step": 2920 + }, + { + "epoch": 4.6736, + "grad_norm": 0.0, + "learning_rate": 0.0004223999999999999, + "loss": 5.0967, + "step": 2921 + }, + { + "epoch": 4.6752, + "grad_norm": 0.0, + "learning_rate": 0.00042219999999999996, + "loss": 5.4827, + "step": 2922 + }, + { + "epoch": 4.6768, + "grad_norm": 0.0, + "learning_rate": 0.00042199999999999996, + "loss": 4.8885, + "step": 2923 + }, + { + "epoch": 4.6784, + "grad_norm": 0.0, + "learning_rate": 0.00042179999999999995, + "loss": 5.8827, + "step": 2924 + }, + { + "epoch": 4.68, + "grad_norm": 0.0, + "learning_rate": 0.00042159999999999995, + "loss": 4.563, + "step": 2925 + }, + { + "epoch": 4.6815999999999995, + "grad_norm": 0.0, + "learning_rate": 0.0004214, + "loss": 4.789, + "step": 2926 + }, + { + "epoch": 4.6832, + "grad_norm": 0.0, + "learning_rate": 0.00042119999999999994, + "loss": 3.8655, + "step": 2927 + }, + { + "epoch": 4.6848, + "grad_norm": 0.0, + "learning_rate": 0.00042099999999999993, + "loss": 6.2099, + "step": 2928 + }, + { + "epoch": 4.6864, + "grad_norm": 0.0, + "learning_rate": 0.0004208, + "loss": 5.1643, + "step": 2929 + }, + { + "epoch": 4.688, + "grad_norm": 0.0, + "learning_rate": 0.0004205999999999999, + "loss": 6.0056, + "step": 2930 + }, + { + "epoch": 4.6896, + "grad_norm": 0.0, + "learning_rate": 0.0004204, + "loss": 5.5561, + "step": 2931 + }, + { + "epoch": 4.6912, + "grad_norm": 0.0, + "learning_rate": 0.00042019999999999997, + "loss": 4.9396, + "step": 2932 + }, + { + "epoch": 4.6928, + "grad_norm": 0.0, + "learning_rate": 0.00041999999999999996, + "loss": 5.0199, + "step": 2933 + }, + { + "epoch": 4.6944, + "grad_norm": 0.0, + "learning_rate": 0.00041979999999999996, + "loss": 5.0728, + "step": 2934 + }, + { + "epoch": 4.696, + "grad_norm": 0.0, + "learning_rate": 0.0004196, + "loss": 5.1721, + "step": 2935 + }, + { + "epoch": 4.6975999999999996, + "grad_norm": 0.0, + "learning_rate": 0.00041939999999999995, + "loss": 5.5109, + "step": 2936 + }, + { + "epoch": 4.6992, + "grad_norm": 0.0, + "learning_rate": 0.00041919999999999994, + "loss": 8.4421, + "step": 2937 + }, + { + "epoch": 4.7008, + "grad_norm": 0.0, + "learning_rate": 0.000419, + "loss": 4.782, + "step": 2938 + }, + { + "epoch": 4.7024, + "grad_norm": 0.0, + "learning_rate": 0.00041879999999999993, + "loss": 4.2826, + "step": 2939 + }, + { + "epoch": 4.704, + "grad_norm": 0.0, + "learning_rate": 0.0004186, + "loss": 5.7527, + "step": 2940 + }, + { + "epoch": 4.7056000000000004, + "grad_norm": 0.0, + "learning_rate": 0.0004184, + "loss": 5.1368, + "step": 2941 + }, + { + "epoch": 4.7072, + "grad_norm": 0.0, + "learning_rate": 0.0004181999999999999, + "loss": 4.1887, + "step": 2942 + }, + { + "epoch": 4.7088, + "grad_norm": 0.0, + "learning_rate": 0.00041799999999999997, + "loss": 3.4475, + "step": 2943 + }, + { + "epoch": 4.7104, + "grad_norm": 0.0, + "learning_rate": 0.00041779999999999996, + "loss": 5.7149, + "step": 2944 + }, + { + "epoch": 4.712, + "grad_norm": 0.0, + "learning_rate": 0.00041759999999999996, + "loss": 5.1234, + "step": 2945 + }, + { + "epoch": 4.7136, + "grad_norm": 0.0, + "learning_rate": 0.00041739999999999995, + "loss": 6.9552, + "step": 2946 + }, + { + "epoch": 4.7152, + "grad_norm": 0.0, + "learning_rate": 0.0004172, + "loss": 7.8605, + "step": 2947 + }, + { + "epoch": 4.7168, + "grad_norm": 0.0, + "learning_rate": 0.00041699999999999994, + "loss": 5.1045, + "step": 2948 + }, + { + "epoch": 4.7184, + "grad_norm": 3.3168752193450928, + "learning_rate": 0.00041679999999999994, + "loss": 4.8862, + "step": 2949 + }, + { + "epoch": 4.72, + "grad_norm": NaN, + "learning_rate": 0.00041679999999999994, + "loss": 4.8903, + "step": 2950 + }, + { + "epoch": 4.7216000000000005, + "grad_norm": 0.0, + "learning_rate": 0.0004166, + "loss": 7.4356, + "step": 2951 + }, + { + "epoch": 4.7232, + "grad_norm": 0.0, + "learning_rate": 0.00041639999999999993, + "loss": 6.7875, + "step": 2952 + }, + { + "epoch": 4.7248, + "grad_norm": 0.0, + "learning_rate": 0.0004162, + "loss": 8.4374, + "step": 2953 + }, + { + "epoch": 4.7264, + "grad_norm": 0.0, + "learning_rate": 0.000416, + "loss": 6.2717, + "step": 2954 + }, + { + "epoch": 4.728, + "grad_norm": 0.0, + "learning_rate": 0.0004157999999999999, + "loss": 5.2044, + "step": 2955 + }, + { + "epoch": 4.7296, + "grad_norm": 0.0, + "learning_rate": 0.00041559999999999996, + "loss": 6.9109, + "step": 2956 + }, + { + "epoch": 4.7312, + "grad_norm": 0.0, + "learning_rate": 0.00041539999999999996, + "loss": 5.2435, + "step": 2957 + }, + { + "epoch": 4.7328, + "grad_norm": 0.0, + "learning_rate": 0.00041519999999999995, + "loss": 6.793, + "step": 2958 + }, + { + "epoch": 4.7344, + "grad_norm": 0.0, + "learning_rate": 0.00041499999999999995, + "loss": 5.6909, + "step": 2959 + }, + { + "epoch": 4.736, + "grad_norm": 0.0, + "learning_rate": 0.0004148, + "loss": 6.6258, + "step": 2960 + }, + { + "epoch": 4.7376000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00041459999999999994, + "loss": 4.6512, + "step": 2961 + }, + { + "epoch": 4.7392, + "grad_norm": 0.0, + "learning_rate": 0.00041439999999999994, + "loss": 5.9544, + "step": 2962 + }, + { + "epoch": 4.7408, + "grad_norm": 0.0, + "learning_rate": 0.0004142, + "loss": 4.9663, + "step": 2963 + }, + { + "epoch": 4.7424, + "grad_norm": 0.0, + "learning_rate": 0.0004139999999999999, + "loss": 5.8425, + "step": 2964 + }, + { + "epoch": 4.744, + "grad_norm": 0.0, + "learning_rate": 0.0004138, + "loss": 5.9031, + "step": 2965 + }, + { + "epoch": 4.7456, + "grad_norm": 0.0, + "learning_rate": 0.00041359999999999997, + "loss": 5.5365, + "step": 2966 + }, + { + "epoch": 4.7472, + "grad_norm": 0.0, + "learning_rate": 0.0004133999999999999, + "loss": 6.1423, + "step": 2967 + }, + { + "epoch": 4.7488, + "grad_norm": 0.0, + "learning_rate": 0.00041319999999999996, + "loss": 6.283, + "step": 2968 + }, + { + "epoch": 4.7504, + "grad_norm": 0.0, + "learning_rate": 0.00041299999999999996, + "loss": 7.0199, + "step": 2969 + }, + { + "epoch": 4.752, + "grad_norm": 0.0, + "learning_rate": 0.00041279999999999995, + "loss": 6.617, + "step": 2970 + }, + { + "epoch": 4.7536000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00041259999999999995, + "loss": 4.9052, + "step": 2971 + }, + { + "epoch": 4.7552, + "grad_norm": 0.0, + "learning_rate": 0.0004124, + "loss": 4.7351, + "step": 2972 + }, + { + "epoch": 4.7568, + "grad_norm": 0.0, + "learning_rate": 0.0004122, + "loss": 6.1535, + "step": 2973 + }, + { + "epoch": 4.7584, + "grad_norm": 0.0, + "learning_rate": 0.00041199999999999993, + "loss": 5.4087, + "step": 2974 + }, + { + "epoch": 4.76, + "grad_norm": 0.0, + "learning_rate": 0.0004118, + "loss": 5.1719, + "step": 2975 + }, + { + "epoch": 4.7616, + "grad_norm": 0.0, + "learning_rate": 0.0004116, + "loss": 4.4333, + "step": 2976 + }, + { + "epoch": 4.7632, + "grad_norm": 0.0, + "learning_rate": 0.00041139999999999997, + "loss": 6.293, + "step": 2977 + }, + { + "epoch": 4.7648, + "grad_norm": 0.0, + "learning_rate": 0.00041119999999999997, + "loss": 5.8174, + "step": 2978 + }, + { + "epoch": 4.7664, + "grad_norm": 0.0, + "learning_rate": 0.000411, + "loss": 4.1179, + "step": 2979 + }, + { + "epoch": 4.768, + "grad_norm": 0.0, + "learning_rate": 0.00041079999999999996, + "loss": 5.873, + "step": 2980 + }, + { + "epoch": 4.7696, + "grad_norm": 0.0, + "learning_rate": 0.00041059999999999995, + "loss": 4.9865, + "step": 2981 + }, + { + "epoch": 4.7712, + "grad_norm": 0.0, + "learning_rate": 0.0004104, + "loss": 5.094, + "step": 2982 + }, + { + "epoch": 4.7728, + "grad_norm": 0.0, + "learning_rate": 0.00041019999999999994, + "loss": 4.8503, + "step": 2983 + }, + { + "epoch": 4.7744, + "grad_norm": 0.0, + "learning_rate": 0.00041, + "loss": 5.678, + "step": 2984 + }, + { + "epoch": 4.776, + "grad_norm": 0.0, + "learning_rate": 0.0004098, + "loss": 7.1912, + "step": 2985 + }, + { + "epoch": 4.7776, + "grad_norm": 0.0, + "learning_rate": 0.00040959999999999993, + "loss": 5.2119, + "step": 2986 + }, + { + "epoch": 4.7792, + "grad_norm": 0.0, + "learning_rate": 0.0004094, + "loss": 4.386, + "step": 2987 + }, + { + "epoch": 4.7808, + "grad_norm": 0.0, + "learning_rate": 0.00040919999999999997, + "loss": 5.1441, + "step": 2988 + }, + { + "epoch": 4.7824, + "grad_norm": 0.0, + "learning_rate": 0.00040899999999999997, + "loss": 5.8762, + "step": 2989 + }, + { + "epoch": 4.784, + "grad_norm": 0.0, + "learning_rate": 0.00040879999999999996, + "loss": 4.5841, + "step": 2990 + }, + { + "epoch": 4.7856, + "grad_norm": 0.0, + "learning_rate": 0.0004086, + "loss": 6.3816, + "step": 2991 + }, + { + "epoch": 4.7872, + "grad_norm": 0.0, + "learning_rate": 0.00040839999999999995, + "loss": 4.907, + "step": 2992 + }, + { + "epoch": 4.7888, + "grad_norm": 0.0, + "learning_rate": 0.00040819999999999995, + "loss": 6.8205, + "step": 2993 + }, + { + "epoch": 4.7904, + "grad_norm": 0.0, + "learning_rate": 0.000408, + "loss": 6.3567, + "step": 2994 + }, + { + "epoch": 4.792, + "grad_norm": 0.0, + "learning_rate": 0.00040779999999999994, + "loss": 3.7933, + "step": 2995 + }, + { + "epoch": 4.7936, + "grad_norm": 0.0, + "learning_rate": 0.0004076, + "loss": 4.7355, + "step": 2996 + }, + { + "epoch": 4.7952, + "grad_norm": 0.0, + "learning_rate": 0.0004074, + "loss": 5.2632, + "step": 2997 + }, + { + "epoch": 4.7968, + "grad_norm": 0.0, + "learning_rate": 0.0004071999999999999, + "loss": 5.1249, + "step": 2998 + }, + { + "epoch": 4.7984, + "grad_norm": 0.0, + "learning_rate": 0.00040699999999999997, + "loss": 6.7669, + "step": 2999 + }, + { + "epoch": 4.8, + "grad_norm": NaN, + "learning_rate": 0.00040699999999999997, + "loss": 5.5318, + "step": 3000 + }, + { + "epoch": 4.8016, + "grad_norm": 0.0, + "learning_rate": 0.00040679999999999997, + "loss": 10.2462, + "step": 3001 + }, + { + "epoch": 4.8032, + "grad_norm": 0.0, + "learning_rate": 0.00040659999999999996, + "loss": 6.9759, + "step": 3002 + }, + { + "epoch": 4.8048, + "grad_norm": 0.0, + "learning_rate": 0.00040639999999999996, + "loss": 5.6984, + "step": 3003 + }, + { + "epoch": 4.8064, + "grad_norm": 0.0, + "learning_rate": 0.0004062, + "loss": 7.0211, + "step": 3004 + }, + { + "epoch": 4.808, + "grad_norm": 0.0, + "learning_rate": 0.00040599999999999995, + "loss": 7.0675, + "step": 3005 + }, + { + "epoch": 4.8096, + "grad_norm": 0.0, + "learning_rate": 0.00040579999999999994, + "loss": 8.8733, + "step": 3006 + }, + { + "epoch": 4.8112, + "grad_norm": 0.0, + "learning_rate": 0.0004056, + "loss": 6.5793, + "step": 3007 + }, + { + "epoch": 4.8128, + "grad_norm": 0.0, + "learning_rate": 0.00040539999999999993, + "loss": 5.8908, + "step": 3008 + }, + { + "epoch": 4.8144, + "grad_norm": 0.0, + "learning_rate": 0.0004052, + "loss": 6.2365, + "step": 3009 + }, + { + "epoch": 4.816, + "grad_norm": 0.0, + "learning_rate": 0.000405, + "loss": 6.5827, + "step": 3010 + }, + { + "epoch": 4.8176, + "grad_norm": 0.0, + "learning_rate": 0.0004048, + "loss": 8.2913, + "step": 3011 + }, + { + "epoch": 4.8192, + "grad_norm": 0.0, + "learning_rate": 0.00040459999999999997, + "loss": 5.7216, + "step": 3012 + }, + { + "epoch": 4.8208, + "grad_norm": 0.0, + "learning_rate": 0.0004044, + "loss": 4.7472, + "step": 3013 + }, + { + "epoch": 4.8224, + "grad_norm": 0.0, + "learning_rate": 0.00040419999999999996, + "loss": 5.3596, + "step": 3014 + }, + { + "epoch": 4.824, + "grad_norm": 0.0, + "learning_rate": 0.00040399999999999995, + "loss": 6.5334, + "step": 3015 + }, + { + "epoch": 4.8256, + "grad_norm": 0.0, + "learning_rate": 0.0004038, + "loss": 6.4598, + "step": 3016 + }, + { + "epoch": 4.8272, + "grad_norm": 0.0, + "learning_rate": 0.00040359999999999994, + "loss": 5.1912, + "step": 3017 + }, + { + "epoch": 4.8288, + "grad_norm": 0.0, + "learning_rate": 0.0004034, + "loss": 4.6177, + "step": 3018 + }, + { + "epoch": 4.8304, + "grad_norm": 0.0, + "learning_rate": 0.0004032, + "loss": 4.7634, + "step": 3019 + }, + { + "epoch": 4.832, + "grad_norm": 0.0, + "learning_rate": 0.00040299999999999993, + "loss": 5.1462, + "step": 3020 + }, + { + "epoch": 4.8336, + "grad_norm": 0.0, + "learning_rate": 0.0004028, + "loss": 4.9172, + "step": 3021 + }, + { + "epoch": 4.8352, + "grad_norm": 0.0, + "learning_rate": 0.0004026, + "loss": 8.2852, + "step": 3022 + }, + { + "epoch": 4.8368, + "grad_norm": 0.0, + "learning_rate": 0.00040239999999999997, + "loss": 4.7602, + "step": 3023 + }, + { + "epoch": 4.8384, + "grad_norm": 0.0, + "learning_rate": 0.00040219999999999996, + "loss": 5.6188, + "step": 3024 + }, + { + "epoch": 4.84, + "grad_norm": 0.0, + "learning_rate": 0.000402, + "loss": 5.4905, + "step": 3025 + }, + { + "epoch": 4.8416, + "grad_norm": 0.0, + "learning_rate": 0.00040179999999999995, + "loss": 5.3078, + "step": 3026 + }, + { + "epoch": 4.8431999999999995, + "grad_norm": 0.0, + "learning_rate": 0.00040159999999999995, + "loss": 4.5428, + "step": 3027 + }, + { + "epoch": 4.8448, + "grad_norm": 0.0, + "learning_rate": 0.0004014, + "loss": 5.7685, + "step": 3028 + }, + { + "epoch": 4.8464, + "grad_norm": 0.0, + "learning_rate": 0.00040119999999999994, + "loss": 4.6266, + "step": 3029 + }, + { + "epoch": 4.848, + "grad_norm": 0.0, + "learning_rate": 0.000401, + "loss": 4.8828, + "step": 3030 + }, + { + "epoch": 4.8496, + "grad_norm": 0.0, + "learning_rate": 0.0004008, + "loss": 6.5834, + "step": 3031 + }, + { + "epoch": 4.8512, + "grad_norm": 0.0, + "learning_rate": 0.0004005999999999999, + "loss": 5.0824, + "step": 3032 + }, + { + "epoch": 4.8528, + "grad_norm": 0.0, + "learning_rate": 0.0004004, + "loss": 4.8299, + "step": 3033 + }, + { + "epoch": 4.8544, + "grad_norm": 0.0, + "learning_rate": 0.00040019999999999997, + "loss": 5.5092, + "step": 3034 + }, + { + "epoch": 4.856, + "grad_norm": 0.0, + "learning_rate": 0.00039999999999999996, + "loss": 5.2468, + "step": 3035 + }, + { + "epoch": 4.8576, + "grad_norm": 0.0, + "learning_rate": 0.00039979999999999996, + "loss": 4.6701, + "step": 3036 + }, + { + "epoch": 4.8591999999999995, + "grad_norm": 0.0, + "learning_rate": 0.0003996, + "loss": 6.5081, + "step": 3037 + }, + { + "epoch": 4.8608, + "grad_norm": 0.0, + "learning_rate": 0.00039939999999999995, + "loss": 4.3119, + "step": 3038 + }, + { + "epoch": 4.8624, + "grad_norm": 0.0, + "learning_rate": 0.00039919999999999995, + "loss": 4.9543, + "step": 3039 + }, + { + "epoch": 4.864, + "grad_norm": 0.0, + "learning_rate": 0.000399, + "loss": 7.465, + "step": 3040 + }, + { + "epoch": 4.8656, + "grad_norm": 0.0, + "learning_rate": 0.00039879999999999994, + "loss": 4.2138, + "step": 3041 + }, + { + "epoch": 4.8672, + "grad_norm": 0.0, + "learning_rate": 0.0003986, + "loss": 4.8901, + "step": 3042 + }, + { + "epoch": 4.8688, + "grad_norm": 0.0, + "learning_rate": 0.0003984, + "loss": 5.0104, + "step": 3043 + }, + { + "epoch": 4.8704, + "grad_norm": 0.0, + "learning_rate": 0.0003981999999999999, + "loss": 4.4113, + "step": 3044 + }, + { + "epoch": 4.872, + "grad_norm": 0.0, + "learning_rate": 0.00039799999999999997, + "loss": 5.7516, + "step": 3045 + }, + { + "epoch": 4.8736, + "grad_norm": 0.0, + "learning_rate": 0.00039779999999999997, + "loss": 5.5562, + "step": 3046 + }, + { + "epoch": 4.8751999999999995, + "grad_norm": 0.0, + "learning_rate": 0.00039759999999999996, + "loss": 5.4988, + "step": 3047 + }, + { + "epoch": 4.8768, + "grad_norm": 0.0, + "learning_rate": 0.00039739999999999996, + "loss": 5.4396, + "step": 3048 + }, + { + "epoch": 4.8784, + "grad_norm": 0.0, + "learning_rate": 0.0003972, + "loss": 4.7456, + "step": 3049 + }, + { + "epoch": 4.88, + "grad_norm": NaN, + "learning_rate": 0.0003972, + "loss": 7.3277, + "step": 3050 + }, + { + "epoch": 4.8816, + "grad_norm": 0.0, + "learning_rate": 0.00039699999999999995, + "loss": 7.5654, + "step": 3051 + }, + { + "epoch": 4.8832, + "grad_norm": 0.0, + "learning_rate": 0.00039679999999999994, + "loss": 11.4621, + "step": 3052 + }, + { + "epoch": 4.8848, + "grad_norm": 0.0, + "learning_rate": 0.0003966, + "loss": 7.8216, + "step": 3053 + }, + { + "epoch": 4.8864, + "grad_norm": 0.0, + "learning_rate": 0.00039639999999999993, + "loss": 6.2116, + "step": 3054 + }, + { + "epoch": 4.888, + "grad_norm": 0.0, + "learning_rate": 0.0003962, + "loss": 5.7363, + "step": 3055 + }, + { + "epoch": 4.8896, + "grad_norm": 0.0, + "learning_rate": 0.000396, + "loss": 6.5144, + "step": 3056 + }, + { + "epoch": 4.8911999999999995, + "grad_norm": 0.0, + "learning_rate": 0.0003957999999999999, + "loss": 5.917, + "step": 3057 + }, + { + "epoch": 4.8928, + "grad_norm": 0.0, + "learning_rate": 0.00039559999999999997, + "loss": 6.1872, + "step": 3058 + }, + { + "epoch": 4.8944, + "grad_norm": 0.0, + "learning_rate": 0.00039539999999999996, + "loss": 6.0313, + "step": 3059 + }, + { + "epoch": 4.896, + "grad_norm": 0.0, + "learning_rate": 0.00039519999999999996, + "loss": 5.5926, + "step": 3060 + }, + { + "epoch": 4.8976, + "grad_norm": 0.0, + "learning_rate": 0.00039499999999999995, + "loss": 6.9679, + "step": 3061 + }, + { + "epoch": 4.8992, + "grad_norm": 0.0, + "learning_rate": 0.0003948, + "loss": 7.5164, + "step": 3062 + }, + { + "epoch": 4.9008, + "grad_norm": 0.0, + "learning_rate": 0.00039459999999999994, + "loss": 5.4189, + "step": 3063 + }, + { + "epoch": 4.9024, + "grad_norm": 0.0, + "learning_rate": 0.00039439999999999994, + "loss": 5.2463, + "step": 3064 + }, + { + "epoch": 4.904, + "grad_norm": 0.0, + "learning_rate": 0.0003942, + "loss": 5.7083, + "step": 3065 + }, + { + "epoch": 4.9056, + "grad_norm": 0.0, + "learning_rate": 0.00039399999999999993, + "loss": 5.7834, + "step": 3066 + }, + { + "epoch": 4.9072, + "grad_norm": 0.0, + "learning_rate": 0.0003938, + "loss": 4.8085, + "step": 3067 + }, + { + "epoch": 4.9088, + "grad_norm": 0.0, + "learning_rate": 0.00039359999999999997, + "loss": 5.291, + "step": 3068 + }, + { + "epoch": 4.9104, + "grad_norm": 0.0, + "learning_rate": 0.0003933999999999999, + "loss": 6.1978, + "step": 3069 + }, + { + "epoch": 4.912, + "grad_norm": 0.0, + "learning_rate": 0.00039319999999999996, + "loss": 5.6269, + "step": 3070 + }, + { + "epoch": 4.9136, + "grad_norm": 0.0, + "learning_rate": 0.00039299999999999996, + "loss": 4.6422, + "step": 3071 + }, + { + "epoch": 4.9152000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00039279999999999995, + "loss": 6.294, + "step": 3072 + }, + { + "epoch": 4.9168, + "grad_norm": 0.0, + "learning_rate": 0.00039259999999999995, + "loss": 6.4443, + "step": 3073 + }, + { + "epoch": 4.9184, + "grad_norm": 0.0, + "learning_rate": 0.0003924, + "loss": 7.8158, + "step": 3074 + }, + { + "epoch": 4.92, + "grad_norm": 0.0, + "learning_rate": 0.00039219999999999994, + "loss": 5.6271, + "step": 3075 + }, + { + "epoch": 4.9216, + "grad_norm": 0.0, + "learning_rate": 0.00039199999999999993, + "loss": 5.1731, + "step": 3076 + }, + { + "epoch": 4.9232, + "grad_norm": 0.0, + "learning_rate": 0.0003918, + "loss": 4.3069, + "step": 3077 + }, + { + "epoch": 4.9248, + "grad_norm": 0.0, + "learning_rate": 0.0003915999999999999, + "loss": 5.0069, + "step": 3078 + }, + { + "epoch": 4.9264, + "grad_norm": 0.0, + "learning_rate": 0.00039139999999999997, + "loss": 5.2292, + "step": 3079 + }, + { + "epoch": 4.928, + "grad_norm": 0.0, + "learning_rate": 0.00039119999999999997, + "loss": 4.2463, + "step": 3080 + }, + { + "epoch": 4.9296, + "grad_norm": 0.0, + "learning_rate": 0.0003909999999999999, + "loss": 5.2186, + "step": 3081 + }, + { + "epoch": 4.9312000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00039079999999999996, + "loss": 4.6326, + "step": 3082 + }, + { + "epoch": 4.9328, + "grad_norm": 0.0, + "learning_rate": 0.00039059999999999995, + "loss": 5.6659, + "step": 3083 + }, + { + "epoch": 4.9344, + "grad_norm": 0.0, + "learning_rate": 0.00039039999999999995, + "loss": 4.5974, + "step": 3084 + }, + { + "epoch": 4.936, + "grad_norm": 0.0, + "learning_rate": 0.00039019999999999994, + "loss": 4.748, + "step": 3085 + }, + { + "epoch": 4.9376, + "grad_norm": 0.0, + "learning_rate": 0.00039, + "loss": 4.8799, + "step": 3086 + }, + { + "epoch": 4.9392, + "grad_norm": 0.0, + "learning_rate": 0.00038979999999999993, + "loss": 5.0342, + "step": 3087 + }, + { + "epoch": 4.9408, + "grad_norm": 0.0, + "learning_rate": 0.0003896, + "loss": 5.5914, + "step": 3088 + }, + { + "epoch": 4.9424, + "grad_norm": 0.0, + "learning_rate": 0.0003894, + "loss": 5.3472, + "step": 3089 + }, + { + "epoch": 4.944, + "grad_norm": 0.0, + "learning_rate": 0.0003891999999999999, + "loss": 5.4005, + "step": 3090 + }, + { + "epoch": 4.9456, + "grad_norm": 0.0, + "learning_rate": 0.00038899999999999997, + "loss": 6.1949, + "step": 3091 + }, + { + "epoch": 4.9472000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00038879999999999996, + "loss": 5.9667, + "step": 3092 + }, + { + "epoch": 4.9488, + "grad_norm": 0.0, + "learning_rate": 0.00038859999999999996, + "loss": 5.5679, + "step": 3093 + }, + { + "epoch": 4.9504, + "grad_norm": 0.0, + "learning_rate": 0.00038839999999999995, + "loss": 5.1527, + "step": 3094 + }, + { + "epoch": 4.952, + "grad_norm": 0.0, + "learning_rate": 0.0003882, + "loss": 4.991, + "step": 3095 + }, + { + "epoch": 4.9536, + "grad_norm": 0.0, + "learning_rate": 0.00038799999999999994, + "loss": 5.4299, + "step": 3096 + }, + { + "epoch": 4.9552, + "grad_norm": 0.0, + "learning_rate": 0.00038779999999999994, + "loss": 7.1702, + "step": 3097 + }, + { + "epoch": 4.9568, + "grad_norm": 0.0, + "learning_rate": 0.0003876, + "loss": 4.5268, + "step": 3098 + }, + { + "epoch": 4.9584, + "grad_norm": 0.0, + "learning_rate": 0.0003874, + "loss": 7.0175, + "step": 3099 + }, + { + "epoch": 4.96, + "grad_norm": NaN, + "learning_rate": 0.0003874, + "loss": 4.9075, + "step": 3100 + }, + { + "epoch": 4.9616, + "grad_norm": 0.0, + "learning_rate": 0.0003872, + "loss": 9.0214, + "step": 3101 + }, + { + "epoch": 4.9632, + "grad_norm": 0.0, + "learning_rate": 0.000387, + "loss": 5.9765, + "step": 3102 + }, + { + "epoch": 4.9648, + "grad_norm": 0.0, + "learning_rate": 0.0003868, + "loss": 7.0184, + "step": 3103 + }, + { + "epoch": 4.9664, + "grad_norm": 0.0, + "learning_rate": 0.00038659999999999996, + "loss": 7.8053, + "step": 3104 + }, + { + "epoch": 4.968, + "grad_norm": 0.0, + "learning_rate": 0.00038639999999999996, + "loss": 5.9001, + "step": 3105 + }, + { + "epoch": 4.9696, + "grad_norm": 0.0, + "learning_rate": 0.0003862, + "loss": 6.0759, + "step": 3106 + }, + { + "epoch": 4.9712, + "grad_norm": 0.0, + "learning_rate": 0.00038599999999999995, + "loss": 6.2509, + "step": 3107 + }, + { + "epoch": 4.9728, + "grad_norm": 0.0, + "learning_rate": 0.0003858, + "loss": 7.205, + "step": 3108 + }, + { + "epoch": 4.9744, + "grad_norm": 0.0, + "learning_rate": 0.0003856, + "loss": 5.5714, + "step": 3109 + }, + { + "epoch": 4.976, + "grad_norm": 0.0, + "learning_rate": 0.00038539999999999994, + "loss": 8.2532, + "step": 3110 + }, + { + "epoch": 4.9776, + "grad_norm": 0.0, + "learning_rate": 0.0003852, + "loss": 5.728, + "step": 3111 + }, + { + "epoch": 4.9792, + "grad_norm": 0.0, + "learning_rate": 0.000385, + "loss": 5.8812, + "step": 3112 + }, + { + "epoch": 4.9808, + "grad_norm": 0.0, + "learning_rate": 0.0003848, + "loss": 4.8908, + "step": 3113 + }, + { + "epoch": 4.9824, + "grad_norm": 0.0, + "learning_rate": 0.00038459999999999997, + "loss": 4.0623, + "step": 3114 + }, + { + "epoch": 4.984, + "grad_norm": 0.0, + "learning_rate": 0.0003844, + "loss": 5.6762, + "step": 3115 + }, + { + "epoch": 4.9856, + "grad_norm": 0.0, + "learning_rate": 0.00038419999999999996, + "loss": 4.9697, + "step": 3116 + }, + { + "epoch": 4.9872, + "grad_norm": 0.0, + "learning_rate": 0.00038399999999999996, + "loss": 4.1973, + "step": 3117 + }, + { + "epoch": 4.9888, + "grad_norm": 0.0, + "learning_rate": 0.0003838, + "loss": 5.2126, + "step": 3118 + }, + { + "epoch": 4.9904, + "grad_norm": 0.0, + "learning_rate": 0.00038359999999999995, + "loss": 4.5909, + "step": 3119 + }, + { + "epoch": 4.992, + "grad_norm": 0.0, + "learning_rate": 0.0003834, + "loss": 5.4625, + "step": 3120 + }, + { + "epoch": 4.9936, + "grad_norm": 0.0, + "learning_rate": 0.0003832, + "loss": 5.0547, + "step": 3121 + }, + { + "epoch": 4.9952, + "grad_norm": 0.0, + "learning_rate": 0.00038299999999999993, + "loss": 5.2914, + "step": 3122 + }, + { + "epoch": 4.9968, + "grad_norm": 0.0, + "learning_rate": 0.0003828, + "loss": 6.4537, + "step": 3123 + }, + { + "epoch": 4.9984, + "grad_norm": 0.0, + "learning_rate": 0.0003826, + "loss": 6.0983, + "step": 3124 + }, + { + "epoch": 5.0, + "grad_norm": 0.0, + "learning_rate": 0.00038239999999999997, + "loss": 5.5575, + "step": 3125 + }, + { + "epoch": 5.0016, + "grad_norm": 0.0, + "learning_rate": 0.00038219999999999997, + "loss": 8.9916, + "step": 3126 + }, + { + "epoch": 5.0032, + "grad_norm": 0.0, + "learning_rate": 0.000382, + "loss": 7.0435, + "step": 3127 + }, + { + "epoch": 5.0048, + "grad_norm": 0.0, + "learning_rate": 0.00038179999999999996, + "loss": 9.3534, + "step": 3128 + }, + { + "epoch": 5.0064, + "grad_norm": 0.0, + "learning_rate": 0.00038159999999999995, + "loss": 8.0404, + "step": 3129 + }, + { + "epoch": 5.008, + "grad_norm": 0.0, + "learning_rate": 0.0003814, + "loss": 6.1829, + "step": 3130 + }, + { + "epoch": 5.0096, + "grad_norm": 0.0, + "learning_rate": 0.00038119999999999994, + "loss": 7.4766, + "step": 3131 + }, + { + "epoch": 5.0112, + "grad_norm": 0.0, + "learning_rate": 0.000381, + "loss": 6.6343, + "step": 3132 + }, + { + "epoch": 5.0128, + "grad_norm": 0.0, + "learning_rate": 0.0003808, + "loss": 5.7863, + "step": 3133 + }, + { + "epoch": 5.0144, + "grad_norm": 0.0, + "learning_rate": 0.0003805999999999999, + "loss": 6.0439, + "step": 3134 + }, + { + "epoch": 5.016, + "grad_norm": 0.0, + "learning_rate": 0.0003804, + "loss": 5.3478, + "step": 3135 + }, + { + "epoch": 5.0176, + "grad_norm": 0.0, + "learning_rate": 0.00038019999999999997, + "loss": 9.3696, + "step": 3136 + }, + { + "epoch": 5.0192, + "grad_norm": 0.0, + "learning_rate": 0.00037999999999999997, + "loss": 6.8533, + "step": 3137 + }, + { + "epoch": 5.0208, + "grad_norm": 0.0, + "learning_rate": 0.00037979999999999996, + "loss": 10.5829, + "step": 3138 + }, + { + "epoch": 5.0224, + "grad_norm": 0.0, + "learning_rate": 0.0003796, + "loss": 7.599, + "step": 3139 + }, + { + "epoch": 5.024, + "grad_norm": 0.0, + "learning_rate": 0.00037939999999999995, + "loss": 4.6454, + "step": 3140 + }, + { + "epoch": 5.0256, + "grad_norm": 0.0, + "learning_rate": 0.00037919999999999995, + "loss": 5.7198, + "step": 3141 + }, + { + "epoch": 5.0272, + "grad_norm": 0.0, + "learning_rate": 0.000379, + "loss": 5.4773, + "step": 3142 + }, + { + "epoch": 5.0288, + "grad_norm": 0.0, + "learning_rate": 0.00037879999999999994, + "loss": 4.9752, + "step": 3143 + }, + { + "epoch": 5.0304, + "grad_norm": 0.0, + "learning_rate": 0.0003786, + "loss": 7.4662, + "step": 3144 + }, + { + "epoch": 5.032, + "grad_norm": 0.0, + "learning_rate": 0.0003784, + "loss": 5.157, + "step": 3145 + }, + { + "epoch": 5.0336, + "grad_norm": 0.0, + "learning_rate": 0.0003781999999999999, + "loss": 6.8937, + "step": 3146 + }, + { + "epoch": 5.0352, + "grad_norm": 0.0, + "learning_rate": 0.00037799999999999997, + "loss": 5.406, + "step": 3147 + }, + { + "epoch": 5.0368, + "grad_norm": 0.0, + "learning_rate": 0.00037779999999999997, + "loss": 5.8694, + "step": 3148 + }, + { + "epoch": 5.0384, + "grad_norm": 0.0, + "learning_rate": 0.00037759999999999996, + "loss": 4.9241, + "step": 3149 + }, + { + "epoch": 5.04, + "grad_norm": 0.0, + "learning_rate": 0.00037739999999999996, + "loss": 5.2262, + "step": 3150 + }, + { + "epoch": 5.0416, + "grad_norm": 0.0, + "learning_rate": 0.0003772, + "loss": 6.398, + "step": 3151 + }, + { + "epoch": 5.0432, + "grad_norm": 0.0, + "learning_rate": 0.00037699999999999995, + "loss": 6.5993, + "step": 3152 + }, + { + "epoch": 5.0448, + "grad_norm": 0.0, + "learning_rate": 0.00037679999999999994, + "loss": 7.0651, + "step": 3153 + }, + { + "epoch": 5.0464, + "grad_norm": 0.0, + "learning_rate": 0.0003766, + "loss": 4.9128, + "step": 3154 + }, + { + "epoch": 5.048, + "grad_norm": 0.0, + "learning_rate": 0.00037639999999999993, + "loss": 5.2534, + "step": 3155 + }, + { + "epoch": 5.0496, + "grad_norm": 0.0, + "learning_rate": 0.0003762, + "loss": 5.4155, + "step": 3156 + }, + { + "epoch": 5.0512, + "grad_norm": 0.0, + "learning_rate": 0.000376, + "loss": 5.4697, + "step": 3157 + }, + { + "epoch": 5.0528, + "grad_norm": 0.0, + "learning_rate": 0.0003757999999999999, + "loss": 5.3402, + "step": 3158 + }, + { + "epoch": 5.0544, + "grad_norm": 0.0, + "learning_rate": 0.00037559999999999997, + "loss": 4.567, + "step": 3159 + }, + { + "epoch": 5.056, + "grad_norm": 0.0, + "learning_rate": 0.00037539999999999996, + "loss": 5.3377, + "step": 3160 + }, + { + "epoch": 5.0576, + "grad_norm": 0.0, + "learning_rate": 0.00037519999999999996, + "loss": 5.9542, + "step": 3161 + }, + { + "epoch": 5.0592, + "grad_norm": 0.0, + "learning_rate": 0.00037499999999999995, + "loss": 6.3136, + "step": 3162 + }, + { + "epoch": 5.0608, + "grad_norm": 0.0, + "learning_rate": 0.0003748, + "loss": 5.353, + "step": 3163 + }, + { + "epoch": 5.0624, + "grad_norm": 0.0, + "learning_rate": 0.00037459999999999994, + "loss": 4.4713, + "step": 3164 + }, + { + "epoch": 5.064, + "grad_norm": 0.0, + "learning_rate": 0.0003744, + "loss": 7.116, + "step": 3165 + }, + { + "epoch": 5.0656, + "grad_norm": 0.0, + "learning_rate": 0.0003742, + "loss": 5.4026, + "step": 3166 + }, + { + "epoch": 5.0672, + "grad_norm": 0.0, + "learning_rate": 0.00037399999999999993, + "loss": 4.4876, + "step": 3167 + }, + { + "epoch": 5.0688, + "grad_norm": 0.0, + "learning_rate": 0.0003738, + "loss": 5.2662, + "step": 3168 + }, + { + "epoch": 5.0704, + "grad_norm": 0.0, + "learning_rate": 0.0003736, + "loss": 5.4257, + "step": 3169 + }, + { + "epoch": 5.072, + "grad_norm": 0.0, + "learning_rate": 0.00037339999999999997, + "loss": 5.0846, + "step": 3170 + }, + { + "epoch": 5.0736, + "grad_norm": 0.0, + "learning_rate": 0.00037319999999999996, + "loss": 6.6244, + "step": 3171 + }, + { + "epoch": 5.0752, + "grad_norm": 0.0, + "learning_rate": 0.000373, + "loss": 5.6759, + "step": 3172 + }, + { + "epoch": 5.0768, + "grad_norm": 0.0, + "learning_rate": 0.00037279999999999995, + "loss": 5.6952, + "step": 3173 + }, + { + "epoch": 5.0784, + "grad_norm": 0.0, + "learning_rate": 0.00037259999999999995, + "loss": 4.549, + "step": 3174 + }, + { + "epoch": 5.08, + "grad_norm": 0.0, + "learning_rate": 0.0003724, + "loss": 5.9356, + "step": 3175 + }, + { + "epoch": 5.0816, + "grad_norm": 0.0, + "learning_rate": 0.00037219999999999994, + "loss": 8.8312, + "step": 3176 + }, + { + "epoch": 5.0832, + "grad_norm": 0.0, + "learning_rate": 0.000372, + "loss": 8.3805, + "step": 3177 + }, + { + "epoch": 5.0848, + "grad_norm": 0.0, + "learning_rate": 0.0003718, + "loss": 6.2337, + "step": 3178 + }, + { + "epoch": 5.0864, + "grad_norm": 0.0, + "learning_rate": 0.0003715999999999999, + "loss": 5.0957, + "step": 3179 + }, + { + "epoch": 5.088, + "grad_norm": 0.0, + "learning_rate": 0.0003714, + "loss": 6.8679, + "step": 3180 + }, + { + "epoch": 5.0896, + "grad_norm": 0.0, + "learning_rate": 0.00037119999999999997, + "loss": 5.7902, + "step": 3181 + }, + { + "epoch": 5.0912, + "grad_norm": 0.0, + "learning_rate": 0.00037099999999999996, + "loss": 7.3836, + "step": 3182 + }, + { + "epoch": 5.0928, + "grad_norm": 0.0, + "learning_rate": 0.00037079999999999996, + "loss": 6.0675, + "step": 3183 + }, + { + "epoch": 5.0944, + "grad_norm": 0.0, + "learning_rate": 0.0003706, + "loss": 5.5288, + "step": 3184 + }, + { + "epoch": 5.096, + "grad_norm": 0.0, + "learning_rate": 0.00037039999999999995, + "loss": 6.7686, + "step": 3185 + }, + { + "epoch": 5.0976, + "grad_norm": 0.0, + "learning_rate": 0.00037019999999999995, + "loss": 7.3085, + "step": 3186 + }, + { + "epoch": 5.0992, + "grad_norm": 0.0, + "learning_rate": 0.00037, + "loss": 6.9464, + "step": 3187 + }, + { + "epoch": 5.1008, + "grad_norm": 0.0, + "learning_rate": 0.00036979999999999994, + "loss": 5.409, + "step": 3188 + }, + { + "epoch": 5.1024, + "grad_norm": 0.0, + "learning_rate": 0.0003696, + "loss": 6.0839, + "step": 3189 + }, + { + "epoch": 5.104, + "grad_norm": 0.0, + "learning_rate": 0.0003694, + "loss": 5.3394, + "step": 3190 + }, + { + "epoch": 5.1056, + "grad_norm": 0.0, + "learning_rate": 0.0003691999999999999, + "loss": 6.4967, + "step": 3191 + }, + { + "epoch": 5.1072, + "grad_norm": 0.0, + "learning_rate": 0.00036899999999999997, + "loss": 5.727, + "step": 3192 + }, + { + "epoch": 5.1088, + "grad_norm": 0.0, + "learning_rate": 0.00036879999999999997, + "loss": 5.8019, + "step": 3193 + }, + { + "epoch": 5.1104, + "grad_norm": 0.0, + "learning_rate": 0.00036859999999999996, + "loss": 8.9153, + "step": 3194 + }, + { + "epoch": 5.112, + "grad_norm": 0.0, + "learning_rate": 0.00036839999999999996, + "loss": 4.3745, + "step": 3195 + }, + { + "epoch": 5.1136, + "grad_norm": 0.0, + "learning_rate": 0.0003682, + "loss": 4.3136, + "step": 3196 + }, + { + "epoch": 5.1152, + "grad_norm": 0.0, + "learning_rate": 0.00036799999999999995, + "loss": 6.0979, + "step": 3197 + }, + { + "epoch": 5.1168, + "grad_norm": 0.0, + "learning_rate": 0.00036779999999999994, + "loss": 5.4874, + "step": 3198 + }, + { + "epoch": 5.1184, + "grad_norm": 0.0, + "learning_rate": 0.0003676, + "loss": 6.3162, + "step": 3199 + }, + { + "epoch": 5.12, + "grad_norm": 0.0, + "learning_rate": 0.00036739999999999993, + "loss": 5.7947, + "step": 3200 + }, + { + "epoch": 5.1216, + "grad_norm": 0.0, + "learning_rate": 0.0003672, + "loss": 4.4182, + "step": 3201 + }, + { + "epoch": 5.1232, + "grad_norm": 0.0, + "learning_rate": 0.000367, + "loss": 6.0254, + "step": 3202 + }, + { + "epoch": 5.1248, + "grad_norm": 0.0, + "learning_rate": 0.0003667999999999999, + "loss": 5.349, + "step": 3203 + }, + { + "epoch": 5.1264, + "grad_norm": 0.0, + "learning_rate": 0.00036659999999999997, + "loss": 4.7453, + "step": 3204 + }, + { + "epoch": 5.128, + "grad_norm": 0.0, + "learning_rate": 0.00036639999999999996, + "loss": 5.2687, + "step": 3205 + }, + { + "epoch": 5.1296, + "grad_norm": 0.0, + "learning_rate": 0.00036619999999999996, + "loss": 3.9506, + "step": 3206 + }, + { + "epoch": 5.1312, + "grad_norm": 0.0, + "learning_rate": 0.00036599999999999995, + "loss": 5.6731, + "step": 3207 + }, + { + "epoch": 5.1328, + "grad_norm": 0.0, + "learning_rate": 0.0003658, + "loss": 7.6111, + "step": 3208 + }, + { + "epoch": 5.1344, + "grad_norm": 0.0, + "learning_rate": 0.00036559999999999994, + "loss": 5.8718, + "step": 3209 + }, + { + "epoch": 5.136, + "grad_norm": 0.0, + "learning_rate": 0.00036539999999999994, + "loss": 6.4942, + "step": 3210 + }, + { + "epoch": 5.1376, + "grad_norm": 0.0, + "learning_rate": 0.0003652, + "loss": 4.4153, + "step": 3211 + }, + { + "epoch": 5.1392, + "grad_norm": 0.0, + "learning_rate": 0.00036499999999999993, + "loss": 6.7634, + "step": 3212 + }, + { + "epoch": 5.1408, + "grad_norm": 0.0, + "learning_rate": 0.0003648, + "loss": 6.3913, + "step": 3213 + }, + { + "epoch": 5.1424, + "grad_norm": 0.0, + "learning_rate": 0.00036459999999999997, + "loss": 7.0869, + "step": 3214 + }, + { + "epoch": 5.144, + "grad_norm": 0.0, + "learning_rate": 0.0003643999999999999, + "loss": 5.1837, + "step": 3215 + }, + { + "epoch": 5.1456, + "grad_norm": 0.0, + "learning_rate": 0.00036419999999999996, + "loss": 4.7044, + "step": 3216 + }, + { + "epoch": 5.1472, + "grad_norm": 0.0, + "learning_rate": 0.00036399999999999996, + "loss": 4.1418, + "step": 3217 + }, + { + "epoch": 5.1488, + "grad_norm": 0.0, + "learning_rate": 0.00036379999999999995, + "loss": 5.5412, + "step": 3218 + }, + { + "epoch": 5.1504, + "grad_norm": 0.0, + "learning_rate": 0.00036359999999999995, + "loss": 5.4164, + "step": 3219 + }, + { + "epoch": 5.152, + "grad_norm": 0.0, + "learning_rate": 0.0003634, + "loss": 7.6755, + "step": 3220 + }, + { + "epoch": 5.1536, + "grad_norm": 0.0, + "learning_rate": 0.00036319999999999994, + "loss": 5.351, + "step": 3221 + }, + { + "epoch": 5.1552, + "grad_norm": 0.0, + "learning_rate": 0.00036299999999999993, + "loss": 5.0267, + "step": 3222 + }, + { + "epoch": 5.1568, + "grad_norm": 0.0, + "learning_rate": 0.0003628, + "loss": 5.2381, + "step": 3223 + }, + { + "epoch": 5.1584, + "grad_norm": 0.0, + "learning_rate": 0.0003625999999999999, + "loss": 5.8054, + "step": 3224 + }, + { + "epoch": 5.16, + "grad_norm": 0.0, + "learning_rate": 0.00036239999999999997, + "loss": 7.8959, + "step": 3225 + }, + { + "epoch": 5.1616, + "grad_norm": 0.0, + "learning_rate": 0.00036219999999999997, + "loss": 9.2238, + "step": 3226 + }, + { + "epoch": 5.1632, + "grad_norm": 0.0, + "learning_rate": 0.000362, + "loss": 7.1367, + "step": 3227 + }, + { + "epoch": 5.1648, + "grad_norm": 0.0, + "learning_rate": 0.00036179999999999996, + "loss": 7.2586, + "step": 3228 + }, + { + "epoch": 5.1664, + "grad_norm": 0.0, + "learning_rate": 0.00036159999999999995, + "loss": 6.4825, + "step": 3229 + }, + { + "epoch": 5.168, + "grad_norm": 0.0, + "learning_rate": 0.0003614, + "loss": 5.8703, + "step": 3230 + }, + { + "epoch": 5.1696, + "grad_norm": 0.0, + "learning_rate": 0.00036119999999999994, + "loss": 7.3977, + "step": 3231 + }, + { + "epoch": 5.1712, + "grad_norm": 0.0, + "learning_rate": 0.000361, + "loss": 6.3445, + "step": 3232 + }, + { + "epoch": 5.1728, + "grad_norm": 0.0, + "learning_rate": 0.0003608, + "loss": 5.2915, + "step": 3233 + }, + { + "epoch": 5.1744, + "grad_norm": 0.0, + "learning_rate": 0.00036059999999999993, + "loss": 5.7432, + "step": 3234 + }, + { + "epoch": 5.176, + "grad_norm": 0.0, + "learning_rate": 0.0003604, + "loss": 8.5188, + "step": 3235 + }, + { + "epoch": 5.1776, + "grad_norm": 0.0, + "learning_rate": 0.0003602, + "loss": 5.4136, + "step": 3236 + }, + { + "epoch": 5.1792, + "grad_norm": 0.0, + "learning_rate": 0.00035999999999999997, + "loss": 7.5626, + "step": 3237 + }, + { + "epoch": 5.1808, + "grad_norm": 0.0, + "learning_rate": 0.00035979999999999996, + "loss": 5.8757, + "step": 3238 + }, + { + "epoch": 5.1824, + "grad_norm": 0.0, + "learning_rate": 0.0003596, + "loss": 5.7759, + "step": 3239 + }, + { + "epoch": 5.184, + "grad_norm": 0.0, + "learning_rate": 0.00035939999999999995, + "loss": 6.4547, + "step": 3240 + }, + { + "epoch": 5.1856, + "grad_norm": 0.0, + "learning_rate": 0.0003592, + "loss": 5.39, + "step": 3241 + }, + { + "epoch": 5.1872, + "grad_norm": 0.0, + "learning_rate": 0.000359, + "loss": 5.5784, + "step": 3242 + }, + { + "epoch": 5.1888, + "grad_norm": 0.0, + "learning_rate": 0.00035879999999999994, + "loss": 6.7785, + "step": 3243 + }, + { + "epoch": 5.1904, + "grad_norm": 0.0, + "learning_rate": 0.0003586, + "loss": 5.3949, + "step": 3244 + }, + { + "epoch": 5.192, + "grad_norm": 0.0, + "learning_rate": 0.0003584, + "loss": 7.2365, + "step": 3245 + }, + { + "epoch": 5.1936, + "grad_norm": 0.0, + "learning_rate": 0.0003582, + "loss": 6.3232, + "step": 3246 + }, + { + "epoch": 5.1952, + "grad_norm": 0.0, + "learning_rate": 0.000358, + "loss": 5.0038, + "step": 3247 + }, + { + "epoch": 5.1968, + "grad_norm": 0.0, + "learning_rate": 0.0003578, + "loss": 5.8545, + "step": 3248 + }, + { + "epoch": 5.1984, + "grad_norm": 0.0, + "learning_rate": 0.00035759999999999996, + "loss": 5.1058, + "step": 3249 + }, + { + "epoch": 5.2, + "grad_norm": 0.0, + "learning_rate": 0.00035739999999999996, + "loss": 4.1752, + "step": 3250 + }, + { + "epoch": 5.2016, + "grad_norm": 0.0, + "learning_rate": 0.0003572, + "loss": 6.5306, + "step": 3251 + }, + { + "epoch": 5.2032, + "grad_norm": 0.0, + "learning_rate": 0.00035699999999999995, + "loss": 6.0615, + "step": 3252 + }, + { + "epoch": 5.2048, + "grad_norm": 0.0, + "learning_rate": 0.0003568, + "loss": 4.6328, + "step": 3253 + }, + { + "epoch": 5.2064, + "grad_norm": 0.0, + "learning_rate": 0.0003566, + "loss": 5.006, + "step": 3254 + }, + { + "epoch": 5.208, + "grad_norm": 0.0, + "learning_rate": 0.00035639999999999994, + "loss": 6.521, + "step": 3255 + }, + { + "epoch": 5.2096, + "grad_norm": 0.0, + "learning_rate": 0.0003562, + "loss": 5.2562, + "step": 3256 + }, + { + "epoch": 5.2112, + "grad_norm": 0.0, + "learning_rate": 0.000356, + "loss": 4.5534, + "step": 3257 + }, + { + "epoch": 5.2128, + "grad_norm": 0.0, + "learning_rate": 0.0003558, + "loss": 6.1229, + "step": 3258 + }, + { + "epoch": 5.2144, + "grad_norm": 0.0, + "learning_rate": 0.00035559999999999997, + "loss": 5.727, + "step": 3259 + }, + { + "epoch": 5.216, + "grad_norm": 0.0, + "learning_rate": 0.0003554, + "loss": 5.9868, + "step": 3260 + }, + { + "epoch": 5.2176, + "grad_norm": 0.0, + "learning_rate": 0.00035519999999999996, + "loss": 5.0845, + "step": 3261 + }, + { + "epoch": 5.2192, + "grad_norm": 0.0, + "learning_rate": 0.00035499999999999996, + "loss": 5.8569, + "step": 3262 + }, + { + "epoch": 5.2208, + "grad_norm": 0.0, + "learning_rate": 0.0003548, + "loss": 4.5071, + "step": 3263 + }, + { + "epoch": 5.2224, + "grad_norm": 0.0, + "learning_rate": 0.00035459999999999995, + "loss": 4.6204, + "step": 3264 + }, + { + "epoch": 5.224, + "grad_norm": 0.0, + "learning_rate": 0.0003544, + "loss": 5.467, + "step": 3265 + }, + { + "epoch": 5.2256, + "grad_norm": 0.0, + "learning_rate": 0.0003542, + "loss": 5.043, + "step": 3266 + }, + { + "epoch": 5.2272, + "grad_norm": 0.0, + "learning_rate": 0.00035399999999999993, + "loss": 4.1488, + "step": 3267 + }, + { + "epoch": 5.2288, + "grad_norm": 0.0, + "learning_rate": 0.0003538, + "loss": 6.4425, + "step": 3268 + }, + { + "epoch": 5.2304, + "grad_norm": 0.0, + "learning_rate": 0.0003536, + "loss": 3.6541, + "step": 3269 + }, + { + "epoch": 5.232, + "grad_norm": 0.0, + "learning_rate": 0.00035339999999999997, + "loss": 4.6929, + "step": 3270 + }, + { + "epoch": 5.2336, + "grad_norm": 0.0, + "learning_rate": 0.00035319999999999997, + "loss": 5.3044, + "step": 3271 + }, + { + "epoch": 5.2352, + "grad_norm": 0.0, + "learning_rate": 0.000353, + "loss": 4.7675, + "step": 3272 + }, + { + "epoch": 5.2368, + "grad_norm": 0.0, + "learning_rate": 0.00035279999999999996, + "loss": 7.0071, + "step": 3273 + }, + { + "epoch": 5.2384, + "grad_norm": 0.0, + "learning_rate": 0.00035259999999999995, + "loss": 7.6142, + "step": 3274 + }, + { + "epoch": 5.24, + "grad_norm": NaN, + "learning_rate": 0.00035259999999999995, + "loss": 4.2145, + "step": 3275 + }, + { + "epoch": 5.2416, + "grad_norm": 0.0, + "learning_rate": 0.0003524, + "loss": 7.5835, + "step": 3276 + }, + { + "epoch": 5.2432, + "grad_norm": 0.0, + "learning_rate": 0.00035219999999999994, + "loss": 6.5222, + "step": 3277 + }, + { + "epoch": 5.2448, + "grad_norm": 0.0, + "learning_rate": 0.000352, + "loss": 7.2513, + "step": 3278 + }, + { + "epoch": 5.2464, + "grad_norm": 0.0, + "learning_rate": 0.0003518, + "loss": 7.4396, + "step": 3279 + }, + { + "epoch": 5.248, + "grad_norm": 0.0, + "learning_rate": 0.0003515999999999999, + "loss": 7.5025, + "step": 3280 + }, + { + "epoch": 5.2496, + "grad_norm": 0.0, + "learning_rate": 0.0003514, + "loss": 7.1106, + "step": 3281 + }, + { + "epoch": 5.2512, + "grad_norm": 0.0, + "learning_rate": 0.00035119999999999997, + "loss": 6.0558, + "step": 3282 + }, + { + "epoch": 5.2528, + "grad_norm": 0.0, + "learning_rate": 0.00035099999999999997, + "loss": 6.2606, + "step": 3283 + }, + { + "epoch": 5.2544, + "grad_norm": 0.0, + "learning_rate": 0.00035079999999999996, + "loss": 6.2598, + "step": 3284 + }, + { + "epoch": 5.256, + "grad_norm": 0.0, + "learning_rate": 0.0003506, + "loss": 5.3476, + "step": 3285 + }, + { + "epoch": 5.2576, + "grad_norm": 0.0, + "learning_rate": 0.00035039999999999995, + "loss": 5.6261, + "step": 3286 + }, + { + "epoch": 5.2592, + "grad_norm": 0.0, + "learning_rate": 0.00035019999999999995, + "loss": 6.2698, + "step": 3287 + }, + { + "epoch": 5.2608, + "grad_norm": 0.0, + "learning_rate": 0.00035, + "loss": 5.8383, + "step": 3288 + }, + { + "epoch": 5.2624, + "grad_norm": 0.0, + "learning_rate": 0.00034979999999999994, + "loss": 5.8442, + "step": 3289 + }, + { + "epoch": 5.264, + "grad_norm": 0.0, + "learning_rate": 0.0003496, + "loss": 5.5433, + "step": 3290 + }, + { + "epoch": 5.2656, + "grad_norm": 0.0, + "learning_rate": 0.0003494, + "loss": 4.4822, + "step": 3291 + }, + { + "epoch": 5.2672, + "grad_norm": 0.0, + "learning_rate": 0.0003491999999999999, + "loss": 4.9684, + "step": 3292 + }, + { + "epoch": 5.2688, + "grad_norm": 0.0, + "learning_rate": 0.00034899999999999997, + "loss": 4.3567, + "step": 3293 + }, + { + "epoch": 5.2704, + "grad_norm": 0.0, + "learning_rate": 0.00034879999999999997, + "loss": 6.8008, + "step": 3294 + }, + { + "epoch": 5.272, + "grad_norm": 0.0, + "learning_rate": 0.00034859999999999996, + "loss": 5.4058, + "step": 3295 + }, + { + "epoch": 5.2736, + "grad_norm": 0.0, + "learning_rate": 0.00034839999999999996, + "loss": 6.1388, + "step": 3296 + }, + { + "epoch": 5.2752, + "grad_norm": 0.0, + "learning_rate": 0.0003482, + "loss": 4.6146, + "step": 3297 + }, + { + "epoch": 5.2768, + "grad_norm": 0.0, + "learning_rate": 0.00034799999999999995, + "loss": 8.8382, + "step": 3298 + }, + { + "epoch": 5.2783999999999995, + "grad_norm": 0.0, + "learning_rate": 0.00034779999999999994, + "loss": 6.4504, + "step": 3299 + }, + { + "epoch": 5.28, + "grad_norm": 0.0, + "learning_rate": 0.0003476, + "loss": 6.3413, + "step": 3300 + }, + { + "epoch": 5.2816, + "grad_norm": 0.0, + "learning_rate": 0.00034739999999999993, + "loss": 5.3272, + "step": 3301 + }, + { + "epoch": 5.2832, + "grad_norm": 0.0, + "learning_rate": 0.0003472, + "loss": 5.3075, + "step": 3302 + }, + { + "epoch": 5.2848, + "grad_norm": 0.0, + "learning_rate": 0.000347, + "loss": 4.318, + "step": 3303 + }, + { + "epoch": 5.2864, + "grad_norm": 0.0, + "learning_rate": 0.0003467999999999999, + "loss": 5.5483, + "step": 3304 + }, + { + "epoch": 5.288, + "grad_norm": 0.0, + "learning_rate": 0.00034659999999999997, + "loss": 5.4227, + "step": 3305 + }, + { + "epoch": 5.2896, + "grad_norm": 0.0, + "learning_rate": 0.00034639999999999996, + "loss": 6.2516, + "step": 3306 + }, + { + "epoch": 5.2912, + "grad_norm": 0.0, + "learning_rate": 0.00034619999999999996, + "loss": 5.937, + "step": 3307 + }, + { + "epoch": 5.2928, + "grad_norm": 0.0, + "learning_rate": 0.00034599999999999995, + "loss": 8.1099, + "step": 3308 + }, + { + "epoch": 5.2943999999999996, + "grad_norm": 0.0, + "learning_rate": 0.0003458, + "loss": 7.4788, + "step": 3309 + }, + { + "epoch": 5.296, + "grad_norm": 0.0, + "learning_rate": 0.00034559999999999994, + "loss": 4.6478, + "step": 3310 + }, + { + "epoch": 5.2976, + "grad_norm": 0.0, + "learning_rate": 0.00034539999999999994, + "loss": 6.6285, + "step": 3311 + }, + { + "epoch": 5.2992, + "grad_norm": 0.0, + "learning_rate": 0.0003452, + "loss": 5.8699, + "step": 3312 + }, + { + "epoch": 5.3008, + "grad_norm": 0.0, + "learning_rate": 0.00034499999999999993, + "loss": 6.5373, + "step": 3313 + }, + { + "epoch": 5.3024000000000004, + "grad_norm": 0.0, + "learning_rate": 0.0003448, + "loss": 6.5422, + "step": 3314 + }, + { + "epoch": 5.304, + "grad_norm": 0.0, + "learning_rate": 0.0003446, + "loss": 4.812, + "step": 3315 + }, + { + "epoch": 5.3056, + "grad_norm": 0.0, + "learning_rate": 0.00034439999999999997, + "loss": 4.6931, + "step": 3316 + }, + { + "epoch": 5.3072, + "grad_norm": 0.0, + "learning_rate": 0.00034419999999999996, + "loss": 5.4772, + "step": 3317 + }, + { + "epoch": 5.3088, + "grad_norm": 0.0, + "learning_rate": 0.000344, + "loss": 5.0108, + "step": 3318 + }, + { + "epoch": 5.3104, + "grad_norm": 0.0, + "learning_rate": 0.00034379999999999995, + "loss": 4.4123, + "step": 3319 + }, + { + "epoch": 5.312, + "grad_norm": 0.0, + "learning_rate": 0.00034359999999999995, + "loss": 5.9074, + "step": 3320 + }, + { + "epoch": 5.3136, + "grad_norm": 0.0, + "learning_rate": 0.0003434, + "loss": 5.5586, + "step": 3321 + }, + { + "epoch": 5.3152, + "grad_norm": 0.0, + "learning_rate": 0.00034319999999999994, + "loss": 5.8787, + "step": 3322 + }, + { + "epoch": 5.3168, + "grad_norm": 0.0, + "learning_rate": 0.000343, + "loss": 7.6869, + "step": 3323 + }, + { + "epoch": 5.3184000000000005, + "grad_norm": 0.0, + "learning_rate": 0.0003428, + "loss": 5.4342, + "step": 3324 + }, + { + "epoch": 5.32, + "grad_norm": NaN, + "learning_rate": 0.0003428, + "loss": 6.3203, + "step": 3325 + }, + { + "epoch": 5.3216, + "grad_norm": 0.0, + "learning_rate": 0.0003425999999999999, + "loss": 10.9897, + "step": 3326 + }, + { + "epoch": 5.3232, + "grad_norm": 0.0, + "learning_rate": 0.0003424, + "loss": 6.4558, + "step": 3327 + }, + { + "epoch": 5.3248, + "grad_norm": 0.0, + "learning_rate": 0.00034219999999999997, + "loss": 10.9902, + "step": 3328 + }, + { + "epoch": 5.3264, + "grad_norm": 0.0, + "learning_rate": 0.00034199999999999996, + "loss": 6.6559, + "step": 3329 + }, + { + "epoch": 5.328, + "grad_norm": 0.0, + "learning_rate": 0.00034179999999999996, + "loss": 5.7364, + "step": 3330 + }, + { + "epoch": 5.3296, + "grad_norm": 0.0, + "learning_rate": 0.0003416, + "loss": 5.6043, + "step": 3331 + }, + { + "epoch": 5.3312, + "grad_norm": 0.0, + "learning_rate": 0.00034139999999999995, + "loss": 6.9407, + "step": 3332 + }, + { + "epoch": 5.3328, + "grad_norm": 0.0, + "learning_rate": 0.00034119999999999995, + "loss": 7.2163, + "step": 3333 + }, + { + "epoch": 5.3344, + "grad_norm": 0.0, + "learning_rate": 0.000341, + "loss": 6.3183, + "step": 3334 + }, + { + "epoch": 5.336, + "grad_norm": 0.0, + "learning_rate": 0.00034079999999999994, + "loss": 6.4988, + "step": 3335 + }, + { + "epoch": 5.3376, + "grad_norm": 0.0, + "learning_rate": 0.0003406, + "loss": 5.2837, + "step": 3336 + }, + { + "epoch": 5.3392, + "grad_norm": 0.0, + "learning_rate": 0.0003404, + "loss": 6.7341, + "step": 3337 + }, + { + "epoch": 5.3408, + "grad_norm": 0.0, + "learning_rate": 0.0003401999999999999, + "loss": 4.9812, + "step": 3338 + }, + { + "epoch": 5.3424, + "grad_norm": 0.0, + "learning_rate": 0.00033999999999999997, + "loss": 5.6653, + "step": 3339 + }, + { + "epoch": 5.344, + "grad_norm": 0.0, + "learning_rate": 0.00033979999999999997, + "loss": 5.2281, + "step": 3340 + }, + { + "epoch": 5.3456, + "grad_norm": 0.0, + "learning_rate": 0.00033959999999999996, + "loss": 6.2072, + "step": 3341 + }, + { + "epoch": 5.3472, + "grad_norm": 0.0, + "learning_rate": 0.00033939999999999996, + "loss": 5.6467, + "step": 3342 + }, + { + "epoch": 5.3488, + "grad_norm": 0.0, + "learning_rate": 0.0003392, + "loss": 5.5247, + "step": 3343 + }, + { + "epoch": 5.3504, + "grad_norm": 0.0, + "learning_rate": 0.00033899999999999995, + "loss": 5.5503, + "step": 3344 + }, + { + "epoch": 5.352, + "grad_norm": 0.0, + "learning_rate": 0.00033879999999999994, + "loss": 4.943, + "step": 3345 + }, + { + "epoch": 5.3536, + "grad_norm": 0.0, + "learning_rate": 0.0003386, + "loss": 5.5551, + "step": 3346 + }, + { + "epoch": 5.3552, + "grad_norm": 0.0, + "learning_rate": 0.00033839999999999993, + "loss": 5.0287, + "step": 3347 + }, + { + "epoch": 5.3568, + "grad_norm": 0.0, + "learning_rate": 0.0003382, + "loss": 5.1007, + "step": 3348 + }, + { + "epoch": 5.3584, + "grad_norm": 0.0, + "learning_rate": 0.000338, + "loss": 4.5669, + "step": 3349 + }, + { + "epoch": 5.36, + "grad_norm": 0.0, + "learning_rate": 0.0003377999999999999, + "loss": 5.3355, + "step": 3350 + }, + { + "epoch": 5.3616, + "grad_norm": 0.0, + "learning_rate": 0.00033759999999999997, + "loss": 4.641, + "step": 3351 + }, + { + "epoch": 5.3632, + "grad_norm": 0.0, + "learning_rate": 0.00033739999999999996, + "loss": 5.8638, + "step": 3352 + }, + { + "epoch": 5.3648, + "grad_norm": 0.0, + "learning_rate": 0.0003372, + "loss": 6.2771, + "step": 3353 + }, + { + "epoch": 5.3664, + "grad_norm": 0.0, + "learning_rate": 0.00033699999999999995, + "loss": 5.096, + "step": 3354 + }, + { + "epoch": 5.368, + "grad_norm": 0.0, + "learning_rate": 0.0003368, + "loss": 4.9151, + "step": 3355 + }, + { + "epoch": 5.3696, + "grad_norm": 0.0, + "learning_rate": 0.0003366, + "loss": 4.7871, + "step": 3356 + }, + { + "epoch": 5.3712, + "grad_norm": 0.0, + "learning_rate": 0.00033639999999999994, + "loss": 5.5332, + "step": 3357 + }, + { + "epoch": 5.3728, + "grad_norm": 0.0, + "learning_rate": 0.0003362, + "loss": 5.0234, + "step": 3358 + }, + { + "epoch": 5.3744, + "grad_norm": 0.0, + "learning_rate": 0.000336, + "loss": 6.6034, + "step": 3359 + }, + { + "epoch": 5.376, + "grad_norm": 0.0, + "learning_rate": 0.0003358, + "loss": 8.1305, + "step": 3360 + }, + { + "epoch": 5.3776, + "grad_norm": 0.0, + "learning_rate": 0.00033559999999999997, + "loss": 5.8665, + "step": 3361 + }, + { + "epoch": 5.3792, + "grad_norm": 0.0, + "learning_rate": 0.0003354, + "loss": 4.6754, + "step": 3362 + }, + { + "epoch": 5.3808, + "grad_norm": 0.0, + "learning_rate": 0.00033519999999999996, + "loss": 4.583, + "step": 3363 + }, + { + "epoch": 5.3824, + "grad_norm": 0.0, + "learning_rate": 0.00033499999999999996, + "loss": 5.3577, + "step": 3364 + }, + { + "epoch": 5.384, + "grad_norm": 0.0, + "learning_rate": 0.0003348, + "loss": 4.7554, + "step": 3365 + }, + { + "epoch": 5.3856, + "grad_norm": 0.0, + "learning_rate": 0.00033459999999999995, + "loss": 4.5869, + "step": 3366 + }, + { + "epoch": 5.3872, + "grad_norm": 0.0, + "learning_rate": 0.0003344, + "loss": 4.0933, + "step": 3367 + }, + { + "epoch": 5.3888, + "grad_norm": 0.0, + "learning_rate": 0.0003342, + "loss": 4.5981, + "step": 3368 + }, + { + "epoch": 5.3904, + "grad_norm": 0.0, + "learning_rate": 0.00033399999999999993, + "loss": 4.7746, + "step": 3369 + }, + { + "epoch": 5.392, + "grad_norm": 0.0, + "learning_rate": 0.0003338, + "loss": 8.1881, + "step": 3370 + }, + { + "epoch": 5.3936, + "grad_norm": 0.0, + "learning_rate": 0.0003336, + "loss": 4.9805, + "step": 3371 + }, + { + "epoch": 5.3952, + "grad_norm": 0.0, + "learning_rate": 0.00033339999999999997, + "loss": 4.8328, + "step": 3372 + }, + { + "epoch": 5.3968, + "grad_norm": 0.0, + "learning_rate": 0.00033319999999999997, + "loss": 4.5362, + "step": 3373 + }, + { + "epoch": 5.3984, + "grad_norm": 0.0, + "learning_rate": 0.000333, + "loss": 6.3657, + "step": 3374 + }, + { + "epoch": 5.4, + "grad_norm": 0.0, + "learning_rate": 0.00033279999999999996, + "loss": 7.0812, + "step": 3375 + }, + { + "epoch": 5.4016, + "grad_norm": 0.0, + "learning_rate": 0.00033259999999999995, + "loss": 8.602, + "step": 3376 + }, + { + "epoch": 5.4032, + "grad_norm": 0.0, + "learning_rate": 0.0003324, + "loss": 5.8295, + "step": 3377 + }, + { + "epoch": 5.4048, + "grad_norm": 0.0, + "learning_rate": 0.00033219999999999994, + "loss": 6.2085, + "step": 3378 + }, + { + "epoch": 5.4064, + "grad_norm": 0.0, + "learning_rate": 0.000332, + "loss": 6.1029, + "step": 3379 + }, + { + "epoch": 5.408, + "grad_norm": 0.0, + "learning_rate": 0.0003318, + "loss": 5.5902, + "step": 3380 + }, + { + "epoch": 5.4096, + "grad_norm": 0.0, + "learning_rate": 0.00033159999999999993, + "loss": 6.9608, + "step": 3381 + }, + { + "epoch": 5.4112, + "grad_norm": 0.0, + "learning_rate": 0.0003314, + "loss": 4.9996, + "step": 3382 + }, + { + "epoch": 5.4128, + "grad_norm": 0.0, + "learning_rate": 0.0003312, + "loss": 6.9153, + "step": 3383 + }, + { + "epoch": 5.4144, + "grad_norm": 0.0, + "learning_rate": 0.00033099999999999997, + "loss": 6.1443, + "step": 3384 + }, + { + "epoch": 5.416, + "grad_norm": 0.0, + "learning_rate": 0.00033079999999999996, + "loss": 5.8148, + "step": 3385 + }, + { + "epoch": 5.4176, + "grad_norm": 0.0, + "learning_rate": 0.0003306, + "loss": 5.6061, + "step": 3386 + }, + { + "epoch": 5.4192, + "grad_norm": 0.0, + "learning_rate": 0.00033039999999999995, + "loss": 6.5339, + "step": 3387 + }, + { + "epoch": 5.4208, + "grad_norm": 0.0, + "learning_rate": 0.00033019999999999995, + "loss": 5.8298, + "step": 3388 + }, + { + "epoch": 5.4224, + "grad_norm": 0.0, + "learning_rate": 0.00033, + "loss": 5.9716, + "step": 3389 + }, + { + "epoch": 5.424, + "grad_norm": 0.0, + "learning_rate": 0.00032979999999999994, + "loss": 5.0324, + "step": 3390 + }, + { + "epoch": 5.4256, + "grad_norm": 0.0, + "learning_rate": 0.0003296, + "loss": 6.0342, + "step": 3391 + }, + { + "epoch": 5.4272, + "grad_norm": 0.0, + "learning_rate": 0.0003294, + "loss": 6.7758, + "step": 3392 + }, + { + "epoch": 5.4288, + "grad_norm": 0.0, + "learning_rate": 0.0003292, + "loss": 5.3112, + "step": 3393 + }, + { + "epoch": 5.4304, + "grad_norm": 0.0, + "learning_rate": 0.000329, + "loss": 4.8849, + "step": 3394 + }, + { + "epoch": 5.432, + "grad_norm": 0.0, + "learning_rate": 0.0003288, + "loss": 5.8174, + "step": 3395 + }, + { + "epoch": 5.4336, + "grad_norm": 0.0, + "learning_rate": 0.00032859999999999996, + "loss": 5.8588, + "step": 3396 + }, + { + "epoch": 5.4352, + "grad_norm": 0.0, + "learning_rate": 0.00032839999999999996, + "loss": 5.3251, + "step": 3397 + }, + { + "epoch": 5.4368, + "grad_norm": 0.0, + "learning_rate": 0.0003282, + "loss": 4.6848, + "step": 3398 + }, + { + "epoch": 5.4384, + "grad_norm": 0.0, + "learning_rate": 0.00032799999999999995, + "loss": 5.1109, + "step": 3399 + }, + { + "epoch": 5.44, + "grad_norm": 0.0, + "learning_rate": 0.0003278, + "loss": 9.1261, + "step": 3400 + }, + { + "epoch": 5.4416, + "grad_norm": 0.0, + "learning_rate": 0.0003276, + "loss": 4.3006, + "step": 3401 + }, + { + "epoch": 5.4432, + "grad_norm": 0.0, + "learning_rate": 0.00032739999999999993, + "loss": 4.6643, + "step": 3402 + }, + { + "epoch": 5.4448, + "grad_norm": 0.0, + "learning_rate": 0.0003272, + "loss": 4.8954, + "step": 3403 + }, + { + "epoch": 5.4464, + "grad_norm": 0.0, + "learning_rate": 0.000327, + "loss": 5.122, + "step": 3404 + }, + { + "epoch": 5.448, + "grad_norm": 0.0, + "learning_rate": 0.0003268, + "loss": 5.6855, + "step": 3405 + }, + { + "epoch": 5.4496, + "grad_norm": 0.0, + "learning_rate": 0.00032659999999999997, + "loss": 5.8883, + "step": 3406 + }, + { + "epoch": 5.4512, + "grad_norm": 0.0, + "learning_rate": 0.0003264, + "loss": 5.5127, + "step": 3407 + }, + { + "epoch": 5.4528, + "grad_norm": 0.0, + "learning_rate": 0.00032619999999999996, + "loss": 6.1715, + "step": 3408 + }, + { + "epoch": 5.4544, + "grad_norm": 0.0, + "learning_rate": 0.00032599999999999996, + "loss": 5.2576, + "step": 3409 + }, + { + "epoch": 5.456, + "grad_norm": 0.0, + "learning_rate": 0.0003258, + "loss": 5.6109, + "step": 3410 + }, + { + "epoch": 5.4576, + "grad_norm": 0.0, + "learning_rate": 0.00032559999999999995, + "loss": 4.5647, + "step": 3411 + }, + { + "epoch": 5.4592, + "grad_norm": 0.0, + "learning_rate": 0.0003254, + "loss": 4.5388, + "step": 3412 + }, + { + "epoch": 5.4608, + "grad_norm": 0.0, + "learning_rate": 0.0003252, + "loss": 4.3299, + "step": 3413 + }, + { + "epoch": 5.4624, + "grad_norm": 0.0, + "learning_rate": 0.00032499999999999993, + "loss": 5.0299, + "step": 3414 + }, + { + "epoch": 5.464, + "grad_norm": 0.0, + "learning_rate": 0.0003248, + "loss": 3.8678, + "step": 3415 + }, + { + "epoch": 5.4656, + "grad_norm": 0.0, + "learning_rate": 0.0003246, + "loss": 5.7612, + "step": 3416 + }, + { + "epoch": 5.4672, + "grad_norm": 0.0, + "learning_rate": 0.00032439999999999997, + "loss": 7.5586, + "step": 3417 + }, + { + "epoch": 5.4688, + "grad_norm": 0.0, + "learning_rate": 0.00032419999999999997, + "loss": 4.0426, + "step": 3418 + }, + { + "epoch": 5.4704, + "grad_norm": 0.0, + "learning_rate": 0.000324, + "loss": 5.2368, + "step": 3419 + }, + { + "epoch": 5.4719999999999995, + "grad_norm": 0.0, + "learning_rate": 0.00032379999999999996, + "loss": 6.5466, + "step": 3420 + }, + { + "epoch": 5.4736, + "grad_norm": 0.0, + "learning_rate": 0.00032359999999999995, + "loss": 6.2514, + "step": 3421 + }, + { + "epoch": 5.4752, + "grad_norm": 0.0, + "learning_rate": 0.0003234, + "loss": 5.6582, + "step": 3422 + }, + { + "epoch": 5.4768, + "grad_norm": 0.0, + "learning_rate": 0.00032319999999999994, + "loss": 5.2433, + "step": 3423 + }, + { + "epoch": 5.4784, + "grad_norm": 0.0, + "learning_rate": 0.000323, + "loss": 3.9395, + "step": 3424 + }, + { + "epoch": 5.48, + "grad_norm": NaN, + "learning_rate": 0.000323, + "loss": 5.3993, + "step": 3425 + }, + { + "epoch": 5.4816, + "grad_norm": 0.0, + "learning_rate": 0.0003228, + "loss": 8.6974, + "step": 3426 + }, + { + "epoch": 5.4832, + "grad_norm": 0.0, + "learning_rate": 0.0003225999999999999, + "loss": 8.4708, + "step": 3427 + }, + { + "epoch": 5.4848, + "grad_norm": 0.0, + "learning_rate": 0.0003224, + "loss": 6.4114, + "step": 3428 + }, + { + "epoch": 5.4864, + "grad_norm": 0.0, + "learning_rate": 0.00032219999999999997, + "loss": 8.877, + "step": 3429 + }, + { + "epoch": 5.4879999999999995, + "grad_norm": 0.0, + "learning_rate": 0.00032199999999999997, + "loss": 6.088, + "step": 3430 + }, + { + "epoch": 5.4896, + "grad_norm": 0.0, + "learning_rate": 0.00032179999999999996, + "loss": 10.0941, + "step": 3431 + }, + { + "epoch": 5.4912, + "grad_norm": 0.0, + "learning_rate": 0.0003216, + "loss": 5.8752, + "step": 3432 + }, + { + "epoch": 5.4928, + "grad_norm": 0.0, + "learning_rate": 0.00032139999999999995, + "loss": 5.7305, + "step": 3433 + }, + { + "epoch": 5.4944, + "grad_norm": 0.0, + "learning_rate": 0.00032119999999999995, + "loss": 8.352, + "step": 3434 + }, + { + "epoch": 5.496, + "grad_norm": 0.0, + "learning_rate": 0.000321, + "loss": 5.8897, + "step": 3435 + }, + { + "epoch": 5.4976, + "grad_norm": 0.0, + "learning_rate": 0.00032079999999999994, + "loss": 5.2924, + "step": 3436 + }, + { + "epoch": 5.4992, + "grad_norm": 0.0, + "learning_rate": 0.0003206, + "loss": 5.1051, + "step": 3437 + }, + { + "epoch": 5.5008, + "grad_norm": 0.0, + "learning_rate": 0.0003204, + "loss": 5.6145, + "step": 3438 + }, + { + "epoch": 5.5024, + "grad_norm": 0.0, + "learning_rate": 0.0003201999999999999, + "loss": 5.5766, + "step": 3439 + }, + { + "epoch": 5.504, + "grad_norm": 0.0, + "learning_rate": 0.00031999999999999997, + "loss": 5.6915, + "step": 3440 + }, + { + "epoch": 5.5056, + "grad_norm": 0.0, + "learning_rate": 0.00031979999999999997, + "loss": 6.1116, + "step": 3441 + }, + { + "epoch": 5.5072, + "grad_norm": 0.0, + "learning_rate": 0.00031959999999999996, + "loss": 5.745, + "step": 3442 + }, + { + "epoch": 5.5088, + "grad_norm": 0.0, + "learning_rate": 0.00031939999999999996, + "loss": 6.2138, + "step": 3443 + }, + { + "epoch": 5.5104, + "grad_norm": 0.0, + "learning_rate": 0.0003192, + "loss": 5.5046, + "step": 3444 + }, + { + "epoch": 5.5120000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00031899999999999995, + "loss": 5.8324, + "step": 3445 + }, + { + "epoch": 5.5136, + "grad_norm": 0.0, + "learning_rate": 0.00031879999999999994, + "loss": 5.9572, + "step": 3446 + }, + { + "epoch": 5.5152, + "grad_norm": 0.0, + "learning_rate": 0.0003186, + "loss": 5.9537, + "step": 3447 + }, + { + "epoch": 5.5168, + "grad_norm": 0.0, + "learning_rate": 0.00031839999999999993, + "loss": 4.9738, + "step": 3448 + }, + { + "epoch": 5.5184, + "grad_norm": 0.0, + "learning_rate": 0.0003182, + "loss": 5.9412, + "step": 3449 + }, + { + "epoch": 5.52, + "grad_norm": 0.0, + "learning_rate": 0.000318, + "loss": 6.1548, + "step": 3450 + }, + { + "epoch": 5.5216, + "grad_norm": 0.0, + "learning_rate": 0.0003177999999999999, + "loss": 5.6497, + "step": 3451 + }, + { + "epoch": 5.5232, + "grad_norm": 0.0, + "learning_rate": 0.00031759999999999997, + "loss": 5.1785, + "step": 3452 + }, + { + "epoch": 5.5248, + "grad_norm": 0.0, + "learning_rate": 0.00031739999999999996, + "loss": 5.704, + "step": 3453 + }, + { + "epoch": 5.5264, + "grad_norm": 0.0, + "learning_rate": 0.00031719999999999996, + "loss": 4.6606, + "step": 3454 + }, + { + "epoch": 5.5280000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00031699999999999995, + "loss": 5.8791, + "step": 3455 + }, + { + "epoch": 5.5296, + "grad_norm": 0.0, + "learning_rate": 0.0003168, + "loss": 4.7122, + "step": 3456 + }, + { + "epoch": 5.5312, + "grad_norm": 0.0, + "learning_rate": 0.00031659999999999994, + "loss": 4.6218, + "step": 3457 + }, + { + "epoch": 5.5328, + "grad_norm": 0.0, + "learning_rate": 0.00031639999999999994, + "loss": 3.9034, + "step": 3458 + }, + { + "epoch": 5.5344, + "grad_norm": 0.0, + "learning_rate": 0.0003162, + "loss": 5.5434, + "step": 3459 + }, + { + "epoch": 5.536, + "grad_norm": 0.0, + "learning_rate": 0.00031599999999999993, + "loss": 5.7622, + "step": 3460 + }, + { + "epoch": 5.5376, + "grad_norm": 0.0, + "learning_rate": 0.0003158, + "loss": 5.0766, + "step": 3461 + }, + { + "epoch": 5.5392, + "grad_norm": 0.0, + "learning_rate": 0.0003156, + "loss": 6.8144, + "step": 3462 + }, + { + "epoch": 5.5408, + "grad_norm": 0.0, + "learning_rate": 0.00031539999999999997, + "loss": 5.1277, + "step": 3463 + }, + { + "epoch": 5.5424, + "grad_norm": 0.0, + "learning_rate": 0.00031519999999999996, + "loss": 6.1543, + "step": 3464 + }, + { + "epoch": 5.5440000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00031499999999999996, + "loss": 6.665, + "step": 3465 + }, + { + "epoch": 5.5456, + "grad_norm": 0.0, + "learning_rate": 0.00031479999999999995, + "loss": 4.5117, + "step": 3466 + }, + { + "epoch": 5.5472, + "grad_norm": 0.0, + "learning_rate": 0.00031459999999999995, + "loss": 4.858, + "step": 3467 + }, + { + "epoch": 5.5488, + "grad_norm": 0.0, + "learning_rate": 0.0003144, + "loss": 5.9249, + "step": 3468 + }, + { + "epoch": 5.5504, + "grad_norm": 0.0, + "learning_rate": 0.00031419999999999994, + "loss": 5.5568, + "step": 3469 + }, + { + "epoch": 5.552, + "grad_norm": 0.0, + "learning_rate": 0.000314, + "loss": 6.1535, + "step": 3470 + }, + { + "epoch": 5.5536, + "grad_norm": 0.0, + "learning_rate": 0.0003138, + "loss": 4.8313, + "step": 3471 + }, + { + "epoch": 5.5552, + "grad_norm": 0.0, + "learning_rate": 0.0003135999999999999, + "loss": 4.4061, + "step": 3472 + }, + { + "epoch": 5.5568, + "grad_norm": 0.0, + "learning_rate": 0.0003134, + "loss": 6.5979, + "step": 3473 + }, + { + "epoch": 5.5584, + "grad_norm": 0.0, + "learning_rate": 0.00031319999999999997, + "loss": 5.5534, + "step": 3474 + }, + { + "epoch": 5.5600000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00031299999999999996, + "loss": 6.1232, + "step": 3475 + }, + { + "epoch": 5.5616, + "grad_norm": 0.0, + "learning_rate": 0.00031279999999999996, + "loss": 8.8069, + "step": 3476 + }, + { + "epoch": 5.5632, + "grad_norm": 0.0, + "learning_rate": 0.0003126, + "loss": 10.0194, + "step": 3477 + }, + { + "epoch": 5.5648, + "grad_norm": 0.0, + "learning_rate": 0.0003124, + "loss": 7.7735, + "step": 3478 + }, + { + "epoch": 5.5664, + "grad_norm": 0.0, + "learning_rate": 0.00031219999999999995, + "loss": 7.6879, + "step": 3479 + }, + { + "epoch": 5.568, + "grad_norm": 0.0, + "learning_rate": 0.000312, + "loss": 6.2111, + "step": 3480 + }, + { + "epoch": 5.5696, + "grad_norm": 0.0, + "learning_rate": 0.0003118, + "loss": 5.668, + "step": 3481 + }, + { + "epoch": 5.5712, + "grad_norm": 0.0, + "learning_rate": 0.0003116, + "loss": 5.803, + "step": 3482 + }, + { + "epoch": 5.5728, + "grad_norm": 0.0, + "learning_rate": 0.0003114, + "loss": 5.6059, + "step": 3483 + }, + { + "epoch": 5.5744, + "grad_norm": 0.0, + "learning_rate": 0.00031120000000000003, + "loss": 6.5504, + "step": 3484 + }, + { + "epoch": 5.576, + "grad_norm": 0.0, + "learning_rate": 0.00031099999999999997, + "loss": 4.8661, + "step": 3485 + }, + { + "epoch": 5.5776, + "grad_norm": 0.0, + "learning_rate": 0.00031079999999999997, + "loss": 5.4792, + "step": 3486 + }, + { + "epoch": 5.5792, + "grad_norm": 0.0, + "learning_rate": 0.0003106, + "loss": 7.2463, + "step": 3487 + }, + { + "epoch": 5.5808, + "grad_norm": 0.0, + "learning_rate": 0.00031039999999999996, + "loss": 6.0487, + "step": 3488 + }, + { + "epoch": 5.5824, + "grad_norm": 0.0, + "learning_rate": 0.0003102, + "loss": 5.6956, + "step": 3489 + }, + { + "epoch": 5.584, + "grad_norm": 0.0, + "learning_rate": 0.00031, + "loss": 5.734, + "step": 3490 + }, + { + "epoch": 5.5856, + "grad_norm": 0.0, + "learning_rate": 0.00030979999999999994, + "loss": 5.7294, + "step": 3491 + }, + { + "epoch": 5.5872, + "grad_norm": 0.0, + "learning_rate": 0.0003096, + "loss": 4.8868, + "step": 3492 + }, + { + "epoch": 5.5888, + "grad_norm": 0.0, + "learning_rate": 0.0003094, + "loss": 8.0877, + "step": 3493 + }, + { + "epoch": 5.5904, + "grad_norm": 0.0, + "learning_rate": 0.0003092, + "loss": 5.6308, + "step": 3494 + }, + { + "epoch": 5.592, + "grad_norm": 0.0, + "learning_rate": 0.000309, + "loss": 5.894, + "step": 3495 + }, + { + "epoch": 5.5936, + "grad_norm": 0.0, + "learning_rate": 0.0003088, + "loss": 4.8339, + "step": 3496 + }, + { + "epoch": 5.5952, + "grad_norm": 0.0, + "learning_rate": 0.00030859999999999997, + "loss": 6.5512, + "step": 3497 + }, + { + "epoch": 5.5968, + "grad_norm": 0.0, + "learning_rate": 0.00030839999999999996, + "loss": 6.1774, + "step": 3498 + }, + { + "epoch": 5.5984, + "grad_norm": 0.0, + "learning_rate": 0.0003082, + "loss": 5.466, + "step": 3499 + }, + { + "epoch": 5.6, + "grad_norm": 0.0, + "learning_rate": 0.00030799999999999995, + "loss": 5.1911, + "step": 3500 + }, + { + "epoch": 5.6016, + "grad_norm": 0.0, + "learning_rate": 0.0003078, + "loss": 5.6427, + "step": 3501 + }, + { + "epoch": 5.6032, + "grad_norm": 0.0, + "learning_rate": 0.0003076, + "loss": 5.775, + "step": 3502 + }, + { + "epoch": 5.6048, + "grad_norm": 0.0, + "learning_rate": 0.00030739999999999994, + "loss": 5.11, + "step": 3503 + }, + { + "epoch": 5.6064, + "grad_norm": 0.0, + "learning_rate": 0.0003072, + "loss": 4.8973, + "step": 3504 + }, + { + "epoch": 5.608, + "grad_norm": 0.0, + "learning_rate": 0.000307, + "loss": 4.6442, + "step": 3505 + }, + { + "epoch": 5.6096, + "grad_norm": 0.0, + "learning_rate": 0.0003068, + "loss": 4.3129, + "step": 3506 + }, + { + "epoch": 5.6112, + "grad_norm": 0.0, + "learning_rate": 0.00030659999999999997, + "loss": 6.2997, + "step": 3507 + }, + { + "epoch": 5.6128, + "grad_norm": 0.0, + "learning_rate": 0.0003064, + "loss": 5.7445, + "step": 3508 + }, + { + "epoch": 5.6144, + "grad_norm": 0.0, + "learning_rate": 0.00030619999999999996, + "loss": 4.3291, + "step": 3509 + }, + { + "epoch": 5.616, + "grad_norm": 0.0, + "learning_rate": 0.00030599999999999996, + "loss": 4.8229, + "step": 3510 + }, + { + "epoch": 5.6176, + "grad_norm": 0.0, + "learning_rate": 0.0003058, + "loss": 5.8915, + "step": 3511 + }, + { + "epoch": 5.6192, + "grad_norm": 0.0, + "learning_rate": 0.00030559999999999995, + "loss": 5.8624, + "step": 3512 + }, + { + "epoch": 5.6208, + "grad_norm": 0.0, + "learning_rate": 0.0003054, + "loss": 5.0837, + "step": 3513 + }, + { + "epoch": 5.6224, + "grad_norm": 0.0, + "learning_rate": 0.0003052, + "loss": 5.597, + "step": 3514 + }, + { + "epoch": 5.624, + "grad_norm": 0.0, + "learning_rate": 0.00030499999999999993, + "loss": 5.4179, + "step": 3515 + }, + { + "epoch": 5.6256, + "grad_norm": 0.0, + "learning_rate": 0.0003048, + "loss": 4.1584, + "step": 3516 + }, + { + "epoch": 5.6272, + "grad_norm": 0.0, + "learning_rate": 0.0003046, + "loss": 4.7938, + "step": 3517 + }, + { + "epoch": 5.6288, + "grad_norm": 0.0, + "learning_rate": 0.00030439999999999997, + "loss": 5.203, + "step": 3518 + }, + { + "epoch": 5.6304, + "grad_norm": 0.0, + "learning_rate": 0.00030419999999999997, + "loss": 4.6087, + "step": 3519 + }, + { + "epoch": 5.632, + "grad_norm": 0.0, + "learning_rate": 0.000304, + "loss": 5.5019, + "step": 3520 + }, + { + "epoch": 5.6336, + "grad_norm": 0.0, + "learning_rate": 0.00030379999999999996, + "loss": 6.6593, + "step": 3521 + }, + { + "epoch": 5.6352, + "grad_norm": 0.0, + "learning_rate": 0.00030359999999999995, + "loss": 6.3611, + "step": 3522 + }, + { + "epoch": 5.6368, + "grad_norm": 0.0, + "learning_rate": 0.0003034, + "loss": 6.8224, + "step": 3523 + }, + { + "epoch": 5.6384, + "grad_norm": 0.0, + "learning_rate": 0.00030319999999999994, + "loss": 4.3047, + "step": 3524 + }, + { + "epoch": 5.64, + "grad_norm": 0.0, + "learning_rate": 0.000303, + "loss": 6.519, + "step": 3525 + }, + { + "epoch": 5.6416, + "grad_norm": 0.0, + "learning_rate": 0.0003028, + "loss": 6.754, + "step": 3526 + }, + { + "epoch": 5.6432, + "grad_norm": 0.0, + "learning_rate": 0.00030259999999999993, + "loss": 6.9593, + "step": 3527 + }, + { + "epoch": 5.6448, + "grad_norm": 0.0, + "learning_rate": 0.0003024, + "loss": 6.2176, + "step": 3528 + }, + { + "epoch": 5.6464, + "grad_norm": 0.0, + "learning_rate": 0.0003022, + "loss": 6.0432, + "step": 3529 + }, + { + "epoch": 5.648, + "grad_norm": 0.0, + "learning_rate": 0.00030199999999999997, + "loss": 6.6611, + "step": 3530 + }, + { + "epoch": 5.6495999999999995, + "grad_norm": 0.0, + "learning_rate": 0.00030179999999999996, + "loss": 6.1555, + "step": 3531 + }, + { + "epoch": 5.6512, + "grad_norm": 0.0, + "learning_rate": 0.0003016, + "loss": 6.9663, + "step": 3532 + }, + { + "epoch": 5.6528, + "grad_norm": 0.0, + "learning_rate": 0.00030139999999999995, + "loss": 6.7112, + "step": 3533 + }, + { + "epoch": 5.6544, + "grad_norm": 0.0, + "learning_rate": 0.00030119999999999995, + "loss": 6.6022, + "step": 3534 + }, + { + "epoch": 5.656, + "grad_norm": 0.0, + "learning_rate": 0.000301, + "loss": 7.3001, + "step": 3535 + }, + { + "epoch": 5.6576, + "grad_norm": 0.0, + "learning_rate": 0.00030079999999999994, + "loss": 6.3927, + "step": 3536 + }, + { + "epoch": 5.6592, + "grad_norm": 0.0, + "learning_rate": 0.0003006, + "loss": 4.9856, + "step": 3537 + }, + { + "epoch": 5.6608, + "grad_norm": 0.0, + "learning_rate": 0.0003004, + "loss": 7.4152, + "step": 3538 + }, + { + "epoch": 5.6624, + "grad_norm": 0.0, + "learning_rate": 0.0003002, + "loss": 5.0768, + "step": 3539 + }, + { + "epoch": 5.664, + "grad_norm": 0.0, + "learning_rate": 0.0003, + "loss": 4.3194, + "step": 3540 + }, + { + "epoch": 5.6655999999999995, + "grad_norm": 0.0, + "learning_rate": 0.00029979999999999997, + "loss": 4.9624, + "step": 3541 + }, + { + "epoch": 5.6672, + "grad_norm": 0.0, + "learning_rate": 0.00029959999999999996, + "loss": 5.5032, + "step": 3542 + }, + { + "epoch": 5.6688, + "grad_norm": 0.0, + "learning_rate": 0.00029939999999999996, + "loss": 5.5542, + "step": 3543 + }, + { + "epoch": 5.6704, + "grad_norm": 0.0, + "learning_rate": 0.00029919999999999995, + "loss": 5.5015, + "step": 3544 + }, + { + "epoch": 5.672, + "grad_norm": 0.0, + "learning_rate": 0.000299, + "loss": 5.3259, + "step": 3545 + }, + { + "epoch": 5.6736, + "grad_norm": 0.0, + "learning_rate": 0.0002988, + "loss": 6.172, + "step": 3546 + }, + { + "epoch": 5.6752, + "grad_norm": 0.0, + "learning_rate": 0.00029859999999999994, + "loss": 5.1184, + "step": 3547 + }, + { + "epoch": 5.6768, + "grad_norm": 0.0, + "learning_rate": 0.0002984, + "loss": 5.7026, + "step": 3548 + }, + { + "epoch": 5.6784, + "grad_norm": 0.0, + "learning_rate": 0.0002982, + "loss": 5.5878, + "step": 3549 + }, + { + "epoch": 5.68, + "grad_norm": 0.0, + "learning_rate": 0.000298, + "loss": 5.3587, + "step": 3550 + }, + { + "epoch": 5.6815999999999995, + "grad_norm": 0.0, + "learning_rate": 0.0002978, + "loss": 5.5599, + "step": 3551 + }, + { + "epoch": 5.6832, + "grad_norm": 0.0, + "learning_rate": 0.00029759999999999997, + "loss": 5.531, + "step": 3552 + }, + { + "epoch": 5.6848, + "grad_norm": 0.0, + "learning_rate": 0.00029739999999999996, + "loss": 7.3248, + "step": 3553 + }, + { + "epoch": 5.6864, + "grad_norm": 0.0, + "learning_rate": 0.00029719999999999996, + "loss": 5.0091, + "step": 3554 + }, + { + "epoch": 5.688, + "grad_norm": 0.0, + "learning_rate": 0.00029699999999999996, + "loss": 5.243, + "step": 3555 + }, + { + "epoch": 5.6896, + "grad_norm": 0.0, + "learning_rate": 0.00029679999999999995, + "loss": 6.226, + "step": 3556 + }, + { + "epoch": 5.6912, + "grad_norm": 0.0, + "learning_rate": 0.0002966, + "loss": 4.939, + "step": 3557 + }, + { + "epoch": 5.6928, + "grad_norm": 0.0, + "learning_rate": 0.0002964, + "loss": 4.7771, + "step": 3558 + }, + { + "epoch": 5.6944, + "grad_norm": 0.0, + "learning_rate": 0.00029619999999999994, + "loss": 7.1094, + "step": 3559 + }, + { + "epoch": 5.696, + "grad_norm": 0.0, + "learning_rate": 0.000296, + "loss": 4.1986, + "step": 3560 + }, + { + "epoch": 5.6975999999999996, + "grad_norm": 0.0, + "learning_rate": 0.0002958, + "loss": 4.4455, + "step": 3561 + }, + { + "epoch": 5.6992, + "grad_norm": 0.0, + "learning_rate": 0.0002956, + "loss": 5.7532, + "step": 3562 + }, + { + "epoch": 5.7008, + "grad_norm": 0.0, + "learning_rate": 0.00029539999999999997, + "loss": 6.2304, + "step": 3563 + }, + { + "epoch": 5.7024, + "grad_norm": 0.0, + "learning_rate": 0.00029519999999999997, + "loss": 4.4639, + "step": 3564 + }, + { + "epoch": 5.704, + "grad_norm": 0.0, + "learning_rate": 0.00029499999999999996, + "loss": 5.3329, + "step": 3565 + }, + { + "epoch": 5.7056000000000004, + "grad_norm": 0.0, + "learning_rate": 0.00029479999999999996, + "loss": 5.4996, + "step": 3566 + }, + { + "epoch": 5.7072, + "grad_norm": 0.0, + "learning_rate": 0.00029459999999999995, + "loss": 6.3588, + "step": 3567 + }, + { + "epoch": 5.7088, + "grad_norm": 0.0, + "learning_rate": 0.00029439999999999995, + "loss": 5.2107, + "step": 3568 + }, + { + "epoch": 5.7104, + "grad_norm": 0.0, + "learning_rate": 0.0002942, + "loss": 5.7148, + "step": 3569 + }, + { + "epoch": 5.712, + "grad_norm": 0.0, + "learning_rate": 0.000294, + "loss": 4.3807, + "step": 3570 + }, + { + "epoch": 5.7136, + "grad_norm": 0.0, + "learning_rate": 0.00029379999999999993, + "loss": 5.8635, + "step": 3571 + }, + { + "epoch": 5.7152, + "grad_norm": 0.0, + "learning_rate": 0.0002936, + "loss": 5.4345, + "step": 3572 + }, + { + "epoch": 5.7168, + "grad_norm": 0.0, + "learning_rate": 0.0002934, + "loss": 5.1674, + "step": 3573 + }, + { + "epoch": 5.7184, + "grad_norm": 0.0, + "learning_rate": 0.00029319999999999997, + "loss": 5.9964, + "step": 3574 + }, + { + "epoch": 5.72, + "grad_norm": 0.0, + "learning_rate": 0.00029299999999999997, + "loss": 8.3877, + "step": 3575 + }, + { + "epoch": 5.7216000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00029279999999999996, + "loss": 6.5507, + "step": 3576 + }, + { + "epoch": 5.7232, + "grad_norm": 0.0, + "learning_rate": 0.0002926, + "loss": 6.6776, + "step": 3577 + }, + { + "epoch": 5.7248, + "grad_norm": 0.0, + "learning_rate": 0.0002924, + "loss": 6.7916, + "step": 3578 + }, + { + "epoch": 5.7264, + "grad_norm": 0.0, + "learning_rate": 0.00029219999999999995, + "loss": 6.5796, + "step": 3579 + }, + { + "epoch": 5.728, + "grad_norm": 0.0, + "learning_rate": 0.000292, + "loss": 6.1082, + "step": 3580 + }, + { + "epoch": 5.7296, + "grad_norm": 0.0, + "learning_rate": 0.0002918, + "loss": 5.9443, + "step": 3581 + }, + { + "epoch": 5.7312, + "grad_norm": 0.0, + "learning_rate": 0.0002916, + "loss": 6.2337, + "step": 3582 + }, + { + "epoch": 5.7328, + "grad_norm": 0.0, + "learning_rate": 0.0002914, + "loss": 6.095, + "step": 3583 + }, + { + "epoch": 5.7344, + "grad_norm": 0.0, + "learning_rate": 0.0002912, + "loss": 7.4777, + "step": 3584 + }, + { + "epoch": 5.736, + "grad_norm": 0.0, + "learning_rate": 0.00029099999999999997, + "loss": 6.7638, + "step": 3585 + }, + { + "epoch": 5.7376000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00029079999999999997, + "loss": 5.733, + "step": 3586 + }, + { + "epoch": 5.7392, + "grad_norm": 0.0, + "learning_rate": 0.00029059999999999996, + "loss": 4.4245, + "step": 3587 + }, + { + "epoch": 5.7408, + "grad_norm": 0.0, + "learning_rate": 0.00029039999999999996, + "loss": 5.5837, + "step": 3588 + }, + { + "epoch": 5.7424, + "grad_norm": 0.0, + "learning_rate": 0.0002902, + "loss": 6.2073, + "step": 3589 + }, + { + "epoch": 5.744, + "grad_norm": 0.0, + "learning_rate": 0.00029, + "loss": 5.8264, + "step": 3590 + }, + { + "epoch": 5.7456, + "grad_norm": 0.0, + "learning_rate": 0.00028979999999999994, + "loss": 4.6939, + "step": 3591 + }, + { + "epoch": 5.7472, + "grad_norm": 0.0, + "learning_rate": 0.0002896, + "loss": 5.3144, + "step": 3592 + }, + { + "epoch": 5.7488, + "grad_norm": 0.0, + "learning_rate": 0.0002894, + "loss": 6.251, + "step": 3593 + }, + { + "epoch": 5.7504, + "grad_norm": 0.0, + "learning_rate": 0.0002892, + "loss": 6.5238, + "step": 3594 + }, + { + "epoch": 5.752, + "grad_norm": 0.0, + "learning_rate": 0.000289, + "loss": 6.2432, + "step": 3595 + }, + { + "epoch": 5.7536000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00028879999999999997, + "loss": 4.7342, + "step": 3596 + }, + { + "epoch": 5.7552, + "grad_norm": 0.0, + "learning_rate": 0.00028859999999999997, + "loss": 6.8051, + "step": 3597 + }, + { + "epoch": 5.7568, + "grad_norm": 0.0, + "learning_rate": 0.00028839999999999996, + "loss": 6.8365, + "step": 3598 + }, + { + "epoch": 5.7584, + "grad_norm": 0.0, + "learning_rate": 0.00028819999999999996, + "loss": 4.9514, + "step": 3599 + }, + { + "epoch": 5.76, + "grad_norm": 0.0, + "learning_rate": 0.00028799999999999995, + "loss": 5.5533, + "step": 3600 + }, + { + "epoch": 5.7616, + "grad_norm": 0.0, + "learning_rate": 0.0002878, + "loss": 5.4546, + "step": 3601 + }, + { + "epoch": 5.7632, + "grad_norm": 0.0, + "learning_rate": 0.0002876, + "loss": 4.9962, + "step": 3602 + }, + { + "epoch": 5.7648, + "grad_norm": 0.0, + "learning_rate": 0.00028739999999999994, + "loss": 6.338, + "step": 3603 + }, + { + "epoch": 5.7664, + "grad_norm": 0.0, + "learning_rate": 0.0002872, + "loss": 7.0083, + "step": 3604 + }, + { + "epoch": 5.768, + "grad_norm": 0.0, + "learning_rate": 0.000287, + "loss": 4.8454, + "step": 3605 + }, + { + "epoch": 5.7696, + "grad_norm": 0.0, + "learning_rate": 0.0002868, + "loss": 5.3004, + "step": 3606 + }, + { + "epoch": 5.7712, + "grad_norm": 0.0, + "learning_rate": 0.0002866, + "loss": 4.8438, + "step": 3607 + }, + { + "epoch": 5.7728, + "grad_norm": 0.0, + "learning_rate": 0.00028639999999999997, + "loss": 5.2329, + "step": 3608 + }, + { + "epoch": 5.7744, + "grad_norm": 0.0, + "learning_rate": 0.00028619999999999996, + "loss": 4.1446, + "step": 3609 + }, + { + "epoch": 5.776, + "grad_norm": 0.0, + "learning_rate": 0.00028599999999999996, + "loss": 5.5349, + "step": 3610 + }, + { + "epoch": 5.7776, + "grad_norm": 0.0, + "learning_rate": 0.00028579999999999995, + "loss": 5.4936, + "step": 3611 + }, + { + "epoch": 5.7792, + "grad_norm": 0.0, + "learning_rate": 0.00028559999999999995, + "loss": 4.7326, + "step": 3612 + }, + { + "epoch": 5.7808, + "grad_norm": 0.0, + "learning_rate": 0.0002854, + "loss": 5.9447, + "step": 3613 + }, + { + "epoch": 5.7824, + "grad_norm": 0.0, + "learning_rate": 0.0002852, + "loss": 4.9128, + "step": 3614 + }, + { + "epoch": 5.784, + "grad_norm": 0.0, + "learning_rate": 0.000285, + "loss": 5.7036, + "step": 3615 + }, + { + "epoch": 5.7856, + "grad_norm": 0.0, + "learning_rate": 0.0002848, + "loss": 5.1957, + "step": 3616 + }, + { + "epoch": 5.7872, + "grad_norm": 0.0, + "learning_rate": 0.0002846, + "loss": 4.7366, + "step": 3617 + }, + { + "epoch": 5.7888, + "grad_norm": 0.0, + "learning_rate": 0.0002844, + "loss": 5.3741, + "step": 3618 + }, + { + "epoch": 5.7904, + "grad_norm": 0.0, + "learning_rate": 0.00028419999999999997, + "loss": 4.4804, + "step": 3619 + }, + { + "epoch": 5.792, + "grad_norm": 0.0, + "learning_rate": 0.00028399999999999996, + "loss": 4.8407, + "step": 3620 + }, + { + "epoch": 5.7936, + "grad_norm": 0.0, + "learning_rate": 0.00028379999999999996, + "loss": 5.6581, + "step": 3621 + }, + { + "epoch": 5.7952, + "grad_norm": 0.0, + "learning_rate": 0.0002836, + "loss": 8.0049, + "step": 3622 + }, + { + "epoch": 5.7968, + "grad_norm": 0.0, + "learning_rate": 0.00028339999999999995, + "loss": 4.7602, + "step": 3623 + }, + { + "epoch": 5.7984, + "grad_norm": 0.0, + "learning_rate": 0.00028319999999999994, + "loss": 4.9305, + "step": 3624 + }, + { + "epoch": 5.8, + "grad_norm": NaN, + "learning_rate": 0.00028319999999999994, + "loss": 5.7485, + "step": 3625 + }, + { + "epoch": 5.8016, + "grad_norm": 0.0, + "learning_rate": 0.000283, + "loss": 8.3658, + "step": 3626 + }, + { + "epoch": 5.8032, + "grad_norm": 0.0, + "learning_rate": 0.0002828, + "loss": 7.1211, + "step": 3627 + }, + { + "epoch": 5.8048, + "grad_norm": 0.0, + "learning_rate": 0.0002826, + "loss": 7.0322, + "step": 3628 + }, + { + "epoch": 5.8064, + "grad_norm": 0.0, + "learning_rate": 0.0002824, + "loss": 8.3492, + "step": 3629 + }, + { + "epoch": 5.808, + "grad_norm": 0.0, + "learning_rate": 0.0002822, + "loss": 6.3224, + "step": 3630 + }, + { + "epoch": 5.8096, + "grad_norm": 0.0, + "learning_rate": 0.00028199999999999997, + "loss": 6.6665, + "step": 3631 + }, + { + "epoch": 5.8112, + "grad_norm": 0.0, + "learning_rate": 0.00028179999999999997, + "loss": 5.8809, + "step": 3632 + }, + { + "epoch": 5.8128, + "grad_norm": 0.0, + "learning_rate": 0.00028159999999999996, + "loss": 7.3065, + "step": 3633 + }, + { + "epoch": 5.8144, + "grad_norm": 0.0, + "learning_rate": 0.00028139999999999996, + "loss": 5.906, + "step": 3634 + }, + { + "epoch": 5.816, + "grad_norm": 0.0, + "learning_rate": 0.0002812, + "loss": 5.3947, + "step": 3635 + }, + { + "epoch": 5.8176, + "grad_norm": 0.0, + "learning_rate": 0.00028099999999999995, + "loss": 6.9699, + "step": 3636 + }, + { + "epoch": 5.8192, + "grad_norm": 0.0, + "learning_rate": 0.0002808, + "loss": 6.1025, + "step": 3637 + }, + { + "epoch": 5.8208, + "grad_norm": 0.0, + "learning_rate": 0.0002806, + "loss": 5.37, + "step": 3638 + }, + { + "epoch": 5.8224, + "grad_norm": 0.0, + "learning_rate": 0.0002804, + "loss": 6.642, + "step": 3639 + }, + { + "epoch": 5.824, + "grad_norm": 0.0, + "learning_rate": 0.0002802, + "loss": 5.8273, + "step": 3640 + }, + { + "epoch": 5.8256, + "grad_norm": 0.0, + "learning_rate": 0.00028, + "loss": 5.6738, + "step": 3641 + }, + { + "epoch": 5.8272, + "grad_norm": 0.0, + "learning_rate": 0.00027979999999999997, + "loss": 5.8018, + "step": 3642 + }, + { + "epoch": 5.8288, + "grad_norm": 0.0, + "learning_rate": 0.00027959999999999997, + "loss": 5.2876, + "step": 3643 + }, + { + "epoch": 5.8304, + "grad_norm": 0.0, + "learning_rate": 0.00027939999999999996, + "loss": 4.9762, + "step": 3644 + }, + { + "epoch": 5.832, + "grad_norm": 0.0, + "learning_rate": 0.00027919999999999996, + "loss": 5.017, + "step": 3645 + }, + { + "epoch": 5.8336, + "grad_norm": 0.0, + "learning_rate": 0.000279, + "loss": 5.5559, + "step": 3646 + }, + { + "epoch": 5.8352, + "grad_norm": 0.0, + "learning_rate": 0.0002788, + "loss": 5.2275, + "step": 3647 + }, + { + "epoch": 5.8368, + "grad_norm": 0.0, + "learning_rate": 0.00027859999999999994, + "loss": 4.935, + "step": 3648 + }, + { + "epoch": 5.8384, + "grad_norm": 0.0, + "learning_rate": 0.0002784, + "loss": 5.0413, + "step": 3649 + }, + { + "epoch": 5.84, + "grad_norm": 0.0, + "learning_rate": 0.0002782, + "loss": 5.7146, + "step": 3650 + }, + { + "epoch": 5.8416, + "grad_norm": 0.0, + "learning_rate": 0.000278, + "loss": 5.5271, + "step": 3651 + }, + { + "epoch": 5.8431999999999995, + "grad_norm": 0.0, + "learning_rate": 0.0002778, + "loss": 6.8281, + "step": 3652 + }, + { + "epoch": 5.8448, + "grad_norm": 0.0, + "learning_rate": 0.00027759999999999997, + "loss": 5.5622, + "step": 3653 + }, + { + "epoch": 5.8464, + "grad_norm": 0.0, + "learning_rate": 0.00027739999999999997, + "loss": 5.2229, + "step": 3654 + }, + { + "epoch": 5.848, + "grad_norm": 0.0, + "learning_rate": 0.0002772, + "loss": 4.256, + "step": 3655 + }, + { + "epoch": 5.8496, + "grad_norm": 0.0, + "learning_rate": 0.00027699999999999996, + "loss": 4.9199, + "step": 3656 + }, + { + "epoch": 5.8512, + "grad_norm": 0.0, + "learning_rate": 0.00027679999999999995, + "loss": 4.5997, + "step": 3657 + }, + { + "epoch": 5.8528, + "grad_norm": 0.0, + "learning_rate": 0.0002766, + "loss": 6.0303, + "step": 3658 + }, + { + "epoch": 5.8544, + "grad_norm": 0.0, + "learning_rate": 0.0002764, + "loss": 6.3918, + "step": 3659 + }, + { + "epoch": 5.856, + "grad_norm": 0.0, + "learning_rate": 0.0002762, + "loss": 5.9235, + "step": 3660 + }, + { + "epoch": 5.8576, + "grad_norm": 0.0, + "learning_rate": 0.000276, + "loss": 6.4329, + "step": 3661 + }, + { + "epoch": 5.8591999999999995, + "grad_norm": 0.0, + "learning_rate": 0.0002758, + "loss": 5.6884, + "step": 3662 + }, + { + "epoch": 5.8608, + "grad_norm": 0.0, + "learning_rate": 0.0002756, + "loss": 4.744, + "step": 3663 + }, + { + "epoch": 5.8624, + "grad_norm": 0.0, + "learning_rate": 0.00027539999999999997, + "loss": 5.5336, + "step": 3664 + }, + { + "epoch": 5.864, + "grad_norm": 0.0, + "learning_rate": 0.00027519999999999997, + "loss": 5.4217, + "step": 3665 + }, + { + "epoch": 5.8656, + "grad_norm": 0.0, + "learning_rate": 0.00027499999999999996, + "loss": 4.8725, + "step": 3666 + }, + { + "epoch": 5.8672, + "grad_norm": 0.0, + "learning_rate": 0.0002748, + "loss": 5.2555, + "step": 3667 + }, + { + "epoch": 5.8688, + "grad_norm": 0.0, + "learning_rate": 0.00027459999999999995, + "loss": 5.1452, + "step": 3668 + }, + { + "epoch": 5.8704, + "grad_norm": 0.0, + "learning_rate": 0.00027439999999999995, + "loss": 6.4179, + "step": 3669 + }, + { + "epoch": 5.872, + "grad_norm": 0.0, + "learning_rate": 0.0002742, + "loss": 4.8269, + "step": 3670 + }, + { + "epoch": 5.8736, + "grad_norm": 0.0, + "learning_rate": 0.000274, + "loss": 6.6833, + "step": 3671 + }, + { + "epoch": 5.8751999999999995, + "grad_norm": 0.0, + "learning_rate": 0.0002738, + "loss": 4.792, + "step": 3672 + }, + { + "epoch": 5.8768, + "grad_norm": 0.0, + "learning_rate": 0.0002736, + "loss": 5.3367, + "step": 3673 + }, + { + "epoch": 5.8784, + "grad_norm": 0.0, + "learning_rate": 0.0002734, + "loss": 4.7007, + "step": 3674 + }, + { + "epoch": 5.88, + "grad_norm": 0.0, + "learning_rate": 0.00027319999999999997, + "loss": 7.16, + "step": 3675 + }, + { + "epoch": 5.8816, + "grad_norm": 0.0, + "learning_rate": 0.00027299999999999997, + "loss": 6.8287, + "step": 3676 + }, + { + "epoch": 5.8832, + "grad_norm": 0.0, + "learning_rate": 0.00027279999999999996, + "loss": 6.2576, + "step": 3677 + }, + { + "epoch": 5.8848, + "grad_norm": 0.0, + "learning_rate": 0.00027259999999999996, + "loss": 7.6439, + "step": 3678 + }, + { + "epoch": 5.8864, + "grad_norm": 0.0, + "learning_rate": 0.0002724, + "loss": 7.9377, + "step": 3679 + }, + { + "epoch": 5.888, + "grad_norm": 0.0, + "learning_rate": 0.00027219999999999995, + "loss": 6.8442, + "step": 3680 + }, + { + "epoch": 5.8896, + "grad_norm": 0.0, + "learning_rate": 0.00027199999999999994, + "loss": 5.2543, + "step": 3681 + }, + { + "epoch": 5.8911999999999995, + "grad_norm": 0.0, + "learning_rate": 0.0002718, + "loss": 6.3046, + "step": 3682 + }, + { + "epoch": 5.8928, + "grad_norm": 0.0, + "learning_rate": 0.0002716, + "loss": 6.2755, + "step": 3683 + }, + { + "epoch": 5.8944, + "grad_norm": 0.0, + "learning_rate": 0.0002714, + "loss": 6.6268, + "step": 3684 + }, + { + "epoch": 5.896, + "grad_norm": 0.0, + "learning_rate": 0.0002712, + "loss": 5.7255, + "step": 3685 + }, + { + "epoch": 5.8976, + "grad_norm": 0.0, + "learning_rate": 0.000271, + "loss": 5.3728, + "step": 3686 + }, + { + "epoch": 5.8992, + "grad_norm": 0.0, + "learning_rate": 0.00027079999999999997, + "loss": 4.9809, + "step": 3687 + }, + { + "epoch": 5.9008, + "grad_norm": 0.0, + "learning_rate": 0.00027059999999999996, + "loss": 5.3099, + "step": 3688 + }, + { + "epoch": 5.9024, + "grad_norm": 0.0, + "learning_rate": 0.00027039999999999996, + "loss": 5.7011, + "step": 3689 + }, + { + "epoch": 5.904, + "grad_norm": 0.0, + "learning_rate": 0.00027019999999999995, + "loss": 6.6388, + "step": 3690 + }, + { + "epoch": 5.9056, + "grad_norm": 0.0, + "learning_rate": 0.00027, + "loss": 6.7033, + "step": 3691 + }, + { + "epoch": 5.9072, + "grad_norm": 0.0, + "learning_rate": 0.0002698, + "loss": 5.7401, + "step": 3692 + }, + { + "epoch": 5.9088, + "grad_norm": 0.0, + "learning_rate": 0.00026959999999999994, + "loss": 6.2838, + "step": 3693 + }, + { + "epoch": 5.9104, + "grad_norm": 0.0, + "learning_rate": 0.0002694, + "loss": 4.524, + "step": 3694 + }, + { + "epoch": 5.912, + "grad_norm": 0.0, + "learning_rate": 0.0002692, + "loss": 4.595, + "step": 3695 + }, + { + "epoch": 5.9136, + "grad_norm": 0.0, + "learning_rate": 0.000269, + "loss": 5.2411, + "step": 3696 + }, + { + "epoch": 5.9152000000000005, + "grad_norm": 0.0, + "learning_rate": 0.0002688, + "loss": 6.8652, + "step": 3697 + }, + { + "epoch": 5.9168, + "grad_norm": 0.0, + "learning_rate": 0.00026859999999999997, + "loss": 6.3878, + "step": 3698 + }, + { + "epoch": 5.9184, + "grad_norm": 0.0, + "learning_rate": 0.0002684, + "loss": 5.5974, + "step": 3699 + }, + { + "epoch": 5.92, + "grad_norm": 0.0, + "learning_rate": 0.00026819999999999996, + "loss": 5.8375, + "step": 3700 + }, + { + "epoch": 5.9216, + "grad_norm": 0.0, + "learning_rate": 0.00026799999999999995, + "loss": 5.9957, + "step": 3701 + }, + { + "epoch": 5.9232, + "grad_norm": 0.0, + "learning_rate": 0.0002678, + "loss": 5.005, + "step": 3702 + }, + { + "epoch": 5.9248, + "grad_norm": 0.0, + "learning_rate": 0.0002676, + "loss": 6.4649, + "step": 3703 + }, + { + "epoch": 5.9264, + "grad_norm": 0.0, + "learning_rate": 0.0002674, + "loss": 4.3296, + "step": 3704 + }, + { + "epoch": 5.928, + "grad_norm": 0.0, + "learning_rate": 0.0002672, + "loss": 5.4839, + "step": 3705 + }, + { + "epoch": 5.9296, + "grad_norm": 0.0, + "learning_rate": 0.000267, + "loss": 6.3118, + "step": 3706 + }, + { + "epoch": 5.9312000000000005, + "grad_norm": 0.0, + "learning_rate": 0.0002668, + "loss": 5.6869, + "step": 3707 + }, + { + "epoch": 5.9328, + "grad_norm": 0.0, + "learning_rate": 0.0002666, + "loss": 4.7917, + "step": 3708 + }, + { + "epoch": 5.9344, + "grad_norm": 0.0, + "learning_rate": 0.00026639999999999997, + "loss": 3.9526, + "step": 3709 + }, + { + "epoch": 5.936, + "grad_norm": 0.0, + "learning_rate": 0.00026619999999999997, + "loss": 5.9955, + "step": 3710 + }, + { + "epoch": 5.9376, + "grad_norm": 0.0, + "learning_rate": 0.000266, + "loss": 4.7419, + "step": 3711 + }, + { + "epoch": 5.9392, + "grad_norm": 0.0, + "learning_rate": 0.00026579999999999996, + "loss": 5.9813, + "step": 3712 + }, + { + "epoch": 5.9408, + "grad_norm": 0.0, + "learning_rate": 0.00026559999999999995, + "loss": 5.1499, + "step": 3713 + }, + { + "epoch": 5.9424, + "grad_norm": 0.0, + "learning_rate": 0.0002654, + "loss": 4.8764, + "step": 3714 + }, + { + "epoch": 5.944, + "grad_norm": 0.0, + "learning_rate": 0.0002652, + "loss": 4.8173, + "step": 3715 + }, + { + "epoch": 5.9456, + "grad_norm": 0.0, + "learning_rate": 0.000265, + "loss": 5.5797, + "step": 3716 + }, + { + "epoch": 5.9472000000000005, + "grad_norm": 0.0, + "learning_rate": 0.0002648, + "loss": 4.5446, + "step": 3717 + }, + { + "epoch": 5.9488, + "grad_norm": 0.0, + "learning_rate": 0.0002646, + "loss": 3.742, + "step": 3718 + }, + { + "epoch": 5.9504, + "grad_norm": 0.0, + "learning_rate": 0.0002644, + "loss": 5.1998, + "step": 3719 + }, + { + "epoch": 5.952, + "grad_norm": 0.0, + "learning_rate": 0.00026419999999999997, + "loss": 7.5806, + "step": 3720 + }, + { + "epoch": 5.9536, + "grad_norm": 0.0, + "learning_rate": 0.00026399999999999997, + "loss": 5.7512, + "step": 3721 + }, + { + "epoch": 5.9552, + "grad_norm": 0.0, + "learning_rate": 0.00026379999999999996, + "loss": 4.9907, + "step": 3722 + }, + { + "epoch": 5.9568, + "grad_norm": 0.0, + "learning_rate": 0.0002636, + "loss": 5.0717, + "step": 3723 + }, + { + "epoch": 5.9584, + "grad_norm": 0.0, + "learning_rate": 0.00026339999999999995, + "loss": 5.6612, + "step": 3724 + }, + { + "epoch": 5.96, + "grad_norm": NaN, + "learning_rate": 0.00026339999999999995, + "loss": 3.9089, + "step": 3725 + }, + { + "epoch": 5.9616, + "grad_norm": 0.0, + "learning_rate": 0.00026319999999999995, + "loss": 7.2625, + "step": 3726 + }, + { + "epoch": 5.9632, + "grad_norm": 0.0, + "learning_rate": 0.000263, + "loss": 6.0971, + "step": 3727 + }, + { + "epoch": 5.9648, + "grad_norm": 0.0, + "learning_rate": 0.0002628, + "loss": 6.1193, + "step": 3728 + }, + { + "epoch": 5.9664, + "grad_norm": 0.0, + "learning_rate": 0.0002626, + "loss": 6.1007, + "step": 3729 + }, + { + "epoch": 5.968, + "grad_norm": 0.0, + "learning_rate": 0.0002624, + "loss": 6.2722, + "step": 3730 + }, + { + "epoch": 5.9696, + "grad_norm": 0.0, + "learning_rate": 0.0002622, + "loss": 6.1931, + "step": 3731 + }, + { + "epoch": 5.9712, + "grad_norm": 0.0, + "learning_rate": 0.00026199999999999997, + "loss": 6.1803, + "step": 3732 + }, + { + "epoch": 5.9728, + "grad_norm": 0.0, + "learning_rate": 0.00026179999999999997, + "loss": 7.0627, + "step": 3733 + }, + { + "epoch": 5.9744, + "grad_norm": 0.0, + "learning_rate": 0.00026159999999999996, + "loss": 5.5696, + "step": 3734 + }, + { + "epoch": 5.976, + "grad_norm": 0.0, + "learning_rate": 0.00026139999999999996, + "loss": 5.6518, + "step": 3735 + }, + { + "epoch": 5.9776, + "grad_norm": 0.0, + "learning_rate": 0.0002612, + "loss": 5.3313, + "step": 3736 + }, + { + "epoch": 5.9792, + "grad_norm": 0.0, + "learning_rate": 0.000261, + "loss": 6.4509, + "step": 3737 + }, + { + "epoch": 5.9808, + "grad_norm": 0.0, + "learning_rate": 0.00026079999999999994, + "loss": 5.8548, + "step": 3738 + }, + { + "epoch": 5.9824, + "grad_norm": 0.0, + "learning_rate": 0.0002606, + "loss": 5.4002, + "step": 3739 + }, + { + "epoch": 5.984, + "grad_norm": 0.0, + "learning_rate": 0.0002604, + "loss": 4.4642, + "step": 3740 + }, + { + "epoch": 5.9856, + "grad_norm": 0.0, + "learning_rate": 0.0002602, + "loss": 4.807, + "step": 3741 + }, + { + "epoch": 5.9872, + "grad_norm": 0.0, + "learning_rate": 0.00026, + "loss": 5.7274, + "step": 3742 + }, + { + "epoch": 5.9888, + "grad_norm": 0.0, + "learning_rate": 0.00025979999999999997, + "loss": 4.3287, + "step": 3743 + }, + { + "epoch": 5.9904, + "grad_norm": 0.0, + "learning_rate": 0.00025959999999999997, + "loss": 5.3405, + "step": 3744 + }, + { + "epoch": 5.992, + "grad_norm": 0.0, + "learning_rate": 0.00025939999999999996, + "loss": 4.8432, + "step": 3745 + }, + { + "epoch": 5.9936, + "grad_norm": 0.0, + "learning_rate": 0.00025919999999999996, + "loss": 6.0134, + "step": 3746 + }, + { + "epoch": 5.9952, + "grad_norm": 0.0, + "learning_rate": 0.00025899999999999995, + "loss": 5.9685, + "step": 3747 + }, + { + "epoch": 5.9968, + "grad_norm": 0.0, + "learning_rate": 0.0002588, + "loss": 6.4016, + "step": 3748 + }, + { + "epoch": 5.9984, + "grad_norm": 0.0, + "learning_rate": 0.0002586, + "loss": 4.3907, + "step": 3749 + }, + { + "epoch": 6.0, + "grad_norm": 0.0, + "learning_rate": 0.00025839999999999994, + "loss": 4.9855, + "step": 3750 + }, + { + "epoch": 6.0016, + "grad_norm": 0.0, + "learning_rate": 0.0002582, + "loss": 7.903, + "step": 3751 + }, + { + "epoch": 6.0032, + "grad_norm": 0.0, + "learning_rate": 0.000258, + "loss": 8.2425, + "step": 3752 + }, + { + "epoch": 6.0048, + "grad_norm": 0.0, + "learning_rate": 0.0002578, + "loss": 6.6237, + "step": 3753 + }, + { + "epoch": 6.0064, + "grad_norm": 0.0, + "learning_rate": 0.0002576, + "loss": 6.7151, + "step": 3754 + }, + { + "epoch": 6.008, + "grad_norm": 0.0, + "learning_rate": 0.00025739999999999997, + "loss": 6.7659, + "step": 3755 + }, + { + "epoch": 6.0096, + "grad_norm": 0.0, + "learning_rate": 0.00025719999999999996, + "loss": 5.9836, + "step": 3756 + }, + { + "epoch": 6.0112, + "grad_norm": 0.0, + "learning_rate": 0.00025699999999999996, + "loss": 6.175, + "step": 3757 + }, + { + "epoch": 6.0128, + "grad_norm": 0.0, + "learning_rate": 0.00025679999999999995, + "loss": 5.6578, + "step": 3758 + }, + { + "epoch": 6.0144, + "grad_norm": 0.0, + "learning_rate": 0.00025659999999999995, + "loss": 5.2847, + "step": 3759 + }, + { + "epoch": 6.016, + "grad_norm": 0.0, + "learning_rate": 0.0002564, + "loss": 4.9896, + "step": 3760 + }, + { + "epoch": 6.0176, + "grad_norm": 0.0, + "learning_rate": 0.0002562, + "loss": 6.4981, + "step": 3761 + }, + { + "epoch": 6.0192, + "grad_norm": 0.0, + "learning_rate": 0.000256, + "loss": 5.0104, + "step": 3762 + }, + { + "epoch": 6.0208, + "grad_norm": 0.0, + "learning_rate": 0.0002558, + "loss": 6.4033, + "step": 3763 + }, + { + "epoch": 6.0224, + "grad_norm": 0.0, + "learning_rate": 0.0002556, + "loss": 5.2367, + "step": 3764 + }, + { + "epoch": 6.024, + "grad_norm": 0.0, + "learning_rate": 0.0002554, + "loss": 6.7913, + "step": 3765 + }, + { + "epoch": 6.0256, + "grad_norm": 0.0, + "learning_rate": 0.00025519999999999997, + "loss": 6.1005, + "step": 3766 + }, + { + "epoch": 6.0272, + "grad_norm": 0.0, + "learning_rate": 0.00025499999999999996, + "loss": 5.4404, + "step": 3767 + }, + { + "epoch": 6.0288, + "grad_norm": 0.0, + "learning_rate": 0.0002548, + "loss": 5.2762, + "step": 3768 + }, + { + "epoch": 6.0304, + "grad_norm": 0.0, + "learning_rate": 0.0002546, + "loss": 4.5447, + "step": 3769 + }, + { + "epoch": 6.032, + "grad_norm": 0.0, + "learning_rate": 0.00025439999999999995, + "loss": 5.6099, + "step": 3770 + }, + { + "epoch": 6.0336, + "grad_norm": 0.0, + "learning_rate": 0.0002542, + "loss": 5.8155, + "step": 3771 + }, + { + "epoch": 6.0352, + "grad_norm": 0.0, + "learning_rate": 0.000254, + "loss": 4.7098, + "step": 3772 + }, + { + "epoch": 6.0368, + "grad_norm": 0.0, + "learning_rate": 0.0002538, + "loss": 5.0747, + "step": 3773 + }, + { + "epoch": 6.0384, + "grad_norm": 0.0, + "learning_rate": 0.0002536, + "loss": 6.1499, + "step": 3774 + }, + { + "epoch": 6.04, + "grad_norm": 0.0, + "learning_rate": 0.0002534, + "loss": 5.3974, + "step": 3775 + }, + { + "epoch": 6.0416, + "grad_norm": 0.0, + "learning_rate": 0.0002532, + "loss": 5.9723, + "step": 3776 + }, + { + "epoch": 6.0432, + "grad_norm": 0.0, + "learning_rate": 0.00025299999999999997, + "loss": 5.1909, + "step": 3777 + }, + { + "epoch": 6.0448, + "grad_norm": 0.0, + "learning_rate": 0.00025279999999999996, + "loss": 5.6738, + "step": 3778 + }, + { + "epoch": 6.0464, + "grad_norm": 0.0, + "learning_rate": 0.00025259999999999996, + "loss": 4.4649, + "step": 3779 + }, + { + "epoch": 6.048, + "grad_norm": 0.0, + "learning_rate": 0.0002524, + "loss": 5.0202, + "step": 3780 + }, + { + "epoch": 6.0496, + "grad_norm": 0.0, + "learning_rate": 0.0002522, + "loss": 5.1147, + "step": 3781 + }, + { + "epoch": 6.0512, + "grad_norm": 0.0, + "learning_rate": 0.00025199999999999995, + "loss": 5.1592, + "step": 3782 + }, + { + "epoch": 6.0528, + "grad_norm": 0.0, + "learning_rate": 0.0002518, + "loss": 6.1826, + "step": 3783 + }, + { + "epoch": 6.0544, + "grad_norm": 0.0, + "learning_rate": 0.0002516, + "loss": 5.7229, + "step": 3784 + }, + { + "epoch": 6.056, + "grad_norm": 0.0, + "learning_rate": 0.0002514, + "loss": 7.5871, + "step": 3785 + }, + { + "epoch": 6.0576, + "grad_norm": 0.0, + "learning_rate": 0.0002512, + "loss": 5.1573, + "step": 3786 + }, + { + "epoch": 6.0592, + "grad_norm": 0.0, + "learning_rate": 0.000251, + "loss": 4.9079, + "step": 3787 + }, + { + "epoch": 6.0608, + "grad_norm": 0.0, + "learning_rate": 0.00025079999999999997, + "loss": 4.5917, + "step": 3788 + }, + { + "epoch": 6.0624, + "grad_norm": 0.0, + "learning_rate": 0.00025059999999999997, + "loss": 6.7335, + "step": 3789 + }, + { + "epoch": 6.064, + "grad_norm": 0.0, + "learning_rate": 0.00025039999999999996, + "loss": 6.1439, + "step": 3790 + }, + { + "epoch": 6.0656, + "grad_norm": 0.0, + "learning_rate": 0.00025019999999999996, + "loss": 5.16, + "step": 3791 + }, + { + "epoch": 6.0672, + "grad_norm": 0.0, + "learning_rate": 0.00025, + "loss": 5.8172, + "step": 3792 + }, + { + "epoch": 6.0688, + "grad_norm": 0.0, + "learning_rate": 0.0002498, + "loss": 4.9083, + "step": 3793 + }, + { + "epoch": 6.0704, + "grad_norm": 0.0, + "learning_rate": 0.00024959999999999994, + "loss": 5.1364, + "step": 3794 + }, + { + "epoch": 6.072, + "grad_norm": 0.0, + "learning_rate": 0.0002494, + "loss": 4.7678, + "step": 3795 + }, + { + "epoch": 6.0736, + "grad_norm": 0.0, + "learning_rate": 0.0002492, + "loss": 4.9779, + "step": 3796 + }, + { + "epoch": 6.0752, + "grad_norm": 0.0, + "learning_rate": 0.000249, + "loss": 4.7857, + "step": 3797 + }, + { + "epoch": 6.0768, + "grad_norm": 0.0, + "learning_rate": 0.0002488, + "loss": 5.6867, + "step": 3798 + }, + { + "epoch": 6.0784, + "grad_norm": 0.0, + "learning_rate": 0.00024859999999999997, + "loss": 6.0305, + "step": 3799 + }, + { + "epoch": 6.08, + "grad_norm": NaN, + "learning_rate": 0.00024859999999999997, + "loss": 5.1104, + "step": 3800 + }, + { + "epoch": 6.0816, + "grad_norm": 0.0, + "learning_rate": 0.00024839999999999997, + "loss": 9.4278, + "step": 3801 + }, + { + "epoch": 6.0832, + "grad_norm": 0.0, + "learning_rate": 0.00024819999999999996, + "loss": 6.2426, + "step": 3802 + }, + { + "epoch": 6.0848, + "grad_norm": 0.0, + "learning_rate": 0.00024799999999999996, + "loss": 8.4432, + "step": 3803 + }, + { + "epoch": 6.0864, + "grad_norm": 0.0, + "learning_rate": 0.00024779999999999995, + "loss": 7.1244, + "step": 3804 + }, + { + "epoch": 6.088, + "grad_norm": 0.0, + "learning_rate": 0.0002476, + "loss": 7.5285, + "step": 3805 + }, + { + "epoch": 6.0896, + "grad_norm": 0.0, + "learning_rate": 0.0002474, + "loss": 6.0996, + "step": 3806 + }, + { + "epoch": 6.0912, + "grad_norm": 0.0, + "learning_rate": 0.0002472, + "loss": 5.0784, + "step": 3807 + }, + { + "epoch": 6.0928, + "grad_norm": 0.0, + "learning_rate": 0.000247, + "loss": 6.9685, + "step": 3808 + }, + { + "epoch": 6.0944, + "grad_norm": 0.0, + "learning_rate": 0.0002468, + "loss": 6.6873, + "step": 3809 + }, + { + "epoch": 6.096, + "grad_norm": 0.0, + "learning_rate": 0.0002466, + "loss": 6.9079, + "step": 3810 + }, + { + "epoch": 6.0976, + "grad_norm": 0.0, + "learning_rate": 0.00024639999999999997, + "loss": 6.7021, + "step": 3811 + }, + { + "epoch": 6.0992, + "grad_norm": 0.0, + "learning_rate": 0.00024619999999999997, + "loss": 6.1739, + "step": 3812 + }, + { + "epoch": 6.1008, + "grad_norm": 0.0, + "learning_rate": 0.00024599999999999996, + "loss": 7.103, + "step": 3813 + }, + { + "epoch": 6.1024, + "grad_norm": 0.0, + "learning_rate": 0.0002458, + "loss": 5.3594, + "step": 3814 + }, + { + "epoch": 6.104, + "grad_norm": 0.0, + "learning_rate": 0.00024559999999999995, + "loss": 5.6977, + "step": 3815 + }, + { + "epoch": 6.1056, + "grad_norm": 0.0, + "learning_rate": 0.00024539999999999995, + "loss": 4.9392, + "step": 3816 + }, + { + "epoch": 6.1072, + "grad_norm": 0.0, + "learning_rate": 0.0002452, + "loss": 6.8875, + "step": 3817 + }, + { + "epoch": 6.1088, + "grad_norm": 0.0, + "learning_rate": 0.000245, + "loss": 5.0149, + "step": 3818 + }, + { + "epoch": 6.1104, + "grad_norm": 0.0, + "learning_rate": 0.0002448, + "loss": 5.7098, + "step": 3819 + }, + { + "epoch": 6.112, + "grad_norm": 0.0, + "learning_rate": 0.0002446, + "loss": 7.3, + "step": 3820 + }, + { + "epoch": 6.1136, + "grad_norm": 0.0, + "learning_rate": 0.0002444, + "loss": 6.8998, + "step": 3821 + }, + { + "epoch": 6.1152, + "grad_norm": 0.0, + "learning_rate": 0.00024419999999999997, + "loss": 5.5671, + "step": 3822 + }, + { + "epoch": 6.1168, + "grad_norm": 0.0, + "learning_rate": 0.000244, + "loss": 5.0307, + "step": 3823 + }, + { + "epoch": 6.1184, + "grad_norm": 0.0, + "learning_rate": 0.00024379999999999996, + "loss": 4.9341, + "step": 3824 + }, + { + "epoch": 6.12, + "grad_norm": 0.0, + "learning_rate": 0.00024359999999999999, + "loss": 5.5168, + "step": 3825 + }, + { + "epoch": 6.1216, + "grad_norm": 0.0, + "learning_rate": 0.00024339999999999998, + "loss": 6.2582, + "step": 3826 + }, + { + "epoch": 6.1232, + "grad_norm": 0.0, + "learning_rate": 0.00024319999999999998, + "loss": 5.3042, + "step": 3827 + }, + { + "epoch": 6.1248, + "grad_norm": 0.0, + "learning_rate": 0.000243, + "loss": 4.689, + "step": 3828 + }, + { + "epoch": 6.1264, + "grad_norm": 0.0, + "learning_rate": 0.0002428, + "loss": 5.6967, + "step": 3829 + }, + { + "epoch": 6.128, + "grad_norm": 0.0, + "learning_rate": 0.00024259999999999996, + "loss": 5.9225, + "step": 3830 + }, + { + "epoch": 6.1296, + "grad_norm": 0.0, + "learning_rate": 0.00024239999999999998, + "loss": 5.503, + "step": 3831 + }, + { + "epoch": 6.1312, + "grad_norm": 0.0, + "learning_rate": 0.00024219999999999998, + "loss": 3.8699, + "step": 3832 + }, + { + "epoch": 6.1328, + "grad_norm": 0.0, + "learning_rate": 0.00024199999999999997, + "loss": 6.7203, + "step": 3833 + }, + { + "epoch": 6.1344, + "grad_norm": 0.0, + "learning_rate": 0.0002418, + "loss": 4.5486, + "step": 3834 + }, + { + "epoch": 6.136, + "grad_norm": 0.0, + "learning_rate": 0.0002416, + "loss": 5.3035, + "step": 3835 + }, + { + "epoch": 6.1376, + "grad_norm": 0.0, + "learning_rate": 0.00024139999999999996, + "loss": 6.0709, + "step": 3836 + }, + { + "epoch": 6.1392, + "grad_norm": 0.0, + "learning_rate": 0.00024119999999999998, + "loss": 5.2587, + "step": 3837 + }, + { + "epoch": 6.1408, + "grad_norm": 0.0, + "learning_rate": 0.00024099999999999998, + "loss": 4.4856, + "step": 3838 + }, + { + "epoch": 6.1424, + "grad_norm": 0.0, + "learning_rate": 0.00024079999999999997, + "loss": 6.3636, + "step": 3839 + }, + { + "epoch": 6.144, + "grad_norm": 0.0, + "learning_rate": 0.0002406, + "loss": 4.715, + "step": 3840 + }, + { + "epoch": 6.1456, + "grad_norm": 0.0, + "learning_rate": 0.0002404, + "loss": 5.179, + "step": 3841 + }, + { + "epoch": 6.1472, + "grad_norm": 0.0, + "learning_rate": 0.00024019999999999996, + "loss": 5.3661, + "step": 3842 + }, + { + "epoch": 6.1488, + "grad_norm": 0.0, + "learning_rate": 0.00023999999999999998, + "loss": 4.3894, + "step": 3843 + }, + { + "epoch": 6.1504, + "grad_norm": 0.0, + "learning_rate": 0.00023979999999999997, + "loss": 6.9619, + "step": 3844 + }, + { + "epoch": 6.152, + "grad_norm": 0.0, + "learning_rate": 0.00023959999999999997, + "loss": 5.6009, + "step": 3845 + }, + { + "epoch": 6.1536, + "grad_norm": 0.0, + "learning_rate": 0.0002394, + "loss": 5.4545, + "step": 3846 + }, + { + "epoch": 6.1552, + "grad_norm": 0.0, + "learning_rate": 0.0002392, + "loss": 6.2692, + "step": 3847 + }, + { + "epoch": 6.1568, + "grad_norm": 0.0, + "learning_rate": 0.00023899999999999998, + "loss": 6.455, + "step": 3848 + }, + { + "epoch": 6.1584, + "grad_norm": 0.0, + "learning_rate": 0.0002388, + "loss": 4.4629, + "step": 3849 + }, + { + "epoch": 6.16, + "grad_norm": 0.0, + "learning_rate": 0.00023859999999999997, + "loss": 5.9264, + "step": 3850 + }, + { + "epoch": 6.1616, + "grad_norm": 0.0, + "learning_rate": 0.00023839999999999997, + "loss": 8.888, + "step": 3851 + }, + { + "epoch": 6.1632, + "grad_norm": 0.0, + "learning_rate": 0.0002382, + "loss": 7.2334, + "step": 3852 + }, + { + "epoch": 6.1648, + "grad_norm": 0.0, + "learning_rate": 0.00023799999999999998, + "loss": 8.2169, + "step": 3853 + }, + { + "epoch": 6.1664, + "grad_norm": 0.0, + "learning_rate": 0.00023779999999999998, + "loss": 6.3764, + "step": 3854 + }, + { + "epoch": 6.168, + "grad_norm": 0.0, + "learning_rate": 0.0002376, + "loss": 6.8972, + "step": 3855 + }, + { + "epoch": 6.1696, + "grad_norm": 0.0, + "learning_rate": 0.00023739999999999997, + "loss": 5.7974, + "step": 3856 + }, + { + "epoch": 6.1712, + "grad_norm": 0.0, + "learning_rate": 0.00023719999999999997, + "loss": 5.8833, + "step": 3857 + }, + { + "epoch": 6.1728, + "grad_norm": 0.0, + "learning_rate": 0.000237, + "loss": 5.781, + "step": 3858 + }, + { + "epoch": 6.1744, + "grad_norm": 0.0, + "learning_rate": 0.00023679999999999998, + "loss": 5.2432, + "step": 3859 + }, + { + "epoch": 6.176, + "grad_norm": 0.0, + "learning_rate": 0.00023659999999999998, + "loss": 5.9126, + "step": 3860 + }, + { + "epoch": 6.1776, + "grad_norm": 0.0, + "learning_rate": 0.0002364, + "loss": 5.7147, + "step": 3861 + }, + { + "epoch": 6.1792, + "grad_norm": 0.0, + "learning_rate": 0.00023619999999999997, + "loss": 10.5996, + "step": 3862 + }, + { + "epoch": 6.1808, + "grad_norm": 0.0, + "learning_rate": 0.00023599999999999996, + "loss": 7.0736, + "step": 3863 + }, + { + "epoch": 6.1824, + "grad_norm": 0.0, + "learning_rate": 0.00023579999999999999, + "loss": 6.2343, + "step": 3864 + }, + { + "epoch": 6.184, + "grad_norm": 0.0, + "learning_rate": 0.00023559999999999998, + "loss": 6.0624, + "step": 3865 + }, + { + "epoch": 6.1856, + "grad_norm": 0.0, + "learning_rate": 0.00023539999999999998, + "loss": 6.0613, + "step": 3866 + }, + { + "epoch": 6.1872, + "grad_norm": 0.0, + "learning_rate": 0.0002352, + "loss": 5.388, + "step": 3867 + }, + { + "epoch": 6.1888, + "grad_norm": 0.0, + "learning_rate": 0.00023499999999999997, + "loss": 5.8486, + "step": 3868 + }, + { + "epoch": 6.1904, + "grad_norm": 0.0, + "learning_rate": 0.00023479999999999996, + "loss": 4.9837, + "step": 3869 + }, + { + "epoch": 6.192, + "grad_norm": 0.0, + "learning_rate": 0.00023459999999999998, + "loss": 5.7935, + "step": 3870 + }, + { + "epoch": 6.1936, + "grad_norm": 0.0, + "learning_rate": 0.00023439999999999998, + "loss": 5.2261, + "step": 3871 + }, + { + "epoch": 6.1952, + "grad_norm": 0.0, + "learning_rate": 0.00023419999999999997, + "loss": 6.4889, + "step": 3872 + }, + { + "epoch": 6.1968, + "grad_norm": 0.0, + "learning_rate": 0.000234, + "loss": 5.166, + "step": 3873 + }, + { + "epoch": 6.1984, + "grad_norm": 0.0, + "learning_rate": 0.00023379999999999996, + "loss": 4.3852, + "step": 3874 + }, + { + "epoch": 6.2, + "grad_norm": 0.0, + "learning_rate": 0.00023359999999999996, + "loss": 4.5725, + "step": 3875 + }, + { + "epoch": 6.2016, + "grad_norm": 0.0, + "learning_rate": 0.00023339999999999998, + "loss": 6.1091, + "step": 3876 + }, + { + "epoch": 6.2032, + "grad_norm": 0.0, + "learning_rate": 0.00023319999999999998, + "loss": 5.7987, + "step": 3877 + }, + { + "epoch": 6.2048, + "grad_norm": 0.0, + "learning_rate": 0.00023299999999999997, + "loss": 7.822, + "step": 3878 + }, + { + "epoch": 6.2064, + "grad_norm": 0.0, + "learning_rate": 0.0002328, + "loss": 4.9151, + "step": 3879 + }, + { + "epoch": 6.208, + "grad_norm": 0.0, + "learning_rate": 0.00023259999999999996, + "loss": 6.7336, + "step": 3880 + }, + { + "epoch": 6.2096, + "grad_norm": 0.0, + "learning_rate": 0.00023239999999999996, + "loss": 5.2923, + "step": 3881 + }, + { + "epoch": 6.2112, + "grad_norm": 0.0, + "learning_rate": 0.00023219999999999998, + "loss": 5.8514, + "step": 3882 + }, + { + "epoch": 6.2128, + "grad_norm": 0.0, + "learning_rate": 0.00023199999999999997, + "loss": 5.226, + "step": 3883 + }, + { + "epoch": 6.2144, + "grad_norm": 0.0, + "learning_rate": 0.00023179999999999997, + "loss": 4.2487, + "step": 3884 + }, + { + "epoch": 6.216, + "grad_norm": 0.0, + "learning_rate": 0.0002316, + "loss": 4.7346, + "step": 3885 + }, + { + "epoch": 6.2176, + "grad_norm": 0.0, + "learning_rate": 0.0002314, + "loss": 6.2053, + "step": 3886 + }, + { + "epoch": 6.2192, + "grad_norm": 0.0, + "learning_rate": 0.0002312, + "loss": 6.8738, + "step": 3887 + }, + { + "epoch": 6.2208, + "grad_norm": 0.0, + "learning_rate": 0.00023099999999999998, + "loss": 5.0234, + "step": 3888 + }, + { + "epoch": 6.2224, + "grad_norm": 0.0, + "learning_rate": 0.00023079999999999997, + "loss": 5.9592, + "step": 3889 + }, + { + "epoch": 6.224, + "grad_norm": 0.0, + "learning_rate": 0.0002306, + "loss": 5.0649, + "step": 3890 + }, + { + "epoch": 6.2256, + "grad_norm": 0.0, + "learning_rate": 0.0002304, + "loss": 4.5155, + "step": 3891 + }, + { + "epoch": 6.2272, + "grad_norm": 0.0, + "learning_rate": 0.00023019999999999998, + "loss": 4.475, + "step": 3892 + }, + { + "epoch": 6.2288, + "grad_norm": 0.0, + "learning_rate": 0.00023, + "loss": 5.257, + "step": 3893 + }, + { + "epoch": 6.2304, + "grad_norm": 0.0, + "learning_rate": 0.00022979999999999997, + "loss": 6.7025, + "step": 3894 + }, + { + "epoch": 6.232, + "grad_norm": 0.0, + "learning_rate": 0.00022959999999999997, + "loss": 4.8638, + "step": 3895 + }, + { + "epoch": 6.2336, + "grad_norm": 0.0, + "learning_rate": 0.0002294, + "loss": 6.1504, + "step": 3896 + }, + { + "epoch": 6.2352, + "grad_norm": 0.0, + "learning_rate": 0.0002292, + "loss": 6.4339, + "step": 3897 + }, + { + "epoch": 6.2368, + "grad_norm": 0.0, + "learning_rate": 0.00022899999999999998, + "loss": 5.8082, + "step": 3898 + }, + { + "epoch": 6.2384, + "grad_norm": 0.0, + "learning_rate": 0.0002288, + "loss": 4.2343, + "step": 3899 + }, + { + "epoch": 6.24, + "grad_norm": NaN, + "learning_rate": 0.0002288, + "loss": 4.4807, + "step": 3900 + }, + { + "epoch": 6.2416, + "grad_norm": 0.0, + "learning_rate": 0.00022859999999999997, + "loss": 10.1381, + "step": 3901 + }, + { + "epoch": 6.2432, + "grad_norm": 0.0, + "learning_rate": 0.00022839999999999997, + "loss": 7.7384, + "step": 3902 + }, + { + "epoch": 6.2448, + "grad_norm": 0.0, + "learning_rate": 0.0002282, + "loss": 7.6657, + "step": 3903 + }, + { + "epoch": 6.2464, + "grad_norm": 0.0, + "learning_rate": 0.00022799999999999999, + "loss": 6.9036, + "step": 3904 + }, + { + "epoch": 6.248, + "grad_norm": 0.0, + "learning_rate": 0.00022779999999999998, + "loss": 6.1985, + "step": 3905 + }, + { + "epoch": 6.2496, + "grad_norm": 0.0, + "learning_rate": 0.0002276, + "loss": 5.9983, + "step": 3906 + }, + { + "epoch": 6.2512, + "grad_norm": 0.0, + "learning_rate": 0.00022739999999999997, + "loss": 6.286, + "step": 3907 + }, + { + "epoch": 6.2528, + "grad_norm": 0.0, + "learning_rate": 0.00022719999999999997, + "loss": 5.6263, + "step": 3908 + }, + { + "epoch": 6.2544, + "grad_norm": 0.0, + "learning_rate": 0.000227, + "loss": 7.0504, + "step": 3909 + }, + { + "epoch": 6.256, + "grad_norm": 0.0, + "learning_rate": 0.00022679999999999998, + "loss": 6.7953, + "step": 3910 + }, + { + "epoch": 6.2576, + "grad_norm": 0.0, + "learning_rate": 0.00022659999999999998, + "loss": 5.4169, + "step": 3911 + }, + { + "epoch": 6.2592, + "grad_norm": 0.0, + "learning_rate": 0.0002264, + "loss": 6.4364, + "step": 3912 + }, + { + "epoch": 6.2608, + "grad_norm": 0.0, + "learning_rate": 0.00022619999999999997, + "loss": 7.2482, + "step": 3913 + }, + { + "epoch": 6.2624, + "grad_norm": 0.0, + "learning_rate": 0.00022599999999999996, + "loss": 4.7631, + "step": 3914 + }, + { + "epoch": 6.264, + "grad_norm": 0.0, + "learning_rate": 0.00022579999999999999, + "loss": 5.5082, + "step": 3915 + }, + { + "epoch": 6.2656, + "grad_norm": 0.0, + "learning_rate": 0.00022559999999999998, + "loss": 5.6763, + "step": 3916 + }, + { + "epoch": 6.2672, + "grad_norm": 0.0, + "learning_rate": 0.00022539999999999998, + "loss": 5.0926, + "step": 3917 + }, + { + "epoch": 6.2688, + "grad_norm": 0.0, + "learning_rate": 0.0002252, + "loss": 6.2179, + "step": 3918 + }, + { + "epoch": 6.2704, + "grad_norm": 0.0, + "learning_rate": 0.000225, + "loss": 5.484, + "step": 3919 + }, + { + "epoch": 6.272, + "grad_norm": 0.0, + "learning_rate": 0.00022479999999999996, + "loss": 5.6124, + "step": 3920 + }, + { + "epoch": 6.2736, + "grad_norm": 0.0, + "learning_rate": 0.00022459999999999998, + "loss": 6.1253, + "step": 3921 + }, + { + "epoch": 6.2752, + "grad_norm": 0.0, + "learning_rate": 0.00022439999999999998, + "loss": 5.8437, + "step": 3922 + }, + { + "epoch": 6.2768, + "grad_norm": 0.0, + "learning_rate": 0.00022419999999999997, + "loss": 6.4418, + "step": 3923 + }, + { + "epoch": 6.2783999999999995, + "grad_norm": 0.0, + "learning_rate": 0.000224, + "loss": 6.9793, + "step": 3924 + }, + { + "epoch": 6.28, + "grad_norm": 0.0, + "learning_rate": 0.0002238, + "loss": 6.736, + "step": 3925 + }, + { + "epoch": 6.2816, + "grad_norm": 0.0, + "learning_rate": 0.00022359999999999996, + "loss": 8.1995, + "step": 3926 + }, + { + "epoch": 6.2832, + "grad_norm": 0.0, + "learning_rate": 0.00022339999999999998, + "loss": 5.2651, + "step": 3927 + }, + { + "epoch": 6.2848, + "grad_norm": 0.0, + "learning_rate": 0.00022319999999999998, + "loss": 4.5386, + "step": 3928 + }, + { + "epoch": 6.2864, + "grad_norm": 0.0, + "learning_rate": 0.00022299999999999997, + "loss": 5.3976, + "step": 3929 + }, + { + "epoch": 6.288, + "grad_norm": 0.0, + "learning_rate": 0.0002228, + "loss": 7.3623, + "step": 3930 + }, + { + "epoch": 6.2896, + "grad_norm": 0.0, + "learning_rate": 0.0002226, + "loss": 5.0797, + "step": 3931 + }, + { + "epoch": 6.2912, + "grad_norm": 0.0, + "learning_rate": 0.00022239999999999996, + "loss": 5.7404, + "step": 3932 + }, + { + "epoch": 6.2928, + "grad_norm": 0.0, + "learning_rate": 0.00022219999999999998, + "loss": 5.8032, + "step": 3933 + }, + { + "epoch": 6.2943999999999996, + "grad_norm": 0.0, + "learning_rate": 0.00022199999999999998, + "loss": 6.7893, + "step": 3934 + }, + { + "epoch": 6.296, + "grad_norm": 0.0, + "learning_rate": 0.00022179999999999997, + "loss": 5.3799, + "step": 3935 + }, + { + "epoch": 6.2976, + "grad_norm": 0.0, + "learning_rate": 0.0002216, + "loss": 4.5106, + "step": 3936 + }, + { + "epoch": 6.2992, + "grad_norm": 0.0, + "learning_rate": 0.0002214, + "loss": 5.7517, + "step": 3937 + }, + { + "epoch": 6.3008, + "grad_norm": 0.0, + "learning_rate": 0.00022119999999999996, + "loss": 5.4802, + "step": 3938 + }, + { + "epoch": 6.3024000000000004, + "grad_norm": 0.0, + "learning_rate": 0.00022099999999999998, + "loss": 4.6119, + "step": 3939 + }, + { + "epoch": 6.304, + "grad_norm": 0.0, + "learning_rate": 0.00022079999999999997, + "loss": 4.9958, + "step": 3940 + }, + { + "epoch": 6.3056, + "grad_norm": 0.0, + "learning_rate": 0.00022059999999999997, + "loss": 3.9698, + "step": 3941 + }, + { + "epoch": 6.3072, + "grad_norm": 0.0, + "learning_rate": 0.0002204, + "loss": 7.3422, + "step": 3942 + }, + { + "epoch": 6.3088, + "grad_norm": 0.0, + "learning_rate": 0.00022019999999999999, + "loss": 5.0724, + "step": 3943 + }, + { + "epoch": 6.3104, + "grad_norm": 0.0, + "learning_rate": 0.00021999999999999995, + "loss": 5.3685, + "step": 3944 + }, + { + "epoch": 6.312, + "grad_norm": 0.0, + "learning_rate": 0.00021979999999999998, + "loss": 3.855, + "step": 3945 + }, + { + "epoch": 6.3136, + "grad_norm": 0.0, + "learning_rate": 0.00021959999999999997, + "loss": 5.6462, + "step": 3946 + }, + { + "epoch": 6.3152, + "grad_norm": 0.0, + "learning_rate": 0.00021939999999999997, + "loss": 5.5764, + "step": 3947 + }, + { + "epoch": 6.3168, + "grad_norm": 0.0, + "learning_rate": 0.0002192, + "loss": 4.6013, + "step": 3948 + }, + { + "epoch": 6.3184000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00021899999999999998, + "loss": 5.538, + "step": 3949 + }, + { + "epoch": 6.32, + "grad_norm": 0.0, + "learning_rate": 0.00021879999999999995, + "loss": 5.0623, + "step": 3950 + }, + { + "epoch": 6.3216, + "grad_norm": 0.0, + "learning_rate": 0.00021859999999999997, + "loss": 7.6811, + "step": 3951 + }, + { + "epoch": 6.3232, + "grad_norm": 0.0, + "learning_rate": 0.00021839999999999997, + "loss": 7.0391, + "step": 3952 + }, + { + "epoch": 6.3248, + "grad_norm": 0.0, + "learning_rate": 0.0002182, + "loss": 8.1816, + "step": 3953 + }, + { + "epoch": 6.3264, + "grad_norm": 0.0, + "learning_rate": 0.00021799999999999999, + "loss": 6.2525, + "step": 3954 + }, + { + "epoch": 6.328, + "grad_norm": 0.0, + "learning_rate": 0.00021779999999999998, + "loss": 8.074, + "step": 3955 + }, + { + "epoch": 6.3296, + "grad_norm": 0.0, + "learning_rate": 0.0002176, + "loss": 7.0054, + "step": 3956 + }, + { + "epoch": 6.3312, + "grad_norm": 0.0, + "learning_rate": 0.0002174, + "loss": 7.4138, + "step": 3957 + }, + { + "epoch": 6.3328, + "grad_norm": 0.0, + "learning_rate": 0.00021719999999999997, + "loss": 6.3797, + "step": 3958 + }, + { + "epoch": 6.3344, + "grad_norm": 0.0, + "learning_rate": 0.000217, + "loss": 6.5496, + "step": 3959 + }, + { + "epoch": 6.336, + "grad_norm": 0.0, + "learning_rate": 0.00021679999999999998, + "loss": 5.8361, + "step": 3960 + }, + { + "epoch": 6.3376, + "grad_norm": 0.0, + "learning_rate": 0.00021659999999999998, + "loss": 7.2696, + "step": 3961 + }, + { + "epoch": 6.3392, + "grad_norm": 0.0, + "learning_rate": 0.0002164, + "loss": 5.327, + "step": 3962 + }, + { + "epoch": 6.3408, + "grad_norm": 0.0, + "learning_rate": 0.0002162, + "loss": 6.8369, + "step": 3963 + }, + { + "epoch": 6.3424, + "grad_norm": 0.0, + "learning_rate": 0.00021599999999999996, + "loss": 6.5091, + "step": 3964 + }, + { + "epoch": 6.344, + "grad_norm": 0.0, + "learning_rate": 0.0002158, + "loss": 5.7159, + "step": 3965 + }, + { + "epoch": 6.3456, + "grad_norm": 0.0, + "learning_rate": 0.00021559999999999998, + "loss": 4.9117, + "step": 3966 + }, + { + "epoch": 6.3472, + "grad_norm": 0.0, + "learning_rate": 0.00021539999999999998, + "loss": 4.9668, + "step": 3967 + }, + { + "epoch": 6.3488, + "grad_norm": 0.0, + "learning_rate": 0.0002152, + "loss": 4.935, + "step": 3968 + }, + { + "epoch": 6.3504, + "grad_norm": 0.0, + "learning_rate": 0.000215, + "loss": 6.7429, + "step": 3969 + }, + { + "epoch": 6.352, + "grad_norm": 0.0, + "learning_rate": 0.00021479999999999996, + "loss": 5.4014, + "step": 3970 + }, + { + "epoch": 6.3536, + "grad_norm": 0.0, + "learning_rate": 0.00021459999999999998, + "loss": 5.4004, + "step": 3971 + }, + { + "epoch": 6.3552, + "grad_norm": 0.0, + "learning_rate": 0.00021439999999999998, + "loss": 7.0612, + "step": 3972 + }, + { + "epoch": 6.3568, + "grad_norm": 0.0, + "learning_rate": 0.00021419999999999998, + "loss": 5.5819, + "step": 3973 + }, + { + "epoch": 6.3584, + "grad_norm": 0.0, + "learning_rate": 0.000214, + "loss": 6.9518, + "step": 3974 + }, + { + "epoch": 6.36, + "grad_norm": 0.0, + "learning_rate": 0.0002138, + "loss": 6.2981, + "step": 3975 + }, + { + "epoch": 6.3616, + "grad_norm": 0.0, + "learning_rate": 0.00021359999999999996, + "loss": 5.4611, + "step": 3976 + }, + { + "epoch": 6.3632, + "grad_norm": 0.0, + "learning_rate": 0.00021339999999999998, + "loss": 4.2251, + "step": 3977 + }, + { + "epoch": 6.3648, + "grad_norm": 0.0, + "learning_rate": 0.00021319999999999998, + "loss": 4.1801, + "step": 3978 + }, + { + "epoch": 6.3664, + "grad_norm": 0.0, + "learning_rate": 0.00021299999999999997, + "loss": 5.8451, + "step": 3979 + }, + { + "epoch": 6.368, + "grad_norm": 0.0, + "learning_rate": 0.0002128, + "loss": 4.215, + "step": 3980 + }, + { + "epoch": 6.3696, + "grad_norm": 0.0, + "learning_rate": 0.0002126, + "loss": 5.4232, + "step": 3981 + }, + { + "epoch": 6.3712, + "grad_norm": 0.0, + "learning_rate": 0.00021239999999999996, + "loss": 4.472, + "step": 3982 + }, + { + "epoch": 6.3728, + "grad_norm": 0.0, + "learning_rate": 0.00021219999999999998, + "loss": 4.2589, + "step": 3983 + }, + { + "epoch": 6.3744, + "grad_norm": 0.0, + "learning_rate": 0.00021199999999999998, + "loss": 6.2165, + "step": 3984 + }, + { + "epoch": 6.376, + "grad_norm": 0.0, + "learning_rate": 0.00021179999999999997, + "loss": 7.2336, + "step": 3985 + }, + { + "epoch": 6.3776, + "grad_norm": 0.0, + "learning_rate": 0.0002116, + "loss": 4.8142, + "step": 3986 + }, + { + "epoch": 6.3792, + "grad_norm": 0.0, + "learning_rate": 0.0002114, + "loss": 4.3796, + "step": 3987 + }, + { + "epoch": 6.3808, + "grad_norm": 0.0, + "learning_rate": 0.00021119999999999996, + "loss": 5.3237, + "step": 3988 + }, + { + "epoch": 6.3824, + "grad_norm": 0.0, + "learning_rate": 0.00021099999999999998, + "loss": 7.0998, + "step": 3989 + }, + { + "epoch": 6.384, + "grad_norm": 0.0, + "learning_rate": 0.00021079999999999997, + "loss": 10.7843, + "step": 3990 + }, + { + "epoch": 6.3856, + "grad_norm": 0.0, + "learning_rate": 0.00021059999999999997, + "loss": 4.2981, + "step": 3991 + }, + { + "epoch": 6.3872, + "grad_norm": 0.0, + "learning_rate": 0.0002104, + "loss": 5.1016, + "step": 3992 + }, + { + "epoch": 6.3888, + "grad_norm": 0.0, + "learning_rate": 0.0002102, + "loss": 4.9774, + "step": 3993 + }, + { + "epoch": 6.3904, + "grad_norm": 0.0, + "learning_rate": 0.00020999999999999998, + "loss": 5.4372, + "step": 3994 + }, + { + "epoch": 6.392, + "grad_norm": 0.0, + "learning_rate": 0.0002098, + "loss": 5.6382, + "step": 3995 + }, + { + "epoch": 6.3936, + "grad_norm": 0.0, + "learning_rate": 0.00020959999999999997, + "loss": 4.3206, + "step": 3996 + }, + { + "epoch": 6.3952, + "grad_norm": 0.0, + "learning_rate": 0.00020939999999999997, + "loss": 6.4536, + "step": 3997 + }, + { + "epoch": 6.3968, + "grad_norm": 0.0, + "learning_rate": 0.0002092, + "loss": 6.5461, + "step": 3998 + }, + { + "epoch": 6.3984, + "grad_norm": 0.0, + "learning_rate": 0.00020899999999999998, + "loss": 4.1852, + "step": 3999 + }, + { + "epoch": 6.4, + "grad_norm": 0.0, + "learning_rate": 0.00020879999999999998, + "loss": 5.0393, + "step": 4000 + }, + { + "epoch": 6.4016, + "grad_norm": 0.0, + "learning_rate": 0.0002086, + "loss": 8.1948, + "step": 4001 + }, + { + "epoch": 6.4032, + "grad_norm": 0.0, + "learning_rate": 0.00020839999999999997, + "loss": 8.0447, + "step": 4002 + }, + { + "epoch": 6.4048, + "grad_norm": 0.0, + "learning_rate": 0.00020819999999999996, + "loss": 6.3871, + "step": 4003 + }, + { + "epoch": 6.4064, + "grad_norm": 0.0, + "learning_rate": 0.000208, + "loss": 7.2012, + "step": 4004 + }, + { + "epoch": 6.408, + "grad_norm": 0.0, + "learning_rate": 0.00020779999999999998, + "loss": 5.7612, + "step": 4005 + }, + { + "epoch": 6.4096, + "grad_norm": 0.0, + "learning_rate": 0.00020759999999999998, + "loss": 6.3648, + "step": 4006 + }, + { + "epoch": 6.4112, + "grad_norm": 0.0, + "learning_rate": 0.0002074, + "loss": 6.5219, + "step": 4007 + }, + { + "epoch": 6.4128, + "grad_norm": 0.0, + "learning_rate": 0.00020719999999999997, + "loss": 5.8563, + "step": 4008 + }, + { + "epoch": 6.4144, + "grad_norm": 0.0, + "learning_rate": 0.00020699999999999996, + "loss": 6.4152, + "step": 4009 + }, + { + "epoch": 6.416, + "grad_norm": 0.0, + "learning_rate": 0.00020679999999999999, + "loss": 5.8432, + "step": 4010 + }, + { + "epoch": 6.4176, + "grad_norm": 0.0, + "learning_rate": 0.00020659999999999998, + "loss": 5.7853, + "step": 4011 + }, + { + "epoch": 6.4192, + "grad_norm": 0.0, + "learning_rate": 0.00020639999999999998, + "loss": 5.6198, + "step": 4012 + }, + { + "epoch": 6.4208, + "grad_norm": 0.0, + "learning_rate": 0.0002062, + "loss": 5.1304, + "step": 4013 + }, + { + "epoch": 6.4224, + "grad_norm": 0.0, + "learning_rate": 0.00020599999999999997, + "loss": 7.2229, + "step": 4014 + }, + { + "epoch": 6.424, + "grad_norm": 0.0, + "learning_rate": 0.0002058, + "loss": 5.5285, + "step": 4015 + }, + { + "epoch": 6.4256, + "grad_norm": 0.0, + "learning_rate": 0.00020559999999999998, + "loss": 5.1019, + "step": 4016 + }, + { + "epoch": 6.4272, + "grad_norm": 0.0, + "learning_rate": 0.00020539999999999998, + "loss": 5.5576, + "step": 4017 + }, + { + "epoch": 6.4288, + "grad_norm": 0.0, + "learning_rate": 0.0002052, + "loss": 6.8628, + "step": 4018 + }, + { + "epoch": 6.4304, + "grad_norm": 0.0, + "learning_rate": 0.000205, + "loss": 6.0875, + "step": 4019 + }, + { + "epoch": 6.432, + "grad_norm": 0.0, + "learning_rate": 0.00020479999999999996, + "loss": 5.3733, + "step": 4020 + }, + { + "epoch": 6.4336, + "grad_norm": 0.0, + "learning_rate": 0.00020459999999999999, + "loss": 8.7083, + "step": 4021 + }, + { + "epoch": 6.4352, + "grad_norm": 0.0, + "learning_rate": 0.00020439999999999998, + "loss": 5.2477, + "step": 4022 + }, + { + "epoch": 6.4368, + "grad_norm": 0.0, + "learning_rate": 0.00020419999999999998, + "loss": 6.3021, + "step": 4023 + }, + { + "epoch": 6.4384, + "grad_norm": 0.0, + "learning_rate": 0.000204, + "loss": 5.0333, + "step": 4024 + }, + { + "epoch": 6.44, + "grad_norm": 0.0, + "learning_rate": 0.0002038, + "loss": 5.3617, + "step": 4025 + }, + { + "epoch": 6.4416, + "grad_norm": 0.0, + "learning_rate": 0.00020359999999999996, + "loss": 4.8622, + "step": 4026 + }, + { + "epoch": 6.4432, + "grad_norm": 0.0, + "learning_rate": 0.00020339999999999998, + "loss": 3.9343, + "step": 4027 + }, + { + "epoch": 6.4448, + "grad_norm": 0.0, + "learning_rate": 0.00020319999999999998, + "loss": 5.0388, + "step": 4028 + }, + { + "epoch": 6.4464, + "grad_norm": 0.0, + "learning_rate": 0.00020299999999999997, + "loss": 5.4971, + "step": 4029 + }, + { + "epoch": 6.448, + "grad_norm": 0.0, + "learning_rate": 0.0002028, + "loss": 7.1424, + "step": 4030 + }, + { + "epoch": 6.4496, + "grad_norm": 0.0, + "learning_rate": 0.0002026, + "loss": 4.6645, + "step": 4031 + }, + { + "epoch": 6.4512, + "grad_norm": 0.0, + "learning_rate": 0.0002024, + "loss": 5.356, + "step": 4032 + }, + { + "epoch": 6.4528, + "grad_norm": 0.0, + "learning_rate": 0.0002022, + "loss": 5.2623, + "step": 4033 + }, + { + "epoch": 6.4544, + "grad_norm": 0.0, + "learning_rate": 0.00020199999999999998, + "loss": 5.3529, + "step": 4034 + }, + { + "epoch": 6.456, + "grad_norm": 0.0, + "learning_rate": 0.00020179999999999997, + "loss": 5.9766, + "step": 4035 + }, + { + "epoch": 6.4576, + "grad_norm": 0.0, + "learning_rate": 0.0002016, + "loss": 5.8505, + "step": 4036 + }, + { + "epoch": 6.4592, + "grad_norm": 0.0, + "learning_rate": 0.0002014, + "loss": 4.9831, + "step": 4037 + }, + { + "epoch": 6.4608, + "grad_norm": 0.0, + "learning_rate": 0.00020119999999999998, + "loss": 5.3228, + "step": 4038 + }, + { + "epoch": 6.4624, + "grad_norm": 0.0, + "learning_rate": 0.000201, + "loss": 5.0874, + "step": 4039 + }, + { + "epoch": 6.464, + "grad_norm": 0.0, + "learning_rate": 0.00020079999999999997, + "loss": 5.1374, + "step": 4040 + }, + { + "epoch": 6.4656, + "grad_norm": 0.0, + "learning_rate": 0.00020059999999999997, + "loss": 5.4303, + "step": 4041 + }, + { + "epoch": 6.4672, + "grad_norm": 0.0, + "learning_rate": 0.0002004, + "loss": 5.0008, + "step": 4042 + }, + { + "epoch": 6.4688, + "grad_norm": 0.0, + "learning_rate": 0.0002002, + "loss": 4.9524, + "step": 4043 + }, + { + "epoch": 6.4704, + "grad_norm": 0.0, + "learning_rate": 0.00019999999999999998, + "loss": 6.077, + "step": 4044 + }, + { + "epoch": 6.4719999999999995, + "grad_norm": 0.0, + "learning_rate": 0.0001998, + "loss": 4.5066, + "step": 4045 + }, + { + "epoch": 6.4736, + "grad_norm": 0.0, + "learning_rate": 0.00019959999999999997, + "loss": 4.5129, + "step": 4046 + }, + { + "epoch": 6.4752, + "grad_norm": 0.0, + "learning_rate": 0.00019939999999999997, + "loss": 5.7921, + "step": 4047 + }, + { + "epoch": 6.4768, + "grad_norm": 0.0, + "learning_rate": 0.0001992, + "loss": 5.525, + "step": 4048 + }, + { + "epoch": 6.4784, + "grad_norm": 0.0, + "learning_rate": 0.00019899999999999999, + "loss": 6.5494, + "step": 4049 + }, + { + "epoch": 6.48, + "grad_norm": 0.0, + "learning_rate": 0.00019879999999999998, + "loss": 10.3255, + "step": 4050 + }, + { + "epoch": 6.4816, + "grad_norm": 0.0, + "learning_rate": 0.0001986, + "loss": 9.0295, + "step": 4051 + }, + { + "epoch": 6.4832, + "grad_norm": 0.0, + "learning_rate": 0.00019839999999999997, + "loss": 7.7972, + "step": 4052 + }, + { + "epoch": 6.4848, + "grad_norm": 0.0, + "learning_rate": 0.00019819999999999997, + "loss": 5.952, + "step": 4053 + }, + { + "epoch": 6.4864, + "grad_norm": 0.0, + "learning_rate": 0.000198, + "loss": 6.4036, + "step": 4054 + }, + { + "epoch": 6.4879999999999995, + "grad_norm": 0.0, + "learning_rate": 0.00019779999999999998, + "loss": 6.8318, + "step": 4055 + }, + { + "epoch": 6.4896, + "grad_norm": 0.0, + "learning_rate": 0.00019759999999999998, + "loss": 9.6484, + "step": 4056 + }, + { + "epoch": 6.4912, + "grad_norm": 0.0, + "learning_rate": 0.0001974, + "loss": 7.1129, + "step": 4057 + }, + { + "epoch": 6.4928, + "grad_norm": 0.0, + "learning_rate": 0.00019719999999999997, + "loss": 6.464, + "step": 4058 + }, + { + "epoch": 6.4944, + "grad_norm": 0.0, + "learning_rate": 0.00019699999999999996, + "loss": 5.1724, + "step": 4059 + }, + { + "epoch": 6.496, + "grad_norm": 0.0, + "learning_rate": 0.00019679999999999999, + "loss": 4.8296, + "step": 4060 + }, + { + "epoch": 6.4976, + "grad_norm": 0.0, + "learning_rate": 0.00019659999999999998, + "loss": 7.1017, + "step": 4061 + }, + { + "epoch": 6.4992, + "grad_norm": 0.0, + "learning_rate": 0.00019639999999999998, + "loss": 5.4083, + "step": 4062 + }, + { + "epoch": 6.5008, + "grad_norm": 0.0, + "learning_rate": 0.0001962, + "loss": 6.0432, + "step": 4063 + }, + { + "epoch": 6.5024, + "grad_norm": 0.0, + "learning_rate": 0.00019599999999999997, + "loss": 4.9672, + "step": 4064 + }, + { + "epoch": 6.504, + "grad_norm": 0.0, + "learning_rate": 0.00019579999999999996, + "loss": 5.8192, + "step": 4065 + }, + { + "epoch": 6.5056, + "grad_norm": 0.0, + "learning_rate": 0.00019559999999999998, + "loss": 6.0436, + "step": 4066 + }, + { + "epoch": 6.5072, + "grad_norm": 0.0, + "learning_rate": 0.00019539999999999998, + "loss": 6.5994, + "step": 4067 + }, + { + "epoch": 6.5088, + "grad_norm": 0.0, + "learning_rate": 0.00019519999999999997, + "loss": 5.4782, + "step": 4068 + }, + { + "epoch": 6.5104, + "grad_norm": 0.0, + "learning_rate": 0.000195, + "loss": 6.522, + "step": 4069 + }, + { + "epoch": 6.5120000000000005, + "grad_norm": 0.0, + "learning_rate": 0.0001948, + "loss": 4.8754, + "step": 4070 + }, + { + "epoch": 6.5136, + "grad_norm": 0.0, + "learning_rate": 0.00019459999999999996, + "loss": 8.1075, + "step": 4071 + }, + { + "epoch": 6.5152, + "grad_norm": 0.0, + "learning_rate": 0.00019439999999999998, + "loss": 5.2561, + "step": 4072 + }, + { + "epoch": 6.5168, + "grad_norm": 0.0, + "learning_rate": 0.00019419999999999998, + "loss": 4.8187, + "step": 4073 + }, + { + "epoch": 6.5184, + "grad_norm": 0.0, + "learning_rate": 0.00019399999999999997, + "loss": 5.0993, + "step": 4074 + }, + { + "epoch": 6.52, + "grad_norm": 0.0, + "learning_rate": 0.0001938, + "loss": 4.854, + "step": 4075 + }, + { + "epoch": 6.5216, + "grad_norm": 0.0, + "learning_rate": 0.0001936, + "loss": 4.5327, + "step": 4076 + }, + { + "epoch": 6.5232, + "grad_norm": 0.0, + "learning_rate": 0.0001934, + "loss": 5.0608, + "step": 4077 + }, + { + "epoch": 6.5248, + "grad_norm": 0.0, + "learning_rate": 0.00019319999999999998, + "loss": 5.0575, + "step": 4078 + }, + { + "epoch": 6.5264, + "grad_norm": 0.0, + "learning_rate": 0.00019299999999999997, + "loss": 5.2626, + "step": 4079 + }, + { + "epoch": 6.5280000000000005, + "grad_norm": 0.0, + "learning_rate": 0.0001928, + "loss": 5.4511, + "step": 4080 + }, + { + "epoch": 6.5296, + "grad_norm": 0.0, + "learning_rate": 0.0001926, + "loss": 5.7777, + "step": 4081 + }, + { + "epoch": 6.5312, + "grad_norm": 0.0, + "learning_rate": 0.0001924, + "loss": 6.0821, + "step": 4082 + }, + { + "epoch": 6.5328, + "grad_norm": 0.0, + "learning_rate": 0.0001922, + "loss": 6.9274, + "step": 4083 + }, + { + "epoch": 6.5344, + "grad_norm": 0.0, + "learning_rate": 0.00019199999999999998, + "loss": 5.6503, + "step": 4084 + }, + { + "epoch": 6.536, + "grad_norm": 0.0, + "learning_rate": 0.00019179999999999997, + "loss": 4.2726, + "step": 4085 + }, + { + "epoch": 6.5376, + "grad_norm": 0.0, + "learning_rate": 0.0001916, + "loss": 5.2065, + "step": 4086 + }, + { + "epoch": 6.5392, + "grad_norm": 0.0, + "learning_rate": 0.0001914, + "loss": 5.9331, + "step": 4087 + }, + { + "epoch": 6.5408, + "grad_norm": 0.0, + "learning_rate": 0.00019119999999999999, + "loss": 4.5911, + "step": 4088 + }, + { + "epoch": 6.5424, + "grad_norm": 0.0, + "learning_rate": 0.000191, + "loss": 6.1543, + "step": 4089 + }, + { + "epoch": 6.5440000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00019079999999999998, + "loss": 5.2918, + "step": 4090 + }, + { + "epoch": 6.5456, + "grad_norm": 0.0, + "learning_rate": 0.00019059999999999997, + "loss": 5.9222, + "step": 4091 + }, + { + "epoch": 6.5472, + "grad_norm": 0.0, + "learning_rate": 0.0001904, + "loss": 6.6886, + "step": 4092 + }, + { + "epoch": 6.5488, + "grad_norm": 0.0, + "learning_rate": 0.0001902, + "loss": 5.3827, + "step": 4093 + }, + { + "epoch": 6.5504, + "grad_norm": 0.0, + "learning_rate": 0.00018999999999999998, + "loss": 5.7445, + "step": 4094 + }, + { + "epoch": 6.552, + "grad_norm": 0.0, + "learning_rate": 0.0001898, + "loss": 6.8291, + "step": 4095 + }, + { + "epoch": 6.5536, + "grad_norm": 0.0, + "learning_rate": 0.00018959999999999997, + "loss": 5.0083, + "step": 4096 + }, + { + "epoch": 6.5552, + "grad_norm": 0.0, + "learning_rate": 0.00018939999999999997, + "loss": 5.8043, + "step": 4097 + }, + { + "epoch": 6.5568, + "grad_norm": 0.0, + "learning_rate": 0.0001892, + "loss": 4.7089, + "step": 4098 + }, + { + "epoch": 6.5584, + "grad_norm": 0.0, + "learning_rate": 0.00018899999999999999, + "loss": 6.6991, + "step": 4099 + }, + { + "epoch": 6.5600000000000005, + "grad_norm": NaN, + "learning_rate": 0.00018899999999999999, + "loss": 6.1105, + "step": 4100 + }, + { + "epoch": 6.5616, + "grad_norm": 0.0, + "learning_rate": 0.00018879999999999998, + "loss": 9.2272, + "step": 4101 + }, + { + "epoch": 6.5632, + "grad_norm": 0.0, + "learning_rate": 0.0001886, + "loss": 7.4793, + "step": 4102 + }, + { + "epoch": 6.5648, + "grad_norm": 0.0, + "learning_rate": 0.00018839999999999997, + "loss": 6.9356, + "step": 4103 + }, + { + "epoch": 6.5664, + "grad_norm": 0.0, + "learning_rate": 0.00018819999999999997, + "loss": 6.5929, + "step": 4104 + }, + { + "epoch": 6.568, + "grad_norm": 0.0, + "learning_rate": 0.000188, + "loss": 6.0832, + "step": 4105 + }, + { + "epoch": 6.5696, + "grad_norm": 0.0, + "learning_rate": 0.00018779999999999998, + "loss": 6.6209, + "step": 4106 + }, + { + "epoch": 6.5712, + "grad_norm": 0.0, + "learning_rate": 0.00018759999999999998, + "loss": 6.8724, + "step": 4107 + }, + { + "epoch": 6.5728, + "grad_norm": 0.0, + "learning_rate": 0.0001874, + "loss": 6.5322, + "step": 4108 + }, + { + "epoch": 6.5744, + "grad_norm": 0.0, + "learning_rate": 0.0001872, + "loss": 6.646, + "step": 4109 + }, + { + "epoch": 6.576, + "grad_norm": 0.0, + "learning_rate": 0.00018699999999999996, + "loss": 5.9154, + "step": 4110 + }, + { + "epoch": 6.5776, + "grad_norm": 0.0, + "learning_rate": 0.0001868, + "loss": 6.5474, + "step": 4111 + }, + { + "epoch": 6.5792, + "grad_norm": 0.0, + "learning_rate": 0.00018659999999999998, + "loss": 6.5186, + "step": 4112 + }, + { + "epoch": 6.5808, + "grad_norm": 0.0, + "learning_rate": 0.00018639999999999998, + "loss": 4.9023, + "step": 4113 + }, + { + "epoch": 6.5824, + "grad_norm": 0.0, + "learning_rate": 0.0001862, + "loss": 8.0242, + "step": 4114 + }, + { + "epoch": 6.584, + "grad_norm": 0.0, + "learning_rate": 0.000186, + "loss": 4.9214, + "step": 4115 + }, + { + "epoch": 6.5856, + "grad_norm": 0.0, + "learning_rate": 0.00018579999999999996, + "loss": 4.7957, + "step": 4116 + }, + { + "epoch": 6.5872, + "grad_norm": 0.0, + "learning_rate": 0.00018559999999999998, + "loss": 5.8198, + "step": 4117 + }, + { + "epoch": 6.5888, + "grad_norm": 0.0, + "learning_rate": 0.00018539999999999998, + "loss": 4.7452, + "step": 4118 + }, + { + "epoch": 6.5904, + "grad_norm": 0.0, + "learning_rate": 0.00018519999999999998, + "loss": 4.9572, + "step": 4119 + }, + { + "epoch": 6.592, + "grad_norm": 0.0, + "learning_rate": 0.000185, + "loss": 5.5537, + "step": 4120 + }, + { + "epoch": 6.5936, + "grad_norm": 0.0, + "learning_rate": 0.0001848, + "loss": 6.972, + "step": 4121 + }, + { + "epoch": 6.5952, + "grad_norm": 0.0, + "learning_rate": 0.00018459999999999996, + "loss": 4.9161, + "step": 4122 + }, + { + "epoch": 6.5968, + "grad_norm": 0.0, + "learning_rate": 0.00018439999999999998, + "loss": 4.8701, + "step": 4123 + }, + { + "epoch": 6.5984, + "grad_norm": 0.0, + "learning_rate": 0.00018419999999999998, + "loss": 6.4972, + "step": 4124 + }, + { + "epoch": 6.6, + "grad_norm": 0.0, + "learning_rate": 0.00018399999999999997, + "loss": 5.1413, + "step": 4125 + }, + { + "epoch": 6.6016, + "grad_norm": 0.0, + "learning_rate": 0.0001838, + "loss": 4.9516, + "step": 4126 + }, + { + "epoch": 6.6032, + "grad_norm": 0.0, + "learning_rate": 0.0001836, + "loss": 4.9221, + "step": 4127 + }, + { + "epoch": 6.6048, + "grad_norm": 0.0, + "learning_rate": 0.00018339999999999996, + "loss": 5.3704, + "step": 4128 + }, + { + "epoch": 6.6064, + "grad_norm": 0.0, + "learning_rate": 0.00018319999999999998, + "loss": 5.7861, + "step": 4129 + }, + { + "epoch": 6.608, + "grad_norm": 0.0, + "learning_rate": 0.00018299999999999998, + "loss": 4.5468, + "step": 4130 + }, + { + "epoch": 6.6096, + "grad_norm": 0.0, + "learning_rate": 0.00018279999999999997, + "loss": 5.9016, + "step": 4131 + }, + { + "epoch": 6.6112, + "grad_norm": 0.0, + "learning_rate": 0.0001826, + "loss": 4.3465, + "step": 4132 + }, + { + "epoch": 6.6128, + "grad_norm": 0.0, + "learning_rate": 0.0001824, + "loss": 4.9091, + "step": 4133 + }, + { + "epoch": 6.6144, + "grad_norm": 0.0, + "learning_rate": 0.00018219999999999996, + "loss": 5.5963, + "step": 4134 + }, + { + "epoch": 6.616, + "grad_norm": 0.0, + "learning_rate": 0.00018199999999999998, + "loss": 4.1554, + "step": 4135 + }, + { + "epoch": 6.6176, + "grad_norm": 0.0, + "learning_rate": 0.00018179999999999997, + "loss": 7.1316, + "step": 4136 + }, + { + "epoch": 6.6192, + "grad_norm": 0.0, + "learning_rate": 0.00018159999999999997, + "loss": 6.0276, + "step": 4137 + }, + { + "epoch": 6.6208, + "grad_norm": 0.0, + "learning_rate": 0.0001814, + "loss": 8.7214, + "step": 4138 + }, + { + "epoch": 6.6224, + "grad_norm": 0.0, + "learning_rate": 0.00018119999999999999, + "loss": 4.9247, + "step": 4139 + }, + { + "epoch": 6.624, + "grad_norm": 0.0, + "learning_rate": 0.000181, + "loss": 6.2757, + "step": 4140 + }, + { + "epoch": 6.6256, + "grad_norm": 0.0, + "learning_rate": 0.00018079999999999998, + "loss": 6.2719, + "step": 4141 + }, + { + "epoch": 6.6272, + "grad_norm": 0.0, + "learning_rate": 0.00018059999999999997, + "loss": 4.5367, + "step": 4142 + }, + { + "epoch": 6.6288, + "grad_norm": 0.0, + "learning_rate": 0.0001804, + "loss": 5.4118, + "step": 4143 + }, + { + "epoch": 6.6304, + "grad_norm": 0.0, + "learning_rate": 0.0001802, + "loss": 5.4421, + "step": 4144 + }, + { + "epoch": 6.632, + "grad_norm": 0.0, + "learning_rate": 0.00017999999999999998, + "loss": 5.4491, + "step": 4145 + }, + { + "epoch": 6.6336, + "grad_norm": 0.0, + "learning_rate": 0.0001798, + "loss": 4.124, + "step": 4146 + }, + { + "epoch": 6.6352, + "grad_norm": 0.0, + "learning_rate": 0.0001796, + "loss": 5.9915, + "step": 4147 + }, + { + "epoch": 6.6368, + "grad_norm": 0.0, + "learning_rate": 0.00017939999999999997, + "loss": 6.7769, + "step": 4148 + }, + { + "epoch": 6.6384, + "grad_norm": 0.0, + "learning_rate": 0.0001792, + "loss": 4.9773, + "step": 4149 + }, + { + "epoch": 6.64, + "grad_norm": 0.0, + "learning_rate": 0.000179, + "loss": 7.5078, + "step": 4150 + }, + { + "epoch": 6.6416, + "grad_norm": 0.0, + "learning_rate": 0.00017879999999999998, + "loss": 6.2653, + "step": 4151 + }, + { + "epoch": 6.6432, + "grad_norm": 0.0, + "learning_rate": 0.0001786, + "loss": 9.3182, + "step": 4152 + }, + { + "epoch": 6.6448, + "grad_norm": 0.0, + "learning_rate": 0.0001784, + "loss": 7.01, + "step": 4153 + }, + { + "epoch": 6.6464, + "grad_norm": 0.0, + "learning_rate": 0.00017819999999999997, + "loss": 7.6461, + "step": 4154 + }, + { + "epoch": 6.648, + "grad_norm": 0.0, + "learning_rate": 0.000178, + "loss": 6.3456, + "step": 4155 + }, + { + "epoch": 6.6495999999999995, + "grad_norm": 0.0, + "learning_rate": 0.00017779999999999998, + "loss": 6.2202, + "step": 4156 + }, + { + "epoch": 6.6512, + "grad_norm": 0.0, + "learning_rate": 0.00017759999999999998, + "loss": 6.0753, + "step": 4157 + }, + { + "epoch": 6.6528, + "grad_norm": 0.0, + "learning_rate": 0.0001774, + "loss": 6.4579, + "step": 4158 + }, + { + "epoch": 6.6544, + "grad_norm": 0.0, + "learning_rate": 0.0001772, + "loss": 7.1686, + "step": 4159 + }, + { + "epoch": 6.656, + "grad_norm": 0.0, + "learning_rate": 0.00017699999999999997, + "loss": 5.7642, + "step": 4160 + }, + { + "epoch": 6.6576, + "grad_norm": 0.0, + "learning_rate": 0.0001768, + "loss": 7.1619, + "step": 4161 + }, + { + "epoch": 6.6592, + "grad_norm": 0.0, + "learning_rate": 0.00017659999999999998, + "loss": 5.6074, + "step": 4162 + }, + { + "epoch": 6.6608, + "grad_norm": 0.0, + "learning_rate": 0.00017639999999999998, + "loss": 5.8866, + "step": 4163 + }, + { + "epoch": 6.6624, + "grad_norm": 0.0, + "learning_rate": 0.0001762, + "loss": 5.7208, + "step": 4164 + }, + { + "epoch": 6.664, + "grad_norm": 0.0, + "learning_rate": 0.000176, + "loss": 5.5302, + "step": 4165 + }, + { + "epoch": 6.6655999999999995, + "grad_norm": 0.0, + "learning_rate": 0.00017579999999999996, + "loss": 5.1154, + "step": 4166 + }, + { + "epoch": 6.6672, + "grad_norm": 0.0, + "learning_rate": 0.00017559999999999999, + "loss": 5.1785, + "step": 4167 + }, + { + "epoch": 6.6688, + "grad_norm": 0.0, + "learning_rate": 0.00017539999999999998, + "loss": 4.88, + "step": 4168 + }, + { + "epoch": 6.6704, + "grad_norm": 0.0, + "learning_rate": 0.00017519999999999998, + "loss": 4.7664, + "step": 4169 + }, + { + "epoch": 6.672, + "grad_norm": 0.0, + "learning_rate": 0.000175, + "loss": 6.8315, + "step": 4170 + }, + { + "epoch": 6.6736, + "grad_norm": 0.0, + "learning_rate": 0.0001748, + "loss": 5.391, + "step": 4171 + }, + { + "epoch": 6.6752, + "grad_norm": 0.0, + "learning_rate": 0.00017459999999999996, + "loss": 4.8035, + "step": 4172 + }, + { + "epoch": 6.6768, + "grad_norm": 0.0, + "learning_rate": 0.00017439999999999998, + "loss": 5.0824, + "step": 4173 + }, + { + "epoch": 6.6784, + "grad_norm": 0.0, + "learning_rate": 0.00017419999999999998, + "loss": 5.1261, + "step": 4174 + }, + { + "epoch": 6.68, + "grad_norm": 0.0, + "learning_rate": 0.00017399999999999997, + "loss": 4.7606, + "step": 4175 + }, + { + "epoch": 6.6815999999999995, + "grad_norm": 0.0, + "learning_rate": 0.0001738, + "loss": 6.4392, + "step": 4176 + }, + { + "epoch": 6.6832, + "grad_norm": 0.0, + "learning_rate": 0.0001736, + "loss": 4.7189, + "step": 4177 + }, + { + "epoch": 6.6848, + "grad_norm": 0.0, + "learning_rate": 0.00017339999999999996, + "loss": 6.0987, + "step": 4178 + }, + { + "epoch": 6.6864, + "grad_norm": 0.0, + "learning_rate": 0.00017319999999999998, + "loss": 5.8903, + "step": 4179 + }, + { + "epoch": 6.688, + "grad_norm": 0.0, + "learning_rate": 0.00017299999999999998, + "loss": 5.6914, + "step": 4180 + }, + { + "epoch": 6.6896, + "grad_norm": 0.0, + "learning_rate": 0.00017279999999999997, + "loss": 5.9072, + "step": 4181 + }, + { + "epoch": 6.6912, + "grad_norm": 0.0, + "learning_rate": 0.0001726, + "loss": 7.1752, + "step": 4182 + }, + { + "epoch": 6.6928, + "grad_norm": 0.0, + "learning_rate": 0.0001724, + "loss": 5.0147, + "step": 4183 + }, + { + "epoch": 6.6944, + "grad_norm": 0.0, + "learning_rate": 0.00017219999999999998, + "loss": 6.0951, + "step": 4184 + }, + { + "epoch": 6.696, + "grad_norm": 0.0, + "learning_rate": 0.000172, + "loss": 5.668, + "step": 4185 + }, + { + "epoch": 6.6975999999999996, + "grad_norm": 0.0, + "learning_rate": 0.00017179999999999997, + "loss": 4.5645, + "step": 4186 + }, + { + "epoch": 6.6992, + "grad_norm": 0.0, + "learning_rate": 0.00017159999999999997, + "loss": 5.1264, + "step": 4187 + }, + { + "epoch": 6.7008, + "grad_norm": 0.0, + "learning_rate": 0.0001714, + "loss": 4.6198, + "step": 4188 + }, + { + "epoch": 6.7024, + "grad_norm": 0.0, + "learning_rate": 0.0001712, + "loss": 5.0099, + "step": 4189 + }, + { + "epoch": 6.704, + "grad_norm": 0.0, + "learning_rate": 0.00017099999999999998, + "loss": 7.4505, + "step": 4190 + }, + { + "epoch": 6.7056000000000004, + "grad_norm": 0.0, + "learning_rate": 0.0001708, + "loss": 5.4084, + "step": 4191 + }, + { + "epoch": 6.7072, + "grad_norm": 0.0, + "learning_rate": 0.00017059999999999997, + "loss": 5.5187, + "step": 4192 + }, + { + "epoch": 6.7088, + "grad_norm": 0.0, + "learning_rate": 0.00017039999999999997, + "loss": 5.7204, + "step": 4193 + }, + { + "epoch": 6.7104, + "grad_norm": 0.0, + "learning_rate": 0.0001702, + "loss": 4.2576, + "step": 4194 + }, + { + "epoch": 6.712, + "grad_norm": 0.0, + "learning_rate": 0.00016999999999999999, + "loss": 5.0965, + "step": 4195 + }, + { + "epoch": 6.7136, + "grad_norm": 0.0, + "learning_rate": 0.00016979999999999998, + "loss": 5.4783, + "step": 4196 + }, + { + "epoch": 6.7152, + "grad_norm": 0.0, + "learning_rate": 0.0001696, + "loss": 4.2457, + "step": 4197 + }, + { + "epoch": 6.7168, + "grad_norm": 0.0, + "learning_rate": 0.00016939999999999997, + "loss": 3.6698, + "step": 4198 + }, + { + "epoch": 6.7184, + "grad_norm": 0.0, + "learning_rate": 0.00016919999999999997, + "loss": 5.668, + "step": 4199 + }, + { + "epoch": 6.72, + "grad_norm": NaN, + "learning_rate": 0.00016919999999999997, + "loss": 5.0759, + "step": 4200 + }, + { + "epoch": 6.7216000000000005, + "grad_norm": 0.0, + "learning_rate": 0.000169, + "loss": 8.7148, + "step": 4201 + }, + { + "epoch": 6.7232, + "grad_norm": 0.0, + "learning_rate": 0.00016879999999999998, + "loss": 6.4265, + "step": 4202 + }, + { + "epoch": 6.7248, + "grad_norm": 0.0, + "learning_rate": 0.0001686, + "loss": 6.1731, + "step": 4203 + }, + { + "epoch": 6.7264, + "grad_norm": 0.0, + "learning_rate": 0.0001684, + "loss": 6.1314, + "step": 4204 + }, + { + "epoch": 6.728, + "grad_norm": 0.0, + "learning_rate": 0.00016819999999999997, + "loss": 7.3555, + "step": 4205 + }, + { + "epoch": 6.7296, + "grad_norm": 0.0, + "learning_rate": 0.000168, + "loss": 4.8547, + "step": 4206 + }, + { + "epoch": 6.7312, + "grad_norm": 0.0, + "learning_rate": 0.00016779999999999999, + "loss": 4.3774, + "step": 4207 + }, + { + "epoch": 6.7328, + "grad_norm": 0.0, + "learning_rate": 0.00016759999999999998, + "loss": 6.8963, + "step": 4208 + }, + { + "epoch": 6.7344, + "grad_norm": 0.0, + "learning_rate": 0.0001674, + "loss": 5.0592, + "step": 4209 + }, + { + "epoch": 6.736, + "grad_norm": 0.0, + "learning_rate": 0.0001672, + "loss": 5.3738, + "step": 4210 + }, + { + "epoch": 6.7376000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00016699999999999997, + "loss": 5.43, + "step": 4211 + }, + { + "epoch": 6.7392, + "grad_norm": 0.0, + "learning_rate": 0.0001668, + "loss": 6.2524, + "step": 4212 + }, + { + "epoch": 6.7408, + "grad_norm": 0.0, + "learning_rate": 0.00016659999999999998, + "loss": 5.1483, + "step": 4213 + }, + { + "epoch": 6.7424, + "grad_norm": 0.0, + "learning_rate": 0.00016639999999999998, + "loss": 6.0518, + "step": 4214 + }, + { + "epoch": 6.744, + "grad_norm": 0.0, + "learning_rate": 0.0001662, + "loss": 6.0559, + "step": 4215 + }, + { + "epoch": 6.7456, + "grad_norm": 0.0, + "learning_rate": 0.000166, + "loss": 4.7188, + "step": 4216 + }, + { + "epoch": 6.7472, + "grad_norm": 0.0, + "learning_rate": 0.00016579999999999996, + "loss": 4.9477, + "step": 4217 + }, + { + "epoch": 6.7488, + "grad_norm": 0.0, + "learning_rate": 0.0001656, + "loss": 5.7809, + "step": 4218 + }, + { + "epoch": 6.7504, + "grad_norm": 0.0, + "learning_rate": 0.00016539999999999998, + "loss": 6.1003, + "step": 4219 + }, + { + "epoch": 6.752, + "grad_norm": 0.0, + "learning_rate": 0.00016519999999999998, + "loss": 5.268, + "step": 4220 + }, + { + "epoch": 6.7536000000000005, + "grad_norm": 0.0, + "learning_rate": 0.000165, + "loss": 6.0587, + "step": 4221 + }, + { + "epoch": 6.7552, + "grad_norm": 0.0, + "learning_rate": 0.0001648, + "loss": 5.2584, + "step": 4222 + }, + { + "epoch": 6.7568, + "grad_norm": 0.0, + "learning_rate": 0.0001646, + "loss": 5.2158, + "step": 4223 + }, + { + "epoch": 6.7584, + "grad_norm": 0.0, + "learning_rate": 0.0001644, + "loss": 6.5743, + "step": 4224 + }, + { + "epoch": 6.76, + "grad_norm": 0.0, + "learning_rate": 0.00016419999999999998, + "loss": 6.0184, + "step": 4225 + }, + { + "epoch": 6.7616, + "grad_norm": 0.0, + "learning_rate": 0.00016399999999999997, + "loss": 6.0504, + "step": 4226 + }, + { + "epoch": 6.7632, + "grad_norm": 0.0, + "learning_rate": 0.0001638, + "loss": 5.21, + "step": 4227 + }, + { + "epoch": 6.7648, + "grad_norm": 0.0, + "learning_rate": 0.0001636, + "loss": 4.6084, + "step": 4228 + }, + { + "epoch": 6.7664, + "grad_norm": 0.0, + "learning_rate": 0.0001634, + "loss": 4.1066, + "step": 4229 + }, + { + "epoch": 6.768, + "grad_norm": 0.0, + "learning_rate": 0.0001632, + "loss": 4.5355, + "step": 4230 + }, + { + "epoch": 6.7696, + "grad_norm": 0.0, + "learning_rate": 0.00016299999999999998, + "loss": 5.1876, + "step": 4231 + }, + { + "epoch": 6.7712, + "grad_norm": 0.0, + "learning_rate": 0.00016279999999999997, + "loss": 4.274, + "step": 4232 + }, + { + "epoch": 6.7728, + "grad_norm": 0.0, + "learning_rate": 0.0001626, + "loss": 5.8047, + "step": 4233 + }, + { + "epoch": 6.7744, + "grad_norm": 0.0, + "learning_rate": 0.0001624, + "loss": 6.694, + "step": 4234 + }, + { + "epoch": 6.776, + "grad_norm": 0.0, + "learning_rate": 0.00016219999999999999, + "loss": 4.4464, + "step": 4235 + }, + { + "epoch": 6.7776, + "grad_norm": 0.0, + "learning_rate": 0.000162, + "loss": 5.0319, + "step": 4236 + }, + { + "epoch": 6.7792, + "grad_norm": 0.0, + "learning_rate": 0.00016179999999999998, + "loss": 5.3984, + "step": 4237 + }, + { + "epoch": 6.7808, + "grad_norm": 0.0, + "learning_rate": 0.00016159999999999997, + "loss": 5.5782, + "step": 4238 + }, + { + "epoch": 6.7824, + "grad_norm": 0.0, + "learning_rate": 0.0001614, + "loss": 4.3758, + "step": 4239 + }, + { + "epoch": 6.784, + "grad_norm": 0.0, + "learning_rate": 0.0001612, + "loss": 4.5905, + "step": 4240 + }, + { + "epoch": 6.7856, + "grad_norm": 0.0, + "learning_rate": 0.00016099999999999998, + "loss": 4.6257, + "step": 4241 + }, + { + "epoch": 6.7872, + "grad_norm": 0.0, + "learning_rate": 0.0001608, + "loss": 6.4815, + "step": 4242 + }, + { + "epoch": 6.7888, + "grad_norm": 0.0, + "learning_rate": 0.00016059999999999997, + "loss": 4.7451, + "step": 4243 + }, + { + "epoch": 6.7904, + "grad_norm": 0.0, + "learning_rate": 0.00016039999999999997, + "loss": 4.5516, + "step": 4244 + }, + { + "epoch": 6.792, + "grad_norm": 0.0, + "learning_rate": 0.0001602, + "loss": 5.5383, + "step": 4245 + }, + { + "epoch": 6.7936, + "grad_norm": 0.0, + "learning_rate": 0.00015999999999999999, + "loss": 5.4859, + "step": 4246 + }, + { + "epoch": 6.7952, + "grad_norm": 0.0, + "learning_rate": 0.00015979999999999998, + "loss": 4.9607, + "step": 4247 + }, + { + "epoch": 6.7968, + "grad_norm": 0.0, + "learning_rate": 0.0001596, + "loss": 4.7679, + "step": 4248 + }, + { + "epoch": 6.7984, + "grad_norm": 0.0, + "learning_rate": 0.00015939999999999997, + "loss": 6.806, + "step": 4249 + }, + { + "epoch": 6.8, + "grad_norm": 0.0, + "learning_rate": 0.00015919999999999997, + "loss": 8.0953, + "step": 4250 + }, + { + "epoch": 6.8016, + "grad_norm": 0.0, + "learning_rate": 0.000159, + "loss": 8.4137, + "step": 4251 + }, + { + "epoch": 6.8032, + "grad_norm": 0.0, + "learning_rate": 0.00015879999999999998, + "loss": 6.6673, + "step": 4252 + }, + { + "epoch": 6.8048, + "grad_norm": 0.0, + "learning_rate": 0.00015859999999999998, + "loss": 7.8875, + "step": 4253 + }, + { + "epoch": 6.8064, + "grad_norm": 0.0, + "learning_rate": 0.0001584, + "loss": 6.823, + "step": 4254 + }, + { + "epoch": 6.808, + "grad_norm": 0.0, + "learning_rate": 0.00015819999999999997, + "loss": 5.2241, + "step": 4255 + }, + { + "epoch": 6.8096, + "grad_norm": 0.0, + "learning_rate": 0.00015799999999999996, + "loss": 8.0661, + "step": 4256 + }, + { + "epoch": 6.8112, + "grad_norm": 0.0, + "learning_rate": 0.0001578, + "loss": 6.7184, + "step": 4257 + }, + { + "epoch": 6.8128, + "grad_norm": 0.0, + "learning_rate": 0.00015759999999999998, + "loss": 6.3589, + "step": 4258 + }, + { + "epoch": 6.8144, + "grad_norm": 0.0, + "learning_rate": 0.00015739999999999998, + "loss": 6.2219, + "step": 4259 + }, + { + "epoch": 6.816, + "grad_norm": 0.0, + "learning_rate": 0.0001572, + "loss": 6.18, + "step": 4260 + }, + { + "epoch": 6.8176, + "grad_norm": 0.0, + "learning_rate": 0.000157, + "loss": 6.3529, + "step": 4261 + }, + { + "epoch": 6.8192, + "grad_norm": 0.0, + "learning_rate": 0.00015679999999999996, + "loss": 6.6016, + "step": 4262 + }, + { + "epoch": 6.8208, + "grad_norm": 0.0, + "learning_rate": 0.00015659999999999998, + "loss": 4.5632, + "step": 4263 + }, + { + "epoch": 6.8224, + "grad_norm": 0.0, + "learning_rate": 0.00015639999999999998, + "loss": 6.0819, + "step": 4264 + }, + { + "epoch": 6.824, + "grad_norm": 0.0, + "learning_rate": 0.0001562, + "loss": 8.791, + "step": 4265 + }, + { + "epoch": 6.8256, + "grad_norm": 0.0, + "learning_rate": 0.000156, + "loss": 5.3828, + "step": 4266 + }, + { + "epoch": 6.8272, + "grad_norm": 0.0, + "learning_rate": 0.0001558, + "loss": 5.9052, + "step": 4267 + }, + { + "epoch": 6.8288, + "grad_norm": 0.0, + "learning_rate": 0.00015560000000000001, + "loss": 5.381, + "step": 4268 + }, + { + "epoch": 6.8304, + "grad_norm": 0.0, + "learning_rate": 0.00015539999999999998, + "loss": 4.8866, + "step": 4269 + }, + { + "epoch": 6.832, + "grad_norm": 0.0, + "learning_rate": 0.00015519999999999998, + "loss": 5.7885, + "step": 4270 + }, + { + "epoch": 6.8336, + "grad_norm": 0.0, + "learning_rate": 0.000155, + "loss": 5.812, + "step": 4271 + }, + { + "epoch": 6.8352, + "grad_norm": 0.0, + "learning_rate": 0.0001548, + "loss": 5.2621, + "step": 4272 + }, + { + "epoch": 6.8368, + "grad_norm": 0.0, + "learning_rate": 0.0001546, + "loss": 5.6901, + "step": 4273 + }, + { + "epoch": 6.8384, + "grad_norm": 0.0, + "learning_rate": 0.0001544, + "loss": 5.0837, + "step": 4274 + }, + { + "epoch": 6.84, + "grad_norm": 0.0, + "learning_rate": 0.00015419999999999998, + "loss": 4.2116, + "step": 4275 + }, + { + "epoch": 6.8416, + "grad_norm": 0.0, + "learning_rate": 0.00015399999999999998, + "loss": 5.1249, + "step": 4276 + }, + { + "epoch": 6.8431999999999995, + "grad_norm": 0.0, + "learning_rate": 0.0001538, + "loss": 5.3888, + "step": 4277 + }, + { + "epoch": 6.8448, + "grad_norm": 0.0, + "learning_rate": 0.0001536, + "loss": 4.6081, + "step": 4278 + }, + { + "epoch": 6.8464, + "grad_norm": 0.0, + "learning_rate": 0.0001534, + "loss": 6.4538, + "step": 4279 + }, + { + "epoch": 6.848, + "grad_norm": 0.0, + "learning_rate": 0.0001532, + "loss": 4.8083, + "step": 4280 + }, + { + "epoch": 6.8496, + "grad_norm": 0.0, + "learning_rate": 0.00015299999999999998, + "loss": 5.039, + "step": 4281 + }, + { + "epoch": 6.8512, + "grad_norm": 0.0, + "learning_rate": 0.00015279999999999997, + "loss": 5.3073, + "step": 4282 + }, + { + "epoch": 6.8528, + "grad_norm": 0.0, + "learning_rate": 0.0001526, + "loss": 5.2667, + "step": 4283 + }, + { + "epoch": 6.8544, + "grad_norm": 0.0, + "learning_rate": 0.0001524, + "loss": 4.2385, + "step": 4284 + }, + { + "epoch": 6.856, + "grad_norm": 0.0, + "learning_rate": 0.00015219999999999999, + "loss": 6.4816, + "step": 4285 + }, + { + "epoch": 6.8576, + "grad_norm": 0.0, + "learning_rate": 0.000152, + "loss": 3.9428, + "step": 4286 + }, + { + "epoch": 6.8591999999999995, + "grad_norm": 0.0, + "learning_rate": 0.00015179999999999998, + "loss": 4.5685, + "step": 4287 + }, + { + "epoch": 6.8608, + "grad_norm": 0.0, + "learning_rate": 0.00015159999999999997, + "loss": 4.6636, + "step": 4288 + }, + { + "epoch": 6.8624, + "grad_norm": 0.0, + "learning_rate": 0.0001514, + "loss": 5.4109, + "step": 4289 + }, + { + "epoch": 6.864, + "grad_norm": 0.0, + "learning_rate": 0.0001512, + "loss": 4.8747, + "step": 4290 + }, + { + "epoch": 6.8656, + "grad_norm": 0.0, + "learning_rate": 0.00015099999999999998, + "loss": 7.5344, + "step": 4291 + }, + { + "epoch": 6.8672, + "grad_norm": 0.0, + "learning_rate": 0.0001508, + "loss": 5.3078, + "step": 4292 + }, + { + "epoch": 6.8688, + "grad_norm": 0.0, + "learning_rate": 0.00015059999999999997, + "loss": 4.5709, + "step": 4293 + }, + { + "epoch": 6.8704, + "grad_norm": 0.0, + "learning_rate": 0.00015039999999999997, + "loss": 9.5683, + "step": 4294 + }, + { + "epoch": 6.872, + "grad_norm": 0.0, + "learning_rate": 0.0001502, + "loss": 7.8465, + "step": 4295 + }, + { + "epoch": 6.8736, + "grad_norm": 0.0, + "learning_rate": 0.00015, + "loss": 5.9877, + "step": 4296 + }, + { + "epoch": 6.8751999999999995, + "grad_norm": 0.0, + "learning_rate": 0.00014979999999999998, + "loss": 5.6867, + "step": 4297 + }, + { + "epoch": 6.8768, + "grad_norm": 0.0, + "learning_rate": 0.00014959999999999998, + "loss": 5.6825, + "step": 4298 + }, + { + "epoch": 6.8784, + "grad_norm": 0.0, + "learning_rate": 0.0001494, + "loss": 6.1692, + "step": 4299 + }, + { + "epoch": 6.88, + "grad_norm": NaN, + "learning_rate": 0.0001494, + "loss": 8.656, + "step": 4300 + }, + { + "epoch": 6.8816, + "grad_norm": 0.0, + "learning_rate": 0.0001492, + "loss": 6.2364, + "step": 4301 + }, + { + "epoch": 6.8832, + "grad_norm": 0.0, + "learning_rate": 0.000149, + "loss": 9.5798, + "step": 4302 + }, + { + "epoch": 6.8848, + "grad_norm": 0.0, + "learning_rate": 0.00014879999999999998, + "loss": 6.8069, + "step": 4303 + }, + { + "epoch": 6.8864, + "grad_norm": 0.0, + "learning_rate": 0.00014859999999999998, + "loss": 7.6856, + "step": 4304 + }, + { + "epoch": 6.888, + "grad_norm": 0.0, + "learning_rate": 0.00014839999999999998, + "loss": 7.3696, + "step": 4305 + }, + { + "epoch": 6.8896, + "grad_norm": 0.0, + "learning_rate": 0.0001482, + "loss": 8.496, + "step": 4306 + }, + { + "epoch": 6.8911999999999995, + "grad_norm": 0.0, + "learning_rate": 0.000148, + "loss": 4.7344, + "step": 4307 + }, + { + "epoch": 6.8928, + "grad_norm": 0.0, + "learning_rate": 0.0001478, + "loss": 5.9466, + "step": 4308 + }, + { + "epoch": 6.8944, + "grad_norm": 0.0, + "learning_rate": 0.00014759999999999998, + "loss": 6.1564, + "step": 4309 + }, + { + "epoch": 6.896, + "grad_norm": 0.0, + "learning_rate": 0.00014739999999999998, + "loss": 7.5019, + "step": 4310 + }, + { + "epoch": 6.8976, + "grad_norm": 0.0, + "learning_rate": 0.00014719999999999997, + "loss": 6.6649, + "step": 4311 + }, + { + "epoch": 6.8992, + "grad_norm": 0.0, + "learning_rate": 0.000147, + "loss": 6.3927, + "step": 4312 + }, + { + "epoch": 6.9008, + "grad_norm": 0.0, + "learning_rate": 0.0001468, + "loss": 7.2916, + "step": 4313 + }, + { + "epoch": 6.9024, + "grad_norm": 0.0, + "learning_rate": 0.00014659999999999999, + "loss": 6.0807, + "step": 4314 + }, + { + "epoch": 6.904, + "grad_norm": 0.0, + "learning_rate": 0.00014639999999999998, + "loss": 6.3078, + "step": 4315 + }, + { + "epoch": 6.9056, + "grad_norm": 0.0, + "learning_rate": 0.0001462, + "loss": 6.1345, + "step": 4316 + }, + { + "epoch": 6.9072, + "grad_norm": 0.0, + "learning_rate": 0.000146, + "loss": 7.4456, + "step": 4317 + }, + { + "epoch": 6.9088, + "grad_norm": 0.0, + "learning_rate": 0.0001458, + "loss": 5.1236, + "step": 4318 + }, + { + "epoch": 6.9104, + "grad_norm": 0.0, + "learning_rate": 0.0001456, + "loss": 5.5366, + "step": 4319 + }, + { + "epoch": 6.912, + "grad_norm": 0.0, + "learning_rate": 0.00014539999999999998, + "loss": 7.4259, + "step": 4320 + }, + { + "epoch": 6.9136, + "grad_norm": 0.0, + "learning_rate": 0.00014519999999999998, + "loss": 5.9814, + "step": 4321 + }, + { + "epoch": 6.9152000000000005, + "grad_norm": 0.0, + "learning_rate": 0.000145, + "loss": 5.4896, + "step": 4322 + }, + { + "epoch": 6.9168, + "grad_norm": 0.0, + "learning_rate": 0.0001448, + "loss": 4.6442, + "step": 4323 + }, + { + "epoch": 6.9184, + "grad_norm": 0.0, + "learning_rate": 0.0001446, + "loss": 4.8726, + "step": 4324 + }, + { + "epoch": 6.92, + "grad_norm": 0.0, + "learning_rate": 0.00014439999999999999, + "loss": 6.6722, + "step": 4325 + }, + { + "epoch": 6.9216, + "grad_norm": 0.0, + "learning_rate": 0.00014419999999999998, + "loss": 4.7456, + "step": 4326 + }, + { + "epoch": 6.9232, + "grad_norm": 0.0, + "learning_rate": 0.00014399999999999998, + "loss": 5.092, + "step": 4327 + }, + { + "epoch": 6.9248, + "grad_norm": 0.0, + "learning_rate": 0.0001438, + "loss": 5.9932, + "step": 4328 + }, + { + "epoch": 6.9264, + "grad_norm": 0.0, + "learning_rate": 0.0001436, + "loss": 5.1919, + "step": 4329 + }, + { + "epoch": 6.928, + "grad_norm": 0.0, + "learning_rate": 0.0001434, + "loss": 5.0041, + "step": 4330 + }, + { + "epoch": 6.9296, + "grad_norm": 0.0, + "learning_rate": 0.00014319999999999998, + "loss": 4.8493, + "step": 4331 + }, + { + "epoch": 6.9312000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00014299999999999998, + "loss": 5.3269, + "step": 4332 + }, + { + "epoch": 6.9328, + "grad_norm": 0.0, + "learning_rate": 0.00014279999999999997, + "loss": 5.5469, + "step": 4333 + }, + { + "epoch": 6.9344, + "grad_norm": 0.0, + "learning_rate": 0.0001426, + "loss": 6.3032, + "step": 4334 + }, + { + "epoch": 6.936, + "grad_norm": 0.0, + "learning_rate": 0.0001424, + "loss": 4.7999, + "step": 4335 + }, + { + "epoch": 6.9376, + "grad_norm": 0.0, + "learning_rate": 0.0001422, + "loss": 5.418, + "step": 4336 + }, + { + "epoch": 6.9392, + "grad_norm": 0.0, + "learning_rate": 0.00014199999999999998, + "loss": 5.2496, + "step": 4337 + }, + { + "epoch": 6.9408, + "grad_norm": 0.0, + "learning_rate": 0.0001418, + "loss": 5.1538, + "step": 4338 + }, + { + "epoch": 6.9424, + "grad_norm": 0.0, + "learning_rate": 0.00014159999999999997, + "loss": 7.7733, + "step": 4339 + }, + { + "epoch": 6.944, + "grad_norm": 0.0, + "learning_rate": 0.0001414, + "loss": 5.1677, + "step": 4340 + }, + { + "epoch": 6.9456, + "grad_norm": 0.0, + "learning_rate": 0.0001412, + "loss": 5.1715, + "step": 4341 + }, + { + "epoch": 6.9472000000000005, + "grad_norm": 0.0, + "learning_rate": 0.00014099999999999998, + "loss": 5.1323, + "step": 4342 + }, + { + "epoch": 6.9488, + "grad_norm": 0.0, + "learning_rate": 0.00014079999999999998, + "loss": 4.6489, + "step": 4343 + }, + { + "epoch": 6.9504, + "grad_norm": 0.0, + "learning_rate": 0.0001406, + "loss": 5.2453, + "step": 4344 + }, + { + "epoch": 6.952, + "grad_norm": 0.0, + "learning_rate": 0.0001404, + "loss": 5.4791, + "step": 4345 + }, + { + "epoch": 6.9536, + "grad_norm": 0.0, + "learning_rate": 0.0001402, + "loss": 5.1121, + "step": 4346 + }, + { + "epoch": 6.9552, + "grad_norm": 0.0, + "learning_rate": 0.00014, + "loss": 5.8898, + "step": 4347 + }, + { + "epoch": 6.9568, + "grad_norm": 0.0, + "learning_rate": 0.00013979999999999998, + "loss": 4.8784, + "step": 4348 + }, + { + "epoch": 6.9584, + "grad_norm": 0.0, + "learning_rate": 0.00013959999999999998, + "loss": 4.8517, + "step": 4349 + }, + { + "epoch": 6.96, + "grad_norm": NaN, + "learning_rate": 0.00013959999999999998, + "loss": 5.1453, + "step": 4350 + }, + { + "epoch": 6.9616, + "grad_norm": 0.0, + "learning_rate": 0.0001394, + "loss": 6.5692, + "step": 4351 + }, + { + "epoch": 6.9632, + "grad_norm": 0.0, + "learning_rate": 0.0001392, + "loss": 7.1124, + "step": 4352 + }, + { + "epoch": 6.9648, + "grad_norm": 0.0, + "learning_rate": 0.000139, + "loss": 6.1347, + "step": 4353 + }, + { + "epoch": 6.9664, + "grad_norm": 0.0, + "learning_rate": 0.00013879999999999999, + "loss": 5.7037, + "step": 4354 + }, + { + "epoch": 6.968, + "grad_norm": 0.0, + "learning_rate": 0.0001386, + "loss": 6.6176, + "step": 4355 + }, + { + "epoch": 6.9696, + "grad_norm": 0.0, + "learning_rate": 0.00013839999999999998, + "loss": 5.5577, + "step": 4356 + }, + { + "epoch": 6.9712, + "grad_norm": 0.0, + "learning_rate": 0.0001382, + "loss": 6.9877, + "step": 4357 + }, + { + "epoch": 6.9728, + "grad_norm": 0.0, + "learning_rate": 0.000138, + "loss": 6.5335, + "step": 4358 + }, + { + "epoch": 6.9744, + "grad_norm": 0.0, + "learning_rate": 0.0001378, + "loss": 6.6167, + "step": 4359 + }, + { + "epoch": 6.976, + "grad_norm": 0.0, + "learning_rate": 0.00013759999999999998, + "loss": 5.6802, + "step": 4360 + }, + { + "epoch": 6.9776, + "grad_norm": 0.0, + "learning_rate": 0.0001374, + "loss": 5.9226, + "step": 4361 + }, + { + "epoch": 6.9792, + "grad_norm": 0.0, + "learning_rate": 0.00013719999999999997, + "loss": 5.3872, + "step": 4362 + }, + { + "epoch": 6.9808, + "grad_norm": 0.0, + "learning_rate": 0.000137, + "loss": 5.3704, + "step": 4363 + }, + { + "epoch": 6.9824, + "grad_norm": 0.0, + "learning_rate": 0.0001368, + "loss": 5.9309, + "step": 4364 + }, + { + "epoch": 6.984, + "grad_norm": 0.0, + "learning_rate": 0.00013659999999999999, + "loss": 4.652, + "step": 4365 + }, + { + "epoch": 6.9856, + "grad_norm": 0.0, + "learning_rate": 0.00013639999999999998, + "loss": 7.1784, + "step": 4366 + }, + { + "epoch": 6.9872, + "grad_norm": 0.0, + "learning_rate": 0.0001362, + "loss": 6.4351, + "step": 4367 + }, + { + "epoch": 6.9888, + "grad_norm": 0.0, + "learning_rate": 0.00013599999999999997, + "loss": 4.8783, + "step": 4368 + }, + { + "epoch": 6.9904, + "grad_norm": 0.0, + "learning_rate": 0.0001358, + "loss": 4.3315, + "step": 4369 + }, + { + "epoch": 6.992, + "grad_norm": 0.0, + "learning_rate": 0.0001356, + "loss": 6.4382, + "step": 4370 + }, + { + "epoch": 6.9936, + "grad_norm": 0.0, + "learning_rate": 0.00013539999999999998, + "loss": 5.6157, + "step": 4371 + }, + { + "epoch": 6.9952, + "grad_norm": 0.0, + "learning_rate": 0.00013519999999999998, + "loss": 4.3788, + "step": 4372 + }, + { + "epoch": 6.9968, + "grad_norm": 0.0, + "learning_rate": 0.000135, + "loss": 5.1114, + "step": 4373 + }, + { + "epoch": 6.9984, + "grad_norm": 0.0, + "learning_rate": 0.00013479999999999997, + "loss": 4.2126, + "step": 4374 + }, + { + "epoch": 7.0, + "grad_norm": NaN, + "learning_rate": 0.00013479999999999997, + "loss": 4.5977, + "step": 4375 + }, + { + "epoch": 7.0016, + "grad_norm": 0.0, + "learning_rate": 0.0001346, + "loss": 7.5475, + "step": 4376 + }, + { + "epoch": 7.0032, + "grad_norm": 0.0, + "learning_rate": 0.0001344, + "loss": 9.2323, + "step": 4377 + }, + { + "epoch": 7.0048, + "grad_norm": 0.0, + "learning_rate": 0.0001342, + "loss": 6.2872, + "step": 4378 + }, + { + "epoch": 7.0064, + "grad_norm": 0.0, + "learning_rate": 0.00013399999999999998, + "loss": 5.7807, + "step": 4379 + }, + { + "epoch": 7.008, + "grad_norm": 0.0, + "learning_rate": 0.0001338, + "loss": 7.1882, + "step": 4380 + }, + { + "epoch": 7.0096, + "grad_norm": 0.0, + "learning_rate": 0.0001336, + "loss": 6.1936, + "step": 4381 + }, + { + "epoch": 7.0112, + "grad_norm": 0.0, + "learning_rate": 0.0001334, + "loss": 5.5946, + "step": 4382 + }, + { + "epoch": 7.0128, + "grad_norm": 0.0, + "learning_rate": 0.00013319999999999999, + "loss": 8.3462, + "step": 4383 + }, + { + "epoch": 7.0144, + "grad_norm": 0.0, + "learning_rate": 0.000133, + "loss": 5.663, + "step": 4384 + }, + { + "epoch": 7.016, + "grad_norm": 0.0, + "learning_rate": 0.00013279999999999998, + "loss": 6.1503, + "step": 4385 + }, + { + "epoch": 7.0176, + "grad_norm": 0.0, + "learning_rate": 0.0001326, + "loss": 6.0101, + "step": 4386 + }, + { + "epoch": 7.0192, + "grad_norm": 0.0, + "learning_rate": 0.0001324, + "loss": 6.4806, + "step": 4387 + }, + { + "epoch": 7.0208, + "grad_norm": 0.0, + "learning_rate": 0.0001322, + "loss": 5.8324, + "step": 4388 + }, + { + "epoch": 7.0224, + "grad_norm": 0.0, + "learning_rate": 0.00013199999999999998, + "loss": 5.7663, + "step": 4389 + }, + { + "epoch": 7.024, + "grad_norm": 0.0, + "learning_rate": 0.0001318, + "loss": 5.3892, + "step": 4390 + }, + { + "epoch": 7.0256, + "grad_norm": 0.0, + "learning_rate": 0.00013159999999999997, + "loss": 4.1761, + "step": 4391 + }, + { + "epoch": 7.0272, + "grad_norm": 0.0, + "learning_rate": 0.0001314, + "loss": 5.7275, + "step": 4392 + }, + { + "epoch": 7.0288, + "grad_norm": 0.0, + "learning_rate": 0.0001312, + "loss": 6.4893, + "step": 4393 + }, + { + "epoch": 7.0304, + "grad_norm": 0.0, + "learning_rate": 0.00013099999999999999, + "loss": 6.1713, + "step": 4394 + }, + { + "epoch": 7.032, + "grad_norm": 0.0, + "learning_rate": 0.00013079999999999998, + "loss": 7.6139, + "step": 4395 + }, + { + "epoch": 7.0336, + "grad_norm": 0.0, + "learning_rate": 0.0001306, + "loss": 6.0886, + "step": 4396 + }, + { + "epoch": 7.0352, + "grad_norm": 0.0, + "learning_rate": 0.00013039999999999997, + "loss": 6.8868, + "step": 4397 + }, + { + "epoch": 7.0368, + "grad_norm": 0.0, + "learning_rate": 0.0001302, + "loss": 4.8965, + "step": 4398 + }, + { + "epoch": 7.0384, + "grad_norm": 0.0, + "learning_rate": 0.00013, + "loss": 5.3175, + "step": 4399 + }, + { + "epoch": 7.04, + "grad_norm": 0.0, + "learning_rate": 0.00012979999999999998, + "loss": 5.4927, + "step": 4400 + }, + { + "epoch": 7.0416, + "grad_norm": 0.0, + "learning_rate": 0.00012959999999999998, + "loss": 5.9534, + "step": 4401 + }, + { + "epoch": 7.0432, + "grad_norm": 0.0, + "learning_rate": 0.0001294, + "loss": 5.1196, + "step": 4402 + }, + { + "epoch": 7.0448, + "grad_norm": 0.0, + "learning_rate": 0.00012919999999999997, + "loss": 4.6083, + "step": 4403 + }, + { + "epoch": 7.0464, + "grad_norm": 0.0, + "learning_rate": 0.000129, + "loss": 5.7149, + "step": 4404 + }, + { + "epoch": 7.048, + "grad_norm": 0.0, + "learning_rate": 0.0001288, + "loss": 5.131, + "step": 4405 + }, + { + "epoch": 7.0496, + "grad_norm": 0.0, + "learning_rate": 0.00012859999999999998, + "loss": 7.9605, + "step": 4406 + }, + { + "epoch": 7.0512, + "grad_norm": 0.0, + "learning_rate": 0.00012839999999999998, + "loss": 6.0242, + "step": 4407 + }, + { + "epoch": 7.0528, + "grad_norm": 0.0, + "learning_rate": 0.0001282, + "loss": 5.4065, + "step": 4408 + }, + { + "epoch": 7.0544, + "grad_norm": 0.0, + "learning_rate": 0.000128, + "loss": 4.6464, + "step": 4409 + }, + { + "epoch": 7.056, + "grad_norm": 0.0, + "learning_rate": 0.0001278, + "loss": 5.2592, + "step": 4410 + }, + { + "epoch": 7.0576, + "grad_norm": 0.0, + "learning_rate": 0.00012759999999999998, + "loss": 4.1099, + "step": 4411 + }, + { + "epoch": 7.0592, + "grad_norm": 0.0, + "learning_rate": 0.0001274, + "loss": 4.1237, + "step": 4412 + }, + { + "epoch": 7.0608, + "grad_norm": 0.0, + "learning_rate": 0.00012719999999999997, + "loss": 5.7603, + "step": 4413 + }, + { + "epoch": 7.0624, + "grad_norm": 0.0, + "learning_rate": 0.000127, + "loss": 5.6802, + "step": 4414 + }, + { + "epoch": 7.064, + "grad_norm": 0.0, + "learning_rate": 0.0001268, + "loss": 5.062, + "step": 4415 + }, + { + "epoch": 7.0656, + "grad_norm": 0.0, + "learning_rate": 0.0001266, + "loss": 5.1358, + "step": 4416 + }, + { + "epoch": 7.0672, + "grad_norm": 0.0, + "learning_rate": 0.00012639999999999998, + "loss": 5.7706, + "step": 4417 + }, + { + "epoch": 7.0688, + "grad_norm": 0.0, + "learning_rate": 0.0001262, + "loss": 4.1849, + "step": 4418 + }, + { + "epoch": 7.0704, + "grad_norm": 0.0, + "learning_rate": 0.00012599999999999997, + "loss": 5.2318, + "step": 4419 + }, + { + "epoch": 7.072, + "grad_norm": 0.0, + "learning_rate": 0.0001258, + "loss": 3.6845, + "step": 4420 + }, + { + "epoch": 7.0736, + "grad_norm": 0.0, + "learning_rate": 0.0001256, + "loss": 6.4123, + "step": 4421 + }, + { + "epoch": 7.0752, + "grad_norm": 0.0, + "learning_rate": 0.00012539999999999999, + "loss": 4.6083, + "step": 4422 + }, + { + "epoch": 7.0768, + "grad_norm": 0.0, + "learning_rate": 0.00012519999999999998, + "loss": 5.0964, + "step": 4423 + }, + { + "epoch": 7.0784, + "grad_norm": 0.0, + "learning_rate": 0.000125, + "loss": 4.5912, + "step": 4424 + }, + { + "epoch": 7.08, + "grad_norm": 0.0, + "learning_rate": 0.00012479999999999997, + "loss": 8.8139, + "step": 4425 + }, + { + "epoch": 7.0816, + "grad_norm": 0.0, + "learning_rate": 0.0001246, + "loss": 6.2302, + "step": 4426 + }, + { + "epoch": 7.0832, + "grad_norm": 0.0, + "learning_rate": 0.0001244, + "loss": 8.877, + "step": 4427 + }, + { + "epoch": 7.0848, + "grad_norm": 0.0, + "learning_rate": 0.00012419999999999998, + "loss": 8.6141, + "step": 4428 + }, + { + "epoch": 7.0864, + "grad_norm": 0.0, + "learning_rate": 0.00012399999999999998, + "loss": 6.1153, + "step": 4429 + }, + { + "epoch": 7.088, + "grad_norm": 0.0, + "learning_rate": 0.0001238, + "loss": 6.8582, + "step": 4430 + }, + { + "epoch": 7.0896, + "grad_norm": 0.0, + "learning_rate": 0.0001236, + "loss": 6.4544, + "step": 4431 + }, + { + "epoch": 7.0912, + "grad_norm": 0.0, + "learning_rate": 0.0001234, + "loss": 5.2604, + "step": 4432 + }, + { + "epoch": 7.0928, + "grad_norm": 0.0, + "learning_rate": 0.00012319999999999999, + "loss": 6.136, + "step": 4433 + }, + { + "epoch": 7.0944, + "grad_norm": 0.0, + "learning_rate": 0.00012299999999999998, + "loss": 5.2371, + "step": 4434 + }, + { + "epoch": 7.096, + "grad_norm": 0.0, + "learning_rate": 0.00012279999999999998, + "loss": 6.3874, + "step": 4435 + }, + { + "epoch": 7.0976, + "grad_norm": 0.0, + "learning_rate": 0.0001226, + "loss": 6.8492, + "step": 4436 + }, + { + "epoch": 7.0992, + "grad_norm": 0.0, + "learning_rate": 0.0001224, + "loss": 6.9934, + "step": 4437 + }, + { + "epoch": 7.1008, + "grad_norm": 0.0, + "learning_rate": 0.0001222, + "loss": 5.6651, + "step": 4438 + }, + { + "epoch": 7.1024, + "grad_norm": 0.0, + "learning_rate": 0.000122, + "loss": 5.3206, + "step": 4439 + }, + { + "epoch": 7.104, + "grad_norm": 0.0, + "learning_rate": 0.00012179999999999999, + "loss": 5.3081, + "step": 4440 + }, + { + "epoch": 7.1056, + "grad_norm": 0.0, + "learning_rate": 0.00012159999999999999, + "loss": 6.3004, + "step": 4441 + }, + { + "epoch": 7.1072, + "grad_norm": 0.0, + "learning_rate": 0.0001214, + "loss": 5.9301, + "step": 4442 + }, + { + "epoch": 7.1088, + "grad_norm": 0.0, + "learning_rate": 0.00012119999999999999, + "loss": 5.9583, + "step": 4443 + }, + { + "epoch": 7.1104, + "grad_norm": 0.0, + "learning_rate": 0.00012099999999999999, + "loss": 6.3635, + "step": 4444 + }, + { + "epoch": 7.112, + "grad_norm": 0.0, + "learning_rate": 0.0001208, + "loss": 6.3125, + "step": 4445 + }, + { + "epoch": 7.1136, + "grad_norm": 0.0, + "learning_rate": 0.00012059999999999999, + "loss": 6.3657, + "step": 4446 + }, + { + "epoch": 7.1152, + "grad_norm": 0.0, + "learning_rate": 0.00012039999999999999, + "loss": 5.9963, + "step": 4447 + }, + { + "epoch": 7.1168, + "grad_norm": 0.0, + "learning_rate": 0.0001202, + "loss": 6.6274, + "step": 4448 + }, + { + "epoch": 7.1184, + "grad_norm": 0.0, + "learning_rate": 0.00011999999999999999, + "loss": 4.9015, + "step": 4449 + }, + { + "epoch": 7.12, + "grad_norm": 0.0, + "learning_rate": 0.00011979999999999998, + "loss": 4.7555, + "step": 4450 + }, + { + "epoch": 7.1216, + "grad_norm": 0.0, + "learning_rate": 0.0001196, + "loss": 5.3612, + "step": 4451 + }, + { + "epoch": 7.1232, + "grad_norm": 0.0, + "learning_rate": 0.0001194, + "loss": 5.6789, + "step": 4452 + }, + { + "epoch": 7.1248, + "grad_norm": 0.0, + "learning_rate": 0.00011919999999999998, + "loss": 5.6262, + "step": 4453 + }, + { + "epoch": 7.1264, + "grad_norm": 0.0, + "learning_rate": 0.00011899999999999999, + "loss": 4.7289, + "step": 4454 + }, + { + "epoch": 7.128, + "grad_norm": 0.0, + "learning_rate": 0.0001188, + "loss": 4.7936, + "step": 4455 + }, + { + "epoch": 7.1296, + "grad_norm": 0.0, + "learning_rate": 0.00011859999999999998, + "loss": 4.2818, + "step": 4456 + }, + { + "epoch": 7.1312, + "grad_norm": 0.0, + "learning_rate": 0.00011839999999999999, + "loss": 5.1505, + "step": 4457 + }, + { + "epoch": 7.1328, + "grad_norm": 0.0, + "learning_rate": 0.0001182, + "loss": 5.4992, + "step": 4458 + }, + { + "epoch": 7.1344, + "grad_norm": 0.0, + "learning_rate": 0.00011799999999999998, + "loss": 6.3987, + "step": 4459 + }, + { + "epoch": 7.136, + "grad_norm": 0.0, + "learning_rate": 0.00011779999999999999, + "loss": 5.4565, + "step": 4460 + }, + { + "epoch": 7.1376, + "grad_norm": 0.0, + "learning_rate": 0.0001176, + "loss": 5.7194, + "step": 4461 + }, + { + "epoch": 7.1392, + "grad_norm": 0.0, + "learning_rate": 0.00011739999999999998, + "loss": 4.7668, + "step": 4462 + }, + { + "epoch": 7.1408, + "grad_norm": 0.0, + "learning_rate": 0.00011719999999999999, + "loss": 4.9837, + "step": 4463 + }, + { + "epoch": 7.1424, + "grad_norm": 0.0, + "learning_rate": 0.000117, + "loss": 5.1758, + "step": 4464 + }, + { + "epoch": 7.144, + "grad_norm": 0.0, + "learning_rate": 0.00011679999999999998, + "loss": 4.7496, + "step": 4465 + }, + { + "epoch": 7.1456, + "grad_norm": 0.0, + "learning_rate": 0.00011659999999999999, + "loss": 7.8581, + "step": 4466 + }, + { + "epoch": 7.1472, + "grad_norm": 0.0, + "learning_rate": 0.0001164, + "loss": 5.145, + "step": 4467 + }, + { + "epoch": 7.1488, + "grad_norm": 0.0, + "learning_rate": 0.00011619999999999998, + "loss": 4.6992, + "step": 4468 + }, + { + "epoch": 7.1504, + "grad_norm": 0.0, + "learning_rate": 0.00011599999999999999, + "loss": 4.039, + "step": 4469 + }, + { + "epoch": 7.152, + "grad_norm": 0.0, + "learning_rate": 0.0001158, + "loss": 7.8544, + "step": 4470 + }, + { + "epoch": 7.1536, + "grad_norm": 0.0, + "learning_rate": 0.0001156, + "loss": 7.962, + "step": 4471 + }, + { + "epoch": 7.1552, + "grad_norm": 0.0, + "learning_rate": 0.00011539999999999999, + "loss": 4.9886, + "step": 4472 + }, + { + "epoch": 7.1568, + "grad_norm": 0.0, + "learning_rate": 0.0001152, + "loss": 5.7649, + "step": 4473 + }, + { + "epoch": 7.1584, + "grad_norm": 0.0, + "learning_rate": 0.000115, + "loss": 5.0145, + "step": 4474 + }, + { + "epoch": 7.16, + "grad_norm": 0.0, + "learning_rate": 0.00011479999999999999, + "loss": 6.2514, + "step": 4475 + }, + { + "epoch": 7.1616, + "grad_norm": 0.0, + "learning_rate": 0.0001146, + "loss": 9.9221, + "step": 4476 + }, + { + "epoch": 7.1632, + "grad_norm": 0.0, + "learning_rate": 0.0001144, + "loss": 7.1045, + "step": 4477 + }, + { + "epoch": 7.1648, + "grad_norm": 0.0, + "learning_rate": 0.00011419999999999998, + "loss": 8.9454, + "step": 4478 + }, + { + "epoch": 7.1664, + "grad_norm": 0.0, + "learning_rate": 0.00011399999999999999, + "loss": 6.1405, + "step": 4479 + }, + { + "epoch": 7.168, + "grad_norm": 0.0, + "learning_rate": 0.0001138, + "loss": 5.8219, + "step": 4480 + }, + { + "epoch": 7.1696, + "grad_norm": 0.0, + "learning_rate": 0.00011359999999999998, + "loss": 5.4465, + "step": 4481 + }, + { + "epoch": 7.1712, + "grad_norm": 0.0, + "learning_rate": 0.00011339999999999999, + "loss": 4.6502, + "step": 4482 + }, + { + "epoch": 7.1728, + "grad_norm": 0.0, + "learning_rate": 0.0001132, + "loss": 6.7632, + "step": 4483 + }, + { + "epoch": 7.1744, + "grad_norm": 0.0, + "learning_rate": 0.00011299999999999998, + "loss": 6.6573, + "step": 4484 + }, + { + "epoch": 7.176, + "grad_norm": 0.0, + "learning_rate": 0.00011279999999999999, + "loss": 6.581, + "step": 4485 + }, + { + "epoch": 7.1776, + "grad_norm": 0.0, + "learning_rate": 0.0001126, + "loss": 5.9751, + "step": 4486 + }, + { + "epoch": 7.1792, + "grad_norm": 0.0, + "learning_rate": 0.00011239999999999998, + "loss": 5.5023, + "step": 4487 + }, + { + "epoch": 7.1808, + "grad_norm": 0.0, + "learning_rate": 0.00011219999999999999, + "loss": 5.6314, + "step": 4488 + }, + { + "epoch": 7.1824, + "grad_norm": 0.0, + "learning_rate": 0.000112, + "loss": 5.947, + "step": 4489 + }, + { + "epoch": 7.184, + "grad_norm": 0.0, + "learning_rate": 0.00011179999999999998, + "loss": 5.6288, + "step": 4490 + }, + { + "epoch": 7.1856, + "grad_norm": 0.0, + "learning_rate": 0.00011159999999999999, + "loss": 8.1437, + "step": 4491 + }, + { + "epoch": 7.1872, + "grad_norm": 0.0, + "learning_rate": 0.0001114, + "loss": 6.3234, + "step": 4492 + }, + { + "epoch": 7.1888, + "grad_norm": 0.0, + "learning_rate": 0.00011119999999999998, + "loss": 5.0264, + "step": 4493 + }, + { + "epoch": 7.1904, + "grad_norm": 0.0, + "learning_rate": 0.00011099999999999999, + "loss": 4.9791, + "step": 4494 + }, + { + "epoch": 7.192, + "grad_norm": 0.0, + "learning_rate": 0.0001108, + "loss": 4.8003, + "step": 4495 + }, + { + "epoch": 7.1936, + "grad_norm": 0.0, + "learning_rate": 0.00011059999999999998, + "loss": 5.6593, + "step": 4496 + }, + { + "epoch": 7.1952, + "grad_norm": 0.0, + "learning_rate": 0.00011039999999999999, + "loss": 5.3067, + "step": 4497 + }, + { + "epoch": 7.1968, + "grad_norm": 0.0, + "learning_rate": 0.0001102, + "loss": 5.948, + "step": 4498 + }, + { + "epoch": 7.1984, + "grad_norm": 0.0, + "learning_rate": 0.00010999999999999998, + "loss": 5.1466, + "step": 4499 + }, + { + "epoch": 7.2, + "grad_norm": 0.0, + "learning_rate": 0.00010979999999999999, + "loss": 6.0155, + "step": 4500 + }, + { + "epoch": 7.2016, + "grad_norm": 0.0, + "learning_rate": 0.0001096, + "loss": 6.0017, + "step": 4501 + }, + { + "epoch": 7.2032, + "grad_norm": 0.0, + "learning_rate": 0.00010939999999999998, + "loss": 4.6621, + "step": 4502 + }, + { + "epoch": 7.2048, + "grad_norm": 0.0, + "learning_rate": 0.00010919999999999998, + "loss": 4.1596, + "step": 4503 + }, + { + "epoch": 7.2064, + "grad_norm": 0.0, + "learning_rate": 0.00010899999999999999, + "loss": 5.6321, + "step": 4504 + }, + { + "epoch": 7.208, + "grad_norm": 0.0, + "learning_rate": 0.0001088, + "loss": 6.1822, + "step": 4505 + }, + { + "epoch": 7.2096, + "grad_norm": 0.0, + "learning_rate": 0.00010859999999999998, + "loss": 5.5523, + "step": 4506 + }, + { + "epoch": 7.2112, + "grad_norm": 0.0, + "learning_rate": 0.00010839999999999999, + "loss": 4.9534, + "step": 4507 + }, + { + "epoch": 7.2128, + "grad_norm": 0.0, + "learning_rate": 0.0001082, + "loss": 7.3682, + "step": 4508 + }, + { + "epoch": 7.2144, + "grad_norm": 0.0, + "learning_rate": 0.00010799999999999998, + "loss": 4.8959, + "step": 4509 + }, + { + "epoch": 7.216, + "grad_norm": 0.0, + "learning_rate": 0.00010779999999999999, + "loss": 6.0667, + "step": 4510 + }, + { + "epoch": 7.2176, + "grad_norm": 0.0, + "learning_rate": 0.0001076, + "loss": 5.9187, + "step": 4511 + }, + { + "epoch": 7.2192, + "grad_norm": 0.0, + "learning_rate": 0.00010739999999999998, + "loss": 5.4073, + "step": 4512 + }, + { + "epoch": 7.2208, + "grad_norm": 0.0, + "learning_rate": 0.00010719999999999999, + "loss": 5.6722, + "step": 4513 + }, + { + "epoch": 7.2224, + "grad_norm": 0.0, + "learning_rate": 0.000107, + "loss": 5.9317, + "step": 4514 + }, + { + "epoch": 7.224, + "grad_norm": 0.0, + "learning_rate": 0.00010679999999999998, + "loss": 4.132, + "step": 4515 + }, + { + "epoch": 7.2256, + "grad_norm": 0.0, + "learning_rate": 0.00010659999999999999, + "loss": 5.6645, + "step": 4516 + }, + { + "epoch": 7.2272, + "grad_norm": 0.0, + "learning_rate": 0.0001064, + "loss": 4.6923, + "step": 4517 + }, + { + "epoch": 7.2288, + "grad_norm": 0.0, + "learning_rate": 0.00010619999999999998, + "loss": 4.8682, + "step": 4518 + }, + { + "epoch": 7.2304, + "grad_norm": 0.0, + "learning_rate": 0.00010599999999999999, + "loss": 4.191, + "step": 4519 + }, + { + "epoch": 7.232, + "grad_norm": 0.0, + "learning_rate": 0.0001058, + "loss": 5.2527, + "step": 4520 + }, + { + "epoch": 7.2336, + "grad_norm": 0.0, + "learning_rate": 0.00010559999999999998, + "loss": 4.7526, + "step": 4521 + }, + { + "epoch": 7.2352, + "grad_norm": 0.0, + "learning_rate": 0.00010539999999999999, + "loss": 5.121, + "step": 4522 + }, + { + "epoch": 7.2368, + "grad_norm": 0.0, + "learning_rate": 0.0001052, + "loss": 4.6552, + "step": 4523 + }, + { + "epoch": 7.2384, + "grad_norm": 0.0, + "learning_rate": 0.00010499999999999999, + "loss": 5.702, + "step": 4524 + }, + { + "epoch": 7.24, + "grad_norm": 0.0, + "learning_rate": 0.00010479999999999999, + "loss": 5.4976, + "step": 4525 + }, + { + "epoch": 7.2416, + "grad_norm": 0.0, + "learning_rate": 0.0001046, + "loss": 9.1778, + "step": 4526 + }, + { + "epoch": 7.2432, + "grad_norm": 0.0, + "learning_rate": 0.00010439999999999999, + "loss": 6.0219, + "step": 4527 + }, + { + "epoch": 7.2448, + "grad_norm": 0.0, + "learning_rate": 0.00010419999999999998, + "loss": 6.5505, + "step": 4528 + }, + { + "epoch": 7.2464, + "grad_norm": 0.0, + "learning_rate": 0.000104, + "loss": 7.398, + "step": 4529 + }, + { + "epoch": 7.248, + "grad_norm": 0.0, + "learning_rate": 0.00010379999999999999, + "loss": 5.4917, + "step": 4530 + }, + { + "epoch": 7.2496, + "grad_norm": 0.0, + "learning_rate": 0.00010359999999999998, + "loss": 7.0849, + "step": 4531 + }, + { + "epoch": 7.2512, + "grad_norm": 0.0, + "learning_rate": 0.00010339999999999999, + "loss": 7.2476, + "step": 4532 + }, + { + "epoch": 7.2528, + "grad_norm": 0.0, + "learning_rate": 0.00010319999999999999, + "loss": 4.7146, + "step": 4533 + }, + { + "epoch": 7.2544, + "grad_norm": 0.0, + "learning_rate": 0.00010299999999999998, + "loss": 4.9772, + "step": 4534 + }, + { + "epoch": 7.256, + "grad_norm": 0.0, + "learning_rate": 0.00010279999999999999, + "loss": 7.015, + "step": 4535 + }, + { + "epoch": 7.2576, + "grad_norm": 0.0, + "learning_rate": 0.0001026, + "loss": 5.3506, + "step": 4536 + }, + { + "epoch": 7.2592, + "grad_norm": 0.0, + "learning_rate": 0.00010239999999999998, + "loss": 5.5831, + "step": 4537 + }, + { + "epoch": 7.2608, + "grad_norm": 0.0, + "learning_rate": 0.00010219999999999999, + "loss": 6.6254, + "step": 4538 + }, + { + "epoch": 7.2624, + "grad_norm": 0.0, + "learning_rate": 0.000102, + "loss": 4.9123, + "step": 4539 + }, + { + "epoch": 7.264, + "grad_norm": 0.0, + "learning_rate": 0.00010179999999999998, + "loss": 6.0006, + "step": 4540 + }, + { + "epoch": 7.2656, + "grad_norm": 0.0, + "learning_rate": 0.00010159999999999999, + "loss": 5.8829, + "step": 4541 + }, + { + "epoch": 7.2672, + "grad_norm": 0.0, + "learning_rate": 0.0001014, + "loss": 5.986, + "step": 4542 + }, + { + "epoch": 7.2688, + "grad_norm": 0.0, + "learning_rate": 0.0001012, + "loss": 5.4264, + "step": 4543 + }, + { + "epoch": 7.2704, + "grad_norm": 0.0, + "learning_rate": 0.00010099999999999999, + "loss": 4.8874, + "step": 4544 + }, + { + "epoch": 7.272, + "grad_norm": 0.0, + "learning_rate": 0.0001008, + "loss": 5.8265, + "step": 4545 + }, + { + "epoch": 7.2736, + "grad_norm": 0.0, + "learning_rate": 0.00010059999999999999, + "loss": 4.9633, + "step": 4546 + }, + { + "epoch": 7.2752, + "grad_norm": 0.0, + "learning_rate": 0.00010039999999999999, + "loss": 4.3749, + "step": 4547 + }, + { + "epoch": 7.2768, + "grad_norm": 0.0, + "learning_rate": 0.0001002, + "loss": 5.671, + "step": 4548 + }, + { + "epoch": 7.2783999999999995, + "grad_norm": 0.0, + "learning_rate": 9.999999999999999e-05, + "loss": 5.3198, + "step": 4549 + }, + { + "epoch": 7.28, + "grad_norm": 0.0, + "learning_rate": 9.979999999999999e-05, + "loss": 5.775, + "step": 4550 + }, + { + "epoch": 7.2816, + "grad_norm": 0.0, + "learning_rate": 9.96e-05, + "loss": 4.7278, + "step": 4551 + }, + { + "epoch": 7.2832, + "grad_norm": 0.0, + "learning_rate": 9.939999999999999e-05, + "loss": 4.5527, + "step": 4552 + }, + { + "epoch": 7.2848, + "grad_norm": 0.0, + "learning_rate": 9.919999999999999e-05, + "loss": 4.5961, + "step": 4553 + }, + { + "epoch": 7.2864, + "grad_norm": 0.0, + "learning_rate": 9.9e-05, + "loss": 5.379, + "step": 4554 + }, + { + "epoch": 7.288, + "grad_norm": 0.0, + "learning_rate": 9.879999999999999e-05, + "loss": 5.084, + "step": 4555 + }, + { + "epoch": 7.2896, + "grad_norm": 0.0, + "learning_rate": 9.859999999999998e-05, + "loss": 5.7861, + "step": 4556 + }, + { + "epoch": 7.2912, + "grad_norm": 0.0, + "learning_rate": 9.839999999999999e-05, + "loss": 4.1631, + "step": 4557 + }, + { + "epoch": 7.2928, + "grad_norm": 0.0, + "learning_rate": 9.819999999999999e-05, + "loss": 6.3292, + "step": 4558 + }, + { + "epoch": 7.2943999999999996, + "grad_norm": 0.0, + "learning_rate": 9.799999999999998e-05, + "loss": 5.4362, + "step": 4559 + }, + { + "epoch": 7.296, + "grad_norm": 0.0, + "learning_rate": 9.779999999999999e-05, + "loss": 4.5189, + "step": 4560 + }, + { + "epoch": 7.2976, + "grad_norm": 0.0, + "learning_rate": 9.759999999999999e-05, + "loss": 5.9651, + "step": 4561 + }, + { + "epoch": 7.2992, + "grad_norm": 0.0, + "learning_rate": 9.74e-05, + "loss": 4.46, + "step": 4562 + }, + { + "epoch": 7.3008, + "grad_norm": 0.0, + "learning_rate": 9.719999999999999e-05, + "loss": 4.4986, + "step": 4563 + }, + { + "epoch": 7.3024000000000004, + "grad_norm": 0.0, + "learning_rate": 9.699999999999999e-05, + "loss": 5.1565, + "step": 4564 + }, + { + "epoch": 7.304, + "grad_norm": 0.0, + "learning_rate": 9.68e-05, + "loss": 5.3843, + "step": 4565 + }, + { + "epoch": 7.3056, + "grad_norm": 0.0, + "learning_rate": 9.659999999999999e-05, + "loss": 5.6626, + "step": 4566 + }, + { + "epoch": 7.3072, + "grad_norm": 0.0, + "learning_rate": 9.64e-05, + "loss": 5.7662, + "step": 4567 + }, + { + "epoch": 7.3088, + "grad_norm": 0.0, + "learning_rate": 9.62e-05, + "loss": 5.5606, + "step": 4568 + }, + { + "epoch": 7.3104, + "grad_norm": 0.0, + "learning_rate": 9.599999999999999e-05, + "loss": 4.2976, + "step": 4569 + }, + { + "epoch": 7.312, + "grad_norm": 0.0, + "learning_rate": 9.58e-05, + "loss": 5.4516, + "step": 4570 + }, + { + "epoch": 7.3136, + "grad_norm": 0.0, + "learning_rate": 9.559999999999999e-05, + "loss": 6.1555, + "step": 4571 + }, + { + "epoch": 7.3152, + "grad_norm": 0.0, + "learning_rate": 9.539999999999999e-05, + "loss": 7.0566, + "step": 4572 + }, + { + "epoch": 7.3168, + "grad_norm": 0.0, + "learning_rate": 9.52e-05, + "loss": 6.2485, + "step": 4573 + }, + { + "epoch": 7.3184000000000005, + "grad_norm": 0.0, + "learning_rate": 9.499999999999999e-05, + "loss": 5.5418, + "step": 4574 + }, + { + "epoch": 7.32, + "grad_norm": 0.0, + "learning_rate": 9.479999999999999e-05, + "loss": 6.8718, + "step": 4575 + }, + { + "epoch": 7.3216, + "grad_norm": 0.0, + "learning_rate": 9.46e-05, + "loss": 7.5029, + "step": 4576 + }, + { + "epoch": 7.3232, + "grad_norm": 0.0, + "learning_rate": 9.439999999999999e-05, + "loss": 10.0158, + "step": 4577 + }, + { + "epoch": 7.3248, + "grad_norm": 0.0, + "learning_rate": 9.419999999999999e-05, + "loss": 7.9961, + "step": 4578 + }, + { + "epoch": 7.3264, + "grad_norm": 0.0, + "learning_rate": 9.4e-05, + "loss": 5.9108, + "step": 4579 + }, + { + "epoch": 7.328, + "grad_norm": 0.0, + "learning_rate": 9.379999999999999e-05, + "loss": 6.1267, + "step": 4580 + }, + { + "epoch": 7.3296, + "grad_norm": 0.0, + "learning_rate": 9.36e-05, + "loss": 7.2778, + "step": 4581 + }, + { + "epoch": 7.3312, + "grad_norm": 0.0, + "learning_rate": 9.34e-05, + "loss": 7.5165, + "step": 4582 + }, + { + "epoch": 7.3328, + "grad_norm": 0.0, + "learning_rate": 9.319999999999999e-05, + "loss": 6.0881, + "step": 4583 + }, + { + "epoch": 7.3344, + "grad_norm": 0.0, + "learning_rate": 9.3e-05, + "loss": 5.5847, + "step": 4584 + }, + { + "epoch": 7.336, + "grad_norm": 0.0, + "learning_rate": 9.279999999999999e-05, + "loss": 6.7273, + "step": 4585 + }, + { + "epoch": 7.3376, + "grad_norm": 0.0, + "learning_rate": 9.259999999999999e-05, + "loss": 7.1118, + "step": 4586 + }, + { + "epoch": 7.3392, + "grad_norm": 0.0, + "learning_rate": 9.24e-05, + "loss": 5.3542, + "step": 4587 + }, + { + "epoch": 7.3408, + "grad_norm": 0.0, + "learning_rate": 9.219999999999999e-05, + "loss": 5.7144, + "step": 4588 + }, + { + "epoch": 7.3424, + "grad_norm": 0.0, + "learning_rate": 9.199999999999999e-05, + "loss": 5.6577, + "step": 4589 + }, + { + "epoch": 7.344, + "grad_norm": 0.0, + "learning_rate": 9.18e-05, + "loss": 6.0528, + "step": 4590 + }, + { + "epoch": 7.3456, + "grad_norm": 0.0, + "learning_rate": 9.159999999999999e-05, + "loss": 5.3154, + "step": 4591 + }, + { + "epoch": 7.3472, + "grad_norm": 0.0, + "learning_rate": 9.139999999999999e-05, + "loss": 5.4556, + "step": 4592 + }, + { + "epoch": 7.3488, + "grad_norm": 0.0, + "learning_rate": 9.12e-05, + "loss": 6.534, + "step": 4593 + }, + { + "epoch": 7.3504, + "grad_norm": 0.0, + "learning_rate": 9.099999999999999e-05, + "loss": 4.7311, + "step": 4594 + }, + { + "epoch": 7.352, + "grad_norm": 0.0, + "learning_rate": 9.079999999999998e-05, + "loss": 5.2122, + "step": 4595 + }, + { + "epoch": 7.3536, + "grad_norm": 0.0, + "learning_rate": 9.059999999999999e-05, + "loss": 6.5695, + "step": 4596 + }, + { + "epoch": 7.3552, + "grad_norm": 0.0, + "learning_rate": 9.039999999999999e-05, + "loss": 4.9663, + "step": 4597 + }, + { + "epoch": 7.3568, + "grad_norm": 0.0, + "learning_rate": 9.02e-05, + "loss": 5.1939, + "step": 4598 + }, + { + "epoch": 7.3584, + "grad_norm": 0.0, + "learning_rate": 8.999999999999999e-05, + "loss": 4.7028, + "step": 4599 + }, + { + "epoch": 7.36, + "grad_norm": 0.0, + "learning_rate": 8.98e-05, + "loss": 5.3145, + "step": 4600 + }, + { + "epoch": 7.3616, + "grad_norm": 0.0, + "learning_rate": 8.96e-05, + "loss": 5.1801, + "step": 4601 + }, + { + "epoch": 7.3632, + "grad_norm": 0.0, + "learning_rate": 8.939999999999999e-05, + "loss": 4.6629, + "step": 4602 + }, + { + "epoch": 7.3648, + "grad_norm": 0.0, + "learning_rate": 8.92e-05, + "loss": 4.8691, + "step": 4603 + }, + { + "epoch": 7.3664, + "grad_norm": 0.0, + "learning_rate": 8.9e-05, + "loss": 5.7095, + "step": 4604 + }, + { + "epoch": 7.368, + "grad_norm": 0.0, + "learning_rate": 8.879999999999999e-05, + "loss": 5.1561, + "step": 4605 + }, + { + "epoch": 7.3696, + "grad_norm": 0.0, + "learning_rate": 8.86e-05, + "loss": 5.1505, + "step": 4606 + }, + { + "epoch": 7.3712, + "grad_norm": 0.0, + "learning_rate": 8.84e-05, + "loss": 6.3596, + "step": 4607 + }, + { + "epoch": 7.3728, + "grad_norm": 0.0, + "learning_rate": 8.819999999999999e-05, + "loss": 5.2175, + "step": 4608 + }, + { + "epoch": 7.3744, + "grad_norm": 0.0, + "learning_rate": 8.8e-05, + "loss": 6.8514, + "step": 4609 + }, + { + "epoch": 7.376, + "grad_norm": 0.0, + "learning_rate": 8.779999999999999e-05, + "loss": 5.192, + "step": 4610 + }, + { + "epoch": 7.3776, + "grad_norm": 0.0, + "learning_rate": 8.759999999999999e-05, + "loss": 5.23, + "step": 4611 + }, + { + "epoch": 7.3792, + "grad_norm": 0.0, + "learning_rate": 8.74e-05, + "loss": 4.7014, + "step": 4612 + }, + { + "epoch": 7.3808, + "grad_norm": 0.0, + "learning_rate": 8.719999999999999e-05, + "loss": 6.5306, + "step": 4613 + }, + { + "epoch": 7.3824, + "grad_norm": 0.0, + "learning_rate": 8.699999999999999e-05, + "loss": 7.1122, + "step": 4614 + }, + { + "epoch": 7.384, + "grad_norm": 0.0, + "learning_rate": 8.68e-05, + "loss": 5.4285, + "step": 4615 + }, + { + "epoch": 7.3856, + "grad_norm": 0.0, + "learning_rate": 8.659999999999999e-05, + "loss": 4.6387, + "step": 4616 + }, + { + "epoch": 7.3872, + "grad_norm": 0.0, + "learning_rate": 8.639999999999999e-05, + "loss": 5.2592, + "step": 4617 + }, + { + "epoch": 7.3888, + "grad_norm": 0.0, + "learning_rate": 8.62e-05, + "loss": 4.1712, + "step": 4618 + }, + { + "epoch": 7.3904, + "grad_norm": 0.0, + "learning_rate": 8.6e-05, + "loss": 5.242, + "step": 4619 + }, + { + "epoch": 7.392, + "grad_norm": 0.0, + "learning_rate": 8.579999999999998e-05, + "loss": 4.1312, + "step": 4620 + }, + { + "epoch": 7.3936, + "grad_norm": 0.0, + "learning_rate": 8.56e-05, + "loss": 4.5079, + "step": 4621 + }, + { + "epoch": 7.3952, + "grad_norm": 0.0, + "learning_rate": 8.54e-05, + "loss": 5.0698, + "step": 4622 + }, + { + "epoch": 7.3968, + "grad_norm": 0.0, + "learning_rate": 8.519999999999998e-05, + "loss": 4.3451, + "step": 4623 + }, + { + "epoch": 7.3984, + "grad_norm": 0.0, + "learning_rate": 8.499999999999999e-05, + "loss": 5.5572, + "step": 4624 + }, + { + "epoch": 7.4, + "grad_norm": NaN, + "learning_rate": 8.499999999999999e-05, + "loss": 4.5706, + "step": 4625 + }, + { + "epoch": 7.4016, + "grad_norm": 0.0, + "learning_rate": 8.48e-05, + "loss": 9.1471, + "step": 4626 + }, + { + "epoch": 7.4032, + "grad_norm": 0.0, + "learning_rate": 8.459999999999998e-05, + "loss": 5.5431, + "step": 4627 + }, + { + "epoch": 7.4048, + "grad_norm": 0.0, + "learning_rate": 8.439999999999999e-05, + "loss": 7.5224, + "step": 4628 + }, + { + "epoch": 7.4064, + "grad_norm": 0.0, + "learning_rate": 8.42e-05, + "loss": 8.1098, + "step": 4629 + }, + { + "epoch": 7.408, + "grad_norm": 0.0, + "learning_rate": 8.4e-05, + "loss": 6.6215, + "step": 4630 + }, + { + "epoch": 7.4096, + "grad_norm": 0.0, + "learning_rate": 8.379999999999999e-05, + "loss": 6.1875, + "step": 4631 + }, + { + "epoch": 7.4112, + "grad_norm": 0.0, + "learning_rate": 8.36e-05, + "loss": 6.8126, + "step": 4632 + }, + { + "epoch": 7.4128, + "grad_norm": 0.0, + "learning_rate": 8.34e-05, + "loss": 6.6636, + "step": 4633 + }, + { + "epoch": 7.4144, + "grad_norm": 0.0, + "learning_rate": 8.319999999999999e-05, + "loss": 6.3582, + "step": 4634 + }, + { + "epoch": 7.416, + "grad_norm": 0.0, + "learning_rate": 8.3e-05, + "loss": 7.9899, + "step": 4635 + }, + { + "epoch": 7.4176, + "grad_norm": 0.0, + "learning_rate": 8.28e-05, + "loss": 6.6879, + "step": 4636 + }, + { + "epoch": 7.4192, + "grad_norm": 0.0, + "learning_rate": 8.259999999999999e-05, + "loss": 9.5544, + "step": 4637 + }, + { + "epoch": 7.4208, + "grad_norm": 0.0, + "learning_rate": 8.24e-05, + "loss": 5.5785, + "step": 4638 + }, + { + "epoch": 7.4224, + "grad_norm": 0.0, + "learning_rate": 8.22e-05, + "loss": 5.8104, + "step": 4639 + }, + { + "epoch": 7.424, + "grad_norm": 0.0, + "learning_rate": 8.199999999999999e-05, + "loss": 7.2454, + "step": 4640 + }, + { + "epoch": 7.4256, + "grad_norm": 0.0, + "learning_rate": 8.18e-05, + "loss": 5.7244, + "step": 4641 + }, + { + "epoch": 7.4272, + "grad_norm": 0.0, + "learning_rate": 8.16e-05, + "loss": 5.099, + "step": 4642 + }, + { + "epoch": 7.4288, + "grad_norm": 0.0, + "learning_rate": 8.139999999999999e-05, + "loss": 5.5758, + "step": 4643 + }, + { + "epoch": 7.4304, + "grad_norm": 0.0, + "learning_rate": 8.12e-05, + "loss": 5.4576, + "step": 4644 + }, + { + "epoch": 7.432, + "grad_norm": 0.0, + "learning_rate": 8.1e-05, + "loss": 5.7294, + "step": 4645 + }, + { + "epoch": 7.4336, + "grad_norm": 0.0, + "learning_rate": 8.079999999999999e-05, + "loss": 5.7129, + "step": 4646 + }, + { + "epoch": 7.4352, + "grad_norm": 0.0, + "learning_rate": 8.06e-05, + "loss": 6.6869, + "step": 4647 + }, + { + "epoch": 7.4368, + "grad_norm": 0.0, + "learning_rate": 8.04e-05, + "loss": 4.8129, + "step": 4648 + }, + { + "epoch": 7.4384, + "grad_norm": 0.0, + "learning_rate": 8.019999999999998e-05, + "loss": 5.9987, + "step": 4649 + }, + { + "epoch": 7.44, + "grad_norm": 0.0, + "learning_rate": 7.999999999999999e-05, + "loss": 6.4853, + "step": 4650 + }, + { + "epoch": 7.4416, + "grad_norm": 0.0, + "learning_rate": 7.98e-05, + "loss": 4.5373, + "step": 4651 + }, + { + "epoch": 7.4432, + "grad_norm": 0.0, + "learning_rate": 7.959999999999998e-05, + "loss": 6.8861, + "step": 4652 + }, + { + "epoch": 7.4448, + "grad_norm": 0.0, + "learning_rate": 7.939999999999999e-05, + "loss": 6.1501, + "step": 4653 + }, + { + "epoch": 7.4464, + "grad_norm": 0.0, + "learning_rate": 7.92e-05, + "loss": 5.8647, + "step": 4654 + }, + { + "epoch": 7.448, + "grad_norm": 0.0, + "learning_rate": 7.899999999999998e-05, + "loss": 4.9637, + "step": 4655 + }, + { + "epoch": 7.4496, + "grad_norm": 0.0, + "learning_rate": 7.879999999999999e-05, + "loss": 5.0376, + "step": 4656 + }, + { + "epoch": 7.4512, + "grad_norm": 0.0, + "learning_rate": 7.86e-05, + "loss": 6.151, + "step": 4657 + }, + { + "epoch": 7.4528, + "grad_norm": 0.0, + "learning_rate": 7.839999999999998e-05, + "loss": 5.6569, + "step": 4658 + }, + { + "epoch": 7.4544, + "grad_norm": 0.0, + "learning_rate": 7.819999999999999e-05, + "loss": 6.8896, + "step": 4659 + }, + { + "epoch": 7.456, + "grad_norm": 0.0, + "learning_rate": 7.8e-05, + "loss": 8.7722, + "step": 4660 + }, + { + "epoch": 7.4576, + "grad_norm": 0.0, + "learning_rate": 7.780000000000001e-05, + "loss": 5.0723, + "step": 4661 + }, + { + "epoch": 7.4592, + "grad_norm": 0.0, + "learning_rate": 7.759999999999999e-05, + "loss": 6.1052, + "step": 4662 + }, + { + "epoch": 7.4608, + "grad_norm": 0.0, + "learning_rate": 7.74e-05, + "loss": 7.5137, + "step": 4663 + }, + { + "epoch": 7.4624, + "grad_norm": 0.0, + "learning_rate": 7.72e-05, + "loss": 6.1545, + "step": 4664 + }, + { + "epoch": 7.464, + "grad_norm": 0.0, + "learning_rate": 7.699999999999999e-05, + "loss": 5.8375, + "step": 4665 + }, + { + "epoch": 7.4656, + "grad_norm": 0.0, + "learning_rate": 7.68e-05, + "loss": 6.6794, + "step": 4666 + }, + { + "epoch": 7.4672, + "grad_norm": 0.0, + "learning_rate": 7.66e-05, + "loss": 6.124, + "step": 4667 + }, + { + "epoch": 7.4688, + "grad_norm": 0.0, + "learning_rate": 7.639999999999999e-05, + "loss": 5.5258, + "step": 4668 + }, + { + "epoch": 7.4704, + "grad_norm": 0.0, + "learning_rate": 7.62e-05, + "loss": 7.1363, + "step": 4669 + }, + { + "epoch": 7.4719999999999995, + "grad_norm": 0.0, + "learning_rate": 7.6e-05, + "loss": 5.5696, + "step": 4670 + }, + { + "epoch": 7.4736, + "grad_norm": 0.0, + "learning_rate": 7.579999999999999e-05, + "loss": 6.3159, + "step": 4671 + }, + { + "epoch": 7.4752, + "grad_norm": 0.0, + "learning_rate": 7.56e-05, + "loss": 4.9495, + "step": 4672 + }, + { + "epoch": 7.4768, + "grad_norm": 0.0, + "learning_rate": 7.54e-05, + "loss": 8.4961, + "step": 4673 + }, + { + "epoch": 7.4784, + "grad_norm": 0.0, + "learning_rate": 7.519999999999998e-05, + "loss": 5.5302, + "step": 4674 + }, + { + "epoch": 7.48, + "grad_norm": NaN, + "learning_rate": 7.519999999999998e-05, + "loss": 5.4775, + "step": 4675 + }, + { + "epoch": 7.4816, + "grad_norm": 0.0, + "learning_rate": 7.5e-05, + "loss": 8.5038, + "step": 4676 + }, + { + "epoch": 7.4832, + "grad_norm": 0.0, + "learning_rate": 7.479999999999999e-05, + "loss": 9.0773, + "step": 4677 + }, + { + "epoch": 7.4848, + "grad_norm": 0.0, + "learning_rate": 7.46e-05, + "loss": 5.5849, + "step": 4678 + }, + { + "epoch": 7.4864, + "grad_norm": 0.0, + "learning_rate": 7.439999999999999e-05, + "loss": 6.4928, + "step": 4679 + }, + { + "epoch": 7.4879999999999995, + "grad_norm": 0.0, + "learning_rate": 7.419999999999999e-05, + "loss": 6.92, + "step": 4680 + }, + { + "epoch": 7.4896, + "grad_norm": 0.0, + "learning_rate": 7.4e-05, + "loss": 9.2248, + "step": 4681 + }, + { + "epoch": 7.4912, + "grad_norm": 0.0, + "learning_rate": 7.379999999999999e-05, + "loss": 6.1579, + "step": 4682 + }, + { + "epoch": 7.4928, + "grad_norm": 0.0, + "learning_rate": 7.359999999999999e-05, + "loss": 5.7149, + "step": 4683 + }, + { + "epoch": 7.4944, + "grad_norm": 0.0, + "learning_rate": 7.34e-05, + "loss": 5.8885, + "step": 4684 + }, + { + "epoch": 7.496, + "grad_norm": 0.0, + "learning_rate": 7.319999999999999e-05, + "loss": 6.8303, + "step": 4685 + }, + { + "epoch": 7.4976, + "grad_norm": 0.0, + "learning_rate": 7.3e-05, + "loss": 5.4559, + "step": 4686 + }, + { + "epoch": 7.4992, + "grad_norm": 0.0, + "learning_rate": 7.28e-05, + "loss": 5.6105, + "step": 4687 + }, + { + "epoch": 7.5008, + "grad_norm": 0.0, + "learning_rate": 7.259999999999999e-05, + "loss": 6.1776, + "step": 4688 + }, + { + "epoch": 7.5024, + "grad_norm": 0.0, + "learning_rate": 7.24e-05, + "loss": 5.752, + "step": 4689 + }, + { + "epoch": 7.504, + "grad_norm": 0.0, + "learning_rate": 7.219999999999999e-05, + "loss": 4.8259, + "step": 4690 + }, + { + "epoch": 7.5056, + "grad_norm": 0.0, + "learning_rate": 7.199999999999999e-05, + "loss": 6.6351, + "step": 4691 + }, + { + "epoch": 7.5072, + "grad_norm": 0.0, + "learning_rate": 7.18e-05, + "loss": 4.9969, + "step": 4692 + }, + { + "epoch": 7.5088, + "grad_norm": 0.0, + "learning_rate": 7.159999999999999e-05, + "loss": 6.3686, + "step": 4693 + }, + { + "epoch": 7.5104, + "grad_norm": 0.0, + "learning_rate": 7.139999999999999e-05, + "loss": 5.6769, + "step": 4694 + }, + { + "epoch": 7.5120000000000005, + "grad_norm": 0.0, + "learning_rate": 7.12e-05, + "loss": 4.8072, + "step": 4695 + }, + { + "epoch": 7.5136, + "grad_norm": 0.0, + "learning_rate": 7.099999999999999e-05, + "loss": 4.6194, + "step": 4696 + }, + { + "epoch": 7.5152, + "grad_norm": 0.0, + "learning_rate": 7.079999999999999e-05, + "loss": 6.4218, + "step": 4697 + }, + { + "epoch": 7.5168, + "grad_norm": 0.0, + "learning_rate": 7.06e-05, + "loss": 6.6693, + "step": 4698 + }, + { + "epoch": 7.5184, + "grad_norm": 0.0, + "learning_rate": 7.039999999999999e-05, + "loss": 5.2383, + "step": 4699 + }, + { + "epoch": 7.52, + "grad_norm": 0.0, + "learning_rate": 7.02e-05, + "loss": 5.0857, + "step": 4700 + }, + { + "epoch": 7.5216, + "grad_norm": 0.0, + "learning_rate": 7e-05, + "loss": 5.7426, + "step": 4701 + }, + { + "epoch": 7.5232, + "grad_norm": 0.0, + "learning_rate": 6.979999999999999e-05, + "loss": 6.5368, + "step": 4702 + }, + { + "epoch": 7.5248, + "grad_norm": 0.0, + "learning_rate": 6.96e-05, + "loss": 4.9804, + "step": 4703 + }, + { + "epoch": 7.5264, + "grad_norm": 0.0, + "learning_rate": 6.939999999999999e-05, + "loss": 6.113, + "step": 4704 + }, + { + "epoch": 7.5280000000000005, + "grad_norm": 0.0, + "learning_rate": 6.919999999999999e-05, + "loss": 5.4554, + "step": 4705 + }, + { + "epoch": 7.5296, + "grad_norm": 0.0, + "learning_rate": 6.9e-05, + "loss": 5.2051, + "step": 4706 + }, + { + "epoch": 7.5312, + "grad_norm": 0.0, + "learning_rate": 6.879999999999999e-05, + "loss": 3.9168, + "step": 4707 + }, + { + "epoch": 7.5328, + "grad_norm": 0.0, + "learning_rate": 6.859999999999999e-05, + "loss": 5.3433, + "step": 4708 + }, + { + "epoch": 7.5344, + "grad_norm": 0.0, + "learning_rate": 6.84e-05, + "loss": 5.7255, + "step": 4709 + }, + { + "epoch": 7.536, + "grad_norm": 0.0, + "learning_rate": 6.819999999999999e-05, + "loss": 5.2953, + "step": 4710 + }, + { + "epoch": 7.5376, + "grad_norm": 0.0, + "learning_rate": 6.799999999999999e-05, + "loss": 3.8887, + "step": 4711 + }, + { + "epoch": 7.5392, + "grad_norm": 0.0, + "learning_rate": 6.78e-05, + "loss": 5.8622, + "step": 4712 + }, + { + "epoch": 7.5408, + "grad_norm": 0.0, + "learning_rate": 6.759999999999999e-05, + "loss": 6.3661, + "step": 4713 + }, + { + "epoch": 7.5424, + "grad_norm": 0.0, + "learning_rate": 6.739999999999998e-05, + "loss": 4.3511, + "step": 4714 + }, + { + "epoch": 7.5440000000000005, + "grad_norm": 0.0, + "learning_rate": 6.72e-05, + "loss": 4.4677, + "step": 4715 + }, + { + "epoch": 7.5456, + "grad_norm": 0.0, + "learning_rate": 6.699999999999999e-05, + "loss": 4.8682, + "step": 4716 + }, + { + "epoch": 7.5472, + "grad_norm": 0.0, + "learning_rate": 6.68e-05, + "loss": 5.301, + "step": 4717 + }, + { + "epoch": 7.5488, + "grad_norm": 0.0, + "learning_rate": 6.659999999999999e-05, + "loss": 5.2216, + "step": 4718 + }, + { + "epoch": 7.5504, + "grad_norm": 0.0, + "learning_rate": 6.639999999999999e-05, + "loss": 6.1162, + "step": 4719 + }, + { + "epoch": 7.552, + "grad_norm": 0.0, + "learning_rate": 6.62e-05, + "loss": 5.4302, + "step": 4720 + }, + { + "epoch": 7.5536, + "grad_norm": 0.0, + "learning_rate": 6.599999999999999e-05, + "loss": 6.0123, + "step": 4721 + }, + { + "epoch": 7.5552, + "grad_norm": 0.0, + "learning_rate": 6.579999999999999e-05, + "loss": 5.5139, + "step": 4722 + }, + { + "epoch": 7.5568, + "grad_norm": 0.0, + "learning_rate": 6.56e-05, + "loss": 4.56, + "step": 4723 + }, + { + "epoch": 7.5584, + "grad_norm": 0.0, + "learning_rate": 6.539999999999999e-05, + "loss": 5.2422, + "step": 4724 + }, + { + "epoch": 7.5600000000000005, + "grad_norm": NaN, + "learning_rate": 6.539999999999999e-05, + "loss": 6.8643, + "step": 4725 + }, + { + "epoch": 7.5616, + "grad_norm": 0.0, + "learning_rate": 6.519999999999999e-05, + "loss": 7.8714, + "step": 4726 + }, + { + "epoch": 7.5632, + "grad_norm": 0.0, + "learning_rate": 6.5e-05, + "loss": 6.7089, + "step": 4727 + }, + { + "epoch": 7.5648, + "grad_norm": 0.0, + "learning_rate": 6.479999999999999e-05, + "loss": 6.5865, + "step": 4728 + }, + { + "epoch": 7.5664, + "grad_norm": 0.0, + "learning_rate": 6.459999999999998e-05, + "loss": 6.3169, + "step": 4729 + }, + { + "epoch": 7.568, + "grad_norm": 0.0, + "learning_rate": 6.44e-05, + "loss": 7.2559, + "step": 4730 + }, + { + "epoch": 7.5696, + "grad_norm": 0.0, + "learning_rate": 6.419999999999999e-05, + "loss": 5.5722, + "step": 4731 + }, + { + "epoch": 7.5712, + "grad_norm": 0.0, + "learning_rate": 6.4e-05, + "loss": 6.0926, + "step": 4732 + }, + { + "epoch": 7.5728, + "grad_norm": 0.0, + "learning_rate": 6.379999999999999e-05, + "loss": 7.0853, + "step": 4733 + }, + { + "epoch": 7.5744, + "grad_norm": 0.0, + "learning_rate": 6.359999999999999e-05, + "loss": 7.875, + "step": 4734 + }, + { + "epoch": 7.576, + "grad_norm": 0.0, + "learning_rate": 6.34e-05, + "loss": 5.6398, + "step": 4735 + }, + { + "epoch": 7.5776, + "grad_norm": 0.0, + "learning_rate": 6.319999999999999e-05, + "loss": 5.8603, + "step": 4736 + }, + { + "epoch": 7.5792, + "grad_norm": 0.0, + "learning_rate": 6.299999999999999e-05, + "loss": 6.0807, + "step": 4737 + }, + { + "epoch": 7.5808, + "grad_norm": 0.0, + "learning_rate": 6.28e-05, + "loss": 6.1367, + "step": 4738 + }, + { + "epoch": 7.5824, + "grad_norm": 0.0, + "learning_rate": 6.259999999999999e-05, + "loss": 6.0631, + "step": 4739 + }, + { + "epoch": 7.584, + "grad_norm": 0.0, + "learning_rate": 6.239999999999999e-05, + "loss": 8.0581, + "step": 4740 + }, + { + "epoch": 7.5856, + "grad_norm": 0.0, + "learning_rate": 6.22e-05, + "loss": 6.0168, + "step": 4741 + }, + { + "epoch": 7.5872, + "grad_norm": 0.0, + "learning_rate": 6.199999999999999e-05, + "loss": 4.8141, + "step": 4742 + }, + { + "epoch": 7.5888, + "grad_norm": 0.0, + "learning_rate": 6.18e-05, + "loss": 5.3598, + "step": 4743 + }, + { + "epoch": 7.5904, + "grad_norm": 0.0, + "learning_rate": 6.159999999999999e-05, + "loss": 6.5929, + "step": 4744 + }, + { + "epoch": 7.592, + "grad_norm": 0.0, + "learning_rate": 6.139999999999999e-05, + "loss": 6.7241, + "step": 4745 + }, + { + "epoch": 7.5936, + "grad_norm": 0.0, + "learning_rate": 6.12e-05, + "loss": 5.0923, + "step": 4746 + }, + { + "epoch": 7.5952, + "grad_norm": 0.0, + "learning_rate": 6.1e-05, + "loss": 5.385, + "step": 4747 + }, + { + "epoch": 7.5968, + "grad_norm": 0.0, + "learning_rate": 6.0799999999999994e-05, + "loss": 6.4209, + "step": 4748 + }, + { + "epoch": 7.5984, + "grad_norm": 0.0, + "learning_rate": 6.0599999999999996e-05, + "loss": 5.168, + "step": 4749 + }, + { + "epoch": 7.6, + "grad_norm": 0.0, + "learning_rate": 6.04e-05, + "loss": 6.4378, + "step": 4750 + }, + { + "epoch": 7.6016, + "grad_norm": 0.0, + "learning_rate": 6.019999999999999e-05, + "loss": 6.2172, + "step": 4751 + }, + { + "epoch": 7.6032, + "grad_norm": 0.0, + "learning_rate": 5.9999999999999995e-05, + "loss": 6.6248, + "step": 4752 + }, + { + "epoch": 7.6048, + "grad_norm": 0.0, + "learning_rate": 5.98e-05, + "loss": 5.5064, + "step": 4753 + }, + { + "epoch": 7.6064, + "grad_norm": 0.0, + "learning_rate": 5.959999999999999e-05, + "loss": 4.5904, + "step": 4754 + }, + { + "epoch": 7.608, + "grad_norm": 0.0, + "learning_rate": 5.94e-05, + "loss": 5.6806, + "step": 4755 + }, + { + "epoch": 7.6096, + "grad_norm": 0.0, + "learning_rate": 5.9199999999999996e-05, + "loss": 5.9776, + "step": 4756 + }, + { + "epoch": 7.6112, + "grad_norm": 0.0, + "learning_rate": 5.899999999999999e-05, + "loss": 4.1508, + "step": 4757 + }, + { + "epoch": 7.6128, + "grad_norm": 0.0, + "learning_rate": 5.88e-05, + "loss": 5.5701, + "step": 4758 + }, + { + "epoch": 7.6144, + "grad_norm": 0.0, + "learning_rate": 5.8599999999999995e-05, + "loss": 4.0837, + "step": 4759 + }, + { + "epoch": 7.616, + "grad_norm": 0.0, + "learning_rate": 5.839999999999999e-05, + "loss": 5.0102, + "step": 4760 + }, + { + "epoch": 7.6176, + "grad_norm": 0.0, + "learning_rate": 5.82e-05, + "loss": 5.8299, + "step": 4761 + }, + { + "epoch": 7.6192, + "grad_norm": 0.0, + "learning_rate": 5.7999999999999994e-05, + "loss": 5.4028, + "step": 4762 + }, + { + "epoch": 7.6208, + "grad_norm": 0.0, + "learning_rate": 5.78e-05, + "loss": 6.6224, + "step": 4763 + }, + { + "epoch": 7.6224, + "grad_norm": 0.0, + "learning_rate": 5.76e-05, + "loss": 8.957, + "step": 4764 + }, + { + "epoch": 7.624, + "grad_norm": 0.0, + "learning_rate": 5.739999999999999e-05, + "loss": 5.6631, + "step": 4765 + }, + { + "epoch": 7.6256, + "grad_norm": 0.0, + "learning_rate": 5.72e-05, + "loss": 4.9659, + "step": 4766 + }, + { + "epoch": 7.6272, + "grad_norm": 0.0, + "learning_rate": 5.6999999999999996e-05, + "loss": 4.7382, + "step": 4767 + }, + { + "epoch": 7.6288, + "grad_norm": 0.0, + "learning_rate": 5.679999999999999e-05, + "loss": 5.591, + "step": 4768 + }, + { + "epoch": 7.6304, + "grad_norm": 0.0, + "learning_rate": 5.66e-05, + "loss": 6.0642, + "step": 4769 + }, + { + "epoch": 7.632, + "grad_norm": 0.0, + "learning_rate": 5.6399999999999995e-05, + "loss": 4.8258, + "step": 4770 + }, + { + "epoch": 7.6336, + "grad_norm": 0.0, + "learning_rate": 5.619999999999999e-05, + "loss": 5.002, + "step": 4771 + }, + { + "epoch": 7.6352, + "grad_norm": 0.0, + "learning_rate": 5.6e-05, + "loss": 5.5322, + "step": 4772 + }, + { + "epoch": 7.6368, + "grad_norm": 0.0, + "learning_rate": 5.5799999999999994e-05, + "loss": 4.1246, + "step": 4773 + }, + { + "epoch": 7.6384, + "grad_norm": 0.0, + "learning_rate": 5.559999999999999e-05, + "loss": 7.411, + "step": 4774 + }, + { + "epoch": 7.64, + "grad_norm": NaN, + "learning_rate": 5.559999999999999e-05, + "loss": 6.6256, + "step": 4775 + }, + { + "epoch": 7.6416, + "grad_norm": 0.0, + "learning_rate": 5.54e-05, + "loss": 7.4721, + "step": 4776 + }, + { + "epoch": 7.6432, + "grad_norm": 0.0, + "learning_rate": 5.519999999999999e-05, + "loss": 8.1493, + "step": 4777 + }, + { + "epoch": 7.6448, + "grad_norm": 0.0, + "learning_rate": 5.499999999999999e-05, + "loss": 6.7723, + "step": 4778 + }, + { + "epoch": 7.6464, + "grad_norm": 0.0, + "learning_rate": 5.48e-05, + "loss": 5.6596, + "step": 4779 + }, + { + "epoch": 7.648, + "grad_norm": 0.0, + "learning_rate": 5.459999999999999e-05, + "loss": 5.8716, + "step": 4780 + }, + { + "epoch": 7.6495999999999995, + "grad_norm": 0.0, + "learning_rate": 5.44e-05, + "loss": 7.7081, + "step": 4781 + }, + { + "epoch": 7.6512, + "grad_norm": 0.0, + "learning_rate": 5.4199999999999996e-05, + "loss": 5.1134, + "step": 4782 + }, + { + "epoch": 7.6528, + "grad_norm": 0.0, + "learning_rate": 5.399999999999999e-05, + "loss": 5.7436, + "step": 4783 + }, + { + "epoch": 7.6544, + "grad_norm": 0.0, + "learning_rate": 5.38e-05, + "loss": 5.7095, + "step": 4784 + }, + { + "epoch": 7.656, + "grad_norm": 0.0, + "learning_rate": 5.3599999999999995e-05, + "loss": 6.8807, + "step": 4785 + }, + { + "epoch": 7.6576, + "grad_norm": 0.0, + "learning_rate": 5.339999999999999e-05, + "loss": 4.901, + "step": 4786 + }, + { + "epoch": 7.6592, + "grad_norm": 0.0, + "learning_rate": 5.32e-05, + "loss": 4.5638, + "step": 4787 + }, + { + "epoch": 7.6608, + "grad_norm": 0.0, + "learning_rate": 5.2999999999999994e-05, + "loss": 6.5257, + "step": 4788 + }, + { + "epoch": 7.6624, + "grad_norm": 0.0, + "learning_rate": 5.279999999999999e-05, + "loss": 6.265, + "step": 4789 + }, + { + "epoch": 7.664, + "grad_norm": 0.0, + "learning_rate": 5.26e-05, + "loss": 4.5508, + "step": 4790 + }, + { + "epoch": 7.6655999999999995, + "grad_norm": 0.0, + "learning_rate": 5.239999999999999e-05, + "loss": 5.7981, + "step": 4791 + }, + { + "epoch": 7.6672, + "grad_norm": 0.0, + "learning_rate": 5.2199999999999995e-05, + "loss": 6.1889, + "step": 4792 + }, + { + "epoch": 7.6688, + "grad_norm": 0.0, + "learning_rate": 5.2e-05, + "loss": 6.1205, + "step": 4793 + }, + { + "epoch": 7.6704, + "grad_norm": 0.0, + "learning_rate": 5.179999999999999e-05, + "loss": 8.812, + "step": 4794 + }, + { + "epoch": 7.672, + "grad_norm": 0.0, + "learning_rate": 5.1599999999999994e-05, + "loss": 5.5074, + "step": 4795 + }, + { + "epoch": 7.6736, + "grad_norm": 0.0, + "learning_rate": 5.1399999999999996e-05, + "loss": 5.5454, + "step": 4796 + }, + { + "epoch": 7.6752, + "grad_norm": 0.0, + "learning_rate": 5.119999999999999e-05, + "loss": 6.1127, + "step": 4797 + }, + { + "epoch": 7.6768, + "grad_norm": 0.0, + "learning_rate": 5.1e-05, + "loss": 5.1458, + "step": 4798 + }, + { + "epoch": 7.6784, + "grad_norm": 0.0, + "learning_rate": 5.0799999999999995e-05, + "loss": 5.9438, + "step": 4799 + }, + { + "epoch": 7.68, + "grad_norm": 0.0, + "learning_rate": 5.06e-05, + "loss": 4.0715, + "step": 4800 + }, + { + "epoch": 7.6815999999999995, + "grad_norm": 0.0, + "learning_rate": 5.04e-05, + "loss": 4.9464, + "step": 4801 + }, + { + "epoch": 7.6832, + "grad_norm": 0.0, + "learning_rate": 5.0199999999999994e-05, + "loss": 4.4932, + "step": 4802 + }, + { + "epoch": 7.6848, + "grad_norm": 0.0, + "learning_rate": 4.9999999999999996e-05, + "loss": 4.6418, + "step": 4803 + }, + { + "epoch": 7.6864, + "grad_norm": 0.0, + "learning_rate": 4.98e-05, + "loss": 5.4522, + "step": 4804 + }, + { + "epoch": 7.688, + "grad_norm": 0.0, + "learning_rate": 4.959999999999999e-05, + "loss": 6.5306, + "step": 4805 + }, + { + "epoch": 7.6896, + "grad_norm": 0.0, + "learning_rate": 4.9399999999999995e-05, + "loss": 4.8029, + "step": 4806 + }, + { + "epoch": 7.6912, + "grad_norm": 0.0, + "learning_rate": 4.9199999999999997e-05, + "loss": 5.2924, + "step": 4807 + }, + { + "epoch": 7.6928, + "grad_norm": 0.0, + "learning_rate": 4.899999999999999e-05, + "loss": 9.3218, + "step": 4808 + }, + { + "epoch": 7.6944, + "grad_norm": 0.0, + "learning_rate": 4.8799999999999994e-05, + "loss": 4.6029, + "step": 4809 + }, + { + "epoch": 7.696, + "grad_norm": 0.0, + "learning_rate": 4.8599999999999995e-05, + "loss": 5.043, + "step": 4810 + }, + { + "epoch": 7.6975999999999996, + "grad_norm": 0.0, + "learning_rate": 4.84e-05, + "loss": 5.9198, + "step": 4811 + }, + { + "epoch": 7.6992, + "grad_norm": 0.0, + "learning_rate": 4.82e-05, + "loss": 5.3473, + "step": 4812 + }, + { + "epoch": 7.7008, + "grad_norm": 0.0, + "learning_rate": 4.7999999999999994e-05, + "loss": 5.6345, + "step": 4813 + }, + { + "epoch": 7.7024, + "grad_norm": 0.0, + "learning_rate": 4.7799999999999996e-05, + "loss": 5.4249, + "step": 4814 + }, + { + "epoch": 7.704, + "grad_norm": 0.0, + "learning_rate": 4.76e-05, + "loss": 5.0125, + "step": 4815 + }, + { + "epoch": 7.7056000000000004, + "grad_norm": 0.0, + "learning_rate": 4.7399999999999993e-05, + "loss": 5.9589, + "step": 4816 + }, + { + "epoch": 7.7072, + "grad_norm": 0.0, + "learning_rate": 4.7199999999999995e-05, + "loss": 4.4343, + "step": 4817 + }, + { + "epoch": 7.7088, + "grad_norm": 0.0, + "learning_rate": 4.7e-05, + "loss": 4.4523, + "step": 4818 + }, + { + "epoch": 7.7104, + "grad_norm": 0.0, + "learning_rate": 4.68e-05, + "loss": 4.799, + "step": 4819 + }, + { + "epoch": 7.712, + "grad_norm": 0.0, + "learning_rate": 4.6599999999999994e-05, + "loss": 5.7031, + "step": 4820 + }, + { + "epoch": 7.7136, + "grad_norm": 0.0, + "learning_rate": 4.6399999999999996e-05, + "loss": 4.2676, + "step": 4821 + }, + { + "epoch": 7.7152, + "grad_norm": 0.0, + "learning_rate": 4.62e-05, + "loss": 5.4716, + "step": 4822 + }, + { + "epoch": 7.7168, + "grad_norm": 0.0, + "learning_rate": 4.599999999999999e-05, + "loss": 5.9318, + "step": 4823 + }, + { + "epoch": 7.7184, + "grad_norm": 0.0, + "learning_rate": 4.5799999999999995e-05, + "loss": 4.9803, + "step": 4824 + }, + { + "epoch": 7.72, + "grad_norm": NaN, + "learning_rate": 4.5799999999999995e-05, + "loss": 4.7987, + "step": 4825 + }, + { + "epoch": 7.7216000000000005, + "grad_norm": 0.0, + "learning_rate": 4.56e-05, + "loss": 8.6466, + "step": 4826 + }, + { + "epoch": 7.7232, + "grad_norm": 0.0, + "learning_rate": 4.539999999999999e-05, + "loss": 7.1114, + "step": 4827 + }, + { + "epoch": 7.7248, + "grad_norm": 0.0, + "learning_rate": 4.5199999999999994e-05, + "loss": 7.2577, + "step": 4828 + }, + { + "epoch": 7.7264, + "grad_norm": 0.0, + "learning_rate": 4.4999999999999996e-05, + "loss": 6.5141, + "step": 4829 + }, + { + "epoch": 7.728, + "grad_norm": 0.0, + "learning_rate": 4.48e-05, + "loss": 6.3293, + "step": 4830 + }, + { + "epoch": 7.7296, + "grad_norm": 0.0, + "learning_rate": 4.46e-05, + "loss": 7.6604, + "step": 4831 + }, + { + "epoch": 7.7312, + "grad_norm": 0.0, + "learning_rate": 4.4399999999999995e-05, + "loss": 5.1769, + "step": 4832 + }, + { + "epoch": 7.7328, + "grad_norm": 0.0, + "learning_rate": 4.42e-05, + "loss": 6.9124, + "step": 4833 + }, + { + "epoch": 7.7344, + "grad_norm": 0.0, + "learning_rate": 4.4e-05, + "loss": 4.8715, + "step": 4834 + }, + { + "epoch": 7.736, + "grad_norm": 0.0, + "learning_rate": 4.3799999999999994e-05, + "loss": 7.8529, + "step": 4835 + }, + { + "epoch": 7.7376000000000005, + "grad_norm": 0.0, + "learning_rate": 4.3599999999999996e-05, + "loss": 6.493, + "step": 4836 + }, + { + "epoch": 7.7392, + "grad_norm": 0.0, + "learning_rate": 4.34e-05, + "loss": 6.1691, + "step": 4837 + }, + { + "epoch": 7.7408, + "grad_norm": 0.0, + "learning_rate": 4.319999999999999e-05, + "loss": 5.6671, + "step": 4838 + }, + { + "epoch": 7.7424, + "grad_norm": 0.0, + "learning_rate": 4.3e-05, + "loss": 6.1603, + "step": 4839 + }, + { + "epoch": 7.744, + "grad_norm": 0.0, + "learning_rate": 4.28e-05, + "loss": 6.2462, + "step": 4840 + }, + { + "epoch": 7.7456, + "grad_norm": 0.0, + "learning_rate": 4.259999999999999e-05, + "loss": 4.9183, + "step": 4841 + }, + { + "epoch": 7.7472, + "grad_norm": 0.0, + "learning_rate": 4.24e-05, + "loss": 8.3458, + "step": 4842 + }, + { + "epoch": 7.7488, + "grad_norm": 0.0, + "learning_rate": 4.2199999999999996e-05, + "loss": 5.0508, + "step": 4843 + }, + { + "epoch": 7.7504, + "grad_norm": 0.0, + "learning_rate": 4.2e-05, + "loss": 5.1722, + "step": 4844 + }, + { + "epoch": 7.752, + "grad_norm": 0.0, + "learning_rate": 4.18e-05, + "loss": 5.6128, + "step": 4845 + }, + { + "epoch": 7.7536000000000005, + "grad_norm": 0.0, + "learning_rate": 4.1599999999999995e-05, + "loss": 5.7953, + "step": 4846 + }, + { + "epoch": 7.7552, + "grad_norm": 0.0, + "learning_rate": 4.14e-05, + "loss": 5.9053, + "step": 4847 + }, + { + "epoch": 7.7568, + "grad_norm": 0.0, + "learning_rate": 4.12e-05, + "loss": 4.2809, + "step": 4848 + }, + { + "epoch": 7.7584, + "grad_norm": 0.0, + "learning_rate": 4.0999999999999994e-05, + "loss": 6.285, + "step": 4849 + }, + { + "epoch": 7.76, + "grad_norm": 0.0, + "learning_rate": 4.08e-05, + "loss": 6.2999, + "step": 4850 + }, + { + "epoch": 7.7616, + "grad_norm": 0.0, + "learning_rate": 4.06e-05, + "loss": 5.6332, + "step": 4851 + }, + { + "epoch": 7.7632, + "grad_norm": 0.0, + "learning_rate": 4.039999999999999e-05, + "loss": 5.5603, + "step": 4852 + }, + { + "epoch": 7.7648, + "grad_norm": 0.0, + "learning_rate": 4.02e-05, + "loss": 6.1317, + "step": 4853 + }, + { + "epoch": 7.7664, + "grad_norm": 0.0, + "learning_rate": 3.9999999999999996e-05, + "loss": 4.6544, + "step": 4854 + }, + { + "epoch": 7.768, + "grad_norm": 0.0, + "learning_rate": 3.979999999999999e-05, + "loss": 5.0832, + "step": 4855 + }, + { + "epoch": 7.7696, + "grad_norm": 0.0, + "learning_rate": 3.96e-05, + "loss": 6.3336, + "step": 4856 + }, + { + "epoch": 7.7712, + "grad_norm": 0.0, + "learning_rate": 3.9399999999999995e-05, + "loss": 5.3673, + "step": 4857 + }, + { + "epoch": 7.7728, + "grad_norm": 0.0, + "learning_rate": 3.919999999999999e-05, + "loss": 5.3218, + "step": 4858 + }, + { + "epoch": 7.7744, + "grad_norm": 0.0, + "learning_rate": 3.9e-05, + "loss": 4.1707, + "step": 4859 + }, + { + "epoch": 7.776, + "grad_norm": 0.0, + "learning_rate": 3.8799999999999994e-05, + "loss": 4.8736, + "step": 4860 + }, + { + "epoch": 7.7776, + "grad_norm": 0.0, + "learning_rate": 3.86e-05, + "loss": 4.311, + "step": 4861 + }, + { + "epoch": 7.7792, + "grad_norm": 0.0, + "learning_rate": 3.84e-05, + "loss": 6.2161, + "step": 4862 + }, + { + "epoch": 7.7808, + "grad_norm": 0.0, + "learning_rate": 3.8199999999999993e-05, + "loss": 7.312, + "step": 4863 + }, + { + "epoch": 7.7824, + "grad_norm": 0.0, + "learning_rate": 3.8e-05, + "loss": 5.7529, + "step": 4864 + }, + { + "epoch": 7.784, + "grad_norm": 0.0, + "learning_rate": 3.78e-05, + "loss": 5.6757, + "step": 4865 + }, + { + "epoch": 7.7856, + "grad_norm": 0.0, + "learning_rate": 3.759999999999999e-05, + "loss": 5.6621, + "step": 4866 + }, + { + "epoch": 7.7872, + "grad_norm": 0.0, + "learning_rate": 3.7399999999999994e-05, + "loss": 8.4583, + "step": 4867 + }, + { + "epoch": 7.7888, + "grad_norm": 0.0, + "learning_rate": 3.7199999999999996e-05, + "loss": 5.1848, + "step": 4868 + }, + { + "epoch": 7.7904, + "grad_norm": 0.0, + "learning_rate": 3.7e-05, + "loss": 6.9994, + "step": 4869 + }, + { + "epoch": 7.792, + "grad_norm": 0.0, + "learning_rate": 3.679999999999999e-05, + "loss": 5.8274, + "step": 4870 + }, + { + "epoch": 7.7936, + "grad_norm": 0.0, + "learning_rate": 3.6599999999999995e-05, + "loss": 6.3826, + "step": 4871 + }, + { + "epoch": 7.7952, + "grad_norm": 0.0, + "learning_rate": 3.64e-05, + "loss": 3.619, + "step": 4872 + }, + { + "epoch": 7.7968, + "grad_norm": 0.0, + "learning_rate": 3.62e-05, + "loss": 5.4668, + "step": 4873 + }, + { + "epoch": 7.7984, + "grad_norm": 0.0, + "learning_rate": 3.5999999999999994e-05, + "loss": 5.2385, + "step": 4874 + }, + { + "epoch": 7.8, + "grad_norm": NaN, + "learning_rate": 3.5999999999999994e-05, + "loss": 5.6297, + "step": 4875 + }, + { + "epoch": 7.8016, + "grad_norm": 0.0, + "learning_rate": 3.5799999999999996e-05, + "loss": 8.6246, + "step": 4876 + }, + { + "epoch": 7.8032, + "grad_norm": 0.0, + "learning_rate": 3.56e-05, + "loss": 6.7583, + "step": 4877 + }, + { + "epoch": 7.8048, + "grad_norm": 0.0, + "learning_rate": 3.539999999999999e-05, + "loss": 8.9947, + "step": 4878 + }, + { + "epoch": 7.8064, + "grad_norm": 0.0, + "learning_rate": 3.5199999999999995e-05, + "loss": 7.3183, + "step": 4879 + }, + { + "epoch": 7.808, + "grad_norm": 0.0, + "learning_rate": 3.5e-05, + "loss": 5.5067, + "step": 4880 + }, + { + "epoch": 7.8096, + "grad_norm": 0.0, + "learning_rate": 3.48e-05, + "loss": 6.5109, + "step": 4881 + }, + { + "epoch": 7.8112, + "grad_norm": 0.0, + "learning_rate": 3.4599999999999994e-05, + "loss": 6.328, + "step": 4882 + }, + { + "epoch": 7.8128, + "grad_norm": 0.0, + "learning_rate": 3.4399999999999996e-05, + "loss": 6.1293, + "step": 4883 + }, + { + "epoch": 7.8144, + "grad_norm": 0.0, + "learning_rate": 3.42e-05, + "loss": 5.8605, + "step": 4884 + }, + { + "epoch": 7.816, + "grad_norm": 0.0, + "learning_rate": 3.399999999999999e-05, + "loss": 6.1975, + "step": 4885 + }, + { + "epoch": 7.8176, + "grad_norm": 0.0, + "learning_rate": 3.3799999999999995e-05, + "loss": 5.0323, + "step": 4886 + }, + { + "epoch": 7.8192, + "grad_norm": 0.0, + "learning_rate": 3.36e-05, + "loss": 6.6259, + "step": 4887 + }, + { + "epoch": 7.8208, + "grad_norm": 0.0, + "learning_rate": 3.34e-05, + "loss": 4.7444, + "step": 4888 + }, + { + "epoch": 7.8224, + "grad_norm": 0.0, + "learning_rate": 3.3199999999999994e-05, + "loss": 6.39, + "step": 4889 + }, + { + "epoch": 7.824, + "grad_norm": 0.0, + "learning_rate": 3.2999999999999996e-05, + "loss": 5.5457, + "step": 4890 + }, + { + "epoch": 7.8256, + "grad_norm": 0.0, + "learning_rate": 3.28e-05, + "loss": 5.2619, + "step": 4891 + }, + { + "epoch": 7.8272, + "grad_norm": 0.0, + "learning_rate": 3.259999999999999e-05, + "loss": 5.6246, + "step": 4892 + }, + { + "epoch": 7.8288, + "grad_norm": 0.0, + "learning_rate": 3.2399999999999995e-05, + "loss": 5.2762, + "step": 4893 + }, + { + "epoch": 7.8304, + "grad_norm": 0.0, + "learning_rate": 3.22e-05, + "loss": 4.8512, + "step": 4894 + }, + { + "epoch": 7.832, + "grad_norm": 0.0, + "learning_rate": 3.2e-05, + "loss": 5.2384, + "step": 4895 + }, + { + "epoch": 7.8336, + "grad_norm": 0.0, + "learning_rate": 3.1799999999999994e-05, + "loss": 5.2882, + "step": 4896 + }, + { + "epoch": 7.8352, + "grad_norm": 0.0, + "learning_rate": 3.1599999999999996e-05, + "loss": 5.5722, + "step": 4897 + }, + { + "epoch": 7.8368, + "grad_norm": 0.0, + "learning_rate": 3.14e-05, + "loss": 6.268, + "step": 4898 + }, + { + "epoch": 7.8384, + "grad_norm": 0.0, + "learning_rate": 3.119999999999999e-05, + "loss": 5.8578, + "step": 4899 + }, + { + "epoch": 7.84, + "grad_norm": 0.0, + "learning_rate": 3.0999999999999995e-05, + "loss": 4.8375, + "step": 4900 + }, + { + "epoch": 7.8416, + "grad_norm": 0.0, + "learning_rate": 3.0799999999999996e-05, + "loss": 6.0366, + "step": 4901 + }, + { + "epoch": 7.8431999999999995, + "grad_norm": 0.0, + "learning_rate": 3.06e-05, + "loss": 6.4492, + "step": 4902 + }, + { + "epoch": 7.8448, + "grad_norm": 0.0, + "learning_rate": 3.0399999999999997e-05, + "loss": 5.3843, + "step": 4903 + }, + { + "epoch": 7.8464, + "grad_norm": 0.0, + "learning_rate": 3.02e-05, + "loss": 3.8829, + "step": 4904 + }, + { + "epoch": 7.848, + "grad_norm": 0.0, + "learning_rate": 2.9999999999999997e-05, + "loss": 4.7846, + "step": 4905 + }, + { + "epoch": 7.8496, + "grad_norm": 0.0, + "learning_rate": 2.9799999999999996e-05, + "loss": 6.2207, + "step": 4906 + }, + { + "epoch": 7.8512, + "grad_norm": 0.0, + "learning_rate": 2.9599999999999998e-05, + "loss": 4.7111, + "step": 4907 + }, + { + "epoch": 7.8528, + "grad_norm": 0.0, + "learning_rate": 2.94e-05, + "loss": 5.3292, + "step": 4908 + }, + { + "epoch": 7.8544, + "grad_norm": 0.0, + "learning_rate": 2.9199999999999995e-05, + "loss": 5.7591, + "step": 4909 + }, + { + "epoch": 7.856, + "grad_norm": 0.0, + "learning_rate": 2.8999999999999997e-05, + "loss": 4.7005, + "step": 4910 + }, + { + "epoch": 7.8576, + "grad_norm": 0.0, + "learning_rate": 2.88e-05, + "loss": 5.4882, + "step": 4911 + }, + { + "epoch": 7.8591999999999995, + "grad_norm": 0.0, + "learning_rate": 2.86e-05, + "loss": 5.6661, + "step": 4912 + }, + { + "epoch": 7.8608, + "grad_norm": 0.0, + "learning_rate": 2.8399999999999996e-05, + "loss": 5.7941, + "step": 4913 + }, + { + "epoch": 7.8624, + "grad_norm": 0.0, + "learning_rate": 2.8199999999999998e-05, + "loss": 5.1456, + "step": 4914 + }, + { + "epoch": 7.864, + "grad_norm": 0.0, + "learning_rate": 2.8e-05, + "loss": 4.8107, + "step": 4915 + }, + { + "epoch": 7.8656, + "grad_norm": 0.0, + "learning_rate": 2.7799999999999995e-05, + "loss": 5.4458, + "step": 4916 + }, + { + "epoch": 7.8672, + "grad_norm": 0.0, + "learning_rate": 2.7599999999999997e-05, + "loss": 5.2492, + "step": 4917 + }, + { + "epoch": 7.8688, + "grad_norm": 0.0, + "learning_rate": 2.74e-05, + "loss": 5.1782, + "step": 4918 + }, + { + "epoch": 7.8704, + "grad_norm": 0.0, + "learning_rate": 2.72e-05, + "loss": 4.3236, + "step": 4919 + }, + { + "epoch": 7.872, + "grad_norm": 0.0, + "learning_rate": 2.6999999999999996e-05, + "loss": 8.2734, + "step": 4920 + }, + { + "epoch": 7.8736, + "grad_norm": 0.0, + "learning_rate": 2.6799999999999998e-05, + "loss": 5.6999, + "step": 4921 + }, + { + "epoch": 7.8751999999999995, + "grad_norm": 0.0, + "learning_rate": 2.66e-05, + "loss": 4.5033, + "step": 4922 + }, + { + "epoch": 7.8768, + "grad_norm": 0.0, + "learning_rate": 2.6399999999999995e-05, + "loss": 4.7946, + "step": 4923 + }, + { + "epoch": 7.8784, + "grad_norm": 0.0, + "learning_rate": 2.6199999999999996e-05, + "loss": 6.4384, + "step": 4924 + }, + { + "epoch": 7.88, + "grad_norm": 0.0, + "learning_rate": 2.6e-05, + "loss": 8.1801, + "step": 4925 + }, + { + "epoch": 7.8816, + "grad_norm": 0.0, + "learning_rate": 2.5799999999999997e-05, + "loss": 7.9482, + "step": 4926 + }, + { + "epoch": 7.8832, + "grad_norm": 0.0, + "learning_rate": 2.5599999999999995e-05, + "loss": 6.7129, + "step": 4927 + }, + { + "epoch": 7.8848, + "grad_norm": 0.0, + "learning_rate": 2.5399999999999997e-05, + "loss": 8.4176, + "step": 4928 + }, + { + "epoch": 7.8864, + "grad_norm": 0.0, + "learning_rate": 2.52e-05, + "loss": 6.6931, + "step": 4929 + }, + { + "epoch": 7.888, + "grad_norm": 0.0, + "learning_rate": 2.4999999999999998e-05, + "loss": 6.193, + "step": 4930 + }, + { + "epoch": 7.8896, + "grad_norm": 0.0, + "learning_rate": 2.4799999999999996e-05, + "loss": 6.3482, + "step": 4931 + }, + { + "epoch": 7.8911999999999995, + "grad_norm": 0.0, + "learning_rate": 2.4599999999999998e-05, + "loss": 6.2278, + "step": 4932 + }, + { + "epoch": 7.8928, + "grad_norm": 0.0, + "learning_rate": 2.4399999999999997e-05, + "loss": 6.6975, + "step": 4933 + }, + { + "epoch": 7.8944, + "grad_norm": 0.0, + "learning_rate": 2.42e-05, + "loss": 7.5855, + "step": 4934 + }, + { + "epoch": 7.896, + "grad_norm": 0.0, + "learning_rate": 2.3999999999999997e-05, + "loss": 5.9267, + "step": 4935 + }, + { + "epoch": 7.8976, + "grad_norm": 0.0, + "learning_rate": 2.38e-05, + "loss": 6.6663, + "step": 4936 + }, + { + "epoch": 7.8992, + "grad_norm": 0.0, + "learning_rate": 2.3599999999999998e-05, + "loss": 6.4916, + "step": 4937 + }, + { + "epoch": 7.9008, + "grad_norm": 0.0, + "learning_rate": 2.34e-05, + "loss": 8.3419, + "step": 4938 + }, + { + "epoch": 7.9024, + "grad_norm": 0.0, + "learning_rate": 2.3199999999999998e-05, + "loss": 6.3328, + "step": 4939 + }, + { + "epoch": 7.904, + "grad_norm": 0.0, + "learning_rate": 2.2999999999999997e-05, + "loss": 5.4885, + "step": 4940 + }, + { + "epoch": 7.9056, + "grad_norm": 0.0, + "learning_rate": 2.28e-05, + "loss": 5.7347, + "step": 4941 + }, + { + "epoch": 7.9072, + "grad_norm": 0.0, + "learning_rate": 2.2599999999999997e-05, + "loss": 7.249, + "step": 4942 + }, + { + "epoch": 7.9088, + "grad_norm": 0.0, + "learning_rate": 2.24e-05, + "loss": 6.6369, + "step": 4943 + }, + { + "epoch": 7.9104, + "grad_norm": 0.0, + "learning_rate": 2.2199999999999998e-05, + "loss": 5.0906, + "step": 4944 + }, + { + "epoch": 7.912, + "grad_norm": 0.0, + "learning_rate": 2.2e-05, + "loss": 5.8825, + "step": 4945 + }, + { + "epoch": 7.9136, + "grad_norm": 0.0, + "learning_rate": 2.1799999999999998e-05, + "loss": 4.4625, + "step": 4946 + }, + { + "epoch": 7.9152000000000005, + "grad_norm": 0.0, + "learning_rate": 2.1599999999999996e-05, + "loss": 5.3964, + "step": 4947 + }, + { + "epoch": 7.9168, + "grad_norm": 0.0, + "learning_rate": 2.14e-05, + "loss": 7.4192, + "step": 4948 + }, + { + "epoch": 7.9184, + "grad_norm": 0.0, + "learning_rate": 2.12e-05, + "loss": 5.0459, + "step": 4949 + }, + { + "epoch": 7.92, + "grad_norm": 0.0, + "learning_rate": 2.1e-05, + "loss": 5.4631, + "step": 4950 + }, + { + "epoch": 7.9216, + "grad_norm": 0.0, + "learning_rate": 2.0799999999999997e-05, + "loss": 6.0327, + "step": 4951 + }, + { + "epoch": 7.9232, + "grad_norm": 0.0, + "learning_rate": 2.06e-05, + "loss": 4.5411, + "step": 4952 + }, + { + "epoch": 7.9248, + "grad_norm": 0.0, + "learning_rate": 2.04e-05, + "loss": 5.5531, + "step": 4953 + }, + { + "epoch": 7.9264, + "grad_norm": 0.0, + "learning_rate": 2.0199999999999996e-05, + "loss": 5.3817, + "step": 4954 + }, + { + "epoch": 7.928, + "grad_norm": 0.0, + "learning_rate": 1.9999999999999998e-05, + "loss": 6.9888, + "step": 4955 + }, + { + "epoch": 7.9296, + "grad_norm": 0.0, + "learning_rate": 1.98e-05, + "loss": 4.5361, + "step": 4956 + }, + { + "epoch": 7.9312000000000005, + "grad_norm": 0.0, + "learning_rate": 1.9599999999999995e-05, + "loss": 5.4912, + "step": 4957 + }, + { + "epoch": 7.9328, + "grad_norm": 0.0, + "learning_rate": 1.9399999999999997e-05, + "loss": 6.1554, + "step": 4958 + }, + { + "epoch": 7.9344, + "grad_norm": 0.0, + "learning_rate": 1.92e-05, + "loss": 5.2586, + "step": 4959 + }, + { + "epoch": 7.936, + "grad_norm": 0.0, + "learning_rate": 1.9e-05, + "loss": 4.7612, + "step": 4960 + }, + { + "epoch": 7.9376, + "grad_norm": 0.0, + "learning_rate": 1.8799999999999996e-05, + "loss": 6.0536, + "step": 4961 + }, + { + "epoch": 7.9392, + "grad_norm": 0.0, + "learning_rate": 1.8599999999999998e-05, + "loss": 5.4688, + "step": 4962 + }, + { + "epoch": 7.9408, + "grad_norm": 0.0, + "learning_rate": 1.8399999999999997e-05, + "loss": 4.6197, + "step": 4963 + }, + { + "epoch": 7.9424, + "grad_norm": 0.0, + "learning_rate": 1.82e-05, + "loss": 3.8822, + "step": 4964 + }, + { + "epoch": 7.944, + "grad_norm": 0.0, + "learning_rate": 1.7999999999999997e-05, + "loss": 5.0744, + "step": 4965 + }, + { + "epoch": 7.9456, + "grad_norm": 0.0, + "learning_rate": 1.78e-05, + "loss": 5.1668, + "step": 4966 + }, + { + "epoch": 7.9472000000000005, + "grad_norm": 0.0, + "learning_rate": 1.7599999999999998e-05, + "loss": 4.0673, + "step": 4967 + }, + { + "epoch": 7.9488, + "grad_norm": 0.0, + "learning_rate": 1.74e-05, + "loss": 5.6994, + "step": 4968 + }, + { + "epoch": 7.9504, + "grad_norm": 0.0, + "learning_rate": 1.7199999999999998e-05, + "loss": 5.9044, + "step": 4969 + }, + { + "epoch": 7.952, + "grad_norm": 0.0, + "learning_rate": 1.6999999999999996e-05, + "loss": 4.5149, + "step": 4970 + }, + { + "epoch": 7.9536, + "grad_norm": 0.0, + "learning_rate": 1.68e-05, + "loss": 3.7487, + "step": 4971 + }, + { + "epoch": 7.9552, + "grad_norm": 0.0, + "learning_rate": 1.6599999999999997e-05, + "loss": 5.4141, + "step": 4972 + }, + { + "epoch": 7.9568, + "grad_norm": 0.0, + "learning_rate": 1.64e-05, + "loss": 4.2148, + "step": 4973 + }, + { + "epoch": 7.9584, + "grad_norm": 0.0, + "learning_rate": 1.6199999999999997e-05, + "loss": 6.6993, + "step": 4974 + }, + { + "epoch": 7.96, + "grad_norm": 0.0, + "learning_rate": 1.6e-05, + "loss": 5.5462, + "step": 4975 + }, + { + "epoch": 7.9616, + "grad_norm": 0.0, + "learning_rate": 1.5799999999999998e-05, + "loss": 8.3309, + "step": 4976 + }, + { + "epoch": 7.9632, + "grad_norm": 0.0, + "learning_rate": 1.5599999999999996e-05, + "loss": 6.8767, + "step": 4977 + }, + { + "epoch": 7.9648, + "grad_norm": 0.0, + "learning_rate": 1.5399999999999998e-05, + "loss": 7.5225, + "step": 4978 + }, + { + "epoch": 7.9664, + "grad_norm": 0.0, + "learning_rate": 1.5199999999999998e-05, + "loss": 7.7427, + "step": 4979 + }, + { + "epoch": 7.968, + "grad_norm": 0.0, + "learning_rate": 1.4999999999999999e-05, + "loss": 5.3646, + "step": 4980 + }, + { + "epoch": 7.9696, + "grad_norm": 0.0, + "learning_rate": 1.4799999999999999e-05, + "loss": 6.1092, + "step": 4981 + }, + { + "epoch": 7.9712, + "grad_norm": 0.0, + "learning_rate": 1.4599999999999997e-05, + "loss": 5.952, + "step": 4982 + }, + { + "epoch": 7.9728, + "grad_norm": 0.0, + "learning_rate": 1.44e-05, + "loss": 6.2539, + "step": 4983 + }, + { + "epoch": 7.9744, + "grad_norm": 0.0, + "learning_rate": 1.4199999999999998e-05, + "loss": 5.018, + "step": 4984 + }, + { + "epoch": 7.976, + "grad_norm": 0.0, + "learning_rate": 1.4e-05, + "loss": 6.8668, + "step": 4985 + }, + { + "epoch": 7.9776, + "grad_norm": 0.0, + "learning_rate": 1.3799999999999998e-05, + "loss": 5.1818, + "step": 4986 + }, + { + "epoch": 7.9792, + "grad_norm": 0.0, + "learning_rate": 1.36e-05, + "loss": 6.0284, + "step": 4987 + }, + { + "epoch": 7.9808, + "grad_norm": 0.0, + "learning_rate": 1.3399999999999999e-05, + "loss": 4.4975, + "step": 4988 + }, + { + "epoch": 7.9824, + "grad_norm": 0.0, + "learning_rate": 1.3199999999999997e-05, + "loss": 5.7793, + "step": 4989 + }, + { + "epoch": 7.984, + "grad_norm": 0.0, + "learning_rate": 1.3e-05, + "loss": 4.3563, + "step": 4990 + }, + { + "epoch": 7.9856, + "grad_norm": 0.0, + "learning_rate": 1.2799999999999998e-05, + "loss": 5.4588, + "step": 4991 + }, + { + "epoch": 7.9872, + "grad_norm": 0.0, + "learning_rate": 1.26e-05, + "loss": 4.8431, + "step": 4992 + }, + { + "epoch": 7.9888, + "grad_norm": 0.0, + "learning_rate": 1.2399999999999998e-05, + "loss": 6.4495, + "step": 4993 + }, + { + "epoch": 7.9904, + "grad_norm": 0.0, + "learning_rate": 1.2199999999999998e-05, + "loss": 6.4643, + "step": 4994 + }, + { + "epoch": 7.992, + "grad_norm": 0.0, + "learning_rate": 1.1999999999999999e-05, + "loss": 4.6621, + "step": 4995 + }, + { + "epoch": 7.9936, + "grad_norm": 0.0, + "learning_rate": 1.1799999999999999e-05, + "loss": 5.3752, + "step": 4996 + }, + { + "epoch": 7.9952, + "grad_norm": 0.0, + "learning_rate": 1.1599999999999999e-05, + "loss": 6.2099, + "step": 4997 + }, + { + "epoch": 7.9968, + "grad_norm": 0.0, + "learning_rate": 1.14e-05, + "loss": 4.377, + "step": 4998 + }, + { + "epoch": 7.9984, + "grad_norm": 0.0, + "learning_rate": 1.12e-05, + "loss": 4.8807, + "step": 4999 + }, + { + "epoch": 8.0, + "grad_norm": 0.0, + "learning_rate": 1.1e-05, + "loss": 5.9653, + "step": 5000 + }, + { + "epoch": 8.0, + "eval_cer": 0.7076378272128795, + "eval_loss": 4.9668779373168945, + "eval_runtime": 159.4135, + "eval_samples_per_second": 19.672, + "eval_steps_per_second": 1.23, + "eval_wer": 0.9942297976440556, + "step": 5000 + }, + { + "epoch": 8.0, + "step": 5000, + "total_flos": 3.0616960056652042e+19, + "train_loss": 4.665264035391807, + "train_runtime": 10255.3805, + "train_samples_per_second": 15.602, + "train_steps_per_second": 0.488 + } + ], + "logging_steps": 1.0, + "max_steps": 5000, + "num_input_tokens_seen": 0, + "num_train_epochs": 8, + "save_steps": 2500, + "stateful_callbacks": { + "TrainerControl": { + "args": { + "should_epoch_stop": false, + "should_evaluate": false, + "should_log": false, + "should_save": true, + "should_training_stop": true + }, + "attributes": {} + } + }, + "total_flos": 3.0616960056652042e+19, + "train_batch_size": 16, + "trial_name": null, + "trial_params": null +}