{ "best_metric": 0.22680288553237915, "best_model_checkpoint": "./EulerMath-Mistral-7B-model/checkpoint-272", "epoch": 0.9990817263544536, "eval_steps": 68, "global_step": 272, "is_hyper_param_search": false, "is_local_process_zero": true, "is_world_process_zero": true, "log_history": [ { "epoch": 0.0, "grad_norm": 19.19068191513093, "learning_rate": 5.000000000000001e-07, "loss": 0.707, "step": 1 }, { "epoch": 0.0, "eval_loss": 0.9060535430908203, "eval_runtime": 1745.9683, "eval_samples_per_second": 1.324, "eval_steps_per_second": 0.074, "step": 1 }, { "epoch": 0.01, "grad_norm": 20.035932532601844, "learning_rate": 1.0000000000000002e-06, "loss": 0.7236, "step": 2 }, { "epoch": 0.01, "grad_norm": 19.31513317860667, "learning_rate": 1.5e-06, "loss": 0.7201, "step": 3 }, { "epoch": 0.01, "grad_norm": 16.561326930760348, "learning_rate": 2.0000000000000003e-06, "loss": 0.6717, "step": 4 }, { "epoch": 0.02, "grad_norm": 9.069275733221579, "learning_rate": 2.5e-06, "loss": 0.573, "step": 5 }, { "epoch": 0.02, "grad_norm": 6.0702110208300475, "learning_rate": 3e-06, "loss": 0.4965, "step": 6 }, { "epoch": 0.03, "grad_norm": 6.5389430446896055, "learning_rate": 3.5e-06, "loss": 0.5093, "step": 7 }, { "epoch": 0.03, "grad_norm": 7.709934958779789, "learning_rate": 4.000000000000001e-06, "loss": 0.524, "step": 8 }, { "epoch": 0.03, "grad_norm": 6.1640217934257135, "learning_rate": 4.5e-06, "loss": 0.503, "step": 9 }, { "epoch": 0.04, "grad_norm": 4.079182690080823, "learning_rate": 5e-06, "loss": 0.4787, "step": 10 }, { "epoch": 0.04, "grad_norm": 4.269731620276111, "learning_rate": 4.999956736067563e-06, "loss": 0.4545, "step": 11 }, { "epoch": 0.04, "grad_norm": 4.059214670786909, "learning_rate": 4.999826945767665e-06, "loss": 0.4638, "step": 12 }, { "epoch": 0.05, "grad_norm": 3.583247385116129, "learning_rate": 4.9996106335924965e-06, "loss": 0.4396, "step": 13 }, { "epoch": 0.05, "grad_norm": 3.2077663599892405, "learning_rate": 4.999307807028872e-06, "loss": 0.4287, "step": 14 }, { "epoch": 0.06, "grad_norm": 2.3678816023894513, "learning_rate": 4.998918476557964e-06, "loss": 0.4169, "step": 15 }, { "epoch": 0.06, "grad_norm": 1.9925263681909064, "learning_rate": 4.998442655654946e-06, "loss": 0.4099, "step": 16 }, { "epoch": 0.06, "grad_norm": 1.7706573910428134, "learning_rate": 4.997880360788527e-06, "loss": 0.4003, "step": 17 }, { "epoch": 0.07, "grad_norm": 1.6789390301868525, "learning_rate": 4.997231611420374e-06, "loss": 0.399, "step": 18 }, { "epoch": 0.07, "grad_norm": 1.5622054221426698, "learning_rate": 4.996496430004446e-06, "loss": 0.3885, "step": 19 }, { "epoch": 0.07, "grad_norm": 1.5663787846468284, "learning_rate": 4.995674841986217e-06, "loss": 0.3987, "step": 20 }, { "epoch": 0.08, "grad_norm": 1.4502330087611721, "learning_rate": 4.994766875801789e-06, "loss": 0.3962, "step": 21 }, { "epoch": 0.08, "grad_norm": 1.4188997099391882, "learning_rate": 4.993772562876909e-06, "loss": 0.3845, "step": 22 }, { "epoch": 0.08, "grad_norm": 1.4360806887465898, "learning_rate": 4.992691937625892e-06, "loss": 0.3764, "step": 23 }, { "epoch": 0.09, "grad_norm": 1.4216582090099372, "learning_rate": 4.991525037450412e-06, "loss": 0.3712, "step": 24 }, { "epoch": 0.09, "grad_norm": 1.2856499279799387, "learning_rate": 4.990271902738223e-06, "loss": 0.3603, "step": 25 }, { "epoch": 0.1, "grad_norm": 1.247117404577534, "learning_rate": 4.988932576861754e-06, "loss": 0.3652, "step": 26 }, { "epoch": 0.1, "grad_norm": 1.3197850379000642, "learning_rate": 4.987507106176606e-06, "loss": 0.371, "step": 27 }, { "epoch": 0.1, "grad_norm": 1.243400495941476, "learning_rate": 4.985995540019956e-06, "loss": 0.3599, "step": 28 }, { "epoch": 0.11, "grad_norm": 1.3278566257982103, "learning_rate": 4.984397930708838e-06, "loss": 0.3594, "step": 29 }, { "epoch": 0.11, "grad_norm": 1.337022527470652, "learning_rate": 4.982714333538344e-06, "loss": 0.3477, "step": 30 }, { "epoch": 0.11, "grad_norm": 1.2099362672151601, "learning_rate": 4.980944806779698e-06, "loss": 0.3425, "step": 31 }, { "epoch": 0.12, "grad_norm": 1.2110593150023343, "learning_rate": 4.979089411678252e-06, "loss": 0.3567, "step": 32 }, { "epoch": 0.12, "grad_norm": 1.2334965596913852, "learning_rate": 4.977148212451354e-06, "loss": 0.3526, "step": 33 }, { "epoch": 0.12, "grad_norm": 1.1687161424016368, "learning_rate": 4.975121276286136e-06, "loss": 0.3496, "step": 34 }, { "epoch": 0.13, "grad_norm": 1.1881954676378432, "learning_rate": 4.973008673337181e-06, "loss": 0.3321, "step": 35 }, { "epoch": 0.13, "grad_norm": 1.2174270605971114, "learning_rate": 4.970810476724097e-06, "loss": 0.3446, "step": 36 }, { "epoch": 0.14, "grad_norm": 1.1609330509652702, "learning_rate": 4.968526762528988e-06, "loss": 0.341, "step": 37 }, { "epoch": 0.14, "grad_norm": 1.2149352568793006, "learning_rate": 4.9661576097938205e-06, "loss": 0.3459, "step": 38 }, { "epoch": 0.14, "grad_norm": 1.1885081900677397, "learning_rate": 4.963703100517684e-06, "loss": 0.3425, "step": 39 }, { "epoch": 0.15, "grad_norm": 1.113235885075549, "learning_rate": 4.961163319653959e-06, "loss": 0.339, "step": 40 }, { "epoch": 0.15, "grad_norm": 1.0983562726057154, "learning_rate": 4.958538355107369e-06, "loss": 0.3298, "step": 41 }, { "epoch": 0.15, "grad_norm": 1.1594289217865181, "learning_rate": 4.955828297730949e-06, "loss": 0.3187, "step": 42 }, { "epoch": 0.16, "grad_norm": 1.1714548911644644, "learning_rate": 4.953033241322887e-06, "loss": 0.3373, "step": 43 }, { "epoch": 0.16, "grad_norm": 1.1450397323165031, "learning_rate": 4.950153282623289e-06, "loss": 0.3232, "step": 44 }, { "epoch": 0.17, "grad_norm": 1.1526363934692334, "learning_rate": 4.947188521310827e-06, "loss": 0.3243, "step": 45 }, { "epoch": 0.17, "grad_norm": 1.2175235837438554, "learning_rate": 4.944139059999286e-06, "loss": 0.3252, "step": 46 }, { "epoch": 0.17, "grad_norm": 1.099789045296574, "learning_rate": 4.941005004234019e-06, "loss": 0.3178, "step": 47 }, { "epoch": 0.18, "grad_norm": 1.2219677196886505, "learning_rate": 4.937786462488284e-06, "loss": 0.3185, "step": 48 }, { "epoch": 0.18, "grad_norm": 1.1806399387287625, "learning_rate": 4.9344835461595016e-06, "loss": 0.3131, "step": 49 }, { "epoch": 0.18, "grad_norm": 1.1320527868188186, "learning_rate": 4.93109636956539e-06, "loss": 0.3198, "step": 50 }, { "epoch": 0.19, "grad_norm": 1.2551253674231917, "learning_rate": 4.927625049940013e-06, "loss": 0.3063, "step": 51 }, { "epoch": 0.19, "grad_norm": 1.1131050315591549, "learning_rate": 4.9240697074297205e-06, "loss": 0.3192, "step": 52 }, { "epoch": 0.19, "grad_norm": 1.218025833644298, "learning_rate": 4.920430465088992e-06, "loss": 0.3083, "step": 53 }, { "epoch": 0.2, "grad_norm": 1.090531576651011, "learning_rate": 4.916707448876173e-06, "loss": 0.3076, "step": 54 }, { "epoch": 0.2, "grad_norm": 1.1865422414756877, "learning_rate": 4.912900787649124e-06, "loss": 0.3155, "step": 55 }, { "epoch": 0.21, "grad_norm": 1.1236405558973956, "learning_rate": 4.909010613160751e-06, "loss": 0.306, "step": 56 }, { "epoch": 0.21, "grad_norm": 1.222805799933775, "learning_rate": 4.90503706005445e-06, "loss": 0.3054, "step": 57 }, { "epoch": 0.21, "grad_norm": 1.179814726076065, "learning_rate": 4.900980265859449e-06, "loss": 0.309, "step": 58 }, { "epoch": 0.22, "grad_norm": 1.155763655177263, "learning_rate": 4.896840370986042e-06, "loss": 0.2974, "step": 59 }, { "epoch": 0.22, "grad_norm": 1.1687171308842221, "learning_rate": 4.892617518720737e-06, "loss": 0.3018, "step": 60 }, { "epoch": 0.22, "grad_norm": 1.2240587320323661, "learning_rate": 4.88831185522129e-06, "loss": 0.3066, "step": 61 }, { "epoch": 0.23, "grad_norm": 1.1042960875500205, "learning_rate": 4.883923529511646e-06, "loss": 0.2977, "step": 62 }, { "epoch": 0.23, "grad_norm": 1.1885949614868223, "learning_rate": 4.87945269347679e-06, "loss": 0.3087, "step": 63 }, { "epoch": 0.24, "grad_norm": 1.1420656757477574, "learning_rate": 4.874899501857477e-06, "loss": 0.2904, "step": 64 }, { "epoch": 0.24, "grad_norm": 1.1453980260713446, "learning_rate": 4.87026411224489e-06, "loss": 0.306, "step": 65 }, { "epoch": 0.24, "grad_norm": 1.2729287210416769, "learning_rate": 4.865546685075174e-06, "loss": 0.2938, "step": 66 }, { "epoch": 0.25, "grad_norm": 1.2052792222072466, "learning_rate": 4.860747383623889e-06, "loss": 0.2977, "step": 67 }, { "epoch": 0.25, "grad_norm": 1.2657508580603682, "learning_rate": 4.85586637400036e-06, "loss": 0.3011, "step": 68 }, { "epoch": 0.25, "eval_loss": 0.32630813121795654, "eval_runtime": 1744.5857, "eval_samples_per_second": 1.325, "eval_steps_per_second": 0.074, "step": 68 }, { "epoch": 0.25, "grad_norm": 1.1832834131492187, "learning_rate": 4.85090382514192e-06, "loss": 0.2972, "step": 69 }, { "epoch": 0.26, "grad_norm": 1.255475532117491, "learning_rate": 4.845859908808074e-06, "loss": 0.302, "step": 70 }, { "epoch": 0.26, "grad_norm": 1.298818409489401, "learning_rate": 4.8407347995745465e-06, "loss": 0.2935, "step": 71 }, { "epoch": 0.26, "grad_norm": 1.3499885398461409, "learning_rate": 4.8355286748272405e-06, "loss": 0.295, "step": 72 }, { "epoch": 0.27, "grad_norm": 1.3446382549398914, "learning_rate": 4.830241714756099e-06, "loss": 0.2824, "step": 73 }, { "epoch": 0.27, "grad_norm": 1.2082987304246777, "learning_rate": 4.8248741023488705e-06, "loss": 0.3026, "step": 74 }, { "epoch": 0.28, "grad_norm": 1.3432457490726049, "learning_rate": 4.81942602338477e-06, "loss": 0.2985, "step": 75 }, { "epoch": 0.28, "grad_norm": 1.170337150254348, "learning_rate": 4.813897666428054e-06, "loss": 0.2969, "step": 76 }, { "epoch": 0.28, "grad_norm": 1.339414484466056, "learning_rate": 4.808289222821491e-06, "loss": 0.2985, "step": 77 }, { "epoch": 0.29, "grad_norm": 1.1944077580462804, "learning_rate": 4.802600886679743e-06, "loss": 0.2852, "step": 78 }, { "epoch": 0.29, "grad_norm": 1.357246876413576, "learning_rate": 4.79683285488264e-06, "loss": 0.2904, "step": 79 }, { "epoch": 0.29, "grad_norm": 1.4115119936533302, "learning_rate": 4.790985327068376e-06, "loss": 0.3079, "step": 80 }, { "epoch": 0.3, "grad_norm": 1.285315536324781, "learning_rate": 4.7850585056265866e-06, "loss": 0.2816, "step": 81 }, { "epoch": 0.3, "grad_norm": 1.3631452273406317, "learning_rate": 4.779052595691355e-06, "loss": 0.2865, "step": 82 }, { "epoch": 0.3, "grad_norm": 1.196518391890594, "learning_rate": 4.772967805134106e-06, "loss": 0.2793, "step": 83 }, { "epoch": 0.31, "grad_norm": 1.2485622601747421, "learning_rate": 4.766804344556414e-06, "loss": 0.2827, "step": 84 }, { "epoch": 0.31, "grad_norm": 1.2945099002171803, "learning_rate": 4.7605624272827125e-06, "loss": 0.2854, "step": 85 }, { "epoch": 0.32, "grad_norm": 1.224576498812201, "learning_rate": 4.754242269352911e-06, "loss": 0.2875, "step": 86 }, { "epoch": 0.32, "grad_norm": 1.2535747430861524, "learning_rate": 4.747844089514919e-06, "loss": 0.2807, "step": 87 }, { "epoch": 0.32, "grad_norm": 1.171951212608294, "learning_rate": 4.741368109217072e-06, "loss": 0.2761, "step": 88 }, { "epoch": 0.33, "grad_norm": 1.2123280755320154, "learning_rate": 4.734814552600469e-06, "loss": 0.2832, "step": 89 }, { "epoch": 0.33, "grad_norm": 1.1358700523339582, "learning_rate": 4.728183646491215e-06, "loss": 0.2871, "step": 90 }, { "epoch": 0.33, "grad_norm": 1.1484698203958048, "learning_rate": 4.721475620392567e-06, "loss": 0.2806, "step": 91 }, { "epoch": 0.34, "grad_norm": 1.1887290775946084, "learning_rate": 4.714690706477e-06, "loss": 0.2858, "step": 92 }, { "epoch": 0.34, "grad_norm": 1.1568061250650739, "learning_rate": 4.707829139578156e-06, "loss": 0.2888, "step": 93 }, { "epoch": 0.35, "grad_norm": 1.176832058354239, "learning_rate": 4.700891157182729e-06, "loss": 0.2829, "step": 94 }, { "epoch": 0.35, "grad_norm": 1.138549309431515, "learning_rate": 4.693876999422241e-06, "loss": 0.2763, "step": 95 }, { "epoch": 0.35, "grad_norm": 1.1479926100837645, "learning_rate": 4.68678690906473e-06, "loss": 0.2686, "step": 96 }, { "epoch": 0.36, "grad_norm": 1.1771516377197246, "learning_rate": 4.679621131506347e-06, "loss": 0.2814, "step": 97 }, { "epoch": 0.36, "grad_norm": 1.2184996974539424, "learning_rate": 4.672379914762867e-06, "loss": 0.2822, "step": 98 }, { "epoch": 0.36, "grad_norm": 1.1792108348242942, "learning_rate": 4.665063509461098e-06, "loss": 0.282, "step": 99 }, { "epoch": 0.37, "grad_norm": 1.2850683815489914, "learning_rate": 4.657672168830211e-06, "loss": 0.2776, "step": 100 }, { "epoch": 0.37, "grad_norm": 1.2508897770511975, "learning_rate": 4.650206148692977e-06, "loss": 0.2787, "step": 101 }, { "epoch": 0.37, "grad_norm": 1.2031990746786907, "learning_rate": 4.642665707456908e-06, "loss": 0.2719, "step": 102 }, { "epoch": 0.38, "grad_norm": 1.1842474930123255, "learning_rate": 4.635051106105316e-06, "loss": 0.2732, "step": 103 }, { "epoch": 0.38, "grad_norm": 1.2596970412015132, "learning_rate": 4.627362608188281e-06, "loss": 0.2731, "step": 104 }, { "epoch": 0.39, "grad_norm": 1.4294759311096437, "learning_rate": 4.619600479813524e-06, "loss": 0.2738, "step": 105 }, { "epoch": 0.39, "grad_norm": 1.31619095423113, "learning_rate": 4.6117649896372055e-06, "loss": 0.2764, "step": 106 }, { "epoch": 0.39, "grad_norm": 1.2349728666776751, "learning_rate": 4.6038564088546185e-06, "loss": 0.2722, "step": 107 }, { "epoch": 0.4, "grad_norm": 1.2418477065252158, "learning_rate": 4.5958750111908065e-06, "loss": 0.271, "step": 108 }, { "epoch": 0.4, "grad_norm": 1.3529322240859796, "learning_rate": 4.587821072891089e-06, "loss": 0.276, "step": 109 }, { "epoch": 0.4, "grad_norm": 1.2671711562594927, "learning_rate": 4.579694872711501e-06, "loss": 0.2706, "step": 110 }, { "epoch": 0.41, "grad_norm": 1.238356873891121, "learning_rate": 4.571496691909142e-06, "loss": 0.2749, "step": 111 }, { "epoch": 0.41, "grad_norm": 1.2059912760303926, "learning_rate": 4.563226814232444e-06, "loss": 0.2676, "step": 112 }, { "epoch": 0.42, "grad_norm": 1.1876458610423755, "learning_rate": 4.554885525911351e-06, "loss": 0.2743, "step": 113 }, { "epoch": 0.42, "grad_norm": 1.1715592937521375, "learning_rate": 4.54647311564741e-06, "loss": 0.2734, "step": 114 }, { "epoch": 0.42, "grad_norm": 1.236329928620471, "learning_rate": 4.53798987460378e-06, "loss": 0.2855, "step": 115 }, { "epoch": 0.43, "grad_norm": 1.1717820999866062, "learning_rate": 4.529436096395157e-06, "loss": 0.2699, "step": 116 }, { "epoch": 0.43, "grad_norm": 1.3490101744641771, "learning_rate": 4.520812077077604e-06, "loss": 0.2731, "step": 117 }, { "epoch": 0.43, "grad_norm": 1.192962777526519, "learning_rate": 4.512118115138315e-06, "loss": 0.2719, "step": 118 }, { "epoch": 0.44, "grad_norm": 1.2384657820337475, "learning_rate": 4.5033545114852734e-06, "loss": 0.2647, "step": 119 }, { "epoch": 0.44, "grad_norm": 1.2128578058956592, "learning_rate": 4.494521569436845e-06, "loss": 0.2615, "step": 120 }, { "epoch": 0.44, "grad_norm": 1.3237640584842072, "learning_rate": 4.485619594711278e-06, "loss": 0.2663, "step": 121 }, { "epoch": 0.45, "grad_norm": 1.2691929068372239, "learning_rate": 4.476648895416116e-06, "loss": 0.2614, "step": 122 }, { "epoch": 0.45, "grad_norm": 1.2606618599832538, "learning_rate": 4.467609782037543e-06, "loss": 0.2606, "step": 123 }, { "epoch": 0.46, "grad_norm": 1.3048381409549332, "learning_rate": 4.4585025674296315e-06, "loss": 0.2601, "step": 124 }, { "epoch": 0.46, "grad_norm": 1.3022768451107203, "learning_rate": 4.449327566803515e-06, "loss": 0.2683, "step": 125 }, { "epoch": 0.46, "grad_norm": 1.3820289309230962, "learning_rate": 4.44008509771648e-06, "loss": 0.2681, "step": 126 }, { "epoch": 0.47, "grad_norm": 1.2802354999925132, "learning_rate": 4.430775480060973e-06, "loss": 0.2648, "step": 127 }, { "epoch": 0.47, "grad_norm": 1.3242106497833372, "learning_rate": 4.4213990360535274e-06, "loss": 0.268, "step": 128 }, { "epoch": 0.47, "grad_norm": 1.3009976864959876, "learning_rate": 4.411956090223618e-06, "loss": 0.2662, "step": 129 }, { "epoch": 0.48, "grad_norm": 1.3212829688401424, "learning_rate": 4.4024469694024194e-06, "loss": 0.2605, "step": 130 }, { "epoch": 0.48, "grad_norm": 1.2123869956343973, "learning_rate": 4.3928720027115015e-06, "loss": 0.2604, "step": 131 }, { "epoch": 0.48, "grad_norm": 1.284537459167204, "learning_rate": 4.383231521551432e-06, "loss": 0.2593, "step": 132 }, { "epoch": 0.49, "grad_norm": 1.443338680183996, "learning_rate": 4.373525859590313e-06, "loss": 0.2561, "step": 133 }, { "epoch": 0.49, "grad_norm": 1.2809230468289576, "learning_rate": 4.3637553527522265e-06, "loss": 0.2599, "step": 134 }, { "epoch": 0.5, "grad_norm": 1.3669470609932883, "learning_rate": 4.3539203392056114e-06, "loss": 0.2587, "step": 135 }, { "epoch": 0.5, "grad_norm": 1.4112940230474231, "learning_rate": 4.3440211593515556e-06, "loss": 0.2585, "step": 136 }, { "epoch": 0.5, "eval_loss": 0.28355109691619873, "eval_runtime": 1744.5175, "eval_samples_per_second": 1.325, "eval_steps_per_second": 0.074, "step": 136 }, { "epoch": 0.5, "grad_norm": 1.3061396480876788, "learning_rate": 4.33405815581202e-06, "loss": 0.2549, "step": 137 }, { "epoch": 0.51, "grad_norm": 1.46460991921356, "learning_rate": 4.324031673417971e-06, "loss": 0.2639, "step": 138 }, { "epoch": 0.51, "grad_norm": 1.211168578821325, "learning_rate": 4.313942059197457e-06, "loss": 0.2581, "step": 139 }, { "epoch": 0.51, "grad_norm": 1.4657150585182341, "learning_rate": 4.303789662363587e-06, "loss": 0.2616, "step": 140 }, { "epoch": 0.52, "grad_norm": 1.4251800081691455, "learning_rate": 4.29357483430245e-06, "loss": 0.2668, "step": 141 }, { "epoch": 0.52, "grad_norm": 1.3599666478045191, "learning_rate": 4.283297928560951e-06, "loss": 0.2598, "step": 142 }, { "epoch": 0.53, "grad_norm": 1.6103346253156021, "learning_rate": 4.272959300834574e-06, "loss": 0.2656, "step": 143 }, { "epoch": 0.53, "grad_norm": 1.2184694580930981, "learning_rate": 4.262559308955072e-06, "loss": 0.2546, "step": 144 }, { "epoch": 0.53, "grad_norm": 1.3362006281948362, "learning_rate": 4.252098312878083e-06, "loss": 0.2557, "step": 145 }, { "epoch": 0.54, "grad_norm": 1.3369296531115935, "learning_rate": 4.241576674670668e-06, "loss": 0.2568, "step": 146 }, { "epoch": 0.54, "grad_norm": 1.4747872641188995, "learning_rate": 4.230994758498783e-06, "loss": 0.2564, "step": 147 }, { "epoch": 0.54, "grad_norm": 1.60778480089848, "learning_rate": 4.220352930614672e-06, "loss": 0.2573, "step": 148 }, { "epoch": 0.55, "grad_norm": 1.188044808018822, "learning_rate": 4.209651559344195e-06, "loss": 0.2525, "step": 149 }, { "epoch": 0.55, "grad_norm": 1.5856639134844415, "learning_rate": 4.198891015074074e-06, "loss": 0.2647, "step": 150 }, { "epoch": 0.55, "grad_norm": 1.2859262024596512, "learning_rate": 4.1880716702390764e-06, "loss": 0.2471, "step": 151 }, { "epoch": 0.56, "grad_norm": 1.4653590828956073, "learning_rate": 4.177193899309127e-06, "loss": 0.2575, "step": 152 }, { "epoch": 0.56, "grad_norm": 1.1821237121686685, "learning_rate": 4.166258078776342e-06, "loss": 0.2493, "step": 153 }, { "epoch": 0.57, "grad_norm": 1.575597475848357, "learning_rate": 4.155264587142002e-06, "loss": 0.2537, "step": 154 }, { "epoch": 0.57, "grad_norm": 1.2702085752651588, "learning_rate": 4.144213804903449e-06, "loss": 0.2493, "step": 155 }, { "epoch": 0.57, "grad_norm": 1.5026735427361002, "learning_rate": 4.133106114540923e-06, "loss": 0.2505, "step": 156 }, { "epoch": 0.58, "grad_norm": 1.5297903686100347, "learning_rate": 4.121941900504316e-06, "loss": 0.2472, "step": 157 }, { "epoch": 0.58, "grad_norm": 1.25258373375573, "learning_rate": 4.110721549199866e-06, "loss": 0.2487, "step": 158 }, { "epoch": 0.58, "grad_norm": 1.5941545034573665, "learning_rate": 4.099445448976793e-06, "loss": 0.2497, "step": 159 }, { "epoch": 0.59, "grad_norm": 1.3096080921873048, "learning_rate": 4.088113990113846e-06, "loss": 0.2439, "step": 160 }, { "epoch": 0.59, "grad_norm": 1.6950266606195492, "learning_rate": 4.076727564805803e-06, "loss": 0.2538, "step": 161 }, { "epoch": 0.6, "grad_norm": 1.440485526817555, "learning_rate": 4.065286567149891e-06, "loss": 0.2613, "step": 162 }, { "epoch": 0.6, "grad_norm": 1.606032223752871, "learning_rate": 4.0537913931321495e-06, "loss": 0.2505, "step": 163 }, { "epoch": 0.6, "grad_norm": 1.5319951141665498, "learning_rate": 4.042242440613724e-06, "loss": 0.256, "step": 164 }, { "epoch": 0.61, "grad_norm": 1.3468098768373629, "learning_rate": 4.030640109317096e-06, "loss": 0.2424, "step": 165 }, { "epoch": 0.61, "grad_norm": 1.6652562481471478, "learning_rate": 4.018984800812248e-06, "loss": 0.2396, "step": 166 }, { "epoch": 0.61, "grad_norm": 1.302975081280886, "learning_rate": 4.007276918502763e-06, "loss": 0.2462, "step": 167 }, { "epoch": 0.62, "grad_norm": 1.623125313268604, "learning_rate": 3.995516867611865e-06, "loss": 0.256, "step": 168 }, { "epoch": 0.62, "grad_norm": 1.3069782036585045, "learning_rate": 3.983705055168391e-06, "loss": 0.2518, "step": 169 }, { "epoch": 0.62, "grad_norm": 1.6527449270834242, "learning_rate": 3.971841889992706e-06, "loss": 0.2544, "step": 170 }, { "epoch": 0.63, "grad_norm": 1.3586948189643275, "learning_rate": 3.959927782682551e-06, "loss": 0.2491, "step": 171 }, { "epoch": 0.63, "grad_norm": 1.3440233460948727, "learning_rate": 3.947963145598833e-06, "loss": 0.2516, "step": 172 }, { "epoch": 0.64, "grad_norm": 1.3389168317613516, "learning_rate": 3.935948392851354e-06, "loss": 0.2541, "step": 173 }, { "epoch": 0.64, "grad_norm": 1.3142664585396417, "learning_rate": 3.923883940284472e-06, "loss": 0.2508, "step": 174 }, { "epoch": 0.64, "grad_norm": 1.2767521320981983, "learning_rate": 3.911770205462717e-06, "loss": 0.2479, "step": 175 }, { "epoch": 0.65, "grad_norm": 1.3281972191838929, "learning_rate": 3.899607607656334e-06, "loss": 0.2501, "step": 176 }, { "epoch": 0.65, "grad_norm": 1.3793116543581005, "learning_rate": 3.887396567826769e-06, "loss": 0.2454, "step": 177 }, { "epoch": 0.65, "grad_norm": 1.3293987156576104, "learning_rate": 3.875137508612104e-06, "loss": 0.249, "step": 178 }, { "epoch": 0.66, "grad_norm": 1.4957835845929142, "learning_rate": 3.862830854312427e-06, "loss": 0.2445, "step": 179 }, { "epoch": 0.66, "grad_norm": 1.2804679875446887, "learning_rate": 3.850477030875147e-06, "loss": 0.2411, "step": 180 }, { "epoch": 0.66, "grad_norm": 1.5611119218300138, "learning_rate": 3.838076465880248e-06, "loss": 0.237, "step": 181 }, { "epoch": 0.67, "grad_norm": 1.3387338916825537, "learning_rate": 3.825629588525498e-06, "loss": 0.2429, "step": 182 }, { "epoch": 0.67, "grad_norm": 1.5091720406707172, "learning_rate": 3.813136829611583e-06, "loss": 0.2428, "step": 183 }, { "epoch": 0.68, "grad_norm": 1.359116281666385, "learning_rate": 3.8005986215272056e-06, "loss": 0.2543, "step": 184 }, { "epoch": 0.68, "grad_norm": 1.4094254259139338, "learning_rate": 3.7880153982341167e-06, "loss": 0.2502, "step": 185 }, { "epoch": 0.68, "grad_norm": 1.2806047483095333, "learning_rate": 3.7753875952520943e-06, "loss": 0.2431, "step": 186 }, { "epoch": 0.69, "grad_norm": 1.409218880016104, "learning_rate": 3.7627156496438686e-06, "loss": 0.2463, "step": 187 }, { "epoch": 0.69, "grad_norm": 1.2466244404207094, "learning_rate": 3.7500000000000005e-06, "loss": 0.2372, "step": 188 }, { "epoch": 0.69, "grad_norm": 1.4192484726979884, "learning_rate": 3.7372410864236954e-06, "loss": 0.2396, "step": 189 }, { "epoch": 0.7, "grad_norm": 1.3260879207799772, "learning_rate": 3.7244393505155713e-06, "loss": 0.241, "step": 190 }, { "epoch": 0.7, "grad_norm": 1.6407257220698948, "learning_rate": 3.7115952353583804e-06, "loss": 0.2552, "step": 191 }, { "epoch": 0.71, "grad_norm": 1.4113760059054485, "learning_rate": 3.6987091855016667e-06, "loss": 0.2513, "step": 192 }, { "epoch": 0.71, "grad_norm": 1.3008883773347888, "learning_rate": 3.6857816469463806e-06, "loss": 0.2361, "step": 193 }, { "epoch": 0.71, "grad_norm": 1.3040857591494066, "learning_rate": 3.6728130671294485e-06, "loss": 0.2491, "step": 194 }, { "epoch": 0.72, "grad_norm": 1.2543618451342111, "learning_rate": 3.6598038949082777e-06, "loss": 0.2309, "step": 195 }, { "epoch": 0.72, "grad_norm": 1.3944108707435374, "learning_rate": 3.6467545805452266e-06, "loss": 0.2426, "step": 196 }, { "epoch": 0.72, "grad_norm": 1.301851485207592, "learning_rate": 3.6336655756920198e-06, "loss": 0.2421, "step": 197 }, { "epoch": 0.73, "grad_norm": 1.3562155385998595, "learning_rate": 3.620537333374114e-06, "loss": 0.2406, "step": 198 }, { "epoch": 0.73, "grad_norm": 1.4263666275672418, "learning_rate": 3.6073703079750204e-06, "loss": 0.2418, "step": 199 }, { "epoch": 0.73, "grad_norm": 1.2767612877970262, "learning_rate": 3.594164955220577e-06, "loss": 0.2353, "step": 200 }, { "epoch": 0.74, "grad_norm": 1.3349267171117716, "learning_rate": 3.5809217321631745e-06, "loss": 0.2348, "step": 201 }, { "epoch": 0.74, "grad_norm": 1.2217693484408796, "learning_rate": 3.5676410971659404e-06, "loss": 0.2287, "step": 202 }, { "epoch": 0.75, "grad_norm": 1.4554473054976789, "learning_rate": 3.5543235098868702e-06, "loss": 0.241, "step": 203 }, { "epoch": 0.75, "grad_norm": 1.184805169962002, "learning_rate": 3.5409694312629193e-06, "loss": 0.2352, "step": 204 }, { "epoch": 0.75, "eval_loss": 0.25444912910461426, "eval_runtime": 1745.7708, "eval_samples_per_second": 1.324, "eval_steps_per_second": 0.074, "step": 204 }, { "epoch": 0.75, "grad_norm": 1.2973792749867632, "learning_rate": 3.527579323494055e-06, "loss": 0.2404, "step": 205 }, { "epoch": 0.76, "grad_norm": 1.390330195755624, "learning_rate": 3.5141536500272494e-06, "loss": 0.2397, "step": 206 }, { "epoch": 0.76, "grad_norm": 1.2415077962351395, "learning_rate": 3.5006928755404467e-06, "loss": 0.2296, "step": 207 }, { "epoch": 0.76, "grad_norm": 1.3223264932925407, "learning_rate": 3.4871974659264786e-06, "loss": 0.2332, "step": 208 }, { "epoch": 0.77, "grad_norm": 1.4376836200586416, "learning_rate": 3.473667888276935e-06, "loss": 0.2361, "step": 209 }, { "epoch": 0.77, "grad_norm": 1.2495709137167788, "learning_rate": 3.4601046108660036e-06, "loss": 0.2351, "step": 210 }, { "epoch": 0.78, "grad_norm": 1.4449247677336339, "learning_rate": 3.446508103134259e-06, "loss": 0.2373, "step": 211 }, { "epoch": 0.78, "grad_norm": 1.3961526866418432, "learning_rate": 3.4328788356724135e-06, "loss": 0.2383, "step": 212 }, { "epoch": 0.78, "grad_norm": 1.2766356071702671, "learning_rate": 3.419217280205032e-06, "loss": 0.2348, "step": 213 }, { "epoch": 0.79, "grad_norm": 1.2201985305952152, "learning_rate": 3.4055239095742067e-06, "loss": 0.236, "step": 214 }, { "epoch": 0.79, "grad_norm": 1.3670381437866368, "learning_rate": 3.3917991977231855e-06, "loss": 0.228, "step": 215 }, { "epoch": 0.79, "grad_norm": 1.2724648753569285, "learning_rate": 3.378043619679974e-06, "loss": 0.2386, "step": 216 }, { "epoch": 0.8, "grad_norm": 1.2826844172302947, "learning_rate": 3.364257651540891e-06, "loss": 0.2366, "step": 217 }, { "epoch": 0.8, "grad_norm": 1.1767059777022655, "learning_rate": 3.3504417704540925e-06, "loss": 0.2251, "step": 218 }, { "epoch": 0.8, "grad_norm": 1.3111513963454882, "learning_rate": 3.3365964546030544e-06, "loss": 0.2396, "step": 219 }, { "epoch": 0.81, "grad_norm": 1.2617225478707708, "learning_rate": 3.322722183190025e-06, "loss": 0.2412, "step": 220 }, { "epoch": 0.81, "grad_norm": 1.2183220743609309, "learning_rate": 3.308819436419437e-06, "loss": 0.2276, "step": 221 }, { "epoch": 0.82, "grad_norm": 1.31561824749082, "learning_rate": 3.2948886954812877e-06, "loss": 0.2404, "step": 222 }, { "epoch": 0.82, "grad_norm": 1.250087552624437, "learning_rate": 3.280930442534486e-06, "loss": 0.2263, "step": 223 }, { "epoch": 0.82, "grad_norm": 1.2524310598377044, "learning_rate": 3.26694516069016e-06, "loss": 0.2368, "step": 224 }, { "epoch": 0.83, "grad_norm": 1.3487266981725987, "learning_rate": 3.252933333994942e-06, "loss": 0.2243, "step": 225 }, { "epoch": 0.83, "grad_norm": 1.2427013509424278, "learning_rate": 3.238895447414211e-06, "loss": 0.2366, "step": 226 }, { "epoch": 0.83, "grad_norm": 1.268723527146989, "learning_rate": 3.2248319868153067e-06, "loss": 0.2262, "step": 227 }, { "epoch": 0.84, "grad_norm": 1.2476040692827028, "learning_rate": 3.210743438950718e-06, "loss": 0.234, "step": 228 }, { "epoch": 0.84, "grad_norm": 1.2944243964732431, "learning_rate": 3.196630291441231e-06, "loss": 0.2261, "step": 229 }, { "epoch": 0.84, "grad_norm": 1.2348938264581308, "learning_rate": 3.182493032759053e-06, "loss": 0.2368, "step": 230 }, { "epoch": 0.85, "grad_norm": 1.3877133957904717, "learning_rate": 3.168332152210909e-06, "loss": 0.2342, "step": 231 }, { "epoch": 0.85, "grad_norm": 1.2088837041711673, "learning_rate": 3.154148139921102e-06, "loss": 0.222, "step": 232 }, { "epoch": 0.86, "grad_norm": 1.4750513048080165, "learning_rate": 3.1399414868145506e-06, "loss": 0.2301, "step": 233 }, { "epoch": 0.86, "grad_norm": 1.2097458338635088, "learning_rate": 3.1257126845998e-06, "loss": 0.2365, "step": 234 }, { "epoch": 0.86, "grad_norm": 1.3570468614316236, "learning_rate": 3.1114622257520004e-06, "loss": 0.2275, "step": 235 }, { "epoch": 0.87, "grad_norm": 1.2331713108579336, "learning_rate": 3.0971906034958616e-06, "loss": 0.2193, "step": 236 }, { "epoch": 0.87, "grad_norm": 1.330924002893457, "learning_rate": 3.0828983117885856e-06, "loss": 0.2258, "step": 237 }, { "epoch": 0.87, "grad_norm": 1.2713775149937143, "learning_rate": 3.0685858453027668e-06, "loss": 0.2287, "step": 238 }, { "epoch": 0.88, "grad_norm": 1.3460227514964078, "learning_rate": 3.05425369940927e-06, "loss": 0.2268, "step": 239 }, { "epoch": 0.88, "grad_norm": 1.3124465221253792, "learning_rate": 3.0399023701600903e-06, "loss": 0.2237, "step": 240 }, { "epoch": 0.89, "grad_norm": 1.2621420000416141, "learning_rate": 3.0255323542711784e-06, "loss": 0.221, "step": 241 }, { "epoch": 0.89, "grad_norm": 1.3207975689997922, "learning_rate": 3.011144149105251e-06, "loss": 0.2177, "step": 242 }, { "epoch": 0.89, "grad_norm": 1.3364690610440046, "learning_rate": 2.996738252654577e-06, "loss": 0.2266, "step": 243 }, { "epoch": 0.9, "grad_norm": 1.3069082882086795, "learning_rate": 2.9823151635237424e-06, "loss": 0.2274, "step": 244 }, { "epoch": 0.9, "grad_norm": 1.402608898892496, "learning_rate": 2.9678753809123884e-06, "loss": 0.233, "step": 245 }, { "epoch": 0.9, "grad_norm": 1.3349783439901974, "learning_rate": 2.9534194045979397e-06, "loss": 0.2198, "step": 246 }, { "epoch": 0.91, "grad_norm": 1.3319911413244738, "learning_rate": 2.938947734918302e-06, "loss": 0.2241, "step": 247 }, { "epoch": 0.91, "grad_norm": 1.2836113523110935, "learning_rate": 2.924460872754547e-06, "loss": 0.2247, "step": 248 }, { "epoch": 0.91, "grad_norm": 1.3420053396118825, "learning_rate": 2.9099593195135743e-06, "loss": 0.2245, "step": 249 }, { "epoch": 0.92, "grad_norm": 1.3018957576647208, "learning_rate": 2.8954435771107604e-06, "loss": 0.2198, "step": 250 }, { "epoch": 0.92, "grad_norm": 1.493108819116986, "learning_rate": 2.8809141479525843e-06, "loss": 0.2261, "step": 251 }, { "epoch": 0.93, "grad_norm": 1.2240817395656585, "learning_rate": 2.8663715349192388e-06, "loss": 0.2182, "step": 252 }, { "epoch": 0.93, "grad_norm": 1.3972966685231503, "learning_rate": 2.8518162413472266e-06, "loss": 0.2289, "step": 253 }, { "epoch": 0.93, "grad_norm": 1.3158850314947335, "learning_rate": 2.8372487710119374e-06, "loss": 0.2286, "step": 254 }, { "epoch": 0.94, "grad_norm": 1.295772538693981, "learning_rate": 2.8226696281102134e-06, "loss": 0.2157, "step": 255 }, { "epoch": 0.94, "grad_norm": 1.34085577207588, "learning_rate": 2.8080793172428965e-06, "loss": 0.2223, "step": 256 }, { "epoch": 0.94, "grad_norm": 1.3610764715193495, "learning_rate": 2.7934783433973672e-06, "loss": 0.2227, "step": 257 }, { "epoch": 0.95, "grad_norm": 1.2629712566442401, "learning_rate": 2.778867211930061e-06, "loss": 0.2263, "step": 258 }, { "epoch": 0.95, "grad_norm": 1.2782582856568219, "learning_rate": 2.764246428548983e-06, "loss": 0.2234, "step": 259 }, { "epoch": 0.96, "grad_norm": 1.2621019245043847, "learning_rate": 2.7496164992961995e-06, "loss": 0.2177, "step": 260 }, { "epoch": 0.96, "grad_norm": 1.2033350046761524, "learning_rate": 2.7349779305303263e-06, "loss": 0.2226, "step": 261 }, { "epoch": 0.96, "grad_norm": 1.361220136423699, "learning_rate": 2.720331228909005e-06, "loss": 0.2179, "step": 262 }, { "epoch": 0.97, "grad_norm": 1.3715434561254194, "learning_rate": 2.7056769013713623e-06, "loss": 0.2231, "step": 263 }, { "epoch": 0.97, "grad_norm": 1.1330086039392537, "learning_rate": 2.691015455120468e-06, "loss": 0.2164, "step": 264 }, { "epoch": 0.97, "grad_norm": 1.2694263709270768, "learning_rate": 2.6763473976057776e-06, "loss": 0.2127, "step": 265 }, { "epoch": 0.98, "grad_norm": 1.3274231972419466, "learning_rate": 2.6616732365055713e-06, "loss": 0.2092, "step": 266 }, { "epoch": 0.98, "grad_norm": 1.276485394682339, "learning_rate": 2.64699347970938e-06, "loss": 0.2206, "step": 267 }, { "epoch": 0.98, "grad_norm": 1.33640777595863, "learning_rate": 2.6323086353004077e-06, "loss": 0.2201, "step": 268 }, { "epoch": 0.99, "grad_norm": 1.2867150222472765, "learning_rate": 2.6176192115379494e-06, "loss": 0.2176, "step": 269 }, { "epoch": 0.99, "grad_norm": 1.220258552427881, "learning_rate": 2.602925716839795e-06, "loss": 0.2131, "step": 270 }, { "epoch": 1.0, "grad_norm": 1.3301323985426015, "learning_rate": 2.588228659764632e-06, "loss": 0.2244, "step": 271 }, { "epoch": 1.0, "grad_norm": 1.2313785507924382, "learning_rate": 2.573528548994449e-06, "loss": 0.2192, "step": 272 }, { "epoch": 1.0, "eval_loss": 0.22680288553237915, "eval_runtime": 1744.6696, "eval_samples_per_second": 1.325, "eval_steps_per_second": 0.074, "step": 272 } ], "logging_steps": 1, "max_steps": 544, "num_input_tokens_seen": 0, "num_train_epochs": 2, "save_steps": 272, "total_flos": 256045146439680.0, "train_batch_size": 2, "trial_name": null, "trial_params": null }