{ "best_global_step": null, "best_metric": null, "best_model_checkpoint": null, "epoch": 1.9333333333333333, "eval_steps": 50, "global_step": 1450, "is_hyper_param_search": false, "is_local_process_zero": true, "is_world_process_zero": true, "log_history": [ { "accuracy": 0.65, "epoch": 0.013333333333333334, "grad_norm": 26.781282424926758, "learning_rate": 4.970000000000001e-06, "loss": 0.6315235137939453, "margin": 0.18203125, "max_reward": -1.005078125, "mean_reward": -1.4798828125, "min_reward": -1.972265625, "num_tokens": 16214.0, "step": 10 }, { "accuracy": 0.35, "epoch": 0.02666666666666667, "grad_norm": 18.78548812866211, "learning_rate": 4.936666666666667e-06, "loss": 0.9058952331542969, "margin": -0.21234130859375, "max_reward": -0.0454345703125, "mean_reward": -1.138067626953125, "min_reward": -2.29296875, "num_tokens": 32750.0, "step": 20 }, { "accuracy": 0.65, "epoch": 0.04, "grad_norm": 23.18710708618164, "learning_rate": 4.9033333333333335e-06, "loss": 0.666615343093872, "margin": 0.1455078125, "max_reward": -0.2720703125, "mean_reward": -0.9361328125, "min_reward": -1.67734375, "num_tokens": 49136.0, "step": 30 }, { "accuracy": 0.55, "epoch": 0.05333333333333334, "grad_norm": 23.498512268066406, "learning_rate": 4.87e-06, "loss": 0.775050687789917, "margin": -0.038720703125, "max_reward": -0.69931640625, "mean_reward": -1.3807861328125, "min_reward": -1.86953125, "num_tokens": 64999.0, "step": 40 }, { "accuracy": 0.7, "epoch": 0.06666666666666667, "grad_norm": 25.616416931152344, "learning_rate": 4.836666666666667e-06, "loss": 0.6898736953735352, "margin": 0.26455078125, "max_reward": -0.5318359375, "mean_reward": -1.313720703125, "min_reward": -2.13203125, "num_tokens": 82563.0, "step": 50 }, { "epoch": 0.06666666666666667, "eval_accuracy": 0.532, "eval_loss": 0.7488015294075012, "eval_margin": 0.0769696044921875, "eval_max_reward": -1.1200030517578126, "eval_mean_reward": -1.4668482055664063, "eval_min_reward": -1.813693359375, "eval_num_tokens": 82563.0, "eval_runtime": 59.7497, "eval_samples_per_second": 8.368, "eval_steps_per_second": 8.368, "step": 50 }, { "accuracy": 0.5, "epoch": 0.08, "grad_norm": 47.123233795166016, "learning_rate": 4.803333333333334e-06, "loss": 0.7043657302856445, "margin": 0.144775390625, "max_reward": -0.82724609375, "mean_reward": -1.5813232421875, "min_reward": -2.2578125, "num_tokens": 100158.0, "step": 60 }, { "accuracy": 0.45, "epoch": 0.09333333333333334, "grad_norm": 43.35136032104492, "learning_rate": 4.77e-06, "loss": 0.7232241630554199, "margin": 0.070654296875, "max_reward": -1.10107421875, "mean_reward": -1.6929443359375, "min_reward": -2.36640625, "num_tokens": 116349.0, "step": 70 }, { "accuracy": 0.35, "epoch": 0.10666666666666667, "grad_norm": 38.70017623901367, "learning_rate": 4.736666666666667e-06, "loss": 0.9364012718200684, "margin": -0.284765625, "max_reward": -0.4794921875, "mean_reward": -1.26298828125, "min_reward": -1.909375, "num_tokens": 132123.0, "step": 80 }, { "accuracy": 0.45, "epoch": 0.12, "grad_norm": 29.024276733398438, "learning_rate": 4.703333333333334e-06, "loss": 0.8137557983398438, "margin": -0.1231597900390625, "max_reward": 0.445391845703125, "mean_reward": -0.22428436279296876, "min_reward": -0.788671875, "num_tokens": 147982.0, "step": 90 }, { "accuracy": 0.45, "epoch": 0.13333333333333333, "grad_norm": 34.12649154663086, "learning_rate": 4.670000000000001e-06, "loss": 0.7763675689697266, "margin": -0.06761474609375, "max_reward": 0.74072265625, "mean_reward": 0.255206298828125, "min_reward": -0.2246826171875, "num_tokens": 164058.0, "step": 100 }, { "epoch": 0.13333333333333333, "eval_accuracy": 0.51, "eval_loss": 0.7375237941741943, "eval_margin": 0.008534343719482421, "eval_max_reward": 0.5244478607177734, "eval_mean_reward": 0.2890134220123291, "eval_min_reward": 0.053578983306884764, "eval_num_tokens": 164058.0, "eval_runtime": 62.906, "eval_samples_per_second": 7.948, "eval_steps_per_second": 7.948, "step": 100 }, { "accuracy": 0.45, "epoch": 0.14666666666666667, "grad_norm": 41.91898727416992, "learning_rate": 4.6366666666666674e-06, "loss": 0.7357148647308349, "margin": -0.05003662109375, "max_reward": 0.8333984375, "mean_reward": 0.398614501953125, "min_reward": -0.0110595703125, "num_tokens": 180287.0, "step": 110 }, { "accuracy": 0.45, "epoch": 0.16, "grad_norm": 27.607086181640625, "learning_rate": 4.603333333333334e-06, "loss": 0.7425424575805664, "margin": -0.05296630859375, "max_reward": 0.73359375, "mean_reward": 0.207928466796875, "min_reward": -0.285498046875, "num_tokens": 195395.0, "step": 120 }, { "accuracy": 0.55, "epoch": 0.17333333333333334, "grad_norm": 18.465009689331055, "learning_rate": 4.57e-06, "loss": 0.7002963066101074, "margin": 0.0166015625, "max_reward": 0.15341796875, "mean_reward": -0.245654296875, "min_reward": -0.6259765625, "num_tokens": 211423.0, "step": 130 }, { "accuracy": 0.85, "epoch": 0.18666666666666668, "grad_norm": 24.311744689941406, "learning_rate": 4.536666666666667e-06, "loss": 0.5668483257293702, "margin": 0.39488706588745115, "max_reward": 0.4626953125, "mean_reward": -0.23407502174377443, "min_reward": -0.8857421875, "num_tokens": 227155.0, "step": 140 }, { "accuracy": 0.55, "epoch": 0.2, "grad_norm": 27.90814971923828, "learning_rate": 4.503333333333333e-06, "loss": 0.9236076354980469, "margin": 0.2251708984375, "max_reward": 1.816796875, "mean_reward": 0.43135986328125, "min_reward": -1.107373046875, "num_tokens": 243609.0, "step": 150 }, { "epoch": 0.2, "eval_accuracy": 0.568, "eval_loss": 0.9139922857284546, "eval_margin": 0.21751553344726562, "eval_max_reward": 1.1517796630859376, "eval_mean_reward": 0.4644862823486328, "eval_min_reward": -0.22280709838867188, "eval_num_tokens": 243609.0, "eval_runtime": 62.3021, "eval_samples_per_second": 8.025, "eval_steps_per_second": 8.025, "step": 150 }, { "accuracy": 0.5, "epoch": 0.21333333333333335, "grad_norm": 47.18056869506836, "learning_rate": 4.47e-06, "loss": 1.0227947235107422, "margin": -0.06168212890625, "max_reward": 1.634716796875, "mean_reward": 0.546588134765625, "min_reward": -0.413671875, "num_tokens": 261657.0, "step": 160 }, { "accuracy": 0.5, "epoch": 0.22666666666666666, "grad_norm": 19.242305755615234, "learning_rate": 4.436666666666667e-06, "loss": 0.6836428642272949, "margin": 0.1833953857421875, "max_reward": 0.95390625, "mean_reward": 0.25375518798828123, "min_reward": -0.5401123046875, "num_tokens": 277075.0, "step": 170 }, { "accuracy": 0.65, "epoch": 0.24, "grad_norm": 13.613568305969238, "learning_rate": 4.403333333333334e-06, "loss": 0.5862503528594971, "margin": 0.335546875, "max_reward": 0.95908203125, "mean_reward": 0.436962890625, "min_reward": -0.14921875, "num_tokens": 291771.0, "step": 180 }, { "accuracy": 0.75, "epoch": 0.25333333333333335, "grad_norm": 16.98612403869629, "learning_rate": 4.3700000000000005e-06, "loss": 0.4962740898132324, "margin": 0.72353515625, "max_reward": 2.7734375, "mean_reward": 1.643115234375, "min_reward": 0.7041015625, "num_tokens": 306869.0, "step": 190 }, { "accuracy": 0.55, "epoch": 0.26666666666666666, "grad_norm": 5.9575724601745605, "learning_rate": 4.336666666666667e-06, "loss": 0.8757614135742188, "margin": 0.180908203125, "max_reward": 3.7546875, "mean_reward": 2.4668212890625, "min_reward": 1.08994140625, "num_tokens": 322209.0, "step": 200 }, { "epoch": 0.26666666666666666, "eval_accuracy": 0.558, "eval_loss": 0.8031588196754456, "eval_margin": 0.211654052734375, "eval_max_reward": 3.3186171875, "eval_mean_reward": 2.7612215576171875, "eval_min_reward": 2.203825927734375, "eval_num_tokens": 322209.0, "eval_runtime": 62.4301, "eval_samples_per_second": 8.009, "eval_steps_per_second": 8.009, "step": 200 }, { "accuracy": 0.5, "epoch": 0.28, "grad_norm": 20.00104331970215, "learning_rate": 4.303333333333334e-06, "loss": 0.742474889755249, "margin": 0.2575439453125, "max_reward": 3.06640625, "mean_reward": 1.81966552734375, "min_reward": 0.197021484375, "num_tokens": 338225.0, "step": 210 }, { "accuracy": 0.5, "epoch": 0.29333333333333333, "grad_norm": 23.415834426879883, "learning_rate": 4.270000000000001e-06, "loss": 0.8180597305297852, "margin": 0.2842529296875, "max_reward": 4.1375, "mean_reward": 2.89720458984375, "min_reward": 1.491162109375, "num_tokens": 352992.0, "step": 220 }, { "accuracy": 0.5, "epoch": 0.30666666666666664, "grad_norm": 28.938772201538086, "learning_rate": 4.236666666666667e-06, "loss": 0.8096456527709961, "margin": -0.066015625, "max_reward": 3.71875, "mean_reward": 2.8822265625, "min_reward": 1.9859375, "num_tokens": 368570.0, "step": 230 }, { "accuracy": 0.65, "epoch": 0.32, "grad_norm": 19.143875122070312, "learning_rate": 4.2033333333333335e-06, "loss": 0.5784501552581787, "margin": 0.43447265625, "max_reward": 2.54921875, "mean_reward": 1.823486328125, "min_reward": 1.0931640625, "num_tokens": 384794.0, "step": 240 }, { "accuracy": 0.6, "epoch": 0.3333333333333333, "grad_norm": 25.47344207763672, "learning_rate": 4.17e-06, "loss": 0.7684556484222412, "margin": 0.04365234375, "max_reward": 3.128125, "mean_reward": 2.072509765625, "min_reward": 1.1900390625, "num_tokens": 401148.0, "step": 250 }, { "epoch": 0.3333333333333333, "eval_accuracy": 0.572, "eval_loss": 0.713194727897644, "eval_margin": 0.1503466796875, "eval_max_reward": 2.57871484375, "eval_mean_reward": 2.22652490234375, "eval_min_reward": 1.8743349609375, "eval_num_tokens": 401148.0, "eval_runtime": 62.3122, "eval_samples_per_second": 8.024, "eval_steps_per_second": 8.024, "step": 250 }, { "accuracy": 0.3, "epoch": 0.3466666666666667, "grad_norm": 27.635066986083984, "learning_rate": 4.136666666666667e-06, "loss": 0.9010297775268554, "margin": -0.2201416015625, "max_reward": 2.42265625, "mean_reward": 1.72039794921875, "min_reward": 0.801123046875, "num_tokens": 416243.0, "step": 260 }, { "accuracy": 0.65, "epoch": 0.36, "grad_norm": 22.39845848083496, "learning_rate": 4.1033333333333336e-06, "loss": 0.7369675159454345, "margin": 0.000390625, "max_reward": 2.31796875, "mean_reward": 1.5876953125, "min_reward": 0.834375, "num_tokens": 432425.0, "step": 270 }, { "accuracy": 0.6, "epoch": 0.37333333333333335, "grad_norm": 30.26738166809082, "learning_rate": 4.07e-06, "loss": 0.7203866004943847, "margin": 0.00186767578125, "max_reward": 1.509375, "mean_reward": 1.016265869140625, "min_reward": 0.5093994140625, "num_tokens": 446709.0, "step": 280 }, { "accuracy": 0.6, "epoch": 0.38666666666666666, "grad_norm": 17.360742568969727, "learning_rate": 4.036666666666667e-06, "loss": 0.6374303817749023, "margin": 0.159130859375, "max_reward": 1.0638671875, "mean_reward": 0.6469970703125, "min_reward": 0.1876953125, "num_tokens": 462422.0, "step": 290 }, { "accuracy": 0.45, "epoch": 0.4, "grad_norm": 24.10837173461914, "learning_rate": 4.003333333333334e-06, "loss": 0.7569143772125244, "margin": 0.0091796875, "max_reward": 1.9375, "mean_reward": 1.299609375, "min_reward": 0.747265625, "num_tokens": 480306.0, "step": 300 }, { "epoch": 0.4, "eval_accuracy": 0.606, "eval_loss": 0.6851577162742615, "eval_margin": 0.110994384765625, "eval_max_reward": 1.68329541015625, "eval_mean_reward": 1.4451339111328125, "eval_min_reward": 1.206972412109375, "eval_num_tokens": 480306.0, "eval_runtime": 62.4657, "eval_samples_per_second": 8.004, "eval_steps_per_second": 8.004, "step": 300 }, { "accuracy": 0.65, "epoch": 0.41333333333333333, "grad_norm": 13.338096618652344, "learning_rate": 3.97e-06, "loss": 0.6534909725189209, "margin": 0.14378662109375, "max_reward": 1.83671875, "mean_reward": 1.229180908203125, "min_reward": 0.6009033203125, "num_tokens": 495683.0, "step": 310 }, { "accuracy": 0.6, "epoch": 0.4266666666666667, "grad_norm": 14.492751121520996, "learning_rate": 3.936666666666667e-06, "loss": 0.6619410991668702, "margin": 0.144091796875, "max_reward": 2.00234375, "mean_reward": 1.4490966796875, "min_reward": 0.80947265625, "num_tokens": 512780.0, "step": 320 }, { "accuracy": 0.6, "epoch": 0.44, "grad_norm": 17.708974838256836, "learning_rate": 3.903333333333334e-06, "loss": 0.6470551013946533, "margin": 0.224090576171875, "max_reward": 2.8828125, "mean_reward": 2.1324859619140626, "min_reward": 1.29166259765625, "num_tokens": 529065.0, "step": 330 }, { "accuracy": 0.5, "epoch": 0.4533333333333333, "grad_norm": 18.278581619262695, "learning_rate": 3.87e-06, "loss": 0.6739425659179688, "margin": 0.33515625, "max_reward": 3.9375, "mean_reward": 2.80546875, "min_reward": 1.79609375, "num_tokens": 546716.0, "step": 340 }, { "accuracy": 0.45, "epoch": 0.4666666666666667, "grad_norm": 38.923805236816406, "learning_rate": 3.836666666666667e-06, "loss": 0.7799544334411621, "margin": -0.078515625, "max_reward": 4.6078125, "mean_reward": 3.6021484375, "min_reward": 2.72265625, "num_tokens": 563762.0, "step": 350 }, { "epoch": 0.4666666666666667, "eval_accuracy": 0.572, "eval_loss": 0.7251431345939636, "eval_margin": 0.1519296875, "eval_max_reward": 4.2361640625, "eval_mean_reward": 3.866208984375, "eval_min_reward": 3.49625390625, "eval_num_tokens": 563762.0, "eval_runtime": 62.2654, "eval_samples_per_second": 8.03, "eval_steps_per_second": 8.03, "step": 350 }, { "accuracy": 0.8, "epoch": 0.48, "grad_norm": 25.30300521850586, "learning_rate": 3.803333333333334e-06, "loss": 0.5838129043579101, "margin": 0.34296875, "max_reward": 4.70625, "mean_reward": 3.857421875, "min_reward": 3.0, "num_tokens": 579498.0, "step": 360 }, { "accuracy": 0.5, "epoch": 0.49333333333333335, "grad_norm": 41.584007263183594, "learning_rate": 3.7700000000000003e-06, "loss": 0.8219932556152344, "margin": 0.094140625, "max_reward": 4.8875, "mean_reward": 4.0658203125, "min_reward": 3.16328125, "num_tokens": 597098.0, "step": 370 }, { "accuracy": 0.65, "epoch": 0.5066666666666667, "grad_norm": 16.238313674926758, "learning_rate": 3.7366666666666667e-06, "loss": 0.6716785430908203, "margin": 0.2427490234375, "max_reward": 3.34375, "mean_reward": 2.30460205078125, "min_reward": 1.168408203125, "num_tokens": 613640.0, "step": 380 }, { "accuracy": 0.6, "epoch": 0.52, "grad_norm": 25.986833572387695, "learning_rate": 3.7033333333333336e-06, "loss": 0.716916561126709, "margin": 0.42451171875, "max_reward": 2.96328125, "mean_reward": 1.55986328125, "min_reward": 0.221484375, "num_tokens": 629441.0, "step": 390 }, { "accuracy": 0.75, "epoch": 0.5333333333333333, "grad_norm": 23.42198371887207, "learning_rate": 3.6700000000000004e-06, "loss": 0.6442339897155762, "margin": 0.3801513671875, "max_reward": 2.6359375, "mean_reward": 1.87711181640625, "min_reward": 1.043212890625, "num_tokens": 645440.0, "step": 400 }, { "epoch": 0.5333333333333333, "eval_accuracy": 0.618, "eval_loss": 0.7101489305496216, "eval_margin": 0.2907686767578125, "eval_max_reward": 2.299792724609375, "eval_mean_reward": 1.8211738891601563, "eval_min_reward": 1.3425550537109374, "eval_num_tokens": 645440.0, "eval_runtime": 62.216, "eval_samples_per_second": 8.037, "eval_steps_per_second": 8.037, "step": 400 }, { "accuracy": 0.6, "epoch": 0.5466666666666666, "grad_norm": 20.917293548583984, "learning_rate": 3.6366666666666673e-06, "loss": 0.8052460670471191, "margin": 0.2086181640625, "max_reward": 2.9359375, "mean_reward": 1.69215087890625, "min_reward": 0.44091796875, "num_tokens": 663799.0, "step": 410 }, { "accuracy": 0.7, "epoch": 0.56, "grad_norm": 16.333770751953125, "learning_rate": 3.6033333333333337e-06, "loss": 0.6448392868041992, "margin": 0.210369873046875, "max_reward": 0.69697265625, "mean_reward": 0.0324127197265625, "min_reward": -0.758642578125, "num_tokens": 678669.0, "step": 420 }, { "accuracy": 0.55, "epoch": 0.5733333333333334, "grad_norm": 29.64971923828125, "learning_rate": 3.57e-06, "loss": 0.6741883277893066, "margin": 0.1592041015625, "max_reward": 0.2900146484375, "mean_reward": -0.4987548828125, "min_reward": -1.176171875, "num_tokens": 694858.0, "step": 430 }, { "accuracy": 0.45, "epoch": 0.5866666666666667, "grad_norm": 26.162715911865234, "learning_rate": 3.536666666666667e-06, "loss": 0.7661936283111572, "margin": -0.02214813232421875, "max_reward": -0.0858306884765625, "mean_reward": -0.7683937072753906, "min_reward": -1.260546875, "num_tokens": 710995.0, "step": 440 }, { "accuracy": 0.75, "epoch": 0.6, "grad_norm": 22.932537078857422, "learning_rate": 3.5033333333333334e-06, "loss": 0.6263980865478516, "margin": 0.1980621337890625, "max_reward": -0.37247314453125, "mean_reward": -0.7766494750976562, "min_reward": -1.234375, "num_tokens": 726473.0, "step": 450 }, { "epoch": 0.6, "eval_accuracy": 0.628, "eval_loss": 0.6893447041511536, "eval_margin": 0.17741348266601562, "eval_max_reward": -0.00065301513671875, "eval_mean_reward": -0.32534046936035155, "eval_min_reward": -0.6500279235839844, "eval_num_tokens": 726473.0, "eval_runtime": 62.4229, "eval_samples_per_second": 8.01, "eval_steps_per_second": 8.01, "step": 450 }, { "accuracy": 0.45, "epoch": 0.6133333333333333, "grad_norm": 23.200468063354492, "learning_rate": 3.4700000000000002e-06, "loss": 0.7532081127166748, "margin": -0.0011383056640625, "max_reward": 0.3897216796875, "mean_reward": -0.43083343505859373, "min_reward": -1.10458984375, "num_tokens": 743418.0, "step": 460 }, { "accuracy": 0.65, "epoch": 0.6266666666666667, "grad_norm": 23.413482666015625, "learning_rate": 3.436666666666667e-06, "loss": 0.6554159641265869, "margin": 0.20365447998046876, "max_reward": -0.0306304931640625, "mean_reward": -0.6653846740722656, "min_reward": -1.2474609375, "num_tokens": 758877.0, "step": 470 }, { "accuracy": 0.45, "epoch": 0.64, "grad_norm": 23.571287155151367, "learning_rate": 3.4033333333333335e-06, "loss": 0.7563723564147949, "margin": 0.1056365966796875, "max_reward": 1.4620849609375, "mean_reward": 0.06235504150390625, "min_reward": -1.196258544921875, "num_tokens": 775285.0, "step": 480 }, { "accuracy": 0.7, "epoch": 0.6533333333333333, "grad_norm": 12.891551971435547, "learning_rate": 3.3700000000000003e-06, "loss": 0.5850713729858399, "margin": 0.73206787109375, "max_reward": 2.0690673828125, "mean_reward": 0.511407470703125, "min_reward": -0.6796875, "num_tokens": 790684.0, "step": 490 }, { "accuracy": 0.8, "epoch": 0.6666666666666666, "grad_norm": 64.09451293945312, "learning_rate": 3.3366666666666668e-06, "loss": 0.498201847076416, "margin": 0.911553955078125, "max_reward": 1.70078125, "mean_reward": 0.3194183349609375, "min_reward": -1.081884765625, "num_tokens": 806487.0, "step": 500 }, { "epoch": 0.6666666666666666, "eval_accuracy": 0.622, "eval_loss": 0.7471536993980408, "eval_margin": 0.40328579330444336, "eval_max_reward": 0.7979255027770996, "eval_mean_reward": 0.18772327995300292, "eval_min_reward": -0.4224789428710938, "eval_num_tokens": 806487.0, "eval_runtime": 62.2921, "eval_samples_per_second": 8.027, "eval_steps_per_second": 8.027, "step": 500 }, { "accuracy": 0.65, "epoch": 0.68, "grad_norm": 122.53633117675781, "learning_rate": 3.303333333333333e-06, "loss": 0.7432568550109864, "margin": 0.4142311096191406, "max_reward": 0.76953125, "mean_reward": -0.3576038360595703, "min_reward": -1.263671875, "num_tokens": 823462.0, "step": 510 }, { "accuracy": 0.75, "epoch": 0.6933333333333334, "grad_norm": 32.80317306518555, "learning_rate": 3.2700000000000005e-06, "loss": 0.41660318374633787, "margin": 1.13740234375, "max_reward": 2.3439453125, "mean_reward": 0.789794921875, "min_reward": -1.080078125, "num_tokens": 838460.0, "step": 520 }, { "accuracy": 0.75, "epoch": 0.7066666666666667, "grad_norm": 17.606361389160156, "learning_rate": 3.236666666666667e-06, "loss": 0.6934970378875732, "margin": 0.973486328125, "max_reward": 4.0046875, "mean_reward": 2.1105224609375, "min_reward": -0.177734375, "num_tokens": 854813.0, "step": 530 }, { "accuracy": 0.55, "epoch": 0.72, "grad_norm": 156.23760986328125, "learning_rate": 3.2033333333333337e-06, "loss": 1.1027880668640138, "margin": 0.20673828125, "max_reward": 5.353125, "mean_reward": 3.049951171875, "min_reward": 0.3279296875, "num_tokens": 870996.0, "step": 540 }, { "accuracy": 0.45, "epoch": 0.7333333333333333, "grad_norm": 34.310646057128906, "learning_rate": 3.17e-06, "loss": 1.0296151161193847, "margin": 0.0447265625, "max_reward": 4.716796875, "mean_reward": 2.67509765625, "min_reward": 0.359375, "num_tokens": 886715.0, "step": 550 }, { "epoch": 0.7333333333333333, "eval_accuracy": 0.598, "eval_loss": 0.8312986493110657, "eval_margin": 0.4697259521484375, "eval_max_reward": 3.234610900878906, "eval_mean_reward": 2.4688244018554686, "eval_min_reward": 1.7030379028320313, "eval_num_tokens": 886715.0, "eval_runtime": 62.3011, "eval_samples_per_second": 8.026, "eval_steps_per_second": 8.026, "step": 550 }, { "accuracy": 0.65, "epoch": 0.7466666666666667, "grad_norm": 2.1129918098449707, "learning_rate": 3.1366666666666666e-06, "loss": 0.6547252178192139, "margin": 0.72010498046875, "max_reward": 3.6328125, "mean_reward": 1.748907470703125, "min_reward": -0.1922607421875, "num_tokens": 901702.0, "step": 560 }, { "accuracy": 0.55, "epoch": 0.76, "grad_norm": 19.6908016204834, "learning_rate": 3.103333333333334e-06, "loss": 0.6725973129272461, "margin": 0.536669921875, "max_reward": 0.89931640625, "mean_reward": -0.4901123046875, "min_reward": -1.93125, "num_tokens": 919528.0, "step": 570 }, { "accuracy": 0.5, "epoch": 0.7733333333333333, "grad_norm": 56.94779586791992, "learning_rate": 3.0700000000000003e-06, "loss": 0.6944450378417969, "margin": 0.448974609375, "max_reward": 0.27529296875, "mean_reward": -1.3425048828125, "min_reward": -2.58046875, "num_tokens": 936620.0, "step": 580 }, { "accuracy": 0.85, "epoch": 0.7866666666666666, "grad_norm": 95.96953582763672, "learning_rate": 3.0366666666666667e-06, "loss": 0.3615936040878296, "margin": 1.50234375, "max_reward": 1.0953125, "mean_reward": -1.33935546875, "min_reward": -3.105078125, "num_tokens": 953455.0, "step": 590 }, { "accuracy": 0.7, "epoch": 0.8, "grad_norm": 88.94745635986328, "learning_rate": 3.0033333333333335e-06, "loss": 0.9391720771789551, "margin": 1.326025390625, "max_reward": 5.046875, "mean_reward": 2.1633544921875, "min_reward": -1.42548828125, "num_tokens": 968502.0, "step": 600 }, { "epoch": 0.8, "eval_accuracy": 0.598, "eval_loss": 1.15407395362854, "eval_margin": 0.8387359619140625, "eval_max_reward": 3.26913037109375, "eval_mean_reward": 1.9263014526367188, "eval_min_reward": 0.5834725341796875, "eval_num_tokens": 968502.0, "eval_runtime": 62.2605, "eval_samples_per_second": 8.031, "eval_steps_per_second": 8.031, "step": 600 }, { "accuracy": 0.5, "epoch": 0.8133333333333334, "grad_norm": 32.30426025390625, "learning_rate": 2.97e-06, "loss": 1.0049471855163574, "margin": 0.078076171875, "max_reward": 4.67255859375, "mean_reward": 1.0933349609375, "min_reward": -2.70625, "num_tokens": 985736.0, "step": 610 }, { "accuracy": 0.7, "epoch": 0.8266666666666667, "grad_norm": 51.47792434692383, "learning_rate": 2.936666666666667e-06, "loss": 0.7427908420562744, "margin": 0.39453125, "max_reward": 5.634375, "mean_reward": 4.305078125, "min_reward": 2.32734375, "num_tokens": 1002008.0, "step": 620 }, { "accuracy": 0.5, "epoch": 0.84, "grad_norm": 23.034889221191406, "learning_rate": 2.9033333333333336e-06, "loss": 0.8153639793395996, "margin": -0.12333984375, "max_reward": 5.821875, "mean_reward": 4.691064453125, "min_reward": 3.4435546875, "num_tokens": 1019028.0, "step": 630 }, { "accuracy": 0.35, "epoch": 0.8533333333333334, "grad_norm": 161.87069702148438, "learning_rate": 2.87e-06, "loss": 0.8048789978027344, "margin": -0.009716796875, "max_reward": 5.628125, "mean_reward": 4.9048583984375, "min_reward": 3.97490234375, "num_tokens": 1033366.0, "step": 640 }, { "accuracy": 0.55, "epoch": 0.8666666666666667, "grad_norm": 15.923206329345703, "learning_rate": 2.836666666666667e-06, "loss": 0.7137910842895507, "margin": 0.00390625, "max_reward": 5.36875, "mean_reward": 4.934765625, "min_reward": 4.5046875, "num_tokens": 1050252.0, "step": 650 }, { "epoch": 0.8666666666666667, "eval_accuracy": 0.526, "eval_loss": 0.7010688185691833, "eval_margin": 0.0384140625, "eval_max_reward": 5.014875, "eval_mean_reward": 4.85937109375, "eval_min_reward": 4.7038671875, "eval_num_tokens": 1050252.0, "eval_runtime": 62.339, "eval_samples_per_second": 8.021, "eval_steps_per_second": 8.021, "step": 650 }, { "accuracy": 0.6, "epoch": 0.88, "grad_norm": 15.554156303405762, "learning_rate": 2.8033333333333333e-06, "loss": 0.65819411277771, "margin": 0.16484375, "max_reward": 5.25, "mean_reward": 4.766015625, "min_reward": 4.1578125, "num_tokens": 1068961.0, "step": 660 }, { "accuracy": 0.55, "epoch": 0.8933333333333333, "grad_norm": 14.689868927001953, "learning_rate": 2.7700000000000006e-06, "loss": 0.68900465965271, "margin": 0.12265625, "max_reward": 4.746875, "mean_reward": 4.285546875, "min_reward": 3.6515625, "num_tokens": 1084178.0, "step": 670 }, { "accuracy": 0.65, "epoch": 0.9066666666666666, "grad_norm": 30.876516342163086, "learning_rate": 2.736666666666667e-06, "loss": 0.6992377281188965, "margin": 0.0375, "max_reward": 4.3859375, "mean_reward": 4.02265625, "min_reward": 3.6703125, "num_tokens": 1100910.0, "step": 680 }, { "accuracy": 0.65, "epoch": 0.92, "grad_norm": 24.66133689880371, "learning_rate": 2.7033333333333334e-06, "loss": 0.6626120567321777, "margin": 0.20625, "max_reward": 4.084375, "mean_reward": 3.6171875, "min_reward": 2.8875, "num_tokens": 1116440.0, "step": 690 }, { "accuracy": 0.45, "epoch": 0.9333333333333333, "grad_norm": 15.120383262634277, "learning_rate": 2.6700000000000003e-06, "loss": 0.696980619430542, "margin": 0.13203125, "max_reward": 4.3109375, "mean_reward": 3.649609375, "min_reward": 3.0734375, "num_tokens": 1132429.0, "step": 700 }, { "epoch": 0.9333333333333333, "eval_accuracy": 0.582, "eval_loss": 0.7070507407188416, "eval_margin": 0.0835546875, "eval_max_reward": 3.954890625, "eval_mean_reward": 3.70844921875, "eval_min_reward": 3.4620078125, "eval_num_tokens": 1132429.0, "eval_runtime": 62.4281, "eval_samples_per_second": 8.009, "eval_steps_per_second": 8.009, "step": 700 }, { "accuracy": 0.55, "epoch": 0.9466666666666667, "grad_norm": 14.82724380493164, "learning_rate": 2.6366666666666667e-06, "loss": 0.6956966400146485, "margin": 0.065625, "max_reward": 4.340625, "mean_reward": 3.9, "min_reward": 3.3796875, "num_tokens": 1148620.0, "step": 710 }, { "accuracy": 0.35, "epoch": 0.96, "grad_norm": 28.879314422607422, "learning_rate": 2.603333333333334e-06, "loss": 0.7410328388214111, "margin": -0.046875, "max_reward": 4.2875, "mean_reward": 3.8375, "min_reward": 3.3515625, "num_tokens": 1164601.0, "step": 720 }, { "accuracy": 0.65, "epoch": 0.9733333333333334, "grad_norm": 17.8446044921875, "learning_rate": 2.5700000000000004e-06, "loss": 0.6587936878204346, "margin": 0.2615234375, "max_reward": 4.5078125, "mean_reward": 3.79892578125, "min_reward": 3.089453125, "num_tokens": 1179554.0, "step": 730 }, { "accuracy": 0.55, "epoch": 0.9866666666666667, "grad_norm": 16.74919891357422, "learning_rate": 2.536666666666667e-06, "loss": 0.9084688186645508, "margin": -0.1103515625, "max_reward": 3.65, "mean_reward": 2.99423828125, "min_reward": 1.926953125, "num_tokens": 1195986.0, "step": 740 }, { "accuracy": 0.6, "epoch": 1.0, "grad_norm": 20.36426544189453, "learning_rate": 2.5033333333333336e-06, "loss": 0.6503221035003662, "margin": 0.196875, "max_reward": 3.8296875, "mean_reward": 3.22890625, "min_reward": 2.6796875, "num_tokens": 1212807.0, "step": 750 }, { "epoch": 1.0, "eval_accuracy": 0.574, "eval_loss": 0.724712073802948, "eval_margin": 0.1419578857421875, "eval_max_reward": 3.170109375, "eval_mean_reward": 2.827115661621094, "eval_min_reward": 2.4841219482421875, "eval_num_tokens": 1212807.0, "eval_runtime": 62.4609, "eval_samples_per_second": 8.005, "eval_steps_per_second": 8.005, "step": 750 }, { "accuracy": 0.8, "epoch": 1.0133333333333334, "grad_norm": 25.44012451171875, "learning_rate": 2.47e-06, "loss": 0.5082518577575683, "margin": 0.634228515625, "max_reward": 3.65, "mean_reward": 2.5745849609375, "min_reward": 1.33837890625, "num_tokens": 1230748.0, "step": 760 }, { "accuracy": 0.6, "epoch": 1.0266666666666666, "grad_norm": 18.013717651367188, "learning_rate": 2.436666666666667e-06, "loss": 0.5603197574615478, "margin": 0.5654541015625, "max_reward": 3.029833984375, "mean_reward": 1.75579833984375, "min_reward": 0.37646484375, "num_tokens": 1245628.0, "step": 770 }, { "accuracy": 0.55, "epoch": 1.04, "grad_norm": 99.6661148071289, "learning_rate": 2.4033333333333338e-06, "loss": 0.8869245529174805, "margin": 0.205218505859375, "max_reward": 2.080810546875, "mean_reward": -0.0414886474609375, "min_reward": -2.2107421875, "num_tokens": 1262533.0, "step": 780 }, { "accuracy": 0.7, "epoch": 1.0533333333333332, "grad_norm": 53.62653732299805, "learning_rate": 2.37e-06, "loss": 0.5512317657470703, "margin": 1.0128662109375, "max_reward": 1.4505859375, "mean_reward": -0.33284912109375, "min_reward": -2.0107421875, "num_tokens": 1277096.0, "step": 790 }, { "accuracy": 0.7, "epoch": 1.0666666666666667, "grad_norm": 95.58021545410156, "learning_rate": 2.3366666666666666e-06, "loss": 0.6172554016113281, "margin": 1.2177459716796875, "max_reward": 1.304345703125, "mean_reward": -1.1034286499023438, "min_reward": -3.26953125, "num_tokens": 1293073.0, "step": 800 }, { "epoch": 1.0666666666666667, "eval_accuracy": 0.596, "eval_loss": 0.8903509378433228, "eval_margin": 0.564534423828125, "eval_max_reward": -0.50098583984375, "eval_mean_reward": -1.4187645263671875, "eval_min_reward": -2.336543212890625, "eval_num_tokens": 1293073.0, "eval_runtime": 62.3781, "eval_samples_per_second": 8.016, "eval_steps_per_second": 8.016, "step": 800 }, { "accuracy": 0.65, "epoch": 1.08, "grad_norm": 34.59308624267578, "learning_rate": 2.3033333333333334e-06, "loss": 0.7249170303344726, "margin": 0.968017578125, "max_reward": 0.56064453125, "mean_reward": -1.4824951171875, "min_reward": -3.8390625, "num_tokens": 1308687.0, "step": 810 }, { "accuracy": 0.8, "epoch": 1.0933333333333333, "grad_norm": 30.92959976196289, "learning_rate": 2.2700000000000003e-06, "loss": 0.39841785430908205, "margin": 1.375030517578125, "max_reward": 1.658984375, "mean_reward": -0.6310943603515625, "min_reward": -3.046484375, "num_tokens": 1324618.0, "step": 820 }, { "accuracy": 0.7, "epoch": 1.1066666666666667, "grad_norm": 208.02574157714844, "learning_rate": 2.236666666666667e-06, "loss": 0.8557153701782226, "margin": 1.5063720703125, "max_reward": 1.581640625, "mean_reward": -0.71844482421875, "min_reward": -3.421484375, "num_tokens": 1339988.0, "step": 830 }, { "accuracy": 0.75, "epoch": 1.12, "grad_norm": 279.54534912109375, "learning_rate": 2.2033333333333336e-06, "loss": 0.7527968406677246, "margin": 2.487060546875, "max_reward": 1.5951171875, "mean_reward": -2.3231201171875, "min_reward": -5.828125, "num_tokens": 1356112.0, "step": 840 }, { "accuracy": 0.7, "epoch": 1.1333333333333333, "grad_norm": 35.76692581176758, "learning_rate": 2.17e-06, "loss": 0.8942340850830078, "margin": 1.6185546875, "max_reward": 0.787890625, "mean_reward": -1.966064453125, "min_reward": -5.40625, "num_tokens": 1372383.0, "step": 850 }, { "epoch": 1.1333333333333333, "eval_accuracy": 0.596, "eval_loss": 1.0435729026794434, "eval_margin": 0.8098057250976562, "eval_max_reward": 0.7392849731445312, "eval_mean_reward": -0.5019579772949219, "eval_min_reward": -1.743200927734375, "eval_num_tokens": 1372383.0, "eval_runtime": 62.4519, "eval_samples_per_second": 8.006, "eval_steps_per_second": 8.006, "step": 850 }, { "accuracy": 0.8, "epoch": 1.1466666666666667, "grad_norm": 18.71978759765625, "learning_rate": 2.136666666666667e-06, "loss": 0.6701197147369384, "margin": 1.843017578125, "max_reward": 3.765625, "mean_reward": 0.3057373046875, "min_reward": -3.39443359375, "num_tokens": 1388575.0, "step": 860 }, { "accuracy": 0.7, "epoch": 1.16, "grad_norm": 150.29055786132812, "learning_rate": 2.1033333333333337e-06, "loss": 0.6811777114868164, "margin": 1.3685546875, "max_reward": 2.4873046875, "mean_reward": 0.19677734375, "min_reward": -2.523046875, "num_tokens": 1403606.0, "step": 870 }, { "accuracy": 0.75, "epoch": 1.1733333333333333, "grad_norm": 187.19081115722656, "learning_rate": 2.07e-06, "loss": 0.8808813095092773, "margin": 1.917291259765625, "max_reward": 4.8203125, "mean_reward": 2.1001434326171875, "min_reward": -1.4076171875, "num_tokens": 1418488.0, "step": 880 }, { "accuracy": 0.8, "epoch": 1.1866666666666668, "grad_norm": 44.864192962646484, "learning_rate": 2.036666666666667e-06, "loss": 0.558538818359375, "margin": 2.50665283203125, "max_reward": 4.6734375, "mean_reward": 1.906243896484375, "min_reward": -2.1400634765625, "num_tokens": 1433529.0, "step": 890 }, { "accuracy": 0.8, "epoch": 1.2, "grad_norm": 46.679744720458984, "learning_rate": 2.0033333333333334e-06, "loss": 0.7218273639678955, "margin": 2.88583984375, "max_reward": 3.57421875, "mean_reward": -0.101611328125, "min_reward": -4.85234375, "num_tokens": 1450110.0, "step": 900 }, { "epoch": 1.2, "eval_accuracy": 0.608, "eval_loss": 1.363513708114624, "eval_margin": 1.084634536743164, "eval_max_reward": 3.07980419921875, "eval_mean_reward": 1.359497184753418, "eval_min_reward": -0.36080982971191405, "eval_num_tokens": 1450110.0, "eval_runtime": 62.4495, "eval_samples_per_second": 8.006, "eval_steps_per_second": 8.006, "step": 900 }, { "accuracy": 0.6, "epoch": 1.2133333333333334, "grad_norm": 190.0553741455078, "learning_rate": 1.97e-06, "loss": 0.7698217391967773, "margin": 1.106640625, "max_reward": 5.125, "mean_reward": 1.506640625, "min_reward": -2.765625, "num_tokens": 1466027.0, "step": 910 }, { "accuracy": 0.6, "epoch": 1.2266666666666666, "grad_norm": 234.94076538085938, "learning_rate": 1.9366666666666666e-06, "loss": 1.6000349044799804, "margin": 1.884661865234375, "max_reward": 1.996875, "mean_reward": -2.3544403076171876, "min_reward": -8.725, "num_tokens": 1481748.0, "step": 920 }, { "accuracy": 0.55, "epoch": 1.24, "grad_norm": 65.35015869140625, "learning_rate": 1.9033333333333335e-06, "loss": 1.3991120338439942, "margin": 1.012890625, "max_reward": -1.74296875, "mean_reward": -5.5919921875, "min_reward": -9.453125, "num_tokens": 1498020.0, "step": 930 }, { "accuracy": 0.75, "epoch": 1.2533333333333334, "grad_norm": 57.545166015625, "learning_rate": 1.87e-06, "loss": 0.49834294319152833, "margin": 2.0171875, "max_reward": -3.440625, "mean_reward": -6.62265625, "min_reward": -9.015625, "num_tokens": 1514583.0, "step": 940 }, { "accuracy": 0.85, "epoch": 1.2666666666666666, "grad_norm": 144.95867919921875, "learning_rate": 1.836666666666667e-06, "loss": 0.5361029624938964, "margin": 3.177197265625, "max_reward": 0.4771484375, "mean_reward": -3.9238525390625, "min_reward": -8.1375, "num_tokens": 1530955.0, "step": 950 }, { "epoch": 1.2666666666666666, "eval_accuracy": 0.596, "eval_loss": 1.3600239753723145, "eval_margin": 1.1310927734375, "eval_max_reward": -1.742294677734375, "eval_mean_reward": -3.521328125, "eval_min_reward": -5.300361572265625, "eval_num_tokens": 1530955.0, "eval_runtime": 62.6012, "eval_samples_per_second": 7.987, "eval_steps_per_second": 7.987, "step": 950 }, { "accuracy": 0.5, "epoch": 1.28, "grad_norm": 319.9293518066406, "learning_rate": 1.8033333333333336e-06, "loss": 1.7453046798706056, "margin": 0.3447265625, "max_reward": 1.01591796875, "mean_reward": -3.259765625, "min_reward": -7.2578125, "num_tokens": 1548332.0, "step": 960 }, { "accuracy": 0.65, "epoch": 1.2933333333333334, "grad_norm": 33.143341064453125, "learning_rate": 1.77e-06, "loss": 0.7040543556213379, "margin": 2.181396484375, "max_reward": -0.30205078125, "mean_reward": -3.7452392578125, "min_reward": -6.275, "num_tokens": 1564625.0, "step": 970 }, { "accuracy": 0.75, "epoch": 1.3066666666666666, "grad_norm": 31.115459442138672, "learning_rate": 1.7366666666666668e-06, "loss": 0.4173430919647217, "margin": 2.164306640625, "max_reward": 2.28095703125, "mean_reward": -0.9407470703125, "min_reward": -4.04375, "num_tokens": 1580000.0, "step": 980 }, { "accuracy": 0.8, "epoch": 1.32, "grad_norm": 168.20628356933594, "learning_rate": 1.7033333333333335e-06, "loss": 0.5366679191589355, "margin": 3.0508544921875, "max_reward": 2.8671875, "mean_reward": -1.78248291015625, "min_reward": -5.2830078125, "num_tokens": 1597302.0, "step": 990 }, { "accuracy": 0.45, "epoch": 1.3333333333333333, "grad_norm": 18.123050689697266, "learning_rate": 1.6700000000000003e-06, "loss": 1.1453736305236817, "margin": 0.8822265625, "max_reward": 4.8984375, "mean_reward": 2.26845703125, "min_reward": -0.678515625, "num_tokens": 1613437.0, "step": 1000 }, { "epoch": 1.3333333333333333, "eval_accuracy": 0.586, "eval_loss": 1.1093744039535522, "eval_margin": 0.682355224609375, "eval_max_reward": 3.630033935546875, "eval_mean_reward": 2.4088892822265624, "eval_min_reward": 1.18774462890625, "eval_num_tokens": 1613437.0, "eval_runtime": 62.4977, "eval_samples_per_second": 8.0, "eval_steps_per_second": 8.0, "step": 1000 }, { "accuracy": 0.75, "epoch": 1.3466666666666667, "grad_norm": 35.67937088012695, "learning_rate": 1.6366666666666667e-06, "loss": 0.675491189956665, "margin": 0.5638671875, "max_reward": 4.290625, "mean_reward": 2.44892578125, "min_reward": 0.67578125, "num_tokens": 1629455.0, "step": 1010 }, { "accuracy": 0.7, "epoch": 1.3599999999999999, "grad_norm": 40.7118034362793, "learning_rate": 1.6033333333333334e-06, "loss": 0.42725315093994143, "margin": 1.8513671875, "max_reward": 4.996875, "mean_reward": 3.25732421875, "min_reward": 0.50625, "num_tokens": 1644785.0, "step": 1020 }, { "accuracy": 0.8, "epoch": 1.3733333333333333, "grad_norm": 136.8484344482422, "learning_rate": 1.5700000000000002e-06, "loss": 0.6001864433288574, "margin": 3.10028076171875, "max_reward": 4.1550537109375, "mean_reward": 0.757366943359375, "min_reward": -3.25009765625, "num_tokens": 1661544.0, "step": 1030 }, { "accuracy": 0.75, "epoch": 1.3866666666666667, "grad_norm": 0.3609694242477417, "learning_rate": 1.5366666666666668e-06, "loss": 0.9377962112426758, "margin": 1.72666015625, "max_reward": 3.7875, "mean_reward": 1.237353515625, "min_reward": -2.774609375, "num_tokens": 1677000.0, "step": 1040 }, { "accuracy": 0.7, "epoch": 1.4, "grad_norm": 16.80458641052246, "learning_rate": 1.5033333333333337e-06, "loss": 0.7020199298858643, "margin": 1.91923828125, "max_reward": 3.17734375, "mean_reward": 0.354052734375, "min_reward": -3.8123046875, "num_tokens": 1691904.0, "step": 1050 }, { "epoch": 1.4, "eval_accuracy": 0.606, "eval_loss": 1.3567456007003784, "eval_margin": 1.0848167724609374, "eval_max_reward": 1.356125732421875, "eval_mean_reward": -0.37665606689453124, "eval_min_reward": -2.1094378662109374, "eval_num_tokens": 1691904.0, "eval_runtime": 62.3936, "eval_samples_per_second": 8.014, "eval_steps_per_second": 8.014, "step": 1050 }, { "accuracy": 0.55, "epoch": 1.4133333333333333, "grad_norm": 117.69019317626953, "learning_rate": 1.4700000000000001e-06, "loss": 1.212506103515625, "margin": 1.3871475219726563, "max_reward": 2.3145721435546873, "mean_reward": -3.046747589111328, "min_reward": -6.746875, "num_tokens": 1708299.0, "step": 1060 }, { "accuracy": 0.6, "epoch": 1.4266666666666667, "grad_norm": 100.34335327148438, "learning_rate": 1.4366666666666667e-06, "loss": 0.9856803894042969, "margin": 1.3234375, "max_reward": 1.334375, "mean_reward": -2.6033203125, "min_reward": -6.49765625, "num_tokens": 1725007.0, "step": 1070 }, { "accuracy": 0.7, "epoch": 1.44, "grad_norm": 185.3729248046875, "learning_rate": 1.4033333333333336e-06, "loss": 1.1272445678710938, "margin": 2.0210784912109374, "max_reward": 3.303125, "mean_reward": -0.05469512939453125, "min_reward": -3.634649658203125, "num_tokens": 1740533.0, "step": 1080 }, { "accuracy": 0.6, "epoch": 1.4533333333333334, "grad_norm": 161.89877319335938, "learning_rate": 1.3700000000000002e-06, "loss": 1.4939114570617675, "margin": 0.5677734375, "max_reward": 1.749609375, "mean_reward": -2.25419921875, "min_reward": -6.180859375, "num_tokens": 1757099.0, "step": 1090 }, { "accuracy": 0.75, "epoch": 1.4666666666666668, "grad_norm": 119.40484619140625, "learning_rate": 1.3366666666666666e-06, "loss": 0.8826108932495117, "margin": 1.85439453125, "max_reward": 2.553125, "mean_reward": -0.852197265625, "min_reward": -4.6171875, "num_tokens": 1775146.0, "step": 1100 }, { "epoch": 1.4666666666666668, "eval_accuracy": 0.604, "eval_loss": 0.9978691339492798, "eval_margin": 0.7300045776367188, "eval_max_reward": 0.6507423095703125, "eval_mean_reward": -0.4976886901855469, "eval_min_reward": -1.6461196899414063, "eval_num_tokens": 1775146.0, "eval_runtime": 62.4188, "eval_samples_per_second": 8.01, "eval_steps_per_second": 8.01, "step": 1100 }, { "accuracy": 0.65, "epoch": 1.48, "grad_norm": 109.5771713256836, "learning_rate": 1.3033333333333335e-06, "loss": 0.9664765357971191, "margin": 0.737762451171875, "max_reward": 1.653662109375, "mean_reward": -0.5663665771484375, "min_reward": -3.6734375, "num_tokens": 1793063.0, "step": 1110 }, { "accuracy": 0.7, "epoch": 1.4933333333333334, "grad_norm": 83.6311264038086, "learning_rate": 1.2700000000000001e-06, "loss": 0.6062012672424316, "margin": 2.137274169921875, "max_reward": 1.9431640625, "mean_reward": -1.1080963134765625, "min_reward": -4.4921875, "num_tokens": 1808403.0, "step": 1120 }, { "accuracy": 0.75, "epoch": 1.5066666666666668, "grad_norm": 60.31748962402344, "learning_rate": 1.2366666666666668e-06, "loss": 0.7049060821533203, "margin": 1.48701171875, "max_reward": 2.08974609375, "mean_reward": -0.200439453125, "min_reward": -2.463671875, "num_tokens": 1824621.0, "step": 1130 }, { "accuracy": 0.55, "epoch": 1.52, "grad_norm": 149.60330200195312, "learning_rate": 1.2033333333333334e-06, "loss": 1.1337081909179687, "margin": 0.48056640625, "max_reward": 1.880859375, "mean_reward": -0.384716796875, "min_reward": -2.83515625, "num_tokens": 1840744.0, "step": 1140 }, { "accuracy": 0.7, "epoch": 1.5333333333333332, "grad_norm": 41.09004592895508, "learning_rate": 1.1700000000000002e-06, "loss": 0.6269871711730957, "margin": 1.2066650390625, "max_reward": 2.41328125, "mean_reward": 0.39190673828125, "min_reward": -1.7814453125, "num_tokens": 1855354.0, "step": 1150 }, { "epoch": 1.5333333333333332, "eval_accuracy": 0.586, "eval_loss": 0.9035583138465881, "eval_margin": 0.5397208251953125, "eval_max_reward": 1.6990680236816407, "eval_mean_reward": 0.7785431518554687, "eval_min_reward": -0.14198171997070314, "eval_num_tokens": 1855354.0, "eval_runtime": 62.4733, "eval_samples_per_second": 8.003, "eval_steps_per_second": 8.003, "step": 1150 }, { "accuracy": 0.85, "epoch": 1.5466666666666666, "grad_norm": 6.0896782875061035, "learning_rate": 1.1366666666666667e-06, "loss": 0.5843002796173096, "margin": 1.83350830078125, "max_reward": 2.8265625, "mean_reward": 0.397064208984375, "min_reward": -2.0440673828125, "num_tokens": 1871860.0, "step": 1160 }, { "accuracy": 0.75, "epoch": 1.56, "grad_norm": 102.23207092285156, "learning_rate": 1.1033333333333335e-06, "loss": 0.5763196468353271, "margin": 1.4167724609375, "max_reward": 3.4875, "mean_reward": 1.43270263671875, "min_reward": -1.278466796875, "num_tokens": 1888601.0, "step": 1170 }, { "accuracy": 0.6, "epoch": 1.5733333333333333, "grad_norm": 109.7595443725586, "learning_rate": 1.0700000000000001e-06, "loss": 0.9253439903259277, "margin": 0.2265625, "max_reward": 2.1390625, "mean_reward": 0.3791015625, "min_reward": -1.2005859375, "num_tokens": 1905756.0, "step": 1180 }, { "accuracy": 0.7, "epoch": 1.5866666666666667, "grad_norm": 27.03172492980957, "learning_rate": 1.0366666666666668e-06, "loss": 0.6028451919555664, "margin": 1.306298828125, "max_reward": 2.61015625, "mean_reward": 0.2985595703125, "min_reward": -1.97734375, "num_tokens": 1922025.0, "step": 1190 }, { "accuracy": 0.75, "epoch": 1.6, "grad_norm": 91.80132293701172, "learning_rate": 1.0033333333333334e-06, "loss": 0.6056195735931397, "margin": 1.401220703125, "max_reward": 2.41943359375, "mean_reward": -0.0202880859375, "min_reward": -2.3765625, "num_tokens": 1938931.0, "step": 1200 }, { "epoch": 1.6, "eval_accuracy": 0.602, "eval_loss": 1.04457688331604, "eval_margin": 0.740804443359375, "eval_max_reward": 1.606439208984375, "eval_mean_reward": 0.3948948974609375, "eval_min_reward": -0.8166494140625, "eval_num_tokens": 1938931.0, "eval_runtime": 62.3655, "eval_samples_per_second": 8.017, "eval_steps_per_second": 8.017, "step": 1200 }, { "accuracy": 0.75, "epoch": 1.6133333333333333, "grad_norm": 0.7046879529953003, "learning_rate": 9.7e-07, "loss": 1.0164477348327636, "margin": 1.867919921875, "max_reward": 3.298828125, "mean_reward": 0.1361083984375, "min_reward": -3.41015625, "num_tokens": 1955127.0, "step": 1210 }, { "accuracy": 0.65, "epoch": 1.6266666666666667, "grad_norm": 25.504459381103516, "learning_rate": 9.366666666666667e-07, "loss": 0.8065451622009278, "margin": 1.255029296875, "max_reward": 2.398828125, "mean_reward": -0.2284912109375, "min_reward": -3.13154296875, "num_tokens": 1972318.0, "step": 1220 }, { "accuracy": 0.65, "epoch": 1.6400000000000001, "grad_norm": 87.41260528564453, "learning_rate": 9.033333333333334e-07, "loss": 1.119236946105957, "margin": 0.7582611083984375, "max_reward": 2.4771484375, "mean_reward": -0.6033004760742188, "min_reward": -3.24765625, "num_tokens": 1987620.0, "step": 1230 }, { "accuracy": 0.65, "epoch": 1.6533333333333333, "grad_norm": 94.04547119140625, "learning_rate": 8.7e-07, "loss": 0.6439452171325684, "margin": 1.82333984375, "max_reward": 2.728125, "mean_reward": 0.210400390625, "min_reward": -2.7384765625, "num_tokens": 2004302.0, "step": 1240 }, { "accuracy": 0.65, "epoch": 1.6666666666666665, "grad_norm": 85.90032958984375, "learning_rate": 8.366666666666668e-07, "loss": 0.5457091331481934, "margin": 1.3715576171875, "max_reward": 2.209033203125, "mean_reward": -0.44652099609375, "min_reward": -2.5546875, "num_tokens": 2021262.0, "step": 1250 }, { "epoch": 1.6666666666666665, "eval_accuracy": 0.612, "eval_loss": 1.0178015232086182, "eval_margin": 0.7143900146484375, "eval_max_reward": 1.167669921875, "eval_mean_reward": 0.00498980712890625, "eval_min_reward": -1.1576903076171876, "eval_num_tokens": 2021262.0, "eval_runtime": 62.4994, "eval_samples_per_second": 8.0, "eval_steps_per_second": 8.0, "step": 1250 }, { "accuracy": 0.85, "epoch": 1.6800000000000002, "grad_norm": 57.4267578125, "learning_rate": 8.033333333333335e-07, "loss": 0.5381040573120117, "margin": 1.5513671875, "max_reward": 2.2685546875, "mean_reward": -0.218896484375, "min_reward": -2.88603515625, "num_tokens": 2037311.0, "step": 1260 }, { "accuracy": 0.65, "epoch": 1.6933333333333334, "grad_norm": 20.574970245361328, "learning_rate": 7.7e-07, "loss": 0.8994636535644531, "margin": 1.5009521484375, "max_reward": 2.207861328125, "mean_reward": -0.48370361328125, "min_reward": -2.67890625, "num_tokens": 2053587.0, "step": 1270 }, { "accuracy": 0.5, "epoch": 1.7066666666666666, "grad_norm": 25.6247501373291, "learning_rate": 7.366666666666668e-07, "loss": 1.4920889854431152, "margin": 0.5493316650390625, "max_reward": 2.29677734375, "mean_reward": -0.34183807373046876, "min_reward": -3.459832763671875, "num_tokens": 2069715.0, "step": 1280 }, { "accuracy": 0.65, "epoch": 1.72, "grad_norm": 28.003578186035156, "learning_rate": 7.033333333333334e-07, "loss": 1.0490900039672852, "margin": 1.029248046875, "max_reward": 2.55390625, "mean_reward": -0.0261474609375, "min_reward": -2.5232421875, "num_tokens": 2083729.0, "step": 1290 }, { "accuracy": 0.6, "epoch": 1.7333333333333334, "grad_norm": 100.13626098632812, "learning_rate": 6.7e-07, "loss": 0.8861888885498047, "margin": 0.58553466796875, "max_reward": 1.53046875, "mean_reward": -0.445611572265625, "min_reward": -2.4734375, "num_tokens": 2099948.0, "step": 1300 }, { "epoch": 1.7333333333333334, "eval_accuracy": 0.606, "eval_loss": 1.0033907890319824, "eval_margin": 0.7165312805175781, "eval_max_reward": 1.2282371826171874, "eval_mean_reward": 0.07899192810058593, "eval_min_reward": -1.0702533264160157, "eval_num_tokens": 2099948.0, "eval_runtime": 62.3657, "eval_samples_per_second": 8.017, "eval_steps_per_second": 8.017, "step": 1300 }, { "accuracy": 0.7, "epoch": 1.7466666666666666, "grad_norm": 144.0567626953125, "learning_rate": 6.366666666666667e-07, "loss": 0.5809046268463135, "margin": 1.460107421875, "max_reward": 1.62353515625, "mean_reward": -0.3951416015625, "min_reward": -2.947265625, "num_tokens": 2115789.0, "step": 1310 }, { "accuracy": 0.7, "epoch": 1.76, "grad_norm": 49.05362319946289, "learning_rate": 6.033333333333334e-07, "loss": 0.87122802734375, "margin": 1.10679931640625, "max_reward": 2.268359375, "mean_reward": -0.050628662109375, "min_reward": -2.1308349609375, "num_tokens": 2133528.0, "step": 1320 }, { "accuracy": 0.65, "epoch": 1.7733333333333334, "grad_norm": 98.22369384765625, "learning_rate": 5.7e-07, "loss": 0.6749707221984863, "margin": 0.971484375, "max_reward": 2.073046875, "mean_reward": -0.6001953125, "min_reward": -3.078125, "num_tokens": 2147275.0, "step": 1330 }, { "accuracy": 0.75, "epoch": 1.7866666666666666, "grad_norm": 178.06886291503906, "learning_rate": 5.366666666666667e-07, "loss": 0.7042564868927002, "margin": 1.1953125, "max_reward": 1.126953125, "mean_reward": -1.3919921875, "min_reward": -3.98046875, "num_tokens": 2165353.0, "step": 1340 }, { "accuracy": 0.65, "epoch": 1.8, "grad_norm": 49.11288070678711, "learning_rate": 5.033333333333334e-07, "loss": 0.9773781776428223, "margin": 0.57353515625, "max_reward": 1.633447265625, "mean_reward": -0.50439453125, "min_reward": -2.768359375, "num_tokens": 2181126.0, "step": 1350 }, { "epoch": 1.8, "eval_accuracy": 0.612, "eval_loss": 0.9391070008277893, "eval_margin": 0.6821452369689941, "eval_max_reward": 0.2567112731933594, "eval_mean_reward": -0.8024592456817627, "eval_min_reward": -1.8616297645568847, "eval_num_tokens": 2181126.0, "eval_runtime": 62.4622, "eval_samples_per_second": 8.005, "eval_steps_per_second": 8.005, "step": 1350 }, { "accuracy": 0.8, "epoch": 1.8133333333333335, "grad_norm": 12.369355201721191, "learning_rate": 4.7000000000000005e-07, "loss": 0.5082510471343994, "margin": 2.278076171875, "max_reward": 2.44150390625, "mean_reward": -0.1749755859375, "min_reward": -3.0359375, "num_tokens": 2198470.0, "step": 1360 }, { "accuracy": 0.6, "epoch": 1.8266666666666667, "grad_norm": 21.943758010864258, "learning_rate": 4.366666666666667e-07, "loss": 0.8553391456604004, "margin": 1.10614013671875, "max_reward": 0.610546875, "mean_reward": -1.453851318359375, "min_reward": -3.56328125, "num_tokens": 2213797.0, "step": 1370 }, { "accuracy": 0.55, "epoch": 1.8399999999999999, "grad_norm": 67.34660339355469, "learning_rate": 4.0333333333333337e-07, "loss": 1.136549472808838, "margin": 0.629931640625, "max_reward": 2.42353515625, "mean_reward": -0.6287841796875, "min_reward": -3.7546875, "num_tokens": 2230299.0, "step": 1380 }, { "accuracy": 0.65, "epoch": 1.8533333333333335, "grad_norm": 14.513836860656738, "learning_rate": 3.7e-07, "loss": 0.5644237518310546, "margin": 1.4962158203125, "max_reward": 1.477099609375, "mean_reward": -1.04271240234375, "min_reward": -3.7123046875, "num_tokens": 2245618.0, "step": 1390 }, { "accuracy": 0.5, "epoch": 1.8666666666666667, "grad_norm": 137.32266235351562, "learning_rate": 3.366666666666667e-07, "loss": 0.7403782367706299, "margin": 1.222882080078125, "max_reward": 1.35396728515625, "mean_reward": -1.0289886474609375, "min_reward": -3.173046875, "num_tokens": 2262798.0, "step": 1400 }, { "epoch": 1.8666666666666667, "eval_accuracy": 0.618, "eval_loss": 0.9525383710861206, "eval_margin": 0.7151371459960938, "eval_max_reward": -0.032703857421875, "eval_mean_reward": -1.126583221435547, "eval_min_reward": -2.220462585449219, "eval_num_tokens": 2262798.0, "eval_runtime": 62.453, "eval_samples_per_second": 8.006, "eval_steps_per_second": 8.006, "step": 1400 }, { "accuracy": 0.75, "epoch": 1.88, "grad_norm": 87.93866729736328, "learning_rate": 3.033333333333334e-07, "loss": 0.7155962944030761, "margin": 1.211572265625, "max_reward": 1.64248046875, "mean_reward": -0.8534912109375, "min_reward": -3.3265625, "num_tokens": 2280625.0, "step": 1410 }, { "accuracy": 0.6, "epoch": 1.8933333333333333, "grad_norm": 136.62033081054688, "learning_rate": 2.7e-07, "loss": 0.7803818702697753, "margin": 1.5741943359375, "max_reward": 1.125, "mean_reward": -1.46397705078125, "min_reward": -3.69921875, "num_tokens": 2297438.0, "step": 1420 }, { "accuracy": 0.8, "epoch": 1.9066666666666667, "grad_norm": 36.08385467529297, "learning_rate": 2.3666666666666667e-07, "loss": 0.44714956283569335, "margin": 1.790771484375, "max_reward": 2.1365234375, "mean_reward": -1.1315673828125, "min_reward": -4.0796875, "num_tokens": 2315559.0, "step": 1430 }, { "accuracy": 0.75, "epoch": 1.92, "grad_norm": 16.93815803527832, "learning_rate": 2.0333333333333333e-07, "loss": 0.6091221332550049, "margin": 1.25640869140625, "max_reward": 0.8255126953125, "mean_reward": -1.083807373046875, "min_reward": -3.1236328125, "num_tokens": 2331603.0, "step": 1440 }, { "accuracy": 0.55, "epoch": 1.9333333333333333, "grad_norm": 53.92049026489258, "learning_rate": 1.7000000000000001e-07, "loss": 0.7587657928466797, "margin": 1.339501953125, "max_reward": 1.06416015625, "mean_reward": -0.8656494140625, "min_reward": -3.60859375, "num_tokens": 2346680.0, "step": 1450 }, { "epoch": 1.9333333333333333, "eval_accuracy": 0.612, "eval_loss": 0.97000652551651, "eval_margin": 0.7449773254394532, "eval_max_reward": 0.27565774536132814, "eval_mean_reward": -0.8554528350830078, "eval_min_reward": -1.9865634155273437, "eval_num_tokens": 2346680.0, "eval_runtime": 62.49, "eval_samples_per_second": 8.001, "eval_steps_per_second": 8.001, "step": 1450 } ], "logging_steps": 10, "max_steps": 1500, "num_input_tokens_seen": 0, "num_train_epochs": 2, "save_steps": 50, "stateful_callbacks": { "TrainerControl": { "args": { "should_epoch_stop": false, "should_evaluate": false, "should_log": false, "should_save": true, "should_training_stop": false }, "attributes": {} } }, "total_flos": 1987610824820736.0, "train_batch_size": 2, "trial_name": null, "trial_params": null }